xref: /freebsd/sys/fs/devfs/devfs_vnops.c (revision a9029fbb59bf682ed0d57511a4a5bf06b17f7e80)
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/time.h>
61 #include <sys/ttycom.h>
62 #include <sys/unistd.h>
63 #include <sys/vnode.h>
64 
65 static struct vop_vector devfs_vnodeops;
66 static struct vop_vector devfs_specops;
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 static int
84 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
85     int *ref)
86 {
87 
88 	*dswp = devvn_refthread(fp->f_vnode, devp, ref);
89 	if (*devp != fp->f_data) {
90 		if (*dswp != NULL)
91 			dev_relthread(*devp, *ref);
92 		return (ENXIO);
93 	}
94 	KASSERT((*devp)->si_refcount > 0,
95 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
96 	if (*dswp == NULL)
97 		return (ENXIO);
98 	curthread->td_fpop = fp;
99 	return (0);
100 }
101 
102 int
103 devfs_get_cdevpriv(void **datap)
104 {
105 	struct file *fp;
106 	struct cdev_privdata *p;
107 	int error;
108 
109 	fp = curthread->td_fpop;
110 	if (fp == NULL)
111 		return (EBADF);
112 	p = fp->f_cdevpriv;
113 	if (p != NULL) {
114 		error = 0;
115 		*datap = p->cdpd_data;
116 	} else
117 		error = ENOENT;
118 	return (error);
119 }
120 
121 int
122 devfs_set_cdevpriv(void *priv, cdevpriv_dtr_t priv_dtr)
123 {
124 	struct file *fp;
125 	struct cdev_priv *cdp;
126 	struct cdev_privdata *p;
127 	int error;
128 
129 	fp = curthread->td_fpop;
130 	if (fp == NULL)
131 		return (ENOENT);
132 	cdp = cdev2priv((struct cdev *)fp->f_data);
133 	p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
134 	p->cdpd_data = priv;
135 	p->cdpd_dtr = priv_dtr;
136 	p->cdpd_fp = fp;
137 	mtx_lock(&cdevpriv_mtx);
138 	if (fp->f_cdevpriv == NULL) {
139 		LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
140 		fp->f_cdevpriv = p;
141 		mtx_unlock(&cdevpriv_mtx);
142 		error = 0;
143 	} else {
144 		mtx_unlock(&cdevpriv_mtx);
145 		free(p, M_CDEVPDATA);
146 		error = EBUSY;
147 	}
148 	return (error);
149 }
150 
151 void
152 devfs_destroy_cdevpriv(struct cdev_privdata *p)
153 {
154 
155 	mtx_assert(&cdevpriv_mtx, MA_OWNED);
156 	p->cdpd_fp->f_cdevpriv = NULL;
157 	LIST_REMOVE(p, cdpd_list);
158 	mtx_unlock(&cdevpriv_mtx);
159 	(p->cdpd_dtr)(p->cdpd_data);
160 	free(p, M_CDEVPDATA);
161 }
162 
163 void
164 devfs_fpdrop(struct file *fp)
165 {
166 	struct cdev_privdata *p;
167 
168 	mtx_lock(&cdevpriv_mtx);
169 	if ((p = fp->f_cdevpriv) == NULL) {
170 		mtx_unlock(&cdevpriv_mtx);
171 		return;
172 	}
173 	devfs_destroy_cdevpriv(p);
174 }
175 
176 void
177 devfs_clear_cdevpriv(void)
178 {
179 	struct file *fp;
180 
181 	fp = curthread->td_fpop;
182 	if (fp == NULL)
183 		return;
184 	devfs_fpdrop(fp);
185 }
186 
187 /*
188  * On success devfs_populate_vp() returns with dmp->dm_lock held.
189  */
190 static int
191 devfs_populate_vp(struct vnode *vp)
192 {
193 	struct devfs_dirent *de;
194 	struct devfs_mount *dmp;
195 	int locked;
196 
197 	ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
198 
199 	dmp = VFSTODEVFS(vp->v_mount);
200 	locked = VOP_ISLOCKED(vp);
201 
202 	sx_xlock(&dmp->dm_lock);
203 	DEVFS_DMP_HOLD(dmp);
204 
205 	/* Can't call devfs_populate() with the vnode lock held. */
206 	VOP_UNLOCK(vp, 0);
207 	devfs_populate(dmp);
208 
209 	sx_xunlock(&dmp->dm_lock);
210 	vn_lock(vp, locked | LK_RETRY);
211 	sx_xlock(&dmp->dm_lock);
212 	if (DEVFS_DMP_DROP(dmp)) {
213 		sx_xunlock(&dmp->dm_lock);
214 		devfs_unmount_final(dmp);
215 		return (EBADF);
216 	}
217 	if ((vp->v_iflag & VI_DOOMED) != 0) {
218 		sx_xunlock(&dmp->dm_lock);
219 		return (EBADF);
220 	}
221 	de = vp->v_data;
222 	KASSERT(de != NULL,
223 	    ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
224 	if ((de->de_flags & DE_DOOMED) != 0) {
225 		sx_xunlock(&dmp->dm_lock);
226 		return (EBADF);
227 	}
228 
229 	return (0);
230 }
231 
232 static int
233 devfs_vptocnp(struct vop_vptocnp_args *ap)
234 {
235 	struct vnode *vp = ap->a_vp;
236 	struct vnode **dvp = ap->a_vpp;
237 	struct devfs_mount *dmp;
238 	char *buf = ap->a_buf;
239 	int *buflen = ap->a_buflen;
240 	struct devfs_dirent *dd, *de;
241 	int i, error;
242 
243 	dmp = VFSTODEVFS(vp->v_mount);
244 
245 	error = devfs_populate_vp(vp);
246 	if (error != 0)
247 		return (error);
248 
249 	i = *buflen;
250 	dd = vp->v_data;
251 
252 	if (vp->v_type == VCHR) {
253 		i -= strlen(dd->de_cdp->cdp_c.si_name);
254 		if (i < 0) {
255 			error = ENOMEM;
256 			goto finished;
257 		}
258 		bcopy(dd->de_cdp->cdp_c.si_name, buf + i,
259 		    strlen(dd->de_cdp->cdp_c.si_name));
260 		de = dd->de_dir;
261 	} else if (vp->v_type == VDIR) {
262 		if (dd == dmp->dm_rootdir) {
263 			*dvp = vp;
264 			vref(*dvp);
265 			goto finished;
266 		}
267 		i -= dd->de_dirent->d_namlen;
268 		if (i < 0) {
269 			error = ENOMEM;
270 			goto finished;
271 		}
272 		bcopy(dd->de_dirent->d_name, buf + i,
273 		    dd->de_dirent->d_namlen);
274 		de = dd;
275 	} else {
276 		error = ENOENT;
277 		goto finished;
278 	}
279 	*buflen = i;
280 	de = devfs_parent_dirent(de);
281 	if (de == NULL) {
282 		error = ENOENT;
283 		goto finished;
284 	}
285 	mtx_lock(&devfs_de_interlock);
286 	*dvp = de->de_vnode;
287 	if (*dvp != NULL) {
288 		VI_LOCK(*dvp);
289 		mtx_unlock(&devfs_de_interlock);
290 		vholdl(*dvp);
291 		VI_UNLOCK(*dvp);
292 		vref(*dvp);
293 		vdrop(*dvp);
294 	} else {
295 		mtx_unlock(&devfs_de_interlock);
296 		error = ENOENT;
297 	}
298 finished:
299 	sx_xunlock(&dmp->dm_lock);
300 	return (error);
301 }
302 
303 /*
304  * Construct the fully qualified path name relative to the mountpoint.
305  * If a NULL cnp is provided, no '/' is appended to the resulting path.
306  */
307 char *
308 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
309     struct componentname *cnp)
310 {
311 	int i;
312 	struct devfs_dirent *de;
313 
314 	sx_assert(&dmp->dm_lock, SA_LOCKED);
315 
316 	i = SPECNAMELEN;
317 	buf[i] = '\0';
318 	if (cnp != NULL)
319 		i -= cnp->cn_namelen;
320 	if (i < 0)
321 		 return (NULL);
322 	if (cnp != NULL)
323 		bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
324 	de = dd;
325 	while (de != dmp->dm_rootdir) {
326 		if (cnp != NULL || i < SPECNAMELEN) {
327 			i--;
328 			if (i < 0)
329 				 return (NULL);
330 			buf[i] = '/';
331 		}
332 		i -= de->de_dirent->d_namlen;
333 		if (i < 0)
334 			 return (NULL);
335 		bcopy(de->de_dirent->d_name, buf + i,
336 		    de->de_dirent->d_namlen);
337 		de = devfs_parent_dirent(de);
338 		if (de == NULL)
339 			return (NULL);
340 	}
341 	return (buf + i);
342 }
343 
344 static int
345 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
346 	struct devfs_dirent *de)
347 {
348 	int not_found;
349 
350 	not_found = 0;
351 	if (de->de_flags & DE_DOOMED)
352 		not_found = 1;
353 	if (DEVFS_DE_DROP(de)) {
354 		KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
355 		devfs_dirent_free(de);
356 	}
357 	if (DEVFS_DMP_DROP(dmp)) {
358 		KASSERT(not_found == 1,
359 			("DEVFS mount struct freed before dirent"));
360 		not_found = 2;
361 		sx_xunlock(&dmp->dm_lock);
362 		devfs_unmount_final(dmp);
363 	}
364 	if (not_found == 1 || (drop_dm_lock && not_found != 2))
365 		sx_unlock(&dmp->dm_lock);
366 	return (not_found);
367 }
368 
369 static void
370 devfs_insmntque_dtr(struct vnode *vp, void *arg)
371 {
372 	struct devfs_dirent *de;
373 
374 	de = (struct devfs_dirent *)arg;
375 	mtx_lock(&devfs_de_interlock);
376 	vp->v_data = NULL;
377 	de->de_vnode = NULL;
378 	mtx_unlock(&devfs_de_interlock);
379 	vgone(vp);
380 	vput(vp);
381 }
382 
383 /*
384  * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
385  * it on return.
386  */
387 int
388 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
389     struct vnode **vpp)
390 {
391 	int error;
392 	struct vnode *vp;
393 	struct cdev *dev;
394 	struct devfs_mount *dmp;
395 	struct cdevsw *dsw;
396 
397 	dmp = VFSTODEVFS(mp);
398 	if (de->de_flags & DE_DOOMED) {
399 		sx_xunlock(&dmp->dm_lock);
400 		return (ENOENT);
401 	}
402 loop:
403 	DEVFS_DE_HOLD(de);
404 	DEVFS_DMP_HOLD(dmp);
405 	mtx_lock(&devfs_de_interlock);
406 	vp = de->de_vnode;
407 	if (vp != NULL) {
408 		VI_LOCK(vp);
409 		mtx_unlock(&devfs_de_interlock);
410 		sx_xunlock(&dmp->dm_lock);
411 		vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread);
412 		sx_xlock(&dmp->dm_lock);
413 		if (devfs_allocv_drop_refs(0, dmp, de)) {
414 			vput(vp);
415 			return (ENOENT);
416 		}
417 		else if ((vp->v_iflag & VI_DOOMED) != 0) {
418 			mtx_lock(&devfs_de_interlock);
419 			if (de->de_vnode == vp) {
420 				de->de_vnode = NULL;
421 				vp->v_data = NULL;
422 			}
423 			mtx_unlock(&devfs_de_interlock);
424 			vput(vp);
425 			goto loop;
426 		}
427 		sx_xunlock(&dmp->dm_lock);
428 		*vpp = vp;
429 		return (0);
430 	}
431 	mtx_unlock(&devfs_de_interlock);
432 	if (de->de_dirent->d_type == DT_CHR) {
433 		if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
434 			devfs_allocv_drop_refs(1, dmp, de);
435 			return (ENOENT);
436 		}
437 		dev = &de->de_cdp->cdp_c;
438 	} else {
439 		dev = NULL;
440 	}
441 	error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
442 	if (error != 0) {
443 		devfs_allocv_drop_refs(1, dmp, de);
444 		printf("devfs_allocv: failed to allocate new vnode\n");
445 		return (error);
446 	}
447 
448 	if (de->de_dirent->d_type == DT_CHR) {
449 		vp->v_type = VCHR;
450 		VI_LOCK(vp);
451 		dev_lock();
452 		dev_refl(dev);
453 		/* XXX: v_rdev should be protect by vnode lock */
454 		vp->v_rdev = dev;
455 		KASSERT(vp->v_usecount == 1,
456 		    ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount));
457 		dev->si_usecount += vp->v_usecount;
458 		/* Special casing of ttys for deadfs.  Probably redundant. */
459 		dsw = dev->si_devsw;
460 		if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
461 			vp->v_vflag |= VV_ISTTY;
462 		dev_unlock();
463 		VI_UNLOCK(vp);
464 		if ((dev->si_flags & SI_ETERNAL) != 0)
465 			vp->v_vflag |= VV_ETERNALDEV;
466 		vp->v_op = &devfs_specops;
467 	} else if (de->de_dirent->d_type == DT_DIR) {
468 		vp->v_type = VDIR;
469 	} else if (de->de_dirent->d_type == DT_LNK) {
470 		vp->v_type = VLNK;
471 	} else {
472 		vp->v_type = VBAD;
473 	}
474 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
475 	VN_LOCK_ASHARE(vp);
476 	mtx_lock(&devfs_de_interlock);
477 	vp->v_data = de;
478 	de->de_vnode = vp;
479 	mtx_unlock(&devfs_de_interlock);
480 	error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
481 	if (error != 0) {
482 		(void) devfs_allocv_drop_refs(1, dmp, de);
483 		return (error);
484 	}
485 	if (devfs_allocv_drop_refs(0, dmp, de)) {
486 		vput(vp);
487 		return (ENOENT);
488 	}
489 #ifdef MAC
490 	mac_devfs_vnode_associate(mp, de, vp);
491 #endif
492 	sx_xunlock(&dmp->dm_lock);
493 	*vpp = vp;
494 	return (0);
495 }
496 
497 static int
498 devfs_access(struct vop_access_args *ap)
499 {
500 	struct vnode *vp = ap->a_vp;
501 	struct devfs_dirent *de;
502 	int error;
503 
504 	de = vp->v_data;
505 	if (vp->v_type == VDIR)
506 		de = de->de_dir;
507 
508 	error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
509 	    ap->a_accmode, ap->a_cred, NULL);
510 	if (error == 0)
511 		return (0);
512 	if (error != EACCES)
513 		return (error);
514 	/* We do, however, allow access to the controlling terminal */
515 	if (!(ap->a_td->td_proc->p_flag & P_CONTROLT))
516 		return (error);
517 	if (ap->a_td->td_proc->p_session->s_ttydp == de->de_cdp)
518 		return (0);
519 	return (error);
520 }
521 
522 /* ARGSUSED */
523 static int
524 devfs_close(struct vop_close_args *ap)
525 {
526 	struct vnode *vp = ap->a_vp, *oldvp;
527 	struct thread *td = ap->a_td;
528 	struct cdev *dev = vp->v_rdev;
529 	struct cdevsw *dsw;
530 	int vp_locked, error, ref;
531 
532 	/*
533 	 * XXX: Don't call d_close() if we were called because of
534 	 * XXX: insmntque1() failure.
535 	 */
536 	if (vp->v_data == NULL)
537 		return (0);
538 
539 	/*
540 	 * Hack: a tty device that is a controlling terminal
541 	 * has a reference from the session structure.
542 	 * We cannot easily tell that a character device is
543 	 * a controlling terminal, unless it is the closing
544 	 * process' controlling terminal.  In that case,
545 	 * if the reference count is 2 (this last descriptor
546 	 * plus the session), release the reference from the session.
547 	 */
548 	oldvp = NULL;
549 	sx_xlock(&proctree_lock);
550 	if (td && vp == td->td_proc->p_session->s_ttyvp) {
551 		SESS_LOCK(td->td_proc->p_session);
552 		VI_LOCK(vp);
553 		if (count_dev(dev) == 2 && (vp->v_iflag & VI_DOOMED) == 0) {
554 			td->td_proc->p_session->s_ttyvp = NULL;
555 			td->td_proc->p_session->s_ttydp = NULL;
556 			oldvp = vp;
557 		}
558 		VI_UNLOCK(vp);
559 		SESS_UNLOCK(td->td_proc->p_session);
560 	}
561 	sx_xunlock(&proctree_lock);
562 	if (oldvp != NULL)
563 		vrele(oldvp);
564 	/*
565 	 * We do not want to really close the device if it
566 	 * is still in use unless we are trying to close it
567 	 * forcibly. Since every use (buffer, vnode, swap, cmap)
568 	 * holds a reference to the vnode, and because we mark
569 	 * any other vnodes that alias this device, when the
570 	 * sum of the reference counts on all the aliased
571 	 * vnodes descends to one, we are on last close.
572 	 */
573 	dsw = dev_refthread(dev, &ref);
574 	if (dsw == NULL)
575 		return (ENXIO);
576 	VI_LOCK(vp);
577 	if (vp->v_iflag & VI_DOOMED) {
578 		/* Forced close. */
579 	} else if (dsw->d_flags & D_TRACKCLOSE) {
580 		/* Keep device updated on status. */
581 	} else if (count_dev(dev) > 1) {
582 		VI_UNLOCK(vp);
583 		dev_relthread(dev, ref);
584 		return (0);
585 	}
586 	vholdl(vp);
587 	VI_UNLOCK(vp);
588 	vp_locked = VOP_ISLOCKED(vp);
589 	VOP_UNLOCK(vp, 0);
590 	KASSERT(dev->si_refcount > 0,
591 	    ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
592 	error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td);
593 	dev_relthread(dev, ref);
594 	vn_lock(vp, vp_locked | LK_RETRY);
595 	vdrop(vp);
596 	return (error);
597 }
598 
599 static int
600 devfs_close_f(struct file *fp, struct thread *td)
601 {
602 	int error;
603 	struct file *fpop;
604 
605 	fpop = td->td_fpop;
606 	td->td_fpop = fp;
607 	error = vnops.fo_close(fp, td);
608 	td->td_fpop = fpop;
609 
610 	/*
611 	 * The f_cdevpriv cannot be assigned non-NULL value while we
612 	 * are destroying the file.
613 	 */
614 	if (fp->f_cdevpriv != NULL)
615 		devfs_fpdrop(fp);
616 	return (error);
617 }
618 
619 static int
620 devfs_fsync(struct vop_fsync_args *ap)
621 {
622 	int error;
623 	struct bufobj *bo;
624 	struct devfs_dirent *de;
625 
626 	if (!vn_isdisk(ap->a_vp, &error)) {
627 		bo = &ap->a_vp->v_bufobj;
628 		de = ap->a_vp->v_data;
629 		if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) {
630 			printf("Device %s went missing before all of the data "
631 			    "could be written to it; expect data loss.\n",
632 			    de->de_dirent->d_name);
633 
634 			error = vop_stdfsync(ap);
635 			if (bo->bo_dirty.bv_cnt != 0 || error != 0)
636 				panic("devfs_fsync: vop_stdfsync failed.");
637 		}
638 
639 		return (0);
640 	}
641 
642 	return (vop_stdfsync(ap));
643 }
644 
645 static int
646 devfs_getattr(struct vop_getattr_args *ap)
647 {
648 	struct vnode *vp = ap->a_vp;
649 	struct vattr *vap = ap->a_vap;
650 	int error;
651 	struct devfs_dirent *de;
652 	struct devfs_mount *dmp;
653 	struct cdev *dev;
654 
655 	error = devfs_populate_vp(vp);
656 	if (error != 0)
657 		return (error);
658 
659 	dmp = VFSTODEVFS(vp->v_mount);
660 	sx_xunlock(&dmp->dm_lock);
661 
662 	de = vp->v_data;
663 	KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
664 	if (vp->v_type == VDIR) {
665 		de = de->de_dir;
666 		KASSERT(de != NULL,
667 		    ("Null dir dirent in devfs_getattr vp=%p", vp));
668 	}
669 	vap->va_uid = de->de_uid;
670 	vap->va_gid = de->de_gid;
671 	vap->va_mode = de->de_mode;
672 	if (vp->v_type == VLNK)
673 		vap->va_size = strlen(de->de_symlink);
674 	else if (vp->v_type == VDIR)
675 		vap->va_size = vap->va_bytes = DEV_BSIZE;
676 	else
677 		vap->va_size = 0;
678 	if (vp->v_type != VDIR)
679 		vap->va_bytes = 0;
680 	vap->va_blocksize = DEV_BSIZE;
681 	vap->va_type = vp->v_type;
682 
683 #define fix(aa)							\
684 	do {							\
685 		if ((aa).tv_sec <= 3600) {			\
686 			(aa).tv_sec = boottime.tv_sec;		\
687 			(aa).tv_nsec = boottime.tv_usec * 1000; \
688 		}						\
689 	} while (0)
690 
691 	if (vp->v_type != VCHR)  {
692 		fix(de->de_atime);
693 		vap->va_atime = de->de_atime;
694 		fix(de->de_mtime);
695 		vap->va_mtime = de->de_mtime;
696 		fix(de->de_ctime);
697 		vap->va_ctime = de->de_ctime;
698 	} else {
699 		dev = vp->v_rdev;
700 		fix(dev->si_atime);
701 		vap->va_atime = dev->si_atime;
702 		fix(dev->si_mtime);
703 		vap->va_mtime = dev->si_mtime;
704 		fix(dev->si_ctime);
705 		vap->va_ctime = dev->si_ctime;
706 
707 		vap->va_rdev = cdev2priv(dev)->cdp_inode;
708 	}
709 	vap->va_gen = 0;
710 	vap->va_flags = 0;
711 	vap->va_filerev = 0;
712 	vap->va_nlink = de->de_links;
713 	vap->va_fileid = de->de_inode;
714 
715 	return (error);
716 }
717 
718 /* ARGSUSED */
719 static int
720 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
721 {
722 	struct cdev *dev;
723 	struct cdevsw *dsw;
724 	struct vnode *vp;
725 	struct vnode *vpold;
726 	int error, i, ref;
727 	const char *p;
728 	struct fiodgname_arg *fgn;
729 	struct file *fpop;
730 
731 	fpop = td->td_fpop;
732 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
733 	if (error)
734 		return (error);
735 
736 	if (com == FIODTYPE) {
737 		*(int *)data = dsw->d_flags & D_TYPEMASK;
738 		td->td_fpop = fpop;
739 		dev_relthread(dev, ref);
740 		return (0);
741 	} else if (com == FIODGNAME) {
742 		fgn = data;
743 		p = devtoname(dev);
744 		i = strlen(p) + 1;
745 		if (i > fgn->len)
746 			error = EINVAL;
747 		else
748 			error = copyout(p, fgn->buf, i);
749 		td->td_fpop = fpop;
750 		dev_relthread(dev, ref);
751 		return (error);
752 	}
753 	error = dsw->d_ioctl(dev, com, data, fp->f_flag, td);
754 	td->td_fpop = NULL;
755 	dev_relthread(dev, ref);
756 	if (error == ENOIOCTL)
757 		error = ENOTTY;
758 	if (error == 0 && com == TIOCSCTTY) {
759 		vp = fp->f_vnode;
760 
761 		/* Do nothing if reassigning same control tty */
762 		sx_slock(&proctree_lock);
763 		if (td->td_proc->p_session->s_ttyvp == vp) {
764 			sx_sunlock(&proctree_lock);
765 			return (0);
766 		}
767 
768 		vpold = td->td_proc->p_session->s_ttyvp;
769 		VREF(vp);
770 		SESS_LOCK(td->td_proc->p_session);
771 		td->td_proc->p_session->s_ttyvp = vp;
772 		td->td_proc->p_session->s_ttydp = cdev2priv(dev);
773 		SESS_UNLOCK(td->td_proc->p_session);
774 
775 		sx_sunlock(&proctree_lock);
776 
777 		/* Get rid of reference to old control tty */
778 		if (vpold)
779 			vrele(vpold);
780 	}
781 	return (error);
782 }
783 
784 /* ARGSUSED */
785 static int
786 devfs_kqfilter_f(struct file *fp, struct knote *kn)
787 {
788 	struct cdev *dev;
789 	struct cdevsw *dsw;
790 	int error, ref;
791 	struct file *fpop;
792 	struct thread *td;
793 
794 	td = curthread;
795 	fpop = td->td_fpop;
796 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
797 	if (error)
798 		return (error);
799 	error = dsw->d_kqfilter(dev, kn);
800 	td->td_fpop = fpop;
801 	dev_relthread(dev, ref);
802 	return (error);
803 }
804 
805 static inline int
806 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
807 {
808 	struct cdev_priv *cdp;
809 	struct ucred *dcr;
810 	int error;
811 
812 	cdp = de->de_cdp;
813 	if (cdp == NULL)
814 		return (0);
815 	dcr = cdp->cdp_c.si_cred;
816 	if (dcr == NULL)
817 		return (0);
818 
819 	error = prison_check(td->td_ucred, dcr);
820 	if (error == 0)
821 		return (0);
822 	/* We do, however, allow access to the controlling terminal */
823 	if (!(td->td_proc->p_flag & P_CONTROLT))
824 		return (error);
825 	if (td->td_proc->p_session->s_ttydp == cdp)
826 		return (0);
827 	return (error);
828 }
829 
830 static int
831 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
832 {
833 	struct componentname *cnp;
834 	struct vnode *dvp, **vpp;
835 	struct thread *td;
836 	struct devfs_dirent *de, *dd;
837 	struct devfs_dirent **dde;
838 	struct devfs_mount *dmp;
839 	struct cdev *cdev;
840 	int error, flags, nameiop, dvplocked;
841 	char specname[SPECNAMELEN + 1], *pname;
842 
843 	cnp = ap->a_cnp;
844 	vpp = ap->a_vpp;
845 	dvp = ap->a_dvp;
846 	pname = cnp->cn_nameptr;
847 	td = cnp->cn_thread;
848 	flags = cnp->cn_flags;
849 	nameiop = cnp->cn_nameiop;
850 	dmp = VFSTODEVFS(dvp->v_mount);
851 	dd = dvp->v_data;
852 	*vpp = NULLVP;
853 
854 	if ((flags & ISLASTCN) && nameiop == RENAME)
855 		return (EOPNOTSUPP);
856 
857 	if (dvp->v_type != VDIR)
858 		return (ENOTDIR);
859 
860 	if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
861 		return (EIO);
862 
863 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
864 	if (error)
865 		return (error);
866 
867 	if (cnp->cn_namelen == 1 && *pname == '.') {
868 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
869 			return (EINVAL);
870 		*vpp = dvp;
871 		VREF(dvp);
872 		return (0);
873 	}
874 
875 	if (flags & ISDOTDOT) {
876 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
877 			return (EINVAL);
878 		de = devfs_parent_dirent(dd);
879 		if (de == NULL)
880 			return (ENOENT);
881 		dvplocked = VOP_ISLOCKED(dvp);
882 		VOP_UNLOCK(dvp, 0);
883 		error = devfs_allocv(de, dvp->v_mount,
884 		    cnp->cn_lkflags & LK_TYPE_MASK, vpp);
885 		*dm_unlock = 0;
886 		vn_lock(dvp, dvplocked | LK_RETRY);
887 		return (error);
888 	}
889 
890 	dd = dvp->v_data;
891 	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
892 	while (de == NULL) {	/* While(...) so we can use break */
893 
894 		if (nameiop == DELETE)
895 			return (ENOENT);
896 
897 		/*
898 		 * OK, we didn't have an entry for the name we were asked for
899 		 * so we try to see if anybody can create it on demand.
900 		 */
901 		pname = devfs_fqpn(specname, dmp, dd, cnp);
902 		if (pname == NULL)
903 			break;
904 
905 		cdev = NULL;
906 		DEVFS_DMP_HOLD(dmp);
907 		sx_xunlock(&dmp->dm_lock);
908 		sx_slock(&clone_drain_lock);
909 		EVENTHANDLER_INVOKE(dev_clone,
910 		    td->td_ucred, pname, strlen(pname), &cdev);
911 		sx_sunlock(&clone_drain_lock);
912 
913 		if (cdev == NULL)
914 			sx_xlock(&dmp->dm_lock);
915 		else if (devfs_populate_vp(dvp) != 0) {
916 			*dm_unlock = 0;
917 			sx_xlock(&dmp->dm_lock);
918 			if (DEVFS_DMP_DROP(dmp)) {
919 				sx_xunlock(&dmp->dm_lock);
920 				devfs_unmount_final(dmp);
921 			} else
922 				sx_xunlock(&dmp->dm_lock);
923 			dev_rel(cdev);
924 			return (ENOENT);
925 		}
926 		if (DEVFS_DMP_DROP(dmp)) {
927 			*dm_unlock = 0;
928 			sx_xunlock(&dmp->dm_lock);
929 			devfs_unmount_final(dmp);
930 			if (cdev != NULL)
931 				dev_rel(cdev);
932 			return (ENOENT);
933 		}
934 
935 		if (cdev == NULL)
936 			break;
937 
938 		dev_lock();
939 		dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
940 		if (dde != NULL && *dde != NULL)
941 			de = *dde;
942 		dev_unlock();
943 		dev_rel(cdev);
944 		break;
945 	}
946 
947 	if (de == NULL || de->de_flags & DE_WHITEOUT) {
948 		if ((nameiop == CREATE || nameiop == RENAME) &&
949 		    (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
950 			cnp->cn_flags |= SAVENAME;
951 			return (EJUSTRETURN);
952 		}
953 		return (ENOENT);
954 	}
955 
956 	if (devfs_prison_check(de, td))
957 		return (ENOENT);
958 
959 	if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
960 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
961 		if (error)
962 			return (error);
963 		if (*vpp == dvp) {
964 			VREF(dvp);
965 			*vpp = dvp;
966 			return (0);
967 		}
968 	}
969 	error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK,
970 	    vpp);
971 	*dm_unlock = 0;
972 	return (error);
973 }
974 
975 static int
976 devfs_lookup(struct vop_lookup_args *ap)
977 {
978 	int j;
979 	struct devfs_mount *dmp;
980 	int dm_unlock;
981 
982 	if (devfs_populate_vp(ap->a_dvp) != 0)
983 		return (ENOTDIR);
984 
985 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
986 	dm_unlock = 1;
987 	j = devfs_lookupx(ap, &dm_unlock);
988 	if (dm_unlock == 1)
989 		sx_xunlock(&dmp->dm_lock);
990 	return (j);
991 }
992 
993 static int
994 devfs_mknod(struct vop_mknod_args *ap)
995 {
996 	struct componentname *cnp;
997 	struct vnode *dvp, **vpp;
998 	struct devfs_dirent *dd, *de;
999 	struct devfs_mount *dmp;
1000 	int error;
1001 
1002 	/*
1003 	 * The only type of node we should be creating here is a
1004 	 * character device, for anything else return EOPNOTSUPP.
1005 	 */
1006 	if (ap->a_vap->va_type != VCHR)
1007 		return (EOPNOTSUPP);
1008 	dvp = ap->a_dvp;
1009 	dmp = VFSTODEVFS(dvp->v_mount);
1010 
1011 	cnp = ap->a_cnp;
1012 	vpp = ap->a_vpp;
1013 	dd = dvp->v_data;
1014 
1015 	error = ENOENT;
1016 	sx_xlock(&dmp->dm_lock);
1017 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1018 		if (cnp->cn_namelen != de->de_dirent->d_namlen)
1019 			continue;
1020 		if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1021 		    de->de_dirent->d_namlen) != 0)
1022 			continue;
1023 		if (de->de_flags & DE_WHITEOUT)
1024 			break;
1025 		goto notfound;
1026 	}
1027 	if (de == NULL)
1028 		goto notfound;
1029 	de->de_flags &= ~DE_WHITEOUT;
1030 	error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1031 	return (error);
1032 notfound:
1033 	sx_xunlock(&dmp->dm_lock);
1034 	return (error);
1035 }
1036 
1037 /* ARGSUSED */
1038 static int
1039 devfs_open(struct vop_open_args *ap)
1040 {
1041 	struct thread *td = ap->a_td;
1042 	struct vnode *vp = ap->a_vp;
1043 	struct cdev *dev = vp->v_rdev;
1044 	struct file *fp = ap->a_fp;
1045 	int error, ref, vlocked;
1046 	struct cdevsw *dsw;
1047 	struct file *fpop;
1048 
1049 	if (vp->v_type == VBLK)
1050 		return (ENXIO);
1051 
1052 	if (dev == NULL)
1053 		return (ENXIO);
1054 
1055 	/* Make this field valid before any I/O in d_open. */
1056 	if (dev->si_iosize_max == 0)
1057 		dev->si_iosize_max = DFLTPHYS;
1058 
1059 	dsw = dev_refthread(dev, &ref);
1060 	if (dsw == NULL)
1061 		return (ENXIO);
1062 	if (fp == NULL && dsw->d_fdopen != NULL) {
1063 		dev_relthread(dev, ref);
1064 		return (ENXIO);
1065 	}
1066 
1067 	vlocked = VOP_ISLOCKED(vp);
1068 	VOP_UNLOCK(vp, 0);
1069 
1070 	fpop = td->td_fpop;
1071 	td->td_fpop = fp;
1072 	if (fp != NULL) {
1073 		fp->f_data = dev;
1074 		fp->f_vnode = vp;
1075 	}
1076 	if (dsw->d_fdopen != NULL)
1077 		error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1078 	else
1079 		error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1080 	td->td_fpop = fpop;
1081 
1082 	vn_lock(vp, vlocked | LK_RETRY);
1083 	dev_relthread(dev, ref);
1084 	if (error)
1085 		return (error);
1086 
1087 #if 0	/* /dev/console */
1088 	KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1089 #else
1090 	if (fp == NULL)
1091 		return (error);
1092 #endif
1093 	if (fp->f_ops == &badfileops)
1094 		finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1095 	return (error);
1096 }
1097 
1098 static int
1099 devfs_pathconf(struct vop_pathconf_args *ap)
1100 {
1101 
1102 	switch (ap->a_name) {
1103 	case _PC_MAC_PRESENT:
1104 #ifdef MAC
1105 		/*
1106 		 * If MAC is enabled, devfs automatically supports
1107 		 * trivial non-persistant label storage.
1108 		 */
1109 		*ap->a_retval = 1;
1110 #else
1111 		*ap->a_retval = 0;
1112 #endif
1113 		return (0);
1114 	default:
1115 		return (vop_stdpathconf(ap));
1116 	}
1117 	/* NOTREACHED */
1118 }
1119 
1120 /* ARGSUSED */
1121 static int
1122 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1123 {
1124 	struct cdev *dev;
1125 	struct cdevsw *dsw;
1126 	int error, ref;
1127 	struct file *fpop;
1128 
1129 	fpop = td->td_fpop;
1130 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1131 	if (error)
1132 		return (poll_no_poll(events));
1133 	error = dsw->d_poll(dev, events, td);
1134 	td->td_fpop = fpop;
1135 	dev_relthread(dev, ref);
1136 	return(error);
1137 }
1138 
1139 /*
1140  * Print out the contents of a special device vnode.
1141  */
1142 static int
1143 devfs_print(struct vop_print_args *ap)
1144 {
1145 
1146 	printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1147 	return (0);
1148 }
1149 
1150 /* ARGSUSED */
1151 static int
1152 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td)
1153 {
1154 	struct cdev *dev;
1155 	int ioflag, error, ref, resid;
1156 	struct cdevsw *dsw;
1157 	struct file *fpop;
1158 
1159 	fpop = td->td_fpop;
1160 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1161 	if (error)
1162 		return (error);
1163 	resid = uio->uio_resid;
1164 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1165 	if (ioflag & O_DIRECT)
1166 		ioflag |= IO_DIRECT;
1167 
1168 	if ((flags & FOF_OFFSET) == 0)
1169 		uio->uio_offset = fp->f_offset;
1170 
1171 	error = dsw->d_read(dev, uio, ioflag);
1172 	if (uio->uio_resid != resid || (error == 0 && resid != 0))
1173 		vfs_timestamp(&dev->si_atime);
1174 	td->td_fpop = fpop;
1175 	dev_relthread(dev, ref);
1176 
1177 	if ((flags & FOF_OFFSET) == 0)
1178 		fp->f_offset = uio->uio_offset;
1179 	fp->f_nextoff = uio->uio_offset;
1180 	return (error);
1181 }
1182 
1183 static int
1184 devfs_readdir(struct vop_readdir_args *ap)
1185 {
1186 	int error;
1187 	struct uio *uio;
1188 	struct dirent *dp;
1189 	struct devfs_dirent *dd;
1190 	struct devfs_dirent *de;
1191 	struct devfs_mount *dmp;
1192 	off_t off;
1193 	int *tmp_ncookies = NULL;
1194 
1195 	if (ap->a_vp->v_type != VDIR)
1196 		return (ENOTDIR);
1197 
1198 	uio = ap->a_uio;
1199 	if (uio->uio_offset < 0)
1200 		return (EINVAL);
1201 
1202 	/*
1203 	 * XXX: This is a temporary hack to get around this filesystem not
1204 	 * supporting cookies. We store the location of the ncookies pointer
1205 	 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1206 	 * and set the number of cookies to 0. We then set the pointer to
1207 	 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1208 	 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1209 	 * pointer to its original location before returning to the caller.
1210 	 */
1211 	if (ap->a_ncookies != NULL) {
1212 		tmp_ncookies = ap->a_ncookies;
1213 		*ap->a_ncookies = 0;
1214 		ap->a_ncookies = NULL;
1215 	}
1216 
1217 	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1218 	if (devfs_populate_vp(ap->a_vp) != 0) {
1219 		if (tmp_ncookies != NULL)
1220 			ap->a_ncookies = tmp_ncookies;
1221 		return (EIO);
1222 	}
1223 	error = 0;
1224 	de = ap->a_vp->v_data;
1225 	off = 0;
1226 	TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1227 		KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1228 		if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1229 			continue;
1230 		if (devfs_prison_check(dd, uio->uio_td))
1231 			continue;
1232 		if (dd->de_dirent->d_type == DT_DIR)
1233 			de = dd->de_dir;
1234 		else
1235 			de = dd;
1236 		dp = dd->de_dirent;
1237 		if (dp->d_reclen > uio->uio_resid)
1238 			break;
1239 		dp->d_fileno = de->de_inode;
1240 		if (off >= uio->uio_offset) {
1241 			error = vfs_read_dirent(ap, dp, off);
1242 			if (error)
1243 				break;
1244 		}
1245 		off += dp->d_reclen;
1246 	}
1247 	sx_xunlock(&dmp->dm_lock);
1248 	uio->uio_offset = off;
1249 
1250 	/*
1251 	 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1252 	 * place.
1253 	 */
1254 	if (tmp_ncookies != NULL)
1255 		ap->a_ncookies = tmp_ncookies;
1256 
1257 	return (error);
1258 }
1259 
1260 static int
1261 devfs_readlink(struct vop_readlink_args *ap)
1262 {
1263 	struct devfs_dirent *de;
1264 
1265 	de = ap->a_vp->v_data;
1266 	return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1267 }
1268 
1269 static int
1270 devfs_reclaim(struct vop_reclaim_args *ap)
1271 {
1272 	struct vnode *vp = ap->a_vp;
1273 	struct devfs_dirent *de;
1274 	struct cdev *dev;
1275 
1276 	mtx_lock(&devfs_de_interlock);
1277 	de = vp->v_data;
1278 	if (de != NULL) {
1279 		de->de_vnode = NULL;
1280 		vp->v_data = NULL;
1281 	}
1282 	mtx_unlock(&devfs_de_interlock);
1283 
1284 	vnode_destroy_vobject(vp);
1285 
1286 	VI_LOCK(vp);
1287 	dev_lock();
1288 	dev = vp->v_rdev;
1289 	vp->v_rdev = NULL;
1290 
1291 	if (dev == NULL) {
1292 		dev_unlock();
1293 		VI_UNLOCK(vp);
1294 		return (0);
1295 	}
1296 
1297 	dev->si_usecount -= vp->v_usecount;
1298 	dev_unlock();
1299 	VI_UNLOCK(vp);
1300 	dev_rel(dev);
1301 	return (0);
1302 }
1303 
1304 static int
1305 devfs_remove(struct vop_remove_args *ap)
1306 {
1307 	struct vnode *dvp = ap->a_dvp;
1308 	struct vnode *vp = ap->a_vp;
1309 	struct devfs_dirent *dd;
1310 	struct devfs_dirent *de, *de_covered;
1311 	struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1312 
1313 	ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1314 	ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1315 
1316 	sx_xlock(&dmp->dm_lock);
1317 	dd = ap->a_dvp->v_data;
1318 	de = vp->v_data;
1319 	if (de->de_cdp == NULL) {
1320 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1321 		if (de->de_dirent->d_type == DT_LNK) {
1322 			de_covered = devfs_find(dd, de->de_dirent->d_name,
1323 			    de->de_dirent->d_namlen, 0);
1324 			if (de_covered != NULL)
1325 				de_covered->de_flags &= ~DE_COVERED;
1326 		}
1327 		/* We need to unlock dvp because devfs_delete() may lock it. */
1328 		VOP_UNLOCK(vp, 0);
1329 		if (dvp != vp)
1330 			VOP_UNLOCK(dvp, 0);
1331 		devfs_delete(dmp, de, 0);
1332 		sx_xunlock(&dmp->dm_lock);
1333 		if (dvp != vp)
1334 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1335 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1336 	} else {
1337 		de->de_flags |= DE_WHITEOUT;
1338 		sx_xunlock(&dmp->dm_lock);
1339 	}
1340 	return (0);
1341 }
1342 
1343 /*
1344  * Revoke is called on a tty when a terminal session ends.  The vnode
1345  * is orphaned by setting v_op to deadfs so we need to let go of it
1346  * as well so that we create a new one next time around.
1347  *
1348  */
1349 static int
1350 devfs_revoke(struct vop_revoke_args *ap)
1351 {
1352 	struct vnode *vp = ap->a_vp, *vp2;
1353 	struct cdev *dev;
1354 	struct cdev_priv *cdp;
1355 	struct devfs_dirent *de;
1356 	int i;
1357 
1358 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1359 
1360 	dev = vp->v_rdev;
1361 	cdp = cdev2priv(dev);
1362 
1363 	dev_lock();
1364 	cdp->cdp_inuse++;
1365 	dev_unlock();
1366 
1367 	vhold(vp);
1368 	vgone(vp);
1369 	vdrop(vp);
1370 
1371 	VOP_UNLOCK(vp,0);
1372  loop:
1373 	for (;;) {
1374 		mtx_lock(&devfs_de_interlock);
1375 		dev_lock();
1376 		vp2 = NULL;
1377 		for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1378 			de = cdp->cdp_dirents[i];
1379 			if (de == NULL)
1380 				continue;
1381 
1382 			vp2 = de->de_vnode;
1383 			if (vp2 != NULL) {
1384 				dev_unlock();
1385 				VI_LOCK(vp2);
1386 				mtx_unlock(&devfs_de_interlock);
1387 				if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1388 				    curthread))
1389 					goto loop;
1390 				vhold(vp2);
1391 				vgone(vp2);
1392 				vdrop(vp2);
1393 				vput(vp2);
1394 				break;
1395 			}
1396 		}
1397 		if (vp2 != NULL) {
1398 			continue;
1399 		}
1400 		dev_unlock();
1401 		mtx_unlock(&devfs_de_interlock);
1402 		break;
1403 	}
1404 	dev_lock();
1405 	cdp->cdp_inuse--;
1406 	if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1407 		TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1408 		dev_unlock();
1409 		dev_rel(&cdp->cdp_c);
1410 	} else
1411 		dev_unlock();
1412 
1413 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1414 	return (0);
1415 }
1416 
1417 static int
1418 devfs_rioctl(struct vop_ioctl_args *ap)
1419 {
1420 	struct vnode *vp;
1421 	struct devfs_mount *dmp;
1422 	int error;
1423 
1424 	vp = ap->a_vp;
1425 	vn_lock(vp, LK_SHARED | LK_RETRY);
1426 	if (vp->v_iflag & VI_DOOMED) {
1427 		VOP_UNLOCK(vp, 0);
1428 		return (EBADF);
1429 	}
1430 	dmp = VFSTODEVFS(vp->v_mount);
1431 	sx_xlock(&dmp->dm_lock);
1432 	VOP_UNLOCK(vp, 0);
1433 	DEVFS_DMP_HOLD(dmp);
1434 	devfs_populate(dmp);
1435 	if (DEVFS_DMP_DROP(dmp)) {
1436 		sx_xunlock(&dmp->dm_lock);
1437 		devfs_unmount_final(dmp);
1438 		return (ENOENT);
1439 	}
1440 	error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1441 	sx_xunlock(&dmp->dm_lock);
1442 	return (error);
1443 }
1444 
1445 static int
1446 devfs_rread(struct vop_read_args *ap)
1447 {
1448 
1449 	if (ap->a_vp->v_type != VDIR)
1450 		return (EINVAL);
1451 	return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1452 }
1453 
1454 static int
1455 devfs_setattr(struct vop_setattr_args *ap)
1456 {
1457 	struct devfs_dirent *de;
1458 	struct vattr *vap;
1459 	struct vnode *vp;
1460 	struct thread *td;
1461 	int c, error;
1462 	uid_t uid;
1463 	gid_t gid;
1464 
1465 	vap = ap->a_vap;
1466 	vp = ap->a_vp;
1467 	td = curthread;
1468 	if ((vap->va_type != VNON) ||
1469 	    (vap->va_nlink != VNOVAL) ||
1470 	    (vap->va_fsid != VNOVAL) ||
1471 	    (vap->va_fileid != VNOVAL) ||
1472 	    (vap->va_blocksize != VNOVAL) ||
1473 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1474 	    (vap->va_rdev != VNOVAL) ||
1475 	    ((int)vap->va_bytes != VNOVAL) ||
1476 	    (vap->va_gen != VNOVAL)) {
1477 		return (EINVAL);
1478 	}
1479 
1480 	de = vp->v_data;
1481 	if (vp->v_type == VDIR)
1482 		de = de->de_dir;
1483 
1484 	error = c = 0;
1485 	if (vap->va_uid == (uid_t)VNOVAL)
1486 		uid = de->de_uid;
1487 	else
1488 		uid = vap->va_uid;
1489 	if (vap->va_gid == (gid_t)VNOVAL)
1490 		gid = de->de_gid;
1491 	else
1492 		gid = vap->va_gid;
1493 	if (uid != de->de_uid || gid != de->de_gid) {
1494 		if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1495 		    (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1496 			error = priv_check(td, PRIV_VFS_CHOWN);
1497 			if (error)
1498 				return (error);
1499 		}
1500 		de->de_uid = uid;
1501 		de->de_gid = gid;
1502 		c = 1;
1503 	}
1504 
1505 	if (vap->va_mode != (mode_t)VNOVAL) {
1506 		if (ap->a_cred->cr_uid != de->de_uid) {
1507 			error = priv_check(td, PRIV_VFS_ADMIN);
1508 			if (error)
1509 				return (error);
1510 		}
1511 		de->de_mode = vap->va_mode;
1512 		c = 1;
1513 	}
1514 
1515 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1516 		/* See the comment in ufs_vnops::ufs_setattr(). */
1517 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) &&
1518 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1519 		    (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td))))
1520 			return (error);
1521 		if (vap->va_atime.tv_sec != VNOVAL) {
1522 			if (vp->v_type == VCHR)
1523 				vp->v_rdev->si_atime = vap->va_atime;
1524 			else
1525 				de->de_atime = vap->va_atime;
1526 		}
1527 		if (vap->va_mtime.tv_sec != VNOVAL) {
1528 			if (vp->v_type == VCHR)
1529 				vp->v_rdev->si_mtime = vap->va_mtime;
1530 			else
1531 				de->de_mtime = vap->va_mtime;
1532 		}
1533 		c = 1;
1534 	}
1535 
1536 	if (c) {
1537 		if (vp->v_type == VCHR)
1538 			vfs_timestamp(&vp->v_rdev->si_ctime);
1539 		else
1540 			vfs_timestamp(&de->de_mtime);
1541 	}
1542 	return (0);
1543 }
1544 
1545 #ifdef MAC
1546 static int
1547 devfs_setlabel(struct vop_setlabel_args *ap)
1548 {
1549 	struct vnode *vp;
1550 	struct devfs_dirent *de;
1551 
1552 	vp = ap->a_vp;
1553 	de = vp->v_data;
1554 
1555 	mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1556 	mac_devfs_update(vp->v_mount, de, vp);
1557 
1558 	return (0);
1559 }
1560 #endif
1561 
1562 static int
1563 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1564 {
1565 
1566 	return (vnops.fo_stat(fp, sb, cred, td));
1567 }
1568 
1569 static int
1570 devfs_symlink(struct vop_symlink_args *ap)
1571 {
1572 	int i, error;
1573 	struct devfs_dirent *dd;
1574 	struct devfs_dirent *de, *de_covered, *de_dotdot;
1575 	struct devfs_mount *dmp;
1576 
1577 	error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1578 	if (error)
1579 		return(error);
1580 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1581 	if (devfs_populate_vp(ap->a_dvp) != 0)
1582 		return (ENOENT);
1583 
1584 	dd = ap->a_dvp->v_data;
1585 	de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1586 	de->de_flags = DE_USER;
1587 	de->de_uid = 0;
1588 	de->de_gid = 0;
1589 	de->de_mode = 0755;
1590 	de->de_inode = alloc_unr(devfs_inos);
1591 	de->de_dir = dd;
1592 	de->de_dirent->d_type = DT_LNK;
1593 	i = strlen(ap->a_target) + 1;
1594 	de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1595 	bcopy(ap->a_target, de->de_symlink, i);
1596 #ifdef MAC
1597 	mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1598 #endif
1599 	de_covered = devfs_find(dd, de->de_dirent->d_name,
1600 	    de->de_dirent->d_namlen, 0);
1601 	if (de_covered != NULL) {
1602 		if ((de_covered->de_flags & DE_USER) != 0) {
1603 			devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1604 			sx_xunlock(&dmp->dm_lock);
1605 			return (EEXIST);
1606 		}
1607 		KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1608 		    ("devfs_symlink: entry %p already covered", de_covered));
1609 		de_covered->de_flags |= DE_COVERED;
1610 	}
1611 
1612 	de_dotdot = TAILQ_FIRST(&dd->de_dlist);		/* "." */
1613 	de_dotdot = TAILQ_NEXT(de_dotdot, de_list);	/* ".." */
1614 	TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1615 	devfs_dir_ref_de(dmp, dd);
1616 	devfs_rules_apply(dmp, de);
1617 
1618 	return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1619 }
1620 
1621 static int
1622 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1623 {
1624 
1625 	return (vnops.fo_truncate(fp, length, cred, td));
1626 }
1627 
1628 /* ARGSUSED */
1629 static int
1630 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td)
1631 {
1632 	struct cdev *dev;
1633 	int error, ioflag, ref, resid;
1634 	struct cdevsw *dsw;
1635 	struct file *fpop;
1636 
1637 	fpop = td->td_fpop;
1638 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1639 	if (error)
1640 		return (error);
1641 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1642 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1643 	if (ioflag & O_DIRECT)
1644 		ioflag |= IO_DIRECT;
1645 	if ((flags & FOF_OFFSET) == 0)
1646 		uio->uio_offset = fp->f_offset;
1647 
1648 	resid = uio->uio_resid;
1649 
1650 	error = dsw->d_write(dev, uio, ioflag);
1651 	if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1652 		vfs_timestamp(&dev->si_ctime);
1653 		dev->si_mtime = dev->si_ctime;
1654 	}
1655 	td->td_fpop = fpop;
1656 	dev_relthread(dev, ref);
1657 
1658 	if ((flags & FOF_OFFSET) == 0)
1659 		fp->f_offset = uio->uio_offset;
1660 	fp->f_nextoff = uio->uio_offset;
1661 	return (error);
1662 }
1663 
1664 dev_t
1665 dev2udev(struct cdev *x)
1666 {
1667 	if (x == NULL)
1668 		return (NODEV);
1669 	return (cdev2priv(x)->cdp_inode);
1670 }
1671 
1672 static struct fileops devfs_ops_f = {
1673 	.fo_read =	devfs_read_f,
1674 	.fo_write =	devfs_write_f,
1675 	.fo_truncate =	devfs_truncate_f,
1676 	.fo_ioctl =	devfs_ioctl_f,
1677 	.fo_poll =	devfs_poll_f,
1678 	.fo_kqfilter =	devfs_kqfilter_f,
1679 	.fo_stat =	devfs_stat_f,
1680 	.fo_close =	devfs_close_f,
1681 	.fo_chmod =	vn_chmod,
1682 	.fo_chown =	vn_chown,
1683 	.fo_flags =	DFLAG_PASSABLE | DFLAG_SEEKABLE
1684 };
1685 
1686 static struct vop_vector devfs_vnodeops = {
1687 	.vop_default =		&default_vnodeops,
1688 
1689 	.vop_access =		devfs_access,
1690 	.vop_getattr =		devfs_getattr,
1691 	.vop_ioctl =		devfs_rioctl,
1692 	.vop_lookup =		devfs_lookup,
1693 	.vop_mknod =		devfs_mknod,
1694 	.vop_pathconf =		devfs_pathconf,
1695 	.vop_read =		devfs_rread,
1696 	.vop_readdir =		devfs_readdir,
1697 	.vop_readlink =		devfs_readlink,
1698 	.vop_reclaim =		devfs_reclaim,
1699 	.vop_remove =		devfs_remove,
1700 	.vop_revoke =		devfs_revoke,
1701 	.vop_setattr =		devfs_setattr,
1702 #ifdef MAC
1703 	.vop_setlabel =		devfs_setlabel,
1704 #endif
1705 	.vop_symlink =		devfs_symlink,
1706 	.vop_vptocnp =		devfs_vptocnp,
1707 };
1708 
1709 static struct vop_vector devfs_specops = {
1710 	.vop_default =		&default_vnodeops,
1711 
1712 	.vop_access =		devfs_access,
1713 	.vop_bmap =		VOP_PANIC,
1714 	.vop_close =		devfs_close,
1715 	.vop_create =		VOP_PANIC,
1716 	.vop_fsync =		devfs_fsync,
1717 	.vop_getattr =		devfs_getattr,
1718 	.vop_link =		VOP_PANIC,
1719 	.vop_mkdir =		VOP_PANIC,
1720 	.vop_mknod =		VOP_PANIC,
1721 	.vop_open =		devfs_open,
1722 	.vop_pathconf =		devfs_pathconf,
1723 	.vop_print =		devfs_print,
1724 	.vop_read =		VOP_PANIC,
1725 	.vop_readdir =		VOP_PANIC,
1726 	.vop_readlink =		VOP_PANIC,
1727 	.vop_reallocblks =	VOP_PANIC,
1728 	.vop_reclaim =		devfs_reclaim,
1729 	.vop_remove =		devfs_remove,
1730 	.vop_rename =		VOP_PANIC,
1731 	.vop_revoke =		devfs_revoke,
1732 	.vop_rmdir =		VOP_PANIC,
1733 	.vop_setattr =		devfs_setattr,
1734 #ifdef MAC
1735 	.vop_setlabel =		devfs_setlabel,
1736 #endif
1737 	.vop_strategy =		VOP_PANIC,
1738 	.vop_symlink =		VOP_PANIC,
1739 	.vop_vptocnp =		devfs_vptocnp,
1740 	.vop_write =		VOP_PANIC,
1741 };
1742 
1743 /*
1744  * Our calling convention to the device drivers used to be that we passed
1745  * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1746  * flags instead since that's what open(), close() and ioctl() takes and
1747  * we don't really want vnode.h in device drivers.
1748  * We solved the source compatibility by redefining some vnode flags to
1749  * be the same as the fcntl ones and by sending down the bitwise OR of
1750  * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
1751  * pulls the rug out under this.
1752  */
1753 CTASSERT(O_NONBLOCK == IO_NDELAY);
1754 CTASSERT(O_FSYNC == IO_SYNC);
1755