xref: /freebsd/sys/fs/devfs/devfs_vnops.c (revision eb6d21b4ca6d668cf89afd99eef7baeafa712197)
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  *	remove empty directories
40  *	mkdir: want it ?
41  */
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/conf.h>
46 #include <sys/dirent.h>
47 #include <sys/fcntl.h>
48 #include <sys/file.h>
49 #include <sys/filedesc.h>
50 #include <sys/filio.h>
51 #include <sys/jail.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/malloc.h>
55 #include <sys/mount.h>
56 #include <sys/namei.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/stat.h>
60 #include <sys/sx.h>
61 #include <sys/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 vop_vector devfs_specops;
68 static struct fileops devfs_ops_f;
69 
70 #include <fs/devfs/devfs.h>
71 #include <fs/devfs/devfs_int.h>
72 
73 #include <security/mac/mac_framework.h>
74 
75 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
76 
77 struct mtx	devfs_de_interlock;
78 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
79 struct sx	clone_drain_lock;
80 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
81 struct mtx	cdevpriv_mtx;
82 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
83 
84 static int
85 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp)
86 {
87 
88 	*dswp = devvn_refthread(fp->f_vnode, devp);
89 	if (*devp != fp->f_data) {
90 		if (*dswp != NULL)
91 			dev_relthread(*devp);
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 static int
188 devfs_vptocnp(struct vop_vptocnp_args *ap)
189 {
190 	struct vnode *vp = ap->a_vp;
191 	struct vnode **dvp = ap->a_vpp;
192 	struct devfs_mount *dmp;
193 	char *buf = ap->a_buf;
194 	int *buflen = ap->a_buflen;
195 	struct devfs_dirent *dd, *de;
196 	int i, error;
197 
198 	dmp = VFSTODEVFS(vp->v_mount);
199 	i = *buflen;
200 	dd = vp->v_data;
201 	error = 0;
202 
203 	sx_xlock(&dmp->dm_lock);
204 
205 	if (vp->v_type == VCHR) {
206 		i -= strlen(dd->de_cdp->cdp_c.si_name);
207 		if (i < 0) {
208 			error = ENOMEM;
209 			goto finished;
210 		}
211 		bcopy(dd->de_cdp->cdp_c.si_name, buf + i,
212 		    strlen(dd->de_cdp->cdp_c.si_name));
213 		de = dd->de_dir;
214 	} else if (vp->v_type == VDIR) {
215 		if (dd == dmp->dm_rootdir) {
216 			*dvp = vp;
217 			vhold(*dvp);
218 			goto finished;
219 		}
220 		i -= dd->de_dirent->d_namlen;
221 		if (i < 0) {
222 			error = ENOMEM;
223 			goto finished;
224 		}
225 		bcopy(dd->de_dirent->d_name, buf + i,
226 		    dd->de_dirent->d_namlen);
227 		de = dd;
228 	} else {
229 		error = ENOENT;
230 		goto finished;
231 	}
232 	*buflen = i;
233 	de = TAILQ_FIRST(&de->de_dlist);	/* "." */
234 	de = TAILQ_NEXT(de, de_list);		/* ".." */
235 	de = de->de_dir;
236 	mtx_lock(&devfs_de_interlock);
237 	*dvp = de->de_vnode;
238 	if (*dvp != NULL) {
239 		VI_LOCK(*dvp);
240 		mtx_unlock(&devfs_de_interlock);
241 		vholdl(*dvp);
242 		VI_UNLOCK(*dvp);
243 	} else {
244 		mtx_unlock(&devfs_de_interlock);
245 		error = ENOENT;
246 	}
247 finished:
248 	sx_xunlock(&dmp->dm_lock);
249 	return (error);
250 }
251 
252 /*
253  * Construct the fully qualified path name relative to the mountpoint
254  */
255 static char *
256 devfs_fqpn(char *buf, struct vnode *dvp, struct componentname *cnp)
257 {
258 	int i;
259 	struct devfs_dirent *de, *dd;
260 	struct devfs_mount *dmp;
261 
262 	dmp = VFSTODEVFS(dvp->v_mount);
263 	dd = dvp->v_data;
264 	i = SPECNAMELEN;
265 	buf[i] = '\0';
266 	i -= cnp->cn_namelen;
267 	if (i < 0)
268 		 return (NULL);
269 	bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
270 	de = dd;
271 	while (de != dmp->dm_rootdir) {
272 		i--;
273 		if (i < 0)
274 			 return (NULL);
275 		buf[i] = '/';
276 		i -= de->de_dirent->d_namlen;
277 		if (i < 0)
278 			 return (NULL);
279 		bcopy(de->de_dirent->d_name, buf + i,
280 		    de->de_dirent->d_namlen);
281 		de = TAILQ_FIRST(&de->de_dlist);	/* "." */
282 		de = TAILQ_NEXT(de, de_list);		/* ".." */
283 		de = de->de_dir;
284 	}
285 	return (buf + i);
286 }
287 
288 static int
289 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
290 	struct devfs_dirent *de)
291 {
292 	int not_found;
293 
294 	not_found = 0;
295 	if (de->de_flags & DE_DOOMED)
296 		not_found = 1;
297 	if (DEVFS_DE_DROP(de)) {
298 		KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
299 		devfs_dirent_free(de);
300 	}
301 	if (DEVFS_DMP_DROP(dmp)) {
302 		KASSERT(not_found == 1,
303 			("DEVFS mount struct freed before dirent"));
304 		not_found = 2;
305 		sx_xunlock(&dmp->dm_lock);
306 		devfs_unmount_final(dmp);
307 	}
308 	if (not_found == 1 || (drop_dm_lock && not_found != 2))
309 		sx_unlock(&dmp->dm_lock);
310 	return (not_found);
311 }
312 
313 static void
314 devfs_insmntque_dtr(struct vnode *vp, void *arg)
315 {
316 	struct devfs_dirent *de;
317 
318 	de = (struct devfs_dirent *)arg;
319 	mtx_lock(&devfs_de_interlock);
320 	vp->v_data = NULL;
321 	de->de_vnode = NULL;
322 	mtx_unlock(&devfs_de_interlock);
323 	vgone(vp);
324 	vput(vp);
325 }
326 
327 /*
328  * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
329  * it on return.
330  */
331 int
332 devfs_allocv(struct devfs_dirent *de, struct mount *mp, struct vnode **vpp)
333 {
334 	int error;
335 	struct vnode *vp;
336 	struct cdev *dev;
337 	struct devfs_mount *dmp;
338 
339 	dmp = VFSTODEVFS(mp);
340 	if (de->de_flags & DE_DOOMED) {
341 		sx_xunlock(&dmp->dm_lock);
342 		return (ENOENT);
343 	}
344 	DEVFS_DE_HOLD(de);
345 	DEVFS_DMP_HOLD(dmp);
346 	mtx_lock(&devfs_de_interlock);
347 	vp = de->de_vnode;
348 	if (vp != NULL) {
349 		VI_LOCK(vp);
350 		mtx_unlock(&devfs_de_interlock);
351 		sx_xunlock(&dmp->dm_lock);
352 		error = vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, curthread);
353 		sx_xlock(&dmp->dm_lock);
354 		if (devfs_allocv_drop_refs(0, dmp, de)) {
355 			if (error == 0)
356 				vput(vp);
357 			return (ENOENT);
358 		}
359 		else if (error) {
360 			sx_xunlock(&dmp->dm_lock);
361 			return (error);
362 		}
363 		sx_xunlock(&dmp->dm_lock);
364 		*vpp = vp;
365 		return (0);
366 	}
367 	mtx_unlock(&devfs_de_interlock);
368 	if (de->de_dirent->d_type == DT_CHR) {
369 		if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
370 			devfs_allocv_drop_refs(1, dmp, de);
371 			return (ENOENT);
372 		}
373 		dev = &de->de_cdp->cdp_c;
374 	} else {
375 		dev = NULL;
376 	}
377 	error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
378 	if (error != 0) {
379 		devfs_allocv_drop_refs(1, dmp, de);
380 		printf("devfs_allocv: failed to allocate new vnode\n");
381 		return (error);
382 	}
383 
384 	if (de->de_dirent->d_type == DT_CHR) {
385 		vp->v_type = VCHR;
386 		VI_LOCK(vp);
387 		dev_lock();
388 		dev_refl(dev);
389 		/* XXX: v_rdev should be protect by vnode lock */
390 		vp->v_rdev = dev;
391 		KASSERT(vp->v_usecount == 1,
392 		    ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount));
393 		dev->si_usecount += vp->v_usecount;
394 		dev_unlock();
395 		VI_UNLOCK(vp);
396 		vp->v_op = &devfs_specops;
397 	} else if (de->de_dirent->d_type == DT_DIR) {
398 		vp->v_type = VDIR;
399 	} else if (de->de_dirent->d_type == DT_LNK) {
400 		vp->v_type = VLNK;
401 	} else {
402 		vp->v_type = VBAD;
403 	}
404 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
405 	mtx_lock(&devfs_de_interlock);
406 	vp->v_data = de;
407 	de->de_vnode = vp;
408 	mtx_unlock(&devfs_de_interlock);
409 	error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
410 	if (error != 0) {
411 		(void) devfs_allocv_drop_refs(1, dmp, de);
412 		return (error);
413 	}
414 	if (devfs_allocv_drop_refs(0, dmp, de)) {
415 		vput(vp);
416 		return (ENOENT);
417 	}
418 #ifdef MAC
419 	mac_devfs_vnode_associate(mp, de, vp);
420 #endif
421 	sx_xunlock(&dmp->dm_lock);
422 	*vpp = vp;
423 	return (0);
424 }
425 
426 static int
427 devfs_access(struct vop_access_args *ap)
428 {
429 	struct vnode *vp = ap->a_vp;
430 	struct devfs_dirent *de;
431 	int error;
432 
433 	de = vp->v_data;
434 	if (vp->v_type == VDIR)
435 		de = de->de_dir;
436 
437 	error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
438 	    ap->a_accmode, ap->a_cred, NULL);
439 	if (!error)
440 		return (error);
441 	if (error != EACCES)
442 		return (error);
443 	/* We do, however, allow access to the controlling terminal */
444 	if (!(ap->a_td->td_proc->p_flag & P_CONTROLT))
445 		return (error);
446 	if (ap->a_td->td_proc->p_session->s_ttyvp == de->de_vnode)
447 		return (0);
448 	return (error);
449 }
450 
451 /* ARGSUSED */
452 static int
453 devfs_close(struct vop_close_args *ap)
454 {
455 	struct vnode *vp = ap->a_vp, *oldvp;
456 	struct thread *td = ap->a_td;
457 	struct cdev *dev = vp->v_rdev;
458 	struct cdevsw *dsw;
459 	int vp_locked, error;
460 
461 	/*
462 	 * Hack: a tty device that is a controlling terminal
463 	 * has a reference from the session structure.
464 	 * We cannot easily tell that a character device is
465 	 * a controlling terminal, unless it is the closing
466 	 * process' controlling terminal.  In that case,
467 	 * if the reference count is 2 (this last descriptor
468 	 * plus the session), release the reference from the session.
469 	 */
470 	oldvp = NULL;
471 	sx_xlock(&proctree_lock);
472 	if (td && vp == td->td_proc->p_session->s_ttyvp) {
473 		SESS_LOCK(td->td_proc->p_session);
474 		VI_LOCK(vp);
475 		if (count_dev(dev) == 2 && (vp->v_iflag & VI_DOOMED) == 0) {
476 			td->td_proc->p_session->s_ttyvp = NULL;
477 			oldvp = vp;
478 		}
479 		VI_UNLOCK(vp);
480 		SESS_UNLOCK(td->td_proc->p_session);
481 	}
482 	sx_xunlock(&proctree_lock);
483 	if (oldvp != NULL)
484 		vrele(oldvp);
485 	/*
486 	 * We do not want to really close the device if it
487 	 * is still in use unless we are trying to close it
488 	 * forcibly. Since every use (buffer, vnode, swap, cmap)
489 	 * holds a reference to the vnode, and because we mark
490 	 * any other vnodes that alias this device, when the
491 	 * sum of the reference counts on all the aliased
492 	 * vnodes descends to one, we are on last close.
493 	 */
494 	dsw = dev_refthread(dev);
495 	if (dsw == NULL)
496 		return (ENXIO);
497 	VI_LOCK(vp);
498 	if (vp->v_iflag & VI_DOOMED) {
499 		/* Forced close. */
500 	} else if (dsw->d_flags & D_TRACKCLOSE) {
501 		/* Keep device updated on status. */
502 	} else if (count_dev(dev) > 1) {
503 		VI_UNLOCK(vp);
504 		dev_relthread(dev);
505 		return (0);
506 	}
507 	vholdl(vp);
508 	VI_UNLOCK(vp);
509 	vp_locked = VOP_ISLOCKED(vp);
510 	VOP_UNLOCK(vp, 0);
511 	KASSERT(dev->si_refcount > 0,
512 	    ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
513 	error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td);
514 	dev_relthread(dev);
515 	vn_lock(vp, vp_locked | LK_RETRY);
516 	vdrop(vp);
517 	return (error);
518 }
519 
520 static int
521 devfs_close_f(struct file *fp, struct thread *td)
522 {
523 	int error;
524 	struct file *fpop;
525 
526 	fpop = td->td_fpop;
527 	td->td_fpop = fp;
528 	error = vnops.fo_close(fp, td);
529 	td->td_fpop = fpop;
530 	return (error);
531 }
532 
533 static int
534 devfs_fsync(struct vop_fsync_args *ap)
535 {
536 	int error;
537 	struct bufobj *bo;
538 	struct devfs_dirent *de;
539 
540 	if (!vn_isdisk(ap->a_vp, &error)) {
541 		bo = &ap->a_vp->v_bufobj;
542 		de = ap->a_vp->v_data;
543 		if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) {
544 			printf("Device %s went missing before all of the data "
545 			    "could be written to it; expect data loss.\n",
546 			    de->de_dirent->d_name);
547 
548 			error = vop_stdfsync(ap);
549 			if (bo->bo_dirty.bv_cnt != 0 || error != 0)
550 				panic("devfs_fsync: vop_stdfsync failed.");
551 		}
552 
553 		return (0);
554 	}
555 
556 	return (vop_stdfsync(ap));
557 }
558 
559 static int
560 devfs_getattr(struct vop_getattr_args *ap)
561 {
562 	struct vnode *vp = ap->a_vp;
563 	struct vattr *vap = ap->a_vap;
564 	int error = 0;
565 	struct devfs_dirent *de;
566 	struct cdev *dev;
567 
568 	de = vp->v_data;
569 	KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
570 	if (vp->v_type == VDIR) {
571 		de = de->de_dir;
572 		KASSERT(de != NULL,
573 		    ("Null dir dirent in devfs_getattr vp=%p", vp));
574 	}
575 	vap->va_uid = de->de_uid;
576 	vap->va_gid = de->de_gid;
577 	vap->va_mode = de->de_mode;
578 	if (vp->v_type == VLNK)
579 		vap->va_size = strlen(de->de_symlink);
580 	else if (vp->v_type == VDIR)
581 		vap->va_size = vap->va_bytes = DEV_BSIZE;
582 	else
583 		vap->va_size = 0;
584 	if (vp->v_type != VDIR)
585 		vap->va_bytes = 0;
586 	vap->va_blocksize = DEV_BSIZE;
587 	vap->va_type = vp->v_type;
588 
589 #define fix(aa)							\
590 	do {							\
591 		if ((aa).tv_sec <= 3600) {			\
592 			(aa).tv_sec = boottime.tv_sec;		\
593 			(aa).tv_nsec = boottime.tv_usec * 1000; \
594 		}						\
595 	} while (0)
596 
597 	if (vp->v_type != VCHR)  {
598 		fix(de->de_atime);
599 		vap->va_atime = de->de_atime;
600 		fix(de->de_mtime);
601 		vap->va_mtime = de->de_mtime;
602 		fix(de->de_ctime);
603 		vap->va_ctime = de->de_ctime;
604 	} else {
605 		dev = vp->v_rdev;
606 		fix(dev->si_atime);
607 		vap->va_atime = dev->si_atime;
608 		fix(dev->si_mtime);
609 		vap->va_mtime = dev->si_mtime;
610 		fix(dev->si_ctime);
611 		vap->va_ctime = dev->si_ctime;
612 
613 		vap->va_rdev = cdev2priv(dev)->cdp_inode;
614 	}
615 	vap->va_gen = 0;
616 	vap->va_flags = 0;
617 	vap->va_filerev = 0;
618 	vap->va_nlink = de->de_links;
619 	vap->va_fileid = de->de_inode;
620 
621 	return (error);
622 }
623 
624 /* ARGSUSED */
625 static int
626 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
627 {
628 	struct cdev *dev;
629 	struct cdevsw *dsw;
630 	struct vnode *vp;
631 	struct vnode *vpold;
632 	int error, i;
633 	const char *p;
634 	struct fiodgname_arg *fgn;
635 	struct file *fpop;
636 
637 	fpop = td->td_fpop;
638 	error = devfs_fp_check(fp, &dev, &dsw);
639 	if (error)
640 		return (error);
641 
642 	if (com == FIODTYPE) {
643 		*(int *)data = dsw->d_flags & D_TYPEMASK;
644 		td->td_fpop = fpop;
645 		dev_relthread(dev);
646 		return (0);
647 	} else if (com == FIODGNAME) {
648 		fgn = data;
649 		p = devtoname(dev);
650 		i = strlen(p) + 1;
651 		if (i > fgn->len)
652 			error = EINVAL;
653 		else
654 			error = copyout(p, fgn->buf, i);
655 		td->td_fpop = fpop;
656 		dev_relthread(dev);
657 		return (error);
658 	}
659 	error = dsw->d_ioctl(dev, com, data, fp->f_flag, td);
660 	td->td_fpop = NULL;
661 	dev_relthread(dev);
662 	if (error == ENOIOCTL)
663 		error = ENOTTY;
664 	if (error == 0 && com == TIOCSCTTY) {
665 		vp = fp->f_vnode;
666 
667 		/* Do nothing if reassigning same control tty */
668 		sx_slock(&proctree_lock);
669 		if (td->td_proc->p_session->s_ttyvp == vp) {
670 			sx_sunlock(&proctree_lock);
671 			return (0);
672 		}
673 
674 		vpold = td->td_proc->p_session->s_ttyvp;
675 		VREF(vp);
676 		SESS_LOCK(td->td_proc->p_session);
677 		td->td_proc->p_session->s_ttyvp = vp;
678 		SESS_UNLOCK(td->td_proc->p_session);
679 
680 		sx_sunlock(&proctree_lock);
681 
682 		/* Get rid of reference to old control tty */
683 		if (vpold)
684 			vrele(vpold);
685 	}
686 	return (error);
687 }
688 
689 /* ARGSUSED */
690 static int
691 devfs_kqfilter_f(struct file *fp, struct knote *kn)
692 {
693 	struct cdev *dev;
694 	struct cdevsw *dsw;
695 	int error;
696 	struct file *fpop;
697 	struct thread *td;
698 
699 	td = curthread;
700 	fpop = td->td_fpop;
701 	error = devfs_fp_check(fp, &dev, &dsw);
702 	if (error)
703 		return (error);
704 	error = dsw->d_kqfilter(dev, kn);
705 	td->td_fpop = fpop;
706 	dev_relthread(dev);
707 	return (error);
708 }
709 
710 static inline int
711 devfs_prison_check(struct devfs_dirent *de, struct ucred *tcr)
712 {
713 	struct cdev_priv *cdp;
714 	struct ucred *dcr;
715 
716 	cdp = de->de_cdp;
717 	if (cdp == NULL)
718 		return (0);
719 	dcr = cdp->cdp_c.si_cred;
720 	if (dcr == NULL)
721 		return (0);
722 
723 	return (prison_check(tcr, dcr));
724 }
725 
726 static int
727 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
728 {
729 	struct componentname *cnp;
730 	struct vnode *dvp, **vpp;
731 	struct thread *td;
732 	struct devfs_dirent *de, *dd;
733 	struct devfs_dirent **dde;
734 	struct devfs_mount *dmp;
735 	struct cdev *cdev;
736 	int error, flags, nameiop;
737 	char specname[SPECNAMELEN + 1], *pname;
738 
739 	cnp = ap->a_cnp;
740 	vpp = ap->a_vpp;
741 	dvp = ap->a_dvp;
742 	pname = cnp->cn_nameptr;
743 	td = cnp->cn_thread;
744 	flags = cnp->cn_flags;
745 	nameiop = cnp->cn_nameiop;
746 	dmp = VFSTODEVFS(dvp->v_mount);
747 	dd = dvp->v_data;
748 	*vpp = NULLVP;
749 
750 	if ((flags & ISLASTCN) && nameiop == RENAME)
751 		return (EOPNOTSUPP);
752 
753 	if (dvp->v_type != VDIR)
754 		return (ENOTDIR);
755 
756 	if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
757 		return (EIO);
758 
759 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
760 	if (error)
761 		return (error);
762 
763 	if (cnp->cn_namelen == 1 && *pname == '.') {
764 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
765 			return (EINVAL);
766 		*vpp = dvp;
767 		VREF(dvp);
768 		return (0);
769 	}
770 
771 	if (flags & ISDOTDOT) {
772 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
773 			return (EINVAL);
774 		VOP_UNLOCK(dvp, 0);
775 		de = TAILQ_FIRST(&dd->de_dlist);	/* "." */
776 		de = TAILQ_NEXT(de, de_list);		/* ".." */
777 		de = de->de_dir;
778 		error = devfs_allocv(de, dvp->v_mount, vpp);
779 		*dm_unlock = 0;
780 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
781 		return (error);
782 	}
783 
784 	DEVFS_DMP_HOLD(dmp);
785 	devfs_populate(dmp);
786 	if (DEVFS_DMP_DROP(dmp)) {
787 		*dm_unlock = 0;
788 		sx_xunlock(&dmp->dm_lock);
789 		devfs_unmount_final(dmp);
790 		return (ENOENT);
791 	}
792 	dd = dvp->v_data;
793 	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen);
794 	while (de == NULL) {	/* While(...) so we can use break */
795 
796 		if (nameiop == DELETE)
797 			return (ENOENT);
798 
799 		/*
800 		 * OK, we didn't have an entry for the name we were asked for
801 		 * so we try to see if anybody can create it on demand.
802 		 */
803 		pname = devfs_fqpn(specname, dvp, cnp);
804 		if (pname == NULL)
805 			break;
806 
807 		cdev = NULL;
808 		DEVFS_DMP_HOLD(dmp);
809 		sx_xunlock(&dmp->dm_lock);
810 		sx_slock(&clone_drain_lock);
811 		EVENTHANDLER_INVOKE(dev_clone,
812 		    td->td_ucred, pname, strlen(pname), &cdev);
813 		sx_sunlock(&clone_drain_lock);
814 		sx_xlock(&dmp->dm_lock);
815 		if (DEVFS_DMP_DROP(dmp)) {
816 			*dm_unlock = 0;
817 			sx_xunlock(&dmp->dm_lock);
818 			devfs_unmount_final(dmp);
819 			return (ENOENT);
820 		}
821 		if (cdev == NULL)
822 			break;
823 
824 		DEVFS_DMP_HOLD(dmp);
825 		devfs_populate(dmp);
826 		if (DEVFS_DMP_DROP(dmp)) {
827 			*dm_unlock = 0;
828 			sx_xunlock(&dmp->dm_lock);
829 			devfs_unmount_final(dmp);
830 			return (ENOENT);
831 		}
832 
833 		dev_lock();
834 		dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
835 		if (dde != NULL && *dde != NULL)
836 			de = *dde;
837 		dev_unlock();
838 		dev_rel(cdev);
839 		break;
840 	}
841 
842 	if (de == NULL || de->de_flags & DE_WHITEOUT) {
843 		if ((nameiop == CREATE || nameiop == RENAME) &&
844 		    (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
845 			cnp->cn_flags |= SAVENAME;
846 			return (EJUSTRETURN);
847 		}
848 		return (ENOENT);
849 	}
850 
851 	if (devfs_prison_check(de, td->td_ucred))
852 		return (ENOENT);
853 
854 	if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
855 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
856 		if (error)
857 			return (error);
858 		if (*vpp == dvp) {
859 			VREF(dvp);
860 			*vpp = dvp;
861 			return (0);
862 		}
863 	}
864 	error = devfs_allocv(de, dvp->v_mount, vpp);
865 	*dm_unlock = 0;
866 	return (error);
867 }
868 
869 static int
870 devfs_lookup(struct vop_lookup_args *ap)
871 {
872 	int j;
873 	struct devfs_mount *dmp;
874 	int dm_unlock;
875 
876 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
877 	dm_unlock = 1;
878 	sx_xlock(&dmp->dm_lock);
879 	j = devfs_lookupx(ap, &dm_unlock);
880 	if (dm_unlock == 1)
881 		sx_xunlock(&dmp->dm_lock);
882 	return (j);
883 }
884 
885 static int
886 devfs_mknod(struct vop_mknod_args *ap)
887 {
888 	struct componentname *cnp;
889 	struct vnode *dvp, **vpp;
890 	struct devfs_dirent *dd, *de;
891 	struct devfs_mount *dmp;
892 	int error;
893 
894 	/*
895 	 * The only type of node we should be creating here is a
896 	 * character device, for anything else return EOPNOTSUPP.
897 	 */
898 	if (ap->a_vap->va_type != VCHR)
899 		return (EOPNOTSUPP);
900 	dvp = ap->a_dvp;
901 	dmp = VFSTODEVFS(dvp->v_mount);
902 
903 	cnp = ap->a_cnp;
904 	vpp = ap->a_vpp;
905 	dd = dvp->v_data;
906 
907 	error = ENOENT;
908 	sx_xlock(&dmp->dm_lock);
909 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
910 		if (cnp->cn_namelen != de->de_dirent->d_namlen)
911 			continue;
912 		if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
913 		    de->de_dirent->d_namlen) != 0)
914 			continue;
915 		if (de->de_flags & DE_WHITEOUT)
916 			break;
917 		goto notfound;
918 	}
919 	if (de == NULL)
920 		goto notfound;
921 	de->de_flags &= ~DE_WHITEOUT;
922 	error = devfs_allocv(de, dvp->v_mount, vpp);
923 	return (error);
924 notfound:
925 	sx_xunlock(&dmp->dm_lock);
926 	return (error);
927 }
928 
929 /* ARGSUSED */
930 static int
931 devfs_open(struct vop_open_args *ap)
932 {
933 	struct thread *td = ap->a_td;
934 	struct vnode *vp = ap->a_vp;
935 	struct cdev *dev = vp->v_rdev;
936 	struct file *fp = ap->a_fp;
937 	int error;
938 	struct cdevsw *dsw;
939 	struct file *fpop;
940 
941 	if (vp->v_type == VBLK)
942 		return (ENXIO);
943 
944 	if (dev == NULL)
945 		return (ENXIO);
946 
947 	/* Make this field valid before any I/O in d_open. */
948 	if (dev->si_iosize_max == 0)
949 		dev->si_iosize_max = DFLTPHYS;
950 
951 	dsw = dev_refthread(dev);
952 	if (dsw == NULL)
953 		return (ENXIO);
954 
955 	/* XXX: Special casing of ttys for deadfs.  Probably redundant. */
956 	if (dsw->d_flags & D_TTY)
957 		vp->v_vflag |= VV_ISTTY;
958 
959 	VOP_UNLOCK(vp, 0);
960 
961 	fpop = td->td_fpop;
962 	td->td_fpop = fp;
963 	if (fp != NULL) {
964 		fp->f_data = dev;
965 		fp->f_vnode = vp;
966 	}
967 	if (dsw->d_fdopen != NULL)
968 		error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
969 	else
970 		error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
971 	td->td_fpop = fpop;
972 
973 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
974 
975 	dev_relthread(dev);
976 
977 	if (error)
978 		return (error);
979 
980 #if 0	/* /dev/console */
981 	KASSERT(fp != NULL,
982 	     ("Could not vnode bypass device on NULL fp"));
983 #else
984 	if(fp == NULL)
985 		return (error);
986 #endif
987 	if (fp->f_ops == &badfileops)
988 		finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
989 	return (error);
990 }
991 
992 static int
993 devfs_pathconf(struct vop_pathconf_args *ap)
994 {
995 
996 	switch (ap->a_name) {
997 	case _PC_MAC_PRESENT:
998 #ifdef MAC
999 		/*
1000 		 * If MAC is enabled, devfs automatically supports
1001 		 * trivial non-persistant label storage.
1002 		 */
1003 		*ap->a_retval = 1;
1004 #else
1005 		*ap->a_retval = 0;
1006 #endif
1007 		return (0);
1008 	default:
1009 		return (vop_stdpathconf(ap));
1010 	}
1011 	/* NOTREACHED */
1012 }
1013 
1014 /* ARGSUSED */
1015 static int
1016 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1017 {
1018 	struct cdev *dev;
1019 	struct cdevsw *dsw;
1020 	int error;
1021 	struct file *fpop;
1022 
1023 	fpop = td->td_fpop;
1024 	error = devfs_fp_check(fp, &dev, &dsw);
1025 	if (error)
1026 		return (poll_no_poll(events));
1027 	error = dsw->d_poll(dev, events, td);
1028 	td->td_fpop = fpop;
1029 	dev_relthread(dev);
1030 	return(error);
1031 }
1032 
1033 /*
1034  * Print out the contents of a special device vnode.
1035  */
1036 static int
1037 devfs_print(struct vop_print_args *ap)
1038 {
1039 
1040 	printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1041 	return (0);
1042 }
1043 
1044 /* ARGSUSED */
1045 static int
1046 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td)
1047 {
1048 	struct cdev *dev;
1049 	int ioflag, error, resid;
1050 	struct cdevsw *dsw;
1051 	struct file *fpop;
1052 
1053 	fpop = td->td_fpop;
1054 	error = devfs_fp_check(fp, &dev, &dsw);
1055 	if (error)
1056 		return (error);
1057 	resid = uio->uio_resid;
1058 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1059 	if (ioflag & O_DIRECT)
1060 		ioflag |= IO_DIRECT;
1061 
1062 	if ((flags & FOF_OFFSET) == 0)
1063 		uio->uio_offset = fp->f_offset;
1064 
1065 	error = dsw->d_read(dev, uio, ioflag);
1066 	if (uio->uio_resid != resid || (error == 0 && resid != 0))
1067 		vfs_timestamp(&dev->si_atime);
1068 	td->td_fpop = fpop;
1069 	dev_relthread(dev);
1070 
1071 	if ((flags & FOF_OFFSET) == 0)
1072 		fp->f_offset = uio->uio_offset;
1073 	fp->f_nextoff = uio->uio_offset;
1074 	return (error);
1075 }
1076 
1077 static int
1078 devfs_readdir(struct vop_readdir_args *ap)
1079 {
1080 	int error;
1081 	struct uio *uio;
1082 	struct dirent *dp;
1083 	struct devfs_dirent *dd;
1084 	struct devfs_dirent *de;
1085 	struct devfs_mount *dmp;
1086 	off_t off;
1087 	int *tmp_ncookies = NULL;
1088 
1089 	if (ap->a_vp->v_type != VDIR)
1090 		return (ENOTDIR);
1091 
1092 	uio = ap->a_uio;
1093 	if (uio->uio_offset < 0)
1094 		return (EINVAL);
1095 
1096 	/*
1097 	 * XXX: This is a temporary hack to get around this filesystem not
1098 	 * supporting cookies. We store the location of the ncookies pointer
1099 	 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1100 	 * and set the number of cookies to 0. We then set the pointer to
1101 	 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1102 	 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1103 	 * pointer to its original location before returning to the caller.
1104 	 */
1105 	if (ap->a_ncookies != NULL) {
1106 		tmp_ncookies = ap->a_ncookies;
1107 		*ap->a_ncookies = 0;
1108 		ap->a_ncookies = NULL;
1109 	}
1110 
1111 	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1112 	sx_xlock(&dmp->dm_lock);
1113 	DEVFS_DMP_HOLD(dmp);
1114 	devfs_populate(dmp);
1115 	if (DEVFS_DMP_DROP(dmp)) {
1116 		sx_xunlock(&dmp->dm_lock);
1117 		devfs_unmount_final(dmp);
1118 		if (tmp_ncookies != NULL)
1119 			ap->a_ncookies = tmp_ncookies;
1120 		return (EIO);
1121 	}
1122 	error = 0;
1123 	de = ap->a_vp->v_data;
1124 	off = 0;
1125 	TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1126 		KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1127 		if (dd->de_flags & DE_WHITEOUT)
1128 			continue;
1129 		if (devfs_prison_check(dd, ap->a_cred))
1130 			continue;
1131 		if (dd->de_dirent->d_type == DT_DIR)
1132 			de = dd->de_dir;
1133 		else
1134 			de = dd;
1135 		dp = dd->de_dirent;
1136 		if (dp->d_reclen > uio->uio_resid)
1137 			break;
1138 		dp->d_fileno = de->de_inode;
1139 		if (off >= uio->uio_offset) {
1140 			error = vfs_read_dirent(ap, dp, off);
1141 			if (error)
1142 				break;
1143 		}
1144 		off += dp->d_reclen;
1145 	}
1146 	sx_xunlock(&dmp->dm_lock);
1147 	uio->uio_offset = off;
1148 
1149 	/*
1150 	 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1151 	 * place.
1152 	 */
1153 	if (tmp_ncookies != NULL)
1154 		ap->a_ncookies = tmp_ncookies;
1155 
1156 	return (error);
1157 }
1158 
1159 static int
1160 devfs_readlink(struct vop_readlink_args *ap)
1161 {
1162 	struct devfs_dirent *de;
1163 
1164 	de = ap->a_vp->v_data;
1165 	return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1166 }
1167 
1168 static int
1169 devfs_reclaim(struct vop_reclaim_args *ap)
1170 {
1171 	struct vnode *vp = ap->a_vp;
1172 	struct devfs_dirent *de;
1173 	struct cdev *dev;
1174 
1175 	mtx_lock(&devfs_de_interlock);
1176 	de = vp->v_data;
1177 	if (de != NULL) {
1178 		de->de_vnode = NULL;
1179 		vp->v_data = NULL;
1180 	}
1181 	mtx_unlock(&devfs_de_interlock);
1182 
1183 	vnode_destroy_vobject(vp);
1184 
1185 	VI_LOCK(vp);
1186 	dev_lock();
1187 	dev = vp->v_rdev;
1188 	vp->v_rdev = NULL;
1189 
1190 	if (dev == NULL) {
1191 		dev_unlock();
1192 		VI_UNLOCK(vp);
1193 		return (0);
1194 	}
1195 
1196 	dev->si_usecount -= vp->v_usecount;
1197 	dev_unlock();
1198 	VI_UNLOCK(vp);
1199 	dev_rel(dev);
1200 	return (0);
1201 }
1202 
1203 static int
1204 devfs_remove(struct vop_remove_args *ap)
1205 {
1206 	struct vnode *vp = ap->a_vp;
1207 	struct devfs_dirent *dd;
1208 	struct devfs_dirent *de;
1209 	struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1210 
1211 	sx_xlock(&dmp->dm_lock);
1212 	dd = ap->a_dvp->v_data;
1213 	de = vp->v_data;
1214 	if (de->de_cdp == NULL) {
1215 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1216 		devfs_delete(dmp, de, 1);
1217 	} else {
1218 		de->de_flags |= DE_WHITEOUT;
1219 	}
1220 	sx_xunlock(&dmp->dm_lock);
1221 	return (0);
1222 }
1223 
1224 /*
1225  * Revoke is called on a tty when a terminal session ends.  The vnode
1226  * is orphaned by setting v_op to deadfs so we need to let go of it
1227  * as well so that we create a new one next time around.
1228  *
1229  */
1230 static int
1231 devfs_revoke(struct vop_revoke_args *ap)
1232 {
1233 	struct vnode *vp = ap->a_vp, *vp2;
1234 	struct cdev *dev;
1235 	struct cdev_priv *cdp;
1236 	struct devfs_dirent *de;
1237 	int i;
1238 
1239 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1240 
1241 	dev = vp->v_rdev;
1242 	cdp = cdev2priv(dev);
1243 
1244 	dev_lock();
1245 	cdp->cdp_inuse++;
1246 	dev_unlock();
1247 
1248 	vhold(vp);
1249 	vgone(vp);
1250 	vdrop(vp);
1251 
1252 	VOP_UNLOCK(vp,0);
1253  loop:
1254 	for (;;) {
1255 		mtx_lock(&devfs_de_interlock);
1256 		dev_lock();
1257 		vp2 = NULL;
1258 		for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1259 			de = cdp->cdp_dirents[i];
1260 			if (de == NULL)
1261 				continue;
1262 
1263 			vp2 = de->de_vnode;
1264 			if (vp2 != NULL) {
1265 				dev_unlock();
1266 				VI_LOCK(vp2);
1267 				mtx_unlock(&devfs_de_interlock);
1268 				if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1269 				    curthread))
1270 					goto loop;
1271 				vhold(vp2);
1272 				vgone(vp2);
1273 				vdrop(vp2);
1274 				vput(vp2);
1275 				break;
1276 			}
1277 		}
1278 		if (vp2 != NULL) {
1279 			continue;
1280 		}
1281 		dev_unlock();
1282 		mtx_unlock(&devfs_de_interlock);
1283 		break;
1284 	}
1285 	dev_lock();
1286 	cdp->cdp_inuse--;
1287 	if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1288 		TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1289 		dev_unlock();
1290 		dev_rel(&cdp->cdp_c);
1291 	} else
1292 		dev_unlock();
1293 
1294 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1295 	return (0);
1296 }
1297 
1298 static int
1299 devfs_rioctl(struct vop_ioctl_args *ap)
1300 {
1301 	struct vnode *vp;
1302 	struct devfs_mount *dmp;
1303 	int error;
1304 
1305 	vp = ap->a_vp;
1306 	vn_lock(vp, LK_SHARED | LK_RETRY);
1307 	if (vp->v_iflag & VI_DOOMED) {
1308 		VOP_UNLOCK(vp, 0);
1309 		return (EBADF);
1310 	}
1311 	dmp = VFSTODEVFS(vp->v_mount);
1312 	sx_xlock(&dmp->dm_lock);
1313 	VOP_UNLOCK(vp, 0);
1314 	DEVFS_DMP_HOLD(dmp);
1315 	devfs_populate(dmp);
1316 	if (DEVFS_DMP_DROP(dmp)) {
1317 		sx_xunlock(&dmp->dm_lock);
1318 		devfs_unmount_final(dmp);
1319 		return (ENOENT);
1320 	}
1321 	error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1322 	sx_xunlock(&dmp->dm_lock);
1323 	return (error);
1324 }
1325 
1326 static int
1327 devfs_rread(struct vop_read_args *ap)
1328 {
1329 
1330 	if (ap->a_vp->v_type != VDIR)
1331 		return (EINVAL);
1332 	return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1333 }
1334 
1335 static int
1336 devfs_setattr(struct vop_setattr_args *ap)
1337 {
1338 	struct devfs_dirent *de;
1339 	struct vattr *vap;
1340 	struct vnode *vp;
1341 	struct thread *td;
1342 	int c, error;
1343 	uid_t uid;
1344 	gid_t gid;
1345 
1346 	vap = ap->a_vap;
1347 	vp = ap->a_vp;
1348 	td = curthread;
1349 	if ((vap->va_type != VNON) ||
1350 	    (vap->va_nlink != VNOVAL) ||
1351 	    (vap->va_fsid != VNOVAL) ||
1352 	    (vap->va_fileid != VNOVAL) ||
1353 	    (vap->va_blocksize != VNOVAL) ||
1354 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1355 	    (vap->va_rdev != VNOVAL) ||
1356 	    ((int)vap->va_bytes != VNOVAL) ||
1357 	    (vap->va_gen != VNOVAL)) {
1358 		return (EINVAL);
1359 	}
1360 
1361 	de = vp->v_data;
1362 	if (vp->v_type == VDIR)
1363 		de = de->de_dir;
1364 
1365 	error = c = 0;
1366 	if (vap->va_uid == (uid_t)VNOVAL)
1367 		uid = de->de_uid;
1368 	else
1369 		uid = vap->va_uid;
1370 	if (vap->va_gid == (gid_t)VNOVAL)
1371 		gid = de->de_gid;
1372 	else
1373 		gid = vap->va_gid;
1374 	if (uid != de->de_uid || gid != de->de_gid) {
1375 		if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1376 		    (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1377 			error = priv_check(td, PRIV_VFS_CHOWN);
1378 			if (error)
1379 				return (error);
1380 		}
1381 		de->de_uid = uid;
1382 		de->de_gid = gid;
1383 		c = 1;
1384 	}
1385 
1386 	if (vap->va_mode != (mode_t)VNOVAL) {
1387 		if (ap->a_cred->cr_uid != de->de_uid) {
1388 			error = priv_check(td, PRIV_VFS_ADMIN);
1389 			if (error)
1390 				return (error);
1391 		}
1392 		de->de_mode = vap->va_mode;
1393 		c = 1;
1394 	}
1395 
1396 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1397 		/* See the comment in ufs_vnops::ufs_setattr(). */
1398 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) &&
1399 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1400 		    (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td))))
1401 			return (error);
1402 		if (vap->va_atime.tv_sec != VNOVAL) {
1403 			if (vp->v_type == VCHR)
1404 				vp->v_rdev->si_atime = vap->va_atime;
1405 			else
1406 				de->de_atime = vap->va_atime;
1407 		}
1408 		if (vap->va_mtime.tv_sec != VNOVAL) {
1409 			if (vp->v_type == VCHR)
1410 				vp->v_rdev->si_mtime = vap->va_mtime;
1411 			else
1412 				de->de_mtime = vap->va_mtime;
1413 		}
1414 		c = 1;
1415 	}
1416 
1417 	if (c) {
1418 		if (vp->v_type == VCHR)
1419 			vfs_timestamp(&vp->v_rdev->si_ctime);
1420 		else
1421 			vfs_timestamp(&de->de_mtime);
1422 	}
1423 	return (0);
1424 }
1425 
1426 #ifdef MAC
1427 static int
1428 devfs_setlabel(struct vop_setlabel_args *ap)
1429 {
1430 	struct vnode *vp;
1431 	struct devfs_dirent *de;
1432 
1433 	vp = ap->a_vp;
1434 	de = vp->v_data;
1435 
1436 	mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1437 	mac_devfs_update(vp->v_mount, de, vp);
1438 
1439 	return (0);
1440 }
1441 #endif
1442 
1443 static int
1444 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1445 {
1446 
1447 	return (vnops.fo_stat(fp, sb, cred, td));
1448 }
1449 
1450 static int
1451 devfs_symlink(struct vop_symlink_args *ap)
1452 {
1453 	int i, error;
1454 	struct devfs_dirent *dd;
1455 	struct devfs_dirent *de;
1456 	struct devfs_mount *dmp;
1457 
1458 	error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1459 	if (error)
1460 		return(error);
1461 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1462 	dd = ap->a_dvp->v_data;
1463 	de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1464 	de->de_uid = 0;
1465 	de->de_gid = 0;
1466 	de->de_mode = 0755;
1467 	de->de_inode = alloc_unr(devfs_inos);
1468 	de->de_dirent->d_type = DT_LNK;
1469 	i = strlen(ap->a_target) + 1;
1470 	de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1471 	bcopy(ap->a_target, de->de_symlink, i);
1472 	sx_xlock(&dmp->dm_lock);
1473 #ifdef MAC
1474 	mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1475 #endif
1476 	TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
1477 	return (devfs_allocv(de, ap->a_dvp->v_mount, ap->a_vpp));
1478 }
1479 
1480 static int
1481 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1482 {
1483 
1484 	return (vnops.fo_truncate(fp, length, cred, td));
1485 }
1486 
1487 /* ARGSUSED */
1488 static int
1489 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td)
1490 {
1491 	struct cdev *dev;
1492 	int error, ioflag, resid;
1493 	struct cdevsw *dsw;
1494 	struct file *fpop;
1495 
1496 	fpop = td->td_fpop;
1497 	error = devfs_fp_check(fp, &dev, &dsw);
1498 	if (error)
1499 		return (error);
1500 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1501 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1502 	if (ioflag & O_DIRECT)
1503 		ioflag |= IO_DIRECT;
1504 	if ((flags & FOF_OFFSET) == 0)
1505 		uio->uio_offset = fp->f_offset;
1506 
1507 	resid = uio->uio_resid;
1508 
1509 	error = dsw->d_write(dev, uio, ioflag);
1510 	if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1511 		vfs_timestamp(&dev->si_ctime);
1512 		dev->si_mtime = dev->si_ctime;
1513 	}
1514 	td->td_fpop = fpop;
1515 	dev_relthread(dev);
1516 
1517 	if ((flags & FOF_OFFSET) == 0)
1518 		fp->f_offset = uio->uio_offset;
1519 	fp->f_nextoff = uio->uio_offset;
1520 	return (error);
1521 }
1522 
1523 dev_t
1524 dev2udev(struct cdev *x)
1525 {
1526 	if (x == NULL)
1527 		return (NODEV);
1528 	return (cdev2priv(x)->cdp_inode);
1529 }
1530 
1531 static struct fileops devfs_ops_f = {
1532 	.fo_read =	devfs_read_f,
1533 	.fo_write =	devfs_write_f,
1534 	.fo_truncate =	devfs_truncate_f,
1535 	.fo_ioctl =	devfs_ioctl_f,
1536 	.fo_poll =	devfs_poll_f,
1537 	.fo_kqfilter =	devfs_kqfilter_f,
1538 	.fo_stat =	devfs_stat_f,
1539 	.fo_close =	devfs_close_f,
1540 	.fo_flags =	DFLAG_PASSABLE | DFLAG_SEEKABLE
1541 };
1542 
1543 static struct vop_vector devfs_vnodeops = {
1544 	.vop_default =		&default_vnodeops,
1545 
1546 	.vop_access =		devfs_access,
1547 	.vop_getattr =		devfs_getattr,
1548 	.vop_ioctl =		devfs_rioctl,
1549 	.vop_lookup =		devfs_lookup,
1550 	.vop_mknod =		devfs_mknod,
1551 	.vop_pathconf =		devfs_pathconf,
1552 	.vop_read =		devfs_rread,
1553 	.vop_readdir =		devfs_readdir,
1554 	.vop_readlink =		devfs_readlink,
1555 	.vop_reclaim =		devfs_reclaim,
1556 	.vop_remove =		devfs_remove,
1557 	.vop_revoke =		devfs_revoke,
1558 	.vop_setattr =		devfs_setattr,
1559 #ifdef MAC
1560 	.vop_setlabel =		devfs_setlabel,
1561 #endif
1562 	.vop_symlink =		devfs_symlink,
1563 	.vop_vptocnp =		devfs_vptocnp,
1564 };
1565 
1566 static struct vop_vector devfs_specops = {
1567 	.vop_default =		&default_vnodeops,
1568 
1569 	.vop_access =		devfs_access,
1570 	.vop_bmap =		VOP_PANIC,
1571 	.vop_close =		devfs_close,
1572 	.vop_create =		VOP_PANIC,
1573 	.vop_fsync =		devfs_fsync,
1574 	.vop_getattr =		devfs_getattr,
1575 	.vop_link =		VOP_PANIC,
1576 	.vop_mkdir =		VOP_PANIC,
1577 	.vop_mknod =		VOP_PANIC,
1578 	.vop_open =		devfs_open,
1579 	.vop_pathconf =		devfs_pathconf,
1580 	.vop_print =		devfs_print,
1581 	.vop_read =		VOP_PANIC,
1582 	.vop_readdir =		VOP_PANIC,
1583 	.vop_readlink =		VOP_PANIC,
1584 	.vop_reallocblks =	VOP_PANIC,
1585 	.vop_reclaim =		devfs_reclaim,
1586 	.vop_remove =		devfs_remove,
1587 	.vop_rename =		VOP_PANIC,
1588 	.vop_revoke =		devfs_revoke,
1589 	.vop_rmdir =		VOP_PANIC,
1590 	.vop_setattr =		devfs_setattr,
1591 #ifdef MAC
1592 	.vop_setlabel =		devfs_setlabel,
1593 #endif
1594 	.vop_strategy =		VOP_PANIC,
1595 	.vop_symlink =		VOP_PANIC,
1596 	.vop_vptocnp =		devfs_vptocnp,
1597 	.vop_write =		VOP_PANIC,
1598 };
1599 
1600 /*
1601  * Our calling convention to the device drivers used to be that we passed
1602  * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1603  * flags instead since that's what open(), close() and ioctl() takes and
1604  * we don't really want vnode.h in device drivers.
1605  * We solved the source compatibility by redefining some vnode flags to
1606  * be the same as the fcntl ones and by sending down the bitwise OR of
1607  * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
1608  * pulls the rug out under this.
1609  */
1610 CTASSERT(O_NONBLOCK == IO_NDELAY);
1611 CTASSERT(O_FSYNC == IO_SYNC);
1612