xref: /freebsd/sys/fs/devfs/devfs_vnops.c (revision b3aaa0cc21c63d388230c7ef2a80abd631ff20d5)
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 "opt_mac.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/conf.h>
48 #include <sys/dirent.h>
49 #include <sys/fcntl.h>
50 #include <sys/file.h>
51 #include <sys/filedesc.h>
52 #include <sys/filio.h>
53 #include <sys/kernel.h>
54 #include <sys/lock.h>
55 #include <sys/malloc.h>
56 #include <sys/mount.h>
57 #include <sys/namei.h>
58 #include <sys/priv.h>
59 #include <sys/proc.h>
60 #include <sys/stat.h>
61 #include <sys/sx.h>
62 #include <sys/time.h>
63 #include <sys/ttycom.h>
64 #include <sys/unistd.h>
65 #include <sys/vnode.h>
66 
67 static struct vop_vector devfs_vnodeops;
68 static struct vop_vector devfs_specops;
69 static struct fileops devfs_ops_f;
70 
71 #include <fs/devfs/devfs.h>
72 #include <fs/devfs/devfs_int.h>
73 
74 #include <security/mac/mac_framework.h>
75 
76 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
77 
78 struct mtx	devfs_de_interlock;
79 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
80 struct sx	clone_drain_lock;
81 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
82 struct mtx	cdevpriv_mtx;
83 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
84 
85 static int
86 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp)
87 {
88 
89 	*dswp = devvn_refthread(fp->f_vnode, devp);
90 	if (*devp != fp->f_data) {
91 		if (*dswp != NULL)
92 			dev_relthread(*devp);
93 		return (ENXIO);
94 	}
95 	KASSERT((*devp)->si_refcount > 0,
96 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
97 	if (*dswp == NULL)
98 		return (ENXIO);
99 	curthread->td_fpop = fp;
100 	return (0);
101 }
102 
103 int
104 devfs_get_cdevpriv(void **datap)
105 {
106 	struct file *fp;
107 	struct cdev_privdata *p;
108 	int error;
109 
110 	fp = curthread->td_fpop;
111 	if (fp == NULL)
112 		return (EBADF);
113 	p = fp->f_cdevpriv;
114 	if (p != NULL) {
115 		error = 0;
116 		*datap = p->cdpd_data;
117 	} else
118 		error = ENOENT;
119 	return (error);
120 }
121 
122 int
123 devfs_set_cdevpriv(void *priv, cdevpriv_dtr_t priv_dtr)
124 {
125 	struct file *fp;
126 	struct cdev_priv *cdp;
127 	struct cdev_privdata *p;
128 	int error;
129 
130 	fp = curthread->td_fpop;
131 	if (fp == NULL)
132 		return (ENOENT);
133 	cdp = cdev2priv((struct cdev *)fp->f_data);
134 	p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
135 	p->cdpd_data = priv;
136 	p->cdpd_dtr = priv_dtr;
137 	p->cdpd_fp = fp;
138 	mtx_lock(&cdevpriv_mtx);
139 	if (fp->f_cdevpriv == NULL) {
140 		LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
141 		fp->f_cdevpriv = p;
142 		mtx_unlock(&cdevpriv_mtx);
143 		error = 0;
144 	} else {
145 		mtx_unlock(&cdevpriv_mtx);
146 		free(p, M_CDEVPDATA);
147 		error = EBUSY;
148 	}
149 	return (error);
150 }
151 
152 void
153 devfs_destroy_cdevpriv(struct cdev_privdata *p)
154 {
155 
156 	mtx_assert(&cdevpriv_mtx, MA_OWNED);
157 	p->cdpd_fp->f_cdevpriv = NULL;
158 	LIST_REMOVE(p, cdpd_list);
159 	mtx_unlock(&cdevpriv_mtx);
160 	(p->cdpd_dtr)(p->cdpd_data);
161 	free(p, M_CDEVPDATA);
162 }
163 
164 void
165 devfs_fpdrop(struct file *fp)
166 {
167 	struct cdev_privdata *p;
168 
169 	mtx_lock(&cdevpriv_mtx);
170 	if ((p = fp->f_cdevpriv) == NULL) {
171 		mtx_unlock(&cdevpriv_mtx);
172 		return;
173 	}
174 	devfs_destroy_cdevpriv(p);
175 }
176 
177 void
178 devfs_clear_cdevpriv(void)
179 {
180 	struct file *fp;
181 
182 	fp = curthread->td_fpop;
183 	if (fp == NULL)
184 		return;
185 	devfs_fpdrop(fp);
186 }
187 
188 static int
189 devfs_vptocnp(struct vop_vptocnp_args *ap)
190 {
191 	struct vnode *vp = ap->a_vp;
192 	struct vnode **dvp = ap->a_vpp;
193 	struct devfs_mount *dmp;
194 	char *buf = ap->a_buf;
195 	int *buflen = ap->a_buflen;
196 	struct devfs_dirent *dd, *de;
197 	int i, error;
198 
199 	dmp = VFSTODEVFS(vp->v_mount);
200 	i = *buflen;
201 	dd = vp->v_data;
202 	error = 0;
203 
204 	sx_xlock(&dmp->dm_lock);
205 
206 	if (vp->v_type == VCHR) {
207 		i -= strlen(dd->de_cdp->cdp_c.si_name);
208 		if (i < 0) {
209 			error = ENOMEM;
210 			goto finished;
211 		}
212 		bcopy(dd->de_cdp->cdp_c.si_name, buf + i,
213 		    strlen(dd->de_cdp->cdp_c.si_name));
214 		de = dd->de_dir;
215 	} else if (vp->v_type == VDIR) {
216 		if (dd == dmp->dm_rootdir) {
217 			*dvp = vp;
218 			vhold(*dvp);
219 			goto finished;
220 		}
221 		i -= dd->de_dirent->d_namlen;
222 		if (i < 0) {
223 			error = ENOMEM;
224 			goto finished;
225 		}
226 		bcopy(dd->de_dirent->d_name, buf + i,
227 		    dd->de_dirent->d_namlen);
228 		de = dd;
229 	} else {
230 		error = ENOENT;
231 		goto finished;
232 	}
233 	*buflen = i;
234 	de = TAILQ_FIRST(&de->de_dlist);	/* "." */
235 	de = TAILQ_NEXT(de, de_list);		/* ".." */
236 	de = de->de_dir;
237 	mtx_lock(&devfs_de_interlock);
238 	*dvp = de->de_vnode;
239 	if (*dvp != NULL) {
240 		VI_LOCK(*dvp);
241 		mtx_unlock(&devfs_de_interlock);
242 		vholdl(*dvp);
243 		VI_UNLOCK(*dvp);
244 	} else {
245 		mtx_unlock(&devfs_de_interlock);
246 		error = ENOENT;
247 	}
248 finished:
249 	sx_xunlock(&dmp->dm_lock);
250 	return (error);
251 }
252 
253 /*
254  * Construct the fully qualified path name relative to the mountpoint
255  */
256 static char *
257 devfs_fqpn(char *buf, struct vnode *dvp, struct componentname *cnp)
258 {
259 	int i;
260 	struct devfs_dirent *de, *dd;
261 	struct devfs_mount *dmp;
262 
263 	dmp = VFSTODEVFS(dvp->v_mount);
264 	dd = dvp->v_data;
265 	i = SPECNAMELEN;
266 	buf[i] = '\0';
267 	i -= cnp->cn_namelen;
268 	if (i < 0)
269 		 return (NULL);
270 	bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
271 	de = dd;
272 	while (de != dmp->dm_rootdir) {
273 		i--;
274 		if (i < 0)
275 			 return (NULL);
276 		buf[i] = '/';
277 		i -= de->de_dirent->d_namlen;
278 		if (i < 0)
279 			 return (NULL);
280 		bcopy(de->de_dirent->d_name, buf + i,
281 		    de->de_dirent->d_namlen);
282 		de = TAILQ_FIRST(&de->de_dlist);	/* "." */
283 		de = TAILQ_NEXT(de, de_list);		/* ".." */
284 		de = de->de_dir;
285 	}
286 	return (buf + i);
287 }
288 
289 static int
290 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
291 	struct devfs_dirent *de)
292 {
293 	int not_found;
294 
295 	not_found = 0;
296 	if (de->de_flags & DE_DOOMED)
297 		not_found = 1;
298 	if (DEVFS_DE_DROP(de)) {
299 		KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
300 		devfs_dirent_free(de);
301 	}
302 	if (DEVFS_DMP_DROP(dmp)) {
303 		KASSERT(not_found == 1,
304 			("DEVFS mount struct freed before dirent"));
305 		not_found = 2;
306 		sx_xunlock(&dmp->dm_lock);
307 		devfs_unmount_final(dmp);
308 	}
309 	if (not_found == 1 || (drop_dm_lock && not_found != 2))
310 		sx_unlock(&dmp->dm_lock);
311 	return (not_found);
312 }
313 
314 static void
315 devfs_insmntque_dtr(struct vnode *vp, void *arg)
316 {
317 	struct devfs_dirent *de;
318 
319 	de = (struct devfs_dirent *)arg;
320 	mtx_lock(&devfs_de_interlock);
321 	vp->v_data = NULL;
322 	de->de_vnode = NULL;
323 	mtx_unlock(&devfs_de_interlock);
324 	vgone(vp);
325 	vput(vp);
326 }
327 
328 /*
329  * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
330  * it on return.
331  */
332 int
333 devfs_allocv(struct devfs_dirent *de, struct mount *mp, struct vnode **vpp, struct thread *td)
334 {
335 	int error;
336 	struct vnode *vp;
337 	struct cdev *dev;
338 	struct devfs_mount *dmp;
339 
340 	KASSERT(td == curthread, ("devfs_allocv: td != curthread"));
341 	dmp = VFSTODEVFS(mp);
342 	if (de->de_flags & DE_DOOMED) {
343 		sx_xunlock(&dmp->dm_lock);
344 		return (ENOENT);
345 	}
346 	DEVFS_DE_HOLD(de);
347 	DEVFS_DMP_HOLD(dmp);
348 	mtx_lock(&devfs_de_interlock);
349 	vp = de->de_vnode;
350 	if (vp != NULL) {
351 		VI_LOCK(vp);
352 		mtx_unlock(&devfs_de_interlock);
353 		sx_xunlock(&dmp->dm_lock);
354 		error = vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td);
355 		sx_xlock(&dmp->dm_lock);
356 		if (devfs_allocv_drop_refs(0, dmp, de)) {
357 			if (error == 0)
358 				vput(vp);
359 			return (ENOENT);
360 		}
361 		else if (error) {
362 			sx_xunlock(&dmp->dm_lock);
363 			return (error);
364 		}
365 		sx_xunlock(&dmp->dm_lock);
366 		*vpp = vp;
367 		return (0);
368 	}
369 	mtx_unlock(&devfs_de_interlock);
370 	if (de->de_dirent->d_type == DT_CHR) {
371 		if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
372 			devfs_allocv_drop_refs(1, dmp, de);
373 			return (ENOENT);
374 		}
375 		dev = &de->de_cdp->cdp_c;
376 	} else {
377 		dev = NULL;
378 	}
379 	error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
380 	if (error != 0) {
381 		devfs_allocv_drop_refs(1, dmp, de);
382 		printf("devfs_allocv: failed to allocate new vnode\n");
383 		return (error);
384 	}
385 
386 	if (de->de_dirent->d_type == DT_CHR) {
387 		vp->v_type = VCHR;
388 		VI_LOCK(vp);
389 		dev_lock();
390 		dev_refl(dev);
391 		/* XXX: v_rdev should be protect by vnode lock */
392 		vp->v_rdev = dev;
393 		KASSERT(vp->v_usecount == 1,
394 		    ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount));
395 		dev->si_usecount += vp->v_usecount;
396 		dev_unlock();
397 		VI_UNLOCK(vp);
398 		vp->v_op = &devfs_specops;
399 	} else if (de->de_dirent->d_type == DT_DIR) {
400 		vp->v_type = VDIR;
401 	} else if (de->de_dirent->d_type == DT_LNK) {
402 		vp->v_type = VLNK;
403 	} else {
404 		vp->v_type = VBAD;
405 	}
406 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
407 	mtx_lock(&devfs_de_interlock);
408 	vp->v_data = de;
409 	de->de_vnode = vp;
410 	mtx_unlock(&devfs_de_interlock);
411 	error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
412 	if (error != 0) {
413 		(void) devfs_allocv_drop_refs(1, dmp, de);
414 		return (error);
415 	}
416 	if (devfs_allocv_drop_refs(0, dmp, de)) {
417 		vput(vp);
418 		return (ENOENT);
419 	}
420 #ifdef MAC
421 	mac_devfs_vnode_associate(mp, de, vp);
422 #endif
423 	sx_xunlock(&dmp->dm_lock);
424 	*vpp = vp;
425 	return (0);
426 }
427 
428 static int
429 devfs_access(struct vop_access_args *ap)
430 {
431 	struct vnode *vp = ap->a_vp;
432 	struct devfs_dirent *de;
433 	int error;
434 
435 	de = vp->v_data;
436 	if (vp->v_type == VDIR)
437 		de = de->de_dir;
438 
439 	error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
440 	    ap->a_accmode, ap->a_cred, NULL);
441 	if (!error)
442 		return (error);
443 	if (error != EACCES)
444 		return (error);
445 	/* We do, however, allow access to the controlling terminal */
446 	if (!(ap->a_td->td_proc->p_flag & P_CONTROLT))
447 		return (error);
448 	if (ap->a_td->td_proc->p_session->s_ttyvp == de->de_vnode)
449 		return (0);
450 	return (error);
451 }
452 
453 /* ARGSUSED */
454 static int
455 devfs_advlock(struct vop_advlock_args *ap)
456 {
457 
458 	return (ap->a_flags & F_FLOCK ? EOPNOTSUPP : EINVAL);
459 }
460 
461 /* ARGSUSED */
462 static int
463 devfs_close(struct vop_close_args *ap)
464 {
465 	struct vnode *vp = ap->a_vp, *oldvp;
466 	struct thread *td = ap->a_td;
467 	struct cdev *dev = vp->v_rdev;
468 	struct cdevsw *dsw;
469 	int vp_locked, error;
470 
471 	/*
472 	 * Hack: a tty device that is a controlling terminal
473 	 * has a reference from the session structure.
474 	 * We cannot easily tell that a character device is
475 	 * a controlling terminal, unless it is the closing
476 	 * process' controlling terminal.  In that case,
477 	 * if the reference count is 2 (this last descriptor
478 	 * plus the session), release the reference from the session.
479 	 */
480 	oldvp = NULL;
481 	sx_xlock(&proctree_lock);
482 	if (td && vp == td->td_proc->p_session->s_ttyvp) {
483 		SESS_LOCK(td->td_proc->p_session);
484 		VI_LOCK(vp);
485 		if (count_dev(dev) == 2 && (vp->v_iflag & VI_DOOMED) == 0) {
486 			td->td_proc->p_session->s_ttyvp = NULL;
487 			oldvp = vp;
488 		}
489 		VI_UNLOCK(vp);
490 		SESS_UNLOCK(td->td_proc->p_session);
491 	}
492 	sx_xunlock(&proctree_lock);
493 	if (oldvp != NULL)
494 		vrele(oldvp);
495 	/*
496 	 * We do not want to really close the device if it
497 	 * is still in use unless we are trying to close it
498 	 * forcibly. Since every use (buffer, vnode, swap, cmap)
499 	 * holds a reference to the vnode, and because we mark
500 	 * any other vnodes that alias this device, when the
501 	 * sum of the reference counts on all the aliased
502 	 * vnodes descends to one, we are on last close.
503 	 */
504 	dsw = dev_refthread(dev);
505 	if (dsw == NULL)
506 		return (ENXIO);
507 	VI_LOCK(vp);
508 	if (vp->v_iflag & VI_DOOMED) {
509 		/* Forced close. */
510 	} else if (dsw->d_flags & D_TRACKCLOSE) {
511 		/* Keep device updated on status. */
512 	} else if (count_dev(dev) > 1) {
513 		VI_UNLOCK(vp);
514 		dev_relthread(dev);
515 		return (0);
516 	}
517 	vholdl(vp);
518 	VI_UNLOCK(vp);
519 	vp_locked = VOP_ISLOCKED(vp);
520 	VOP_UNLOCK(vp, 0);
521 	KASSERT(dev->si_refcount > 0,
522 	    ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
523 	error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td);
524 	dev_relthread(dev);
525 	vn_lock(vp, vp_locked | LK_RETRY);
526 	vdrop(vp);
527 	return (error);
528 }
529 
530 static int
531 devfs_close_f(struct file *fp, struct thread *td)
532 {
533 	int error;
534 	struct file *fpop;
535 
536 	fpop = td->td_fpop;
537 	td->td_fpop = fp;
538 	error = vnops.fo_close(fp, td);
539 	td->td_fpop = fpop;
540 	return (error);
541 }
542 
543 static int
544 devfs_fsync(struct vop_fsync_args *ap)
545 {
546 	int error;
547 	struct bufobj *bo;
548 	struct devfs_dirent *de;
549 
550 	if (!vn_isdisk(ap->a_vp, &error)) {
551 		bo = &ap->a_vp->v_bufobj;
552 		de = ap->a_vp->v_data;
553 		if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) {
554 			printf("Device %s went missing before all of the data "
555 			    "could be written to it; expect data loss.\n",
556 			    de->de_dirent->d_name);
557 
558 			error = vop_stdfsync(ap);
559 			if (bo->bo_dirty.bv_cnt != 0 || error != 0)
560 				panic("devfs_fsync: vop_stdfsync failed.");
561 		}
562 
563 		return (0);
564 	}
565 
566 	return (vop_stdfsync(ap));
567 }
568 
569 static int
570 devfs_getattr(struct vop_getattr_args *ap)
571 {
572 	struct vnode *vp = ap->a_vp;
573 	struct vattr *vap = ap->a_vap;
574 	int error = 0;
575 	struct devfs_dirent *de;
576 	struct cdev *dev;
577 
578 	de = vp->v_data;
579 	KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
580 	if (vp->v_type == VDIR) {
581 		de = de->de_dir;
582 		KASSERT(de != NULL,
583 		    ("Null dir dirent in devfs_getattr vp=%p", vp));
584 	}
585 	vap->va_uid = de->de_uid;
586 	vap->va_gid = de->de_gid;
587 	vap->va_mode = de->de_mode;
588 	if (vp->v_type == VLNK)
589 		vap->va_size = strlen(de->de_symlink);
590 	else if (vp->v_type == VDIR)
591 		vap->va_size = vap->va_bytes = DEV_BSIZE;
592 	else
593 		vap->va_size = 0;
594 	if (vp->v_type != VDIR)
595 		vap->va_bytes = 0;
596 	vap->va_blocksize = DEV_BSIZE;
597 	vap->va_type = vp->v_type;
598 
599 #define fix(aa)							\
600 	do {							\
601 		if ((aa).tv_sec <= 3600) {			\
602 			(aa).tv_sec = boottime.tv_sec;		\
603 			(aa).tv_nsec = boottime.tv_usec * 1000; \
604 		}						\
605 	} while (0)
606 
607 	if (vp->v_type != VCHR)  {
608 		fix(de->de_atime);
609 		vap->va_atime = de->de_atime;
610 		fix(de->de_mtime);
611 		vap->va_mtime = de->de_mtime;
612 		fix(de->de_ctime);
613 		vap->va_ctime = de->de_ctime;
614 	} else {
615 		dev = vp->v_rdev;
616 		fix(dev->si_atime);
617 		vap->va_atime = dev->si_atime;
618 		fix(dev->si_mtime);
619 		vap->va_mtime = dev->si_mtime;
620 		fix(dev->si_ctime);
621 		vap->va_ctime = dev->si_ctime;
622 
623 		vap->va_rdev = cdev2priv(dev)->cdp_inode;
624 	}
625 	vap->va_gen = 0;
626 	vap->va_flags = 0;
627 	vap->va_filerev = 0;
628 	vap->va_nlink = de->de_links;
629 	vap->va_fileid = de->de_inode;
630 
631 	return (error);
632 }
633 
634 /* ARGSUSED */
635 static int
636 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
637 {
638 	struct cdev *dev;
639 	struct cdevsw *dsw;
640 	struct vnode *vp;
641 	struct vnode *vpold;
642 	int error, i;
643 	const char *p;
644 	struct fiodgname_arg *fgn;
645 	struct file *fpop;
646 
647 	fpop = td->td_fpop;
648 	error = devfs_fp_check(fp, &dev, &dsw);
649 	if (error)
650 		return (error);
651 
652 	if (com == FIODTYPE) {
653 		*(int *)data = dsw->d_flags & D_TYPEMASK;
654 		td->td_fpop = fpop;
655 		dev_relthread(dev);
656 		return (0);
657 	} else if (com == FIODGNAME) {
658 		fgn = data;
659 		p = devtoname(dev);
660 		i = strlen(p) + 1;
661 		if (i > fgn->len)
662 			error = EINVAL;
663 		else
664 			error = copyout(p, fgn->buf, i);
665 		td->td_fpop = fpop;
666 		dev_relthread(dev);
667 		return (error);
668 	}
669 	error = dsw->d_ioctl(dev, com, data, fp->f_flag, td);
670 	td->td_fpop = NULL;
671 	dev_relthread(dev);
672 	if (error == ENOIOCTL)
673 		error = ENOTTY;
674 	if (error == 0 && com == TIOCSCTTY) {
675 		vp = fp->f_vnode;
676 
677 		/* Do nothing if reassigning same control tty */
678 		sx_slock(&proctree_lock);
679 		if (td->td_proc->p_session->s_ttyvp == vp) {
680 			sx_sunlock(&proctree_lock);
681 			return (0);
682 		}
683 
684 		vpold = td->td_proc->p_session->s_ttyvp;
685 		VREF(vp);
686 		SESS_LOCK(td->td_proc->p_session);
687 		td->td_proc->p_session->s_ttyvp = vp;
688 		SESS_UNLOCK(td->td_proc->p_session);
689 
690 		sx_sunlock(&proctree_lock);
691 
692 		/* Get rid of reference to old control tty */
693 		if (vpold)
694 			vrele(vpold);
695 	}
696 	return (error);
697 }
698 
699 /* ARGSUSED */
700 static int
701 devfs_kqfilter_f(struct file *fp, struct knote *kn)
702 {
703 	struct cdev *dev;
704 	struct cdevsw *dsw;
705 	int error;
706 	struct file *fpop;
707 	struct thread *td;
708 
709 	td = curthread;
710 	fpop = td->td_fpop;
711 	error = devfs_fp_check(fp, &dev, &dsw);
712 	if (error)
713 		return (error);
714 	error = dsw->d_kqfilter(dev, kn);
715 	td->td_fpop = fpop;
716 	dev_relthread(dev);
717 	return (error);
718 }
719 
720 static int
721 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
722 {
723 	struct componentname *cnp;
724 	struct vnode *dvp, **vpp;
725 	struct thread *td;
726 	struct devfs_dirent *de, *dd;
727 	struct devfs_dirent **dde;
728 	struct devfs_mount *dmp;
729 	struct cdev *cdev;
730 	int error, flags, nameiop;
731 	char specname[SPECNAMELEN + 1], *pname;
732 
733 	cnp = ap->a_cnp;
734 	vpp = ap->a_vpp;
735 	dvp = ap->a_dvp;
736 	pname = cnp->cn_nameptr;
737 	td = cnp->cn_thread;
738 	flags = cnp->cn_flags;
739 	nameiop = cnp->cn_nameiop;
740 	dmp = VFSTODEVFS(dvp->v_mount);
741 	dd = dvp->v_data;
742 	*vpp = NULLVP;
743 
744 	if ((flags & ISLASTCN) && nameiop == RENAME)
745 		return (EOPNOTSUPP);
746 
747 	if (dvp->v_type != VDIR)
748 		return (ENOTDIR);
749 
750 	if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
751 		return (EIO);
752 
753 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
754 	if (error)
755 		return (error);
756 
757 	if (cnp->cn_namelen == 1 && *pname == '.') {
758 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
759 			return (EINVAL);
760 		*vpp = dvp;
761 		VREF(dvp);
762 		return (0);
763 	}
764 
765 	if (flags & ISDOTDOT) {
766 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
767 			return (EINVAL);
768 		VOP_UNLOCK(dvp, 0);
769 		de = TAILQ_FIRST(&dd->de_dlist);	/* "." */
770 		de = TAILQ_NEXT(de, de_list);		/* ".." */
771 		de = de->de_dir;
772 		error = devfs_allocv(de, dvp->v_mount, vpp, td);
773 		*dm_unlock = 0;
774 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
775 		return (error);
776 	}
777 
778 	DEVFS_DMP_HOLD(dmp);
779 	devfs_populate(dmp);
780 	if (DEVFS_DMP_DROP(dmp)) {
781 		*dm_unlock = 0;
782 		sx_xunlock(&dmp->dm_lock);
783 		devfs_unmount_final(dmp);
784 		return (ENOENT);
785 	}
786 	dd = dvp->v_data;
787 	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen);
788 	while (de == NULL) {	/* While(...) so we can use break */
789 
790 		if (nameiop == DELETE)
791 			return (ENOENT);
792 
793 		/*
794 		 * OK, we didn't have an entry for the name we were asked for
795 		 * so we try to see if anybody can create it on demand.
796 		 */
797 		pname = devfs_fqpn(specname, dvp, cnp);
798 		if (pname == NULL)
799 			break;
800 
801 		cdev = NULL;
802 		DEVFS_DMP_HOLD(dmp);
803 		sx_xunlock(&dmp->dm_lock);
804 		sx_slock(&clone_drain_lock);
805 		EVENTHANDLER_INVOKE(dev_clone,
806 		    td->td_ucred, pname, strlen(pname), &cdev);
807 		sx_sunlock(&clone_drain_lock);
808 		sx_xlock(&dmp->dm_lock);
809 		if (DEVFS_DMP_DROP(dmp)) {
810 			*dm_unlock = 0;
811 			sx_xunlock(&dmp->dm_lock);
812 			devfs_unmount_final(dmp);
813 			return (ENOENT);
814 		}
815 		if (cdev == NULL)
816 			break;
817 
818 		DEVFS_DMP_HOLD(dmp);
819 		devfs_populate(dmp);
820 		if (DEVFS_DMP_DROP(dmp)) {
821 			*dm_unlock = 0;
822 			sx_xunlock(&dmp->dm_lock);
823 			devfs_unmount_final(dmp);
824 			return (ENOENT);
825 		}
826 
827 		dev_lock();
828 		dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
829 		if (dde != NULL && *dde != NULL)
830 			de = *dde;
831 		dev_unlock();
832 		dev_rel(cdev);
833 		break;
834 	}
835 
836 	if (de == NULL || de->de_flags & DE_WHITEOUT) {
837 		if ((nameiop == CREATE || nameiop == RENAME) &&
838 		    (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
839 			cnp->cn_flags |= SAVENAME;
840 			return (EJUSTRETURN);
841 		}
842 		return (ENOENT);
843 	}
844 
845 	if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
846 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
847 		if (error)
848 			return (error);
849 		if (*vpp == dvp) {
850 			VREF(dvp);
851 			*vpp = dvp;
852 			return (0);
853 		}
854 	}
855 	error = devfs_allocv(de, dvp->v_mount, vpp, td);
856 	*dm_unlock = 0;
857 	return (error);
858 }
859 
860 static int
861 devfs_lookup(struct vop_lookup_args *ap)
862 {
863 	int j;
864 	struct devfs_mount *dmp;
865 	int dm_unlock;
866 
867 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
868 	dm_unlock = 1;
869 	sx_xlock(&dmp->dm_lock);
870 	j = devfs_lookupx(ap, &dm_unlock);
871 	if (dm_unlock == 1)
872 		sx_xunlock(&dmp->dm_lock);
873 	return (j);
874 }
875 
876 static int
877 devfs_mknod(struct vop_mknod_args *ap)
878 {
879 	struct componentname *cnp;
880 	struct vnode *dvp, **vpp;
881 	struct thread *td;
882 	struct devfs_dirent *dd, *de;
883 	struct devfs_mount *dmp;
884 	int error;
885 
886 	/*
887 	 * The only type of node we should be creating here is a
888 	 * character device, for anything else return EOPNOTSUPP.
889 	 */
890 	if (ap->a_vap->va_type != VCHR)
891 		return (EOPNOTSUPP);
892 	dvp = ap->a_dvp;
893 	dmp = VFSTODEVFS(dvp->v_mount);
894 
895 	cnp = ap->a_cnp;
896 	vpp = ap->a_vpp;
897 	td = cnp->cn_thread;
898 	dd = dvp->v_data;
899 
900 	error = ENOENT;
901 	sx_xlock(&dmp->dm_lock);
902 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
903 		if (cnp->cn_namelen != de->de_dirent->d_namlen)
904 			continue;
905 		if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
906 		    de->de_dirent->d_namlen) != 0)
907 			continue;
908 		if (de->de_flags & DE_WHITEOUT)
909 			break;
910 		goto notfound;
911 	}
912 	if (de == NULL)
913 		goto notfound;
914 	de->de_flags &= ~DE_WHITEOUT;
915 	error = devfs_allocv(de, dvp->v_mount, vpp, td);
916 	return (error);
917 notfound:
918 	sx_xunlock(&dmp->dm_lock);
919 	return (error);
920 }
921 
922 /* ARGSUSED */
923 static int
924 devfs_open(struct vop_open_args *ap)
925 {
926 	struct thread *td = ap->a_td;
927 	struct vnode *vp = ap->a_vp;
928 	struct cdev *dev = vp->v_rdev;
929 	struct file *fp = ap->a_fp;
930 	int error;
931 	struct cdevsw *dsw;
932 	struct file *fpop;
933 
934 	if (vp->v_type == VBLK)
935 		return (ENXIO);
936 
937 	if (dev == NULL)
938 		return (ENXIO);
939 
940 	/* Make this field valid before any I/O in d_open. */
941 	if (dev->si_iosize_max == 0)
942 		dev->si_iosize_max = DFLTPHYS;
943 
944 	dsw = dev_refthread(dev);
945 	if (dsw == NULL)
946 		return (ENXIO);
947 
948 	/* XXX: Special casing of ttys for deadfs.  Probably redundant. */
949 	if (dsw->d_flags & D_TTY)
950 		vp->v_vflag |= VV_ISTTY;
951 
952 	VOP_UNLOCK(vp, 0);
953 
954 	fpop = td->td_fpop;
955 	td->td_fpop = fp;
956 	if (fp != NULL)
957 		fp->f_data = dev;
958 	if (dsw->d_fdopen != NULL)
959 		error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
960 	else
961 		error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
962 	td->td_fpop = fpop;
963 
964 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
965 
966 	dev_relthread(dev);
967 
968 	if (error)
969 		return (error);
970 
971 #if 0	/* /dev/console */
972 	KASSERT(fp != NULL,
973 	     ("Could not vnode bypass device on NULL fp"));
974 #else
975 	if(fp == NULL)
976 		return (error);
977 #endif
978 	if (fp->f_ops == &badfileops)
979 		finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
980 	return (error);
981 }
982 
983 static int
984 devfs_pathconf(struct vop_pathconf_args *ap)
985 {
986 
987 	switch (ap->a_name) {
988 	case _PC_MAC_PRESENT:
989 #ifdef MAC
990 		/*
991 		 * If MAC is enabled, devfs automatically supports
992 		 * trivial non-persistant label storage.
993 		 */
994 		*ap->a_retval = 1;
995 #else
996 		*ap->a_retval = 0;
997 #endif
998 		return (0);
999 	default:
1000 		return (vop_stdpathconf(ap));
1001 	}
1002 	/* NOTREACHED */
1003 }
1004 
1005 /* ARGSUSED */
1006 static int
1007 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1008 {
1009 	struct cdev *dev;
1010 	struct cdevsw *dsw;
1011 	int error;
1012 	struct file *fpop;
1013 
1014 	fpop = td->td_fpop;
1015 	error = devfs_fp_check(fp, &dev, &dsw);
1016 	if (error)
1017 		return (error);
1018 	error = dsw->d_poll(dev, events, td);
1019 	td->td_fpop = fpop;
1020 	dev_relthread(dev);
1021 	return(error);
1022 }
1023 
1024 /*
1025  * Print out the contents of a special device vnode.
1026  */
1027 static int
1028 devfs_print(struct vop_print_args *ap)
1029 {
1030 
1031 	printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1032 	return (0);
1033 }
1034 
1035 /* ARGSUSED */
1036 static int
1037 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td)
1038 {
1039 	struct cdev *dev;
1040 	int ioflag, error, resid;
1041 	struct cdevsw *dsw;
1042 	struct file *fpop;
1043 
1044 	fpop = td->td_fpop;
1045 	error = devfs_fp_check(fp, &dev, &dsw);
1046 	if (error)
1047 		return (error);
1048 	resid = uio->uio_resid;
1049 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1050 	if (ioflag & O_DIRECT)
1051 		ioflag |= IO_DIRECT;
1052 
1053 	if ((flags & FOF_OFFSET) == 0)
1054 		uio->uio_offset = fp->f_offset;
1055 
1056 	error = dsw->d_read(dev, uio, ioflag);
1057 	if (uio->uio_resid != resid || (error == 0 && resid != 0))
1058 		vfs_timestamp(&dev->si_atime);
1059 	td->td_fpop = fpop;
1060 	dev_relthread(dev);
1061 
1062 	if ((flags & FOF_OFFSET) == 0)
1063 		fp->f_offset = uio->uio_offset;
1064 	fp->f_nextoff = uio->uio_offset;
1065 	return (error);
1066 }
1067 
1068 static int
1069 devfs_readdir(struct vop_readdir_args *ap)
1070 {
1071 	int error;
1072 	struct uio *uio;
1073 	struct dirent *dp;
1074 	struct devfs_dirent *dd;
1075 	struct devfs_dirent *de;
1076 	struct devfs_mount *dmp;
1077 	off_t off;
1078 	int *tmp_ncookies = NULL;
1079 
1080 	if (ap->a_vp->v_type != VDIR)
1081 		return (ENOTDIR);
1082 
1083 	uio = ap->a_uio;
1084 	if (uio->uio_offset < 0)
1085 		return (EINVAL);
1086 
1087 	/*
1088 	 * XXX: This is a temporary hack to get around this filesystem not
1089 	 * supporting cookies. We store the location of the ncookies pointer
1090 	 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1091 	 * and set the number of cookies to 0. We then set the pointer to
1092 	 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1093 	 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1094 	 * pointer to its original location before returning to the caller.
1095 	 */
1096 	if (ap->a_ncookies != NULL) {
1097 		tmp_ncookies = ap->a_ncookies;
1098 		*ap->a_ncookies = 0;
1099 		ap->a_ncookies = NULL;
1100 	}
1101 
1102 	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1103 	sx_xlock(&dmp->dm_lock);
1104 	DEVFS_DMP_HOLD(dmp);
1105 	devfs_populate(dmp);
1106 	if (DEVFS_DMP_DROP(dmp)) {
1107 		sx_xunlock(&dmp->dm_lock);
1108 		devfs_unmount_final(dmp);
1109 		if (tmp_ncookies != NULL)
1110 			ap->a_ncookies = tmp_ncookies;
1111 		return (EIO);
1112 	}
1113 	error = 0;
1114 	de = ap->a_vp->v_data;
1115 	off = 0;
1116 	TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1117 		KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1118 		if (dd->de_flags & DE_WHITEOUT)
1119 			continue;
1120 		if (dd->de_dirent->d_type == DT_DIR)
1121 			de = dd->de_dir;
1122 		else
1123 			de = dd;
1124 		dp = dd->de_dirent;
1125 		if (dp->d_reclen > uio->uio_resid)
1126 			break;
1127 		dp->d_fileno = de->de_inode;
1128 		if (off >= uio->uio_offset) {
1129 			error = vfs_read_dirent(ap, dp, off);
1130 			if (error)
1131 				break;
1132 		}
1133 		off += dp->d_reclen;
1134 	}
1135 	sx_xunlock(&dmp->dm_lock);
1136 	uio->uio_offset = off;
1137 
1138 	/*
1139 	 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1140 	 * place.
1141 	 */
1142 	if (tmp_ncookies != NULL)
1143 		ap->a_ncookies = tmp_ncookies;
1144 
1145 	return (error);
1146 }
1147 
1148 static int
1149 devfs_readlink(struct vop_readlink_args *ap)
1150 {
1151 	struct devfs_dirent *de;
1152 
1153 	de = ap->a_vp->v_data;
1154 	return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1155 }
1156 
1157 static int
1158 devfs_reclaim(struct vop_reclaim_args *ap)
1159 {
1160 	struct vnode *vp = ap->a_vp;
1161 	struct devfs_dirent *de;
1162 	struct cdev *dev;
1163 
1164 	mtx_lock(&devfs_de_interlock);
1165 	de = vp->v_data;
1166 	if (de != NULL) {
1167 		de->de_vnode = NULL;
1168 		vp->v_data = NULL;
1169 	}
1170 	mtx_unlock(&devfs_de_interlock);
1171 
1172 	vnode_destroy_vobject(vp);
1173 
1174 	VI_LOCK(vp);
1175 	dev_lock();
1176 	dev = vp->v_rdev;
1177 	vp->v_rdev = NULL;
1178 
1179 	if (dev == NULL) {
1180 		dev_unlock();
1181 		VI_UNLOCK(vp);
1182 		return (0);
1183 	}
1184 
1185 	dev->si_usecount -= vp->v_usecount;
1186 	dev_unlock();
1187 	VI_UNLOCK(vp);
1188 	dev_rel(dev);
1189 	return (0);
1190 }
1191 
1192 static int
1193 devfs_remove(struct vop_remove_args *ap)
1194 {
1195 	struct vnode *vp = ap->a_vp;
1196 	struct devfs_dirent *dd;
1197 	struct devfs_dirent *de;
1198 	struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1199 
1200 	sx_xlock(&dmp->dm_lock);
1201 	dd = ap->a_dvp->v_data;
1202 	de = vp->v_data;
1203 	if (de->de_cdp == NULL) {
1204 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1205 		devfs_delete(dmp, de, 1);
1206 	} else {
1207 		de->de_flags |= DE_WHITEOUT;
1208 	}
1209 	sx_xunlock(&dmp->dm_lock);
1210 	return (0);
1211 }
1212 
1213 /*
1214  * Revoke is called on a tty when a terminal session ends.  The vnode
1215  * is orphaned by setting v_op to deadfs so we need to let go of it
1216  * as well so that we create a new one next time around.
1217  *
1218  */
1219 static int
1220 devfs_revoke(struct vop_revoke_args *ap)
1221 {
1222 	struct vnode *vp = ap->a_vp, *vp2;
1223 	struct cdev *dev;
1224 	struct cdev_priv *cdp;
1225 	struct devfs_dirent *de;
1226 	int i;
1227 
1228 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1229 
1230 	dev = vp->v_rdev;
1231 	cdp = cdev2priv(dev);
1232 
1233 	dev_lock();
1234 	cdp->cdp_inuse++;
1235 	dev_unlock();
1236 
1237 	vhold(vp);
1238 	vgone(vp);
1239 	vdrop(vp);
1240 
1241 	VOP_UNLOCK(vp,0);
1242  loop:
1243 	for (;;) {
1244 		mtx_lock(&devfs_de_interlock);
1245 		dev_lock();
1246 		vp2 = NULL;
1247 		for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1248 			de = cdp->cdp_dirents[i];
1249 			if (de == NULL)
1250 				continue;
1251 
1252 			vp2 = de->de_vnode;
1253 			if (vp2 != NULL) {
1254 				dev_unlock();
1255 				VI_LOCK(vp2);
1256 				mtx_unlock(&devfs_de_interlock);
1257 				if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1258 				    curthread))
1259 					goto loop;
1260 				vhold(vp2);
1261 				vgone(vp2);
1262 				vdrop(vp2);
1263 				vput(vp2);
1264 				break;
1265 			}
1266 		}
1267 		if (vp2 != NULL) {
1268 			continue;
1269 		}
1270 		dev_unlock();
1271 		mtx_unlock(&devfs_de_interlock);
1272 		break;
1273 	}
1274 	dev_lock();
1275 	cdp->cdp_inuse--;
1276 	if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1277 		TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1278 		dev_unlock();
1279 		dev_rel(&cdp->cdp_c);
1280 	} else
1281 		dev_unlock();
1282 
1283 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1284 	return (0);
1285 }
1286 
1287 static int
1288 devfs_rioctl(struct vop_ioctl_args *ap)
1289 {
1290 	int error;
1291 	struct devfs_mount *dmp;
1292 
1293 	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1294 	sx_xlock(&dmp->dm_lock);
1295 	DEVFS_DMP_HOLD(dmp);
1296 	devfs_populate(dmp);
1297 	if (DEVFS_DMP_DROP(dmp)) {
1298 		sx_xunlock(&dmp->dm_lock);
1299 		devfs_unmount_final(dmp);
1300 		return (ENOENT);
1301 	}
1302 	error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1303 	sx_xunlock(&dmp->dm_lock);
1304 	return (error);
1305 }
1306 
1307 static int
1308 devfs_rread(struct vop_read_args *ap)
1309 {
1310 
1311 	if (ap->a_vp->v_type != VDIR)
1312 		return (EINVAL);
1313 	return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1314 }
1315 
1316 static int
1317 devfs_setattr(struct vop_setattr_args *ap)
1318 {
1319 	struct devfs_dirent *de;
1320 	struct vattr *vap;
1321 	struct vnode *vp;
1322 	struct thread *td;
1323 	int c, error;
1324 	uid_t uid;
1325 	gid_t gid;
1326 
1327 	vap = ap->a_vap;
1328 	vp = ap->a_vp;
1329 	td = curthread;
1330 	if ((vap->va_type != VNON) ||
1331 	    (vap->va_nlink != VNOVAL) ||
1332 	    (vap->va_fsid != VNOVAL) ||
1333 	    (vap->va_fileid != VNOVAL) ||
1334 	    (vap->va_blocksize != VNOVAL) ||
1335 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1336 	    (vap->va_rdev != VNOVAL) ||
1337 	    ((int)vap->va_bytes != VNOVAL) ||
1338 	    (vap->va_gen != VNOVAL)) {
1339 		return (EINVAL);
1340 	}
1341 
1342 	de = vp->v_data;
1343 	if (vp->v_type == VDIR)
1344 		de = de->de_dir;
1345 
1346 	error = c = 0;
1347 	if (vap->va_uid == (uid_t)VNOVAL)
1348 		uid = de->de_uid;
1349 	else
1350 		uid = vap->va_uid;
1351 	if (vap->va_gid == (gid_t)VNOVAL)
1352 		gid = de->de_gid;
1353 	else
1354 		gid = vap->va_gid;
1355 	if (uid != de->de_uid || gid != de->de_gid) {
1356 		if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1357 		    (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1358 			error = priv_check(td, PRIV_VFS_CHOWN);
1359 			if (error)
1360 				return (error);
1361 		}
1362 		de->de_uid = uid;
1363 		de->de_gid = gid;
1364 		c = 1;
1365 	}
1366 
1367 	if (vap->va_mode != (mode_t)VNOVAL) {
1368 		if (ap->a_cred->cr_uid != de->de_uid) {
1369 			error = priv_check(td, PRIV_VFS_ADMIN);
1370 			if (error)
1371 				return (error);
1372 		}
1373 		de->de_mode = vap->va_mode;
1374 		c = 1;
1375 	}
1376 
1377 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1378 		/* See the comment in ufs_vnops::ufs_setattr(). */
1379 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) &&
1380 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1381 		    (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td))))
1382 			return (error);
1383 		if (vap->va_atime.tv_sec != VNOVAL) {
1384 			if (vp->v_type == VCHR)
1385 				vp->v_rdev->si_atime = vap->va_atime;
1386 			else
1387 				de->de_atime = vap->va_atime;
1388 		}
1389 		if (vap->va_mtime.tv_sec != VNOVAL) {
1390 			if (vp->v_type == VCHR)
1391 				vp->v_rdev->si_mtime = vap->va_mtime;
1392 			else
1393 				de->de_mtime = vap->va_mtime;
1394 		}
1395 		c = 1;
1396 	}
1397 
1398 	if (c) {
1399 		if (vp->v_type == VCHR)
1400 			vfs_timestamp(&vp->v_rdev->si_ctime);
1401 		else
1402 			vfs_timestamp(&de->de_mtime);
1403 	}
1404 	return (0);
1405 }
1406 
1407 #ifdef MAC
1408 static int
1409 devfs_setlabel(struct vop_setlabel_args *ap)
1410 {
1411 	struct vnode *vp;
1412 	struct devfs_dirent *de;
1413 
1414 	vp = ap->a_vp;
1415 	de = vp->v_data;
1416 
1417 	mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1418 	mac_devfs_update(vp->v_mount, de, vp);
1419 
1420 	return (0);
1421 }
1422 #endif
1423 
1424 static int
1425 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1426 {
1427 
1428 	return (vnops.fo_stat(fp, sb, cred, td));
1429 }
1430 
1431 static int
1432 devfs_symlink(struct vop_symlink_args *ap)
1433 {
1434 	int i, error;
1435 	struct devfs_dirent *dd;
1436 	struct devfs_dirent *de;
1437 	struct devfs_mount *dmp;
1438 	struct thread *td;
1439 
1440 	td = ap->a_cnp->cn_thread;
1441 	KASSERT(td == curthread, ("devfs_symlink: td != curthread"));
1442 
1443 	error = priv_check(td, PRIV_DEVFS_SYMLINK);
1444 	if (error)
1445 		return(error);
1446 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1447 	dd = ap->a_dvp->v_data;
1448 	de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1449 	de->de_uid = 0;
1450 	de->de_gid = 0;
1451 	de->de_mode = 0755;
1452 	de->de_inode = alloc_unr(devfs_inos);
1453 	de->de_dirent->d_type = DT_LNK;
1454 	i = strlen(ap->a_target) + 1;
1455 	de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1456 	bcopy(ap->a_target, de->de_symlink, i);
1457 	sx_xlock(&dmp->dm_lock);
1458 #ifdef MAC
1459 	mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1460 #endif
1461 	TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
1462 	return (devfs_allocv(de, ap->a_dvp->v_mount, ap->a_vpp, td));
1463 }
1464 
1465 static int
1466 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1467 {
1468 
1469 	return (vnops.fo_truncate(fp, length, cred, td));
1470 }
1471 
1472 /* ARGSUSED */
1473 static int
1474 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td)
1475 {
1476 	struct cdev *dev;
1477 	int error, ioflag, resid;
1478 	struct cdevsw *dsw;
1479 	struct file *fpop;
1480 
1481 	fpop = td->td_fpop;
1482 	error = devfs_fp_check(fp, &dev, &dsw);
1483 	if (error)
1484 		return (error);
1485 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1486 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1487 	if (ioflag & O_DIRECT)
1488 		ioflag |= IO_DIRECT;
1489 	if ((flags & FOF_OFFSET) == 0)
1490 		uio->uio_offset = fp->f_offset;
1491 
1492 	resid = uio->uio_resid;
1493 
1494 	error = dsw->d_write(dev, uio, ioflag);
1495 	if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1496 		vfs_timestamp(&dev->si_ctime);
1497 		dev->si_mtime = dev->si_ctime;
1498 	}
1499 	td->td_fpop = fpop;
1500 	dev_relthread(dev);
1501 
1502 	if ((flags & FOF_OFFSET) == 0)
1503 		fp->f_offset = uio->uio_offset;
1504 	fp->f_nextoff = uio->uio_offset;
1505 	return (error);
1506 }
1507 
1508 dev_t
1509 dev2udev(struct cdev *x)
1510 {
1511 	if (x == NULL)
1512 		return (NODEV);
1513 	return (cdev2priv(x)->cdp_inode);
1514 }
1515 
1516 static struct fileops devfs_ops_f = {
1517 	.fo_read =	devfs_read_f,
1518 	.fo_write =	devfs_write_f,
1519 	.fo_truncate =	devfs_truncate_f,
1520 	.fo_ioctl =	devfs_ioctl_f,
1521 	.fo_poll =	devfs_poll_f,
1522 	.fo_kqfilter =	devfs_kqfilter_f,
1523 	.fo_stat =	devfs_stat_f,
1524 	.fo_close =	devfs_close_f,
1525 	.fo_flags =	DFLAG_PASSABLE | DFLAG_SEEKABLE
1526 };
1527 
1528 static struct vop_vector devfs_vnodeops = {
1529 	.vop_default =		&default_vnodeops,
1530 
1531 	.vop_access =		devfs_access,
1532 	.vop_getattr =		devfs_getattr,
1533 	.vop_ioctl =		devfs_rioctl,
1534 	.vop_lookup =		devfs_lookup,
1535 	.vop_mknod =		devfs_mknod,
1536 	.vop_pathconf =		devfs_pathconf,
1537 	.vop_read =		devfs_rread,
1538 	.vop_readdir =		devfs_readdir,
1539 	.vop_readlink =		devfs_readlink,
1540 	.vop_reclaim =		devfs_reclaim,
1541 	.vop_remove =		devfs_remove,
1542 	.vop_revoke =		devfs_revoke,
1543 	.vop_setattr =		devfs_setattr,
1544 #ifdef MAC
1545 	.vop_setlabel =		devfs_setlabel,
1546 #endif
1547 	.vop_symlink =		devfs_symlink,
1548 	.vop_vptocnp =		devfs_vptocnp,
1549 };
1550 
1551 static struct vop_vector devfs_specops = {
1552 	.vop_default =		&default_vnodeops,
1553 
1554 	.vop_access =		devfs_access,
1555 	.vop_advlock =		devfs_advlock,
1556 	.vop_bmap =		VOP_PANIC,
1557 	.vop_close =		devfs_close,
1558 	.vop_create =		VOP_PANIC,
1559 	.vop_fsync =		devfs_fsync,
1560 	.vop_getattr =		devfs_getattr,
1561 	.vop_lease =		VOP_NULL,
1562 	.vop_link =		VOP_PANIC,
1563 	.vop_mkdir =		VOP_PANIC,
1564 	.vop_mknod =		VOP_PANIC,
1565 	.vop_open =		devfs_open,
1566 	.vop_pathconf =		devfs_pathconf,
1567 	.vop_print =		devfs_print,
1568 	.vop_read =		VOP_PANIC,
1569 	.vop_readdir =		VOP_PANIC,
1570 	.vop_readlink =		VOP_PANIC,
1571 	.vop_reallocblks =	VOP_PANIC,
1572 	.vop_reclaim =		devfs_reclaim,
1573 	.vop_remove =		devfs_remove,
1574 	.vop_rename =		VOP_PANIC,
1575 	.vop_revoke =		devfs_revoke,
1576 	.vop_rmdir =		VOP_PANIC,
1577 	.vop_setattr =		devfs_setattr,
1578 #ifdef MAC
1579 	.vop_setlabel =		devfs_setlabel,
1580 #endif
1581 	.vop_strategy =		VOP_PANIC,
1582 	.vop_symlink =		VOP_PANIC,
1583 	.vop_vptocnp =		devfs_vptocnp,
1584 	.vop_write =		VOP_PANIC,
1585 };
1586 
1587 /*
1588  * Our calling convention to the device drivers used to be that we passed
1589  * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1590  * flags instead since that's what open(), close() and ioctl() takes and
1591  * we don't really want vnode.h in device drivers.
1592  * We solved the source compatibility by redefining some vnode flags to
1593  * be the same as the fcntl ones and by sending down the bitwise OR of
1594  * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
1595  * pulls the rug out under this.
1596  */
1597 CTASSERT(O_NONBLOCK == IO_NDELAY);
1598 CTASSERT(O_FSYNC == IO_SYNC);
1599