xref: /freebsd/sys/fs/devfs/devfs_vnops.c (revision 2e1417489338b971e5fd599ff48b5f65df9e8d3b)
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 	/*
606 	 * NB: td may be NULL if this descriptor is closed due to
607 	 * garbage collection from a closed UNIX domain socket.
608 	 */
609 	fpop = curthread->td_fpop;
610 	curthread->td_fpop = fp;
611 	error = vnops.fo_close(fp, td);
612 	curthread->td_fpop = fpop;
613 
614 	/*
615 	 * The f_cdevpriv cannot be assigned non-NULL value while we
616 	 * are destroying the file.
617 	 */
618 	if (fp->f_cdevpriv != NULL)
619 		devfs_fpdrop(fp);
620 	return (error);
621 }
622 
623 static int
624 devfs_fsync(struct vop_fsync_args *ap)
625 {
626 	int error;
627 	struct bufobj *bo;
628 	struct devfs_dirent *de;
629 
630 	if (!vn_isdisk(ap->a_vp, &error)) {
631 		bo = &ap->a_vp->v_bufobj;
632 		de = ap->a_vp->v_data;
633 		if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) {
634 			printf("Device %s went missing before all of the data "
635 			    "could be written to it; expect data loss.\n",
636 			    de->de_dirent->d_name);
637 
638 			error = vop_stdfsync(ap);
639 			if (bo->bo_dirty.bv_cnt != 0 || error != 0)
640 				panic("devfs_fsync: vop_stdfsync failed.");
641 		}
642 
643 		return (0);
644 	}
645 
646 	return (vop_stdfsync(ap));
647 }
648 
649 static int
650 devfs_getattr(struct vop_getattr_args *ap)
651 {
652 	struct vnode *vp = ap->a_vp;
653 	struct vattr *vap = ap->a_vap;
654 	int error;
655 	struct devfs_dirent *de;
656 	struct devfs_mount *dmp;
657 	struct cdev *dev;
658 
659 	error = devfs_populate_vp(vp);
660 	if (error != 0)
661 		return (error);
662 
663 	dmp = VFSTODEVFS(vp->v_mount);
664 	sx_xunlock(&dmp->dm_lock);
665 
666 	de = vp->v_data;
667 	KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
668 	if (vp->v_type == VDIR) {
669 		de = de->de_dir;
670 		KASSERT(de != NULL,
671 		    ("Null dir dirent in devfs_getattr vp=%p", vp));
672 	}
673 	vap->va_uid = de->de_uid;
674 	vap->va_gid = de->de_gid;
675 	vap->va_mode = de->de_mode;
676 	if (vp->v_type == VLNK)
677 		vap->va_size = strlen(de->de_symlink);
678 	else if (vp->v_type == VDIR)
679 		vap->va_size = vap->va_bytes = DEV_BSIZE;
680 	else
681 		vap->va_size = 0;
682 	if (vp->v_type != VDIR)
683 		vap->va_bytes = 0;
684 	vap->va_blocksize = DEV_BSIZE;
685 	vap->va_type = vp->v_type;
686 
687 #define fix(aa)							\
688 	do {							\
689 		if ((aa).tv_sec <= 3600) {			\
690 			(aa).tv_sec = boottime.tv_sec;		\
691 			(aa).tv_nsec = boottime.tv_usec * 1000; \
692 		}						\
693 	} while (0)
694 
695 	if (vp->v_type != VCHR)  {
696 		fix(de->de_atime);
697 		vap->va_atime = de->de_atime;
698 		fix(de->de_mtime);
699 		vap->va_mtime = de->de_mtime;
700 		fix(de->de_ctime);
701 		vap->va_ctime = de->de_ctime;
702 	} else {
703 		dev = vp->v_rdev;
704 		fix(dev->si_atime);
705 		vap->va_atime = dev->si_atime;
706 		fix(dev->si_mtime);
707 		vap->va_mtime = dev->si_mtime;
708 		fix(dev->si_ctime);
709 		vap->va_ctime = dev->si_ctime;
710 
711 		vap->va_rdev = cdev2priv(dev)->cdp_inode;
712 	}
713 	vap->va_gen = 0;
714 	vap->va_flags = 0;
715 	vap->va_filerev = 0;
716 	vap->va_nlink = de->de_links;
717 	vap->va_fileid = de->de_inode;
718 
719 	return (error);
720 }
721 
722 /* ARGSUSED */
723 static int
724 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
725 {
726 	struct cdev *dev;
727 	struct cdevsw *dsw;
728 	struct vnode *vp;
729 	struct vnode *vpold;
730 	int error, i, ref;
731 	const char *p;
732 	struct fiodgname_arg *fgn;
733 	struct file *fpop;
734 
735 	fpop = td->td_fpop;
736 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
737 	if (error)
738 		return (error);
739 
740 	if (com == FIODTYPE) {
741 		*(int *)data = dsw->d_flags & D_TYPEMASK;
742 		td->td_fpop = fpop;
743 		dev_relthread(dev, ref);
744 		return (0);
745 	} else if (com == FIODGNAME) {
746 		fgn = data;
747 		p = devtoname(dev);
748 		i = strlen(p) + 1;
749 		if (i > fgn->len)
750 			error = EINVAL;
751 		else
752 			error = copyout(p, fgn->buf, i);
753 		td->td_fpop = fpop;
754 		dev_relthread(dev, ref);
755 		return (error);
756 	}
757 	error = dsw->d_ioctl(dev, com, data, fp->f_flag, td);
758 	td->td_fpop = NULL;
759 	dev_relthread(dev, ref);
760 	if (error == ENOIOCTL)
761 		error = ENOTTY;
762 	if (error == 0 && com == TIOCSCTTY) {
763 		vp = fp->f_vnode;
764 
765 		/* Do nothing if reassigning same control tty */
766 		sx_slock(&proctree_lock);
767 		if (td->td_proc->p_session->s_ttyvp == vp) {
768 			sx_sunlock(&proctree_lock);
769 			return (0);
770 		}
771 
772 		vpold = td->td_proc->p_session->s_ttyvp;
773 		VREF(vp);
774 		SESS_LOCK(td->td_proc->p_session);
775 		td->td_proc->p_session->s_ttyvp = vp;
776 		td->td_proc->p_session->s_ttydp = cdev2priv(dev);
777 		SESS_UNLOCK(td->td_proc->p_session);
778 
779 		sx_sunlock(&proctree_lock);
780 
781 		/* Get rid of reference to old control tty */
782 		if (vpold)
783 			vrele(vpold);
784 	}
785 	return (error);
786 }
787 
788 /* ARGSUSED */
789 static int
790 devfs_kqfilter_f(struct file *fp, struct knote *kn)
791 {
792 	struct cdev *dev;
793 	struct cdevsw *dsw;
794 	int error, ref;
795 	struct file *fpop;
796 	struct thread *td;
797 
798 	td = curthread;
799 	fpop = td->td_fpop;
800 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
801 	if (error)
802 		return (error);
803 	error = dsw->d_kqfilter(dev, kn);
804 	td->td_fpop = fpop;
805 	dev_relthread(dev, ref);
806 	return (error);
807 }
808 
809 static inline int
810 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
811 {
812 	struct cdev_priv *cdp;
813 	struct ucred *dcr;
814 	int error;
815 
816 	cdp = de->de_cdp;
817 	if (cdp == NULL)
818 		return (0);
819 	dcr = cdp->cdp_c.si_cred;
820 	if (dcr == NULL)
821 		return (0);
822 
823 	error = prison_check(td->td_ucred, dcr);
824 	if (error == 0)
825 		return (0);
826 	/* We do, however, allow access to the controlling terminal */
827 	if (!(td->td_proc->p_flag & P_CONTROLT))
828 		return (error);
829 	if (td->td_proc->p_session->s_ttydp == cdp)
830 		return (0);
831 	return (error);
832 }
833 
834 static int
835 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
836 {
837 	struct componentname *cnp;
838 	struct vnode *dvp, **vpp;
839 	struct thread *td;
840 	struct devfs_dirent *de, *dd;
841 	struct devfs_dirent **dde;
842 	struct devfs_mount *dmp;
843 	struct cdev *cdev;
844 	int error, flags, nameiop, dvplocked;
845 	char specname[SPECNAMELEN + 1], *pname;
846 
847 	cnp = ap->a_cnp;
848 	vpp = ap->a_vpp;
849 	dvp = ap->a_dvp;
850 	pname = cnp->cn_nameptr;
851 	td = cnp->cn_thread;
852 	flags = cnp->cn_flags;
853 	nameiop = cnp->cn_nameiop;
854 	dmp = VFSTODEVFS(dvp->v_mount);
855 	dd = dvp->v_data;
856 	*vpp = NULLVP;
857 
858 	if ((flags & ISLASTCN) && nameiop == RENAME)
859 		return (EOPNOTSUPP);
860 
861 	if (dvp->v_type != VDIR)
862 		return (ENOTDIR);
863 
864 	if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
865 		return (EIO);
866 
867 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
868 	if (error)
869 		return (error);
870 
871 	if (cnp->cn_namelen == 1 && *pname == '.') {
872 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
873 			return (EINVAL);
874 		*vpp = dvp;
875 		VREF(dvp);
876 		return (0);
877 	}
878 
879 	if (flags & ISDOTDOT) {
880 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
881 			return (EINVAL);
882 		de = devfs_parent_dirent(dd);
883 		if (de == NULL)
884 			return (ENOENT);
885 		dvplocked = VOP_ISLOCKED(dvp);
886 		VOP_UNLOCK(dvp, 0);
887 		error = devfs_allocv(de, dvp->v_mount,
888 		    cnp->cn_lkflags & LK_TYPE_MASK, vpp);
889 		*dm_unlock = 0;
890 		vn_lock(dvp, dvplocked | LK_RETRY);
891 		return (error);
892 	}
893 
894 	dd = dvp->v_data;
895 	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
896 	while (de == NULL) {	/* While(...) so we can use break */
897 
898 		if (nameiop == DELETE)
899 			return (ENOENT);
900 
901 		/*
902 		 * OK, we didn't have an entry for the name we were asked for
903 		 * so we try to see if anybody can create it on demand.
904 		 */
905 		pname = devfs_fqpn(specname, dmp, dd, cnp);
906 		if (pname == NULL)
907 			break;
908 
909 		cdev = NULL;
910 		DEVFS_DMP_HOLD(dmp);
911 		sx_xunlock(&dmp->dm_lock);
912 		sx_slock(&clone_drain_lock);
913 		EVENTHANDLER_INVOKE(dev_clone,
914 		    td->td_ucred, pname, strlen(pname), &cdev);
915 		sx_sunlock(&clone_drain_lock);
916 
917 		if (cdev == NULL)
918 			sx_xlock(&dmp->dm_lock);
919 		else if (devfs_populate_vp(dvp) != 0) {
920 			*dm_unlock = 0;
921 			sx_xlock(&dmp->dm_lock);
922 			if (DEVFS_DMP_DROP(dmp)) {
923 				sx_xunlock(&dmp->dm_lock);
924 				devfs_unmount_final(dmp);
925 			} else
926 				sx_xunlock(&dmp->dm_lock);
927 			dev_rel(cdev);
928 			return (ENOENT);
929 		}
930 		if (DEVFS_DMP_DROP(dmp)) {
931 			*dm_unlock = 0;
932 			sx_xunlock(&dmp->dm_lock);
933 			devfs_unmount_final(dmp);
934 			if (cdev != NULL)
935 				dev_rel(cdev);
936 			return (ENOENT);
937 		}
938 
939 		if (cdev == NULL)
940 			break;
941 
942 		dev_lock();
943 		dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
944 		if (dde != NULL && *dde != NULL)
945 			de = *dde;
946 		dev_unlock();
947 		dev_rel(cdev);
948 		break;
949 	}
950 
951 	if (de == NULL || de->de_flags & DE_WHITEOUT) {
952 		if ((nameiop == CREATE || nameiop == RENAME) &&
953 		    (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
954 			cnp->cn_flags |= SAVENAME;
955 			return (EJUSTRETURN);
956 		}
957 		return (ENOENT);
958 	}
959 
960 	if (devfs_prison_check(de, td))
961 		return (ENOENT);
962 
963 	if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
964 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
965 		if (error)
966 			return (error);
967 		if (*vpp == dvp) {
968 			VREF(dvp);
969 			*vpp = dvp;
970 			return (0);
971 		}
972 	}
973 	error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK,
974 	    vpp);
975 	*dm_unlock = 0;
976 	return (error);
977 }
978 
979 static int
980 devfs_lookup(struct vop_lookup_args *ap)
981 {
982 	int j;
983 	struct devfs_mount *dmp;
984 	int dm_unlock;
985 
986 	if (devfs_populate_vp(ap->a_dvp) != 0)
987 		return (ENOTDIR);
988 
989 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
990 	dm_unlock = 1;
991 	j = devfs_lookupx(ap, &dm_unlock);
992 	if (dm_unlock == 1)
993 		sx_xunlock(&dmp->dm_lock);
994 	return (j);
995 }
996 
997 static int
998 devfs_mknod(struct vop_mknod_args *ap)
999 {
1000 	struct componentname *cnp;
1001 	struct vnode *dvp, **vpp;
1002 	struct devfs_dirent *dd, *de;
1003 	struct devfs_mount *dmp;
1004 	int error;
1005 
1006 	/*
1007 	 * The only type of node we should be creating here is a
1008 	 * character device, for anything else return EOPNOTSUPP.
1009 	 */
1010 	if (ap->a_vap->va_type != VCHR)
1011 		return (EOPNOTSUPP);
1012 	dvp = ap->a_dvp;
1013 	dmp = VFSTODEVFS(dvp->v_mount);
1014 
1015 	cnp = ap->a_cnp;
1016 	vpp = ap->a_vpp;
1017 	dd = dvp->v_data;
1018 
1019 	error = ENOENT;
1020 	sx_xlock(&dmp->dm_lock);
1021 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1022 		if (cnp->cn_namelen != de->de_dirent->d_namlen)
1023 			continue;
1024 		if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1025 		    de->de_dirent->d_namlen) != 0)
1026 			continue;
1027 		if (de->de_flags & DE_WHITEOUT)
1028 			break;
1029 		goto notfound;
1030 	}
1031 	if (de == NULL)
1032 		goto notfound;
1033 	de->de_flags &= ~DE_WHITEOUT;
1034 	error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1035 	return (error);
1036 notfound:
1037 	sx_xunlock(&dmp->dm_lock);
1038 	return (error);
1039 }
1040 
1041 /* ARGSUSED */
1042 static int
1043 devfs_open(struct vop_open_args *ap)
1044 {
1045 	struct thread *td = ap->a_td;
1046 	struct vnode *vp = ap->a_vp;
1047 	struct cdev *dev = vp->v_rdev;
1048 	struct file *fp = ap->a_fp;
1049 	int error, ref, vlocked;
1050 	struct cdevsw *dsw;
1051 	struct file *fpop;
1052 
1053 	if (vp->v_type == VBLK)
1054 		return (ENXIO);
1055 
1056 	if (dev == NULL)
1057 		return (ENXIO);
1058 
1059 	/* Make this field valid before any I/O in d_open. */
1060 	if (dev->si_iosize_max == 0)
1061 		dev->si_iosize_max = DFLTPHYS;
1062 
1063 	dsw = dev_refthread(dev, &ref);
1064 	if (dsw == NULL)
1065 		return (ENXIO);
1066 	if (fp == NULL && dsw->d_fdopen != NULL) {
1067 		dev_relthread(dev, ref);
1068 		return (ENXIO);
1069 	}
1070 
1071 	vlocked = VOP_ISLOCKED(vp);
1072 	VOP_UNLOCK(vp, 0);
1073 
1074 	fpop = td->td_fpop;
1075 	td->td_fpop = fp;
1076 	if (fp != NULL) {
1077 		fp->f_data = dev;
1078 		fp->f_vnode = vp;
1079 	}
1080 	if (dsw->d_fdopen != NULL)
1081 		error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1082 	else
1083 		error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1084 	td->td_fpop = fpop;
1085 
1086 	vn_lock(vp, vlocked | LK_RETRY);
1087 	dev_relthread(dev, ref);
1088 	if (error)
1089 		return (error);
1090 
1091 #if 0	/* /dev/console */
1092 	KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1093 #else
1094 	if (fp == NULL)
1095 		return (error);
1096 #endif
1097 	if (fp->f_ops == &badfileops)
1098 		finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1099 	return (error);
1100 }
1101 
1102 static int
1103 devfs_pathconf(struct vop_pathconf_args *ap)
1104 {
1105 
1106 	switch (ap->a_name) {
1107 	case _PC_MAC_PRESENT:
1108 #ifdef MAC
1109 		/*
1110 		 * If MAC is enabled, devfs automatically supports
1111 		 * trivial non-persistant label storage.
1112 		 */
1113 		*ap->a_retval = 1;
1114 #else
1115 		*ap->a_retval = 0;
1116 #endif
1117 		return (0);
1118 	default:
1119 		return (vop_stdpathconf(ap));
1120 	}
1121 	/* NOTREACHED */
1122 }
1123 
1124 /* ARGSUSED */
1125 static int
1126 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1127 {
1128 	struct cdev *dev;
1129 	struct cdevsw *dsw;
1130 	int error, ref;
1131 	struct file *fpop;
1132 
1133 	fpop = td->td_fpop;
1134 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1135 	if (error)
1136 		return (poll_no_poll(events));
1137 	error = dsw->d_poll(dev, events, td);
1138 	td->td_fpop = fpop;
1139 	dev_relthread(dev, ref);
1140 	return(error);
1141 }
1142 
1143 /*
1144  * Print out the contents of a special device vnode.
1145  */
1146 static int
1147 devfs_print(struct vop_print_args *ap)
1148 {
1149 
1150 	printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1151 	return (0);
1152 }
1153 
1154 /* ARGSUSED */
1155 static int
1156 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td)
1157 {
1158 	struct cdev *dev;
1159 	int ioflag, error, ref, resid;
1160 	struct cdevsw *dsw;
1161 	struct file *fpop;
1162 
1163 	fpop = td->td_fpop;
1164 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1165 	if (error)
1166 		return (error);
1167 	resid = uio->uio_resid;
1168 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1169 	if (ioflag & O_DIRECT)
1170 		ioflag |= IO_DIRECT;
1171 
1172 	if ((flags & FOF_OFFSET) == 0)
1173 		uio->uio_offset = fp->f_offset;
1174 
1175 	error = dsw->d_read(dev, uio, ioflag);
1176 	if (uio->uio_resid != resid || (error == 0 && resid != 0))
1177 		vfs_timestamp(&dev->si_atime);
1178 	td->td_fpop = fpop;
1179 	dev_relthread(dev, ref);
1180 
1181 	if ((flags & FOF_OFFSET) == 0)
1182 		fp->f_offset = uio->uio_offset;
1183 	fp->f_nextoff = uio->uio_offset;
1184 	return (error);
1185 }
1186 
1187 static int
1188 devfs_readdir(struct vop_readdir_args *ap)
1189 {
1190 	int error;
1191 	struct uio *uio;
1192 	struct dirent *dp;
1193 	struct devfs_dirent *dd;
1194 	struct devfs_dirent *de;
1195 	struct devfs_mount *dmp;
1196 	off_t off;
1197 	int *tmp_ncookies = NULL;
1198 
1199 	if (ap->a_vp->v_type != VDIR)
1200 		return (ENOTDIR);
1201 
1202 	uio = ap->a_uio;
1203 	if (uio->uio_offset < 0)
1204 		return (EINVAL);
1205 
1206 	/*
1207 	 * XXX: This is a temporary hack to get around this filesystem not
1208 	 * supporting cookies. We store the location of the ncookies pointer
1209 	 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1210 	 * and set the number of cookies to 0. We then set the pointer to
1211 	 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1212 	 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1213 	 * pointer to its original location before returning to the caller.
1214 	 */
1215 	if (ap->a_ncookies != NULL) {
1216 		tmp_ncookies = ap->a_ncookies;
1217 		*ap->a_ncookies = 0;
1218 		ap->a_ncookies = NULL;
1219 	}
1220 
1221 	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1222 	if (devfs_populate_vp(ap->a_vp) != 0) {
1223 		if (tmp_ncookies != NULL)
1224 			ap->a_ncookies = tmp_ncookies;
1225 		return (EIO);
1226 	}
1227 	error = 0;
1228 	de = ap->a_vp->v_data;
1229 	off = 0;
1230 	TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1231 		KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1232 		if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1233 			continue;
1234 		if (devfs_prison_check(dd, uio->uio_td))
1235 			continue;
1236 		if (dd->de_dirent->d_type == DT_DIR)
1237 			de = dd->de_dir;
1238 		else
1239 			de = dd;
1240 		dp = dd->de_dirent;
1241 		if (dp->d_reclen > uio->uio_resid)
1242 			break;
1243 		dp->d_fileno = de->de_inode;
1244 		if (off >= uio->uio_offset) {
1245 			error = vfs_read_dirent(ap, dp, off);
1246 			if (error)
1247 				break;
1248 		}
1249 		off += dp->d_reclen;
1250 	}
1251 	sx_xunlock(&dmp->dm_lock);
1252 	uio->uio_offset = off;
1253 
1254 	/*
1255 	 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1256 	 * place.
1257 	 */
1258 	if (tmp_ncookies != NULL)
1259 		ap->a_ncookies = tmp_ncookies;
1260 
1261 	return (error);
1262 }
1263 
1264 static int
1265 devfs_readlink(struct vop_readlink_args *ap)
1266 {
1267 	struct devfs_dirent *de;
1268 
1269 	de = ap->a_vp->v_data;
1270 	return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1271 }
1272 
1273 static int
1274 devfs_reclaim(struct vop_reclaim_args *ap)
1275 {
1276 	struct vnode *vp = ap->a_vp;
1277 	struct devfs_dirent *de;
1278 	struct cdev *dev;
1279 
1280 	mtx_lock(&devfs_de_interlock);
1281 	de = vp->v_data;
1282 	if (de != NULL) {
1283 		de->de_vnode = NULL;
1284 		vp->v_data = NULL;
1285 	}
1286 	mtx_unlock(&devfs_de_interlock);
1287 
1288 	vnode_destroy_vobject(vp);
1289 
1290 	VI_LOCK(vp);
1291 	dev_lock();
1292 	dev = vp->v_rdev;
1293 	vp->v_rdev = NULL;
1294 
1295 	if (dev == NULL) {
1296 		dev_unlock();
1297 		VI_UNLOCK(vp);
1298 		return (0);
1299 	}
1300 
1301 	dev->si_usecount -= vp->v_usecount;
1302 	dev_unlock();
1303 	VI_UNLOCK(vp);
1304 	dev_rel(dev);
1305 	return (0);
1306 }
1307 
1308 static int
1309 devfs_remove(struct vop_remove_args *ap)
1310 {
1311 	struct vnode *dvp = ap->a_dvp;
1312 	struct vnode *vp = ap->a_vp;
1313 	struct devfs_dirent *dd;
1314 	struct devfs_dirent *de, *de_covered;
1315 	struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1316 
1317 	ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1318 	ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1319 
1320 	sx_xlock(&dmp->dm_lock);
1321 	dd = ap->a_dvp->v_data;
1322 	de = vp->v_data;
1323 	if (de->de_cdp == NULL) {
1324 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1325 		if (de->de_dirent->d_type == DT_LNK) {
1326 			de_covered = devfs_find(dd, de->de_dirent->d_name,
1327 			    de->de_dirent->d_namlen, 0);
1328 			if (de_covered != NULL)
1329 				de_covered->de_flags &= ~DE_COVERED;
1330 		}
1331 		/* We need to unlock dvp because devfs_delete() may lock it. */
1332 		VOP_UNLOCK(vp, 0);
1333 		if (dvp != vp)
1334 			VOP_UNLOCK(dvp, 0);
1335 		devfs_delete(dmp, de, 0);
1336 		sx_xunlock(&dmp->dm_lock);
1337 		if (dvp != vp)
1338 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1339 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1340 	} else {
1341 		de->de_flags |= DE_WHITEOUT;
1342 		sx_xunlock(&dmp->dm_lock);
1343 	}
1344 	return (0);
1345 }
1346 
1347 /*
1348  * Revoke is called on a tty when a terminal session ends.  The vnode
1349  * is orphaned by setting v_op to deadfs so we need to let go of it
1350  * as well so that we create a new one next time around.
1351  *
1352  */
1353 static int
1354 devfs_revoke(struct vop_revoke_args *ap)
1355 {
1356 	struct vnode *vp = ap->a_vp, *vp2;
1357 	struct cdev *dev;
1358 	struct cdev_priv *cdp;
1359 	struct devfs_dirent *de;
1360 	int i;
1361 
1362 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1363 
1364 	dev = vp->v_rdev;
1365 	cdp = cdev2priv(dev);
1366 
1367 	dev_lock();
1368 	cdp->cdp_inuse++;
1369 	dev_unlock();
1370 
1371 	vhold(vp);
1372 	vgone(vp);
1373 	vdrop(vp);
1374 
1375 	VOP_UNLOCK(vp,0);
1376  loop:
1377 	for (;;) {
1378 		mtx_lock(&devfs_de_interlock);
1379 		dev_lock();
1380 		vp2 = NULL;
1381 		for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1382 			de = cdp->cdp_dirents[i];
1383 			if (de == NULL)
1384 				continue;
1385 
1386 			vp2 = de->de_vnode;
1387 			if (vp2 != NULL) {
1388 				dev_unlock();
1389 				VI_LOCK(vp2);
1390 				mtx_unlock(&devfs_de_interlock);
1391 				if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1392 				    curthread))
1393 					goto loop;
1394 				vhold(vp2);
1395 				vgone(vp2);
1396 				vdrop(vp2);
1397 				vput(vp2);
1398 				break;
1399 			}
1400 		}
1401 		if (vp2 != NULL) {
1402 			continue;
1403 		}
1404 		dev_unlock();
1405 		mtx_unlock(&devfs_de_interlock);
1406 		break;
1407 	}
1408 	dev_lock();
1409 	cdp->cdp_inuse--;
1410 	if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1411 		TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1412 		dev_unlock();
1413 		dev_rel(&cdp->cdp_c);
1414 	} else
1415 		dev_unlock();
1416 
1417 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1418 	return (0);
1419 }
1420 
1421 static int
1422 devfs_rioctl(struct vop_ioctl_args *ap)
1423 {
1424 	struct vnode *vp;
1425 	struct devfs_mount *dmp;
1426 	int error;
1427 
1428 	vp = ap->a_vp;
1429 	vn_lock(vp, LK_SHARED | LK_RETRY);
1430 	if (vp->v_iflag & VI_DOOMED) {
1431 		VOP_UNLOCK(vp, 0);
1432 		return (EBADF);
1433 	}
1434 	dmp = VFSTODEVFS(vp->v_mount);
1435 	sx_xlock(&dmp->dm_lock);
1436 	VOP_UNLOCK(vp, 0);
1437 	DEVFS_DMP_HOLD(dmp);
1438 	devfs_populate(dmp);
1439 	if (DEVFS_DMP_DROP(dmp)) {
1440 		sx_xunlock(&dmp->dm_lock);
1441 		devfs_unmount_final(dmp);
1442 		return (ENOENT);
1443 	}
1444 	error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1445 	sx_xunlock(&dmp->dm_lock);
1446 	return (error);
1447 }
1448 
1449 static int
1450 devfs_rread(struct vop_read_args *ap)
1451 {
1452 
1453 	if (ap->a_vp->v_type != VDIR)
1454 		return (EINVAL);
1455 	return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1456 }
1457 
1458 static int
1459 devfs_setattr(struct vop_setattr_args *ap)
1460 {
1461 	struct devfs_dirent *de;
1462 	struct vattr *vap;
1463 	struct vnode *vp;
1464 	struct thread *td;
1465 	int c, error;
1466 	uid_t uid;
1467 	gid_t gid;
1468 
1469 	vap = ap->a_vap;
1470 	vp = ap->a_vp;
1471 	td = curthread;
1472 	if ((vap->va_type != VNON) ||
1473 	    (vap->va_nlink != VNOVAL) ||
1474 	    (vap->va_fsid != VNOVAL) ||
1475 	    (vap->va_fileid != VNOVAL) ||
1476 	    (vap->va_blocksize != VNOVAL) ||
1477 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1478 	    (vap->va_rdev != VNOVAL) ||
1479 	    ((int)vap->va_bytes != VNOVAL) ||
1480 	    (vap->va_gen != VNOVAL)) {
1481 		return (EINVAL);
1482 	}
1483 
1484 	de = vp->v_data;
1485 	if (vp->v_type == VDIR)
1486 		de = de->de_dir;
1487 
1488 	error = c = 0;
1489 	if (vap->va_uid == (uid_t)VNOVAL)
1490 		uid = de->de_uid;
1491 	else
1492 		uid = vap->va_uid;
1493 	if (vap->va_gid == (gid_t)VNOVAL)
1494 		gid = de->de_gid;
1495 	else
1496 		gid = vap->va_gid;
1497 	if (uid != de->de_uid || gid != de->de_gid) {
1498 		if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1499 		    (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1500 			error = priv_check(td, PRIV_VFS_CHOWN);
1501 			if (error)
1502 				return (error);
1503 		}
1504 		de->de_uid = uid;
1505 		de->de_gid = gid;
1506 		c = 1;
1507 	}
1508 
1509 	if (vap->va_mode != (mode_t)VNOVAL) {
1510 		if (ap->a_cred->cr_uid != de->de_uid) {
1511 			error = priv_check(td, PRIV_VFS_ADMIN);
1512 			if (error)
1513 				return (error);
1514 		}
1515 		de->de_mode = vap->va_mode;
1516 		c = 1;
1517 	}
1518 
1519 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1520 		/* See the comment in ufs_vnops::ufs_setattr(). */
1521 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) &&
1522 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1523 		    (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td))))
1524 			return (error);
1525 		if (vap->va_atime.tv_sec != VNOVAL) {
1526 			if (vp->v_type == VCHR)
1527 				vp->v_rdev->si_atime = vap->va_atime;
1528 			else
1529 				de->de_atime = vap->va_atime;
1530 		}
1531 		if (vap->va_mtime.tv_sec != VNOVAL) {
1532 			if (vp->v_type == VCHR)
1533 				vp->v_rdev->si_mtime = vap->va_mtime;
1534 			else
1535 				de->de_mtime = vap->va_mtime;
1536 		}
1537 		c = 1;
1538 	}
1539 
1540 	if (c) {
1541 		if (vp->v_type == VCHR)
1542 			vfs_timestamp(&vp->v_rdev->si_ctime);
1543 		else
1544 			vfs_timestamp(&de->de_mtime);
1545 	}
1546 	return (0);
1547 }
1548 
1549 #ifdef MAC
1550 static int
1551 devfs_setlabel(struct vop_setlabel_args *ap)
1552 {
1553 	struct vnode *vp;
1554 	struct devfs_dirent *de;
1555 
1556 	vp = ap->a_vp;
1557 	de = vp->v_data;
1558 
1559 	mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1560 	mac_devfs_update(vp->v_mount, de, vp);
1561 
1562 	return (0);
1563 }
1564 #endif
1565 
1566 static int
1567 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1568 {
1569 
1570 	return (vnops.fo_stat(fp, sb, cred, td));
1571 }
1572 
1573 static int
1574 devfs_symlink(struct vop_symlink_args *ap)
1575 {
1576 	int i, error;
1577 	struct devfs_dirent *dd;
1578 	struct devfs_dirent *de, *de_covered, *de_dotdot;
1579 	struct devfs_mount *dmp;
1580 
1581 	error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1582 	if (error)
1583 		return(error);
1584 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1585 	if (devfs_populate_vp(ap->a_dvp) != 0)
1586 		return (ENOENT);
1587 
1588 	dd = ap->a_dvp->v_data;
1589 	de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1590 	de->de_flags = DE_USER;
1591 	de->de_uid = 0;
1592 	de->de_gid = 0;
1593 	de->de_mode = 0755;
1594 	de->de_inode = alloc_unr(devfs_inos);
1595 	de->de_dir = dd;
1596 	de->de_dirent->d_type = DT_LNK;
1597 	i = strlen(ap->a_target) + 1;
1598 	de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1599 	bcopy(ap->a_target, de->de_symlink, i);
1600 #ifdef MAC
1601 	mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1602 #endif
1603 	de_covered = devfs_find(dd, de->de_dirent->d_name,
1604 	    de->de_dirent->d_namlen, 0);
1605 	if (de_covered != NULL) {
1606 		if ((de_covered->de_flags & DE_USER) != 0) {
1607 			devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1608 			sx_xunlock(&dmp->dm_lock);
1609 			return (EEXIST);
1610 		}
1611 		KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1612 		    ("devfs_symlink: entry %p already covered", de_covered));
1613 		de_covered->de_flags |= DE_COVERED;
1614 	}
1615 
1616 	de_dotdot = TAILQ_FIRST(&dd->de_dlist);		/* "." */
1617 	de_dotdot = TAILQ_NEXT(de_dotdot, de_list);	/* ".." */
1618 	TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1619 	devfs_dir_ref_de(dmp, dd);
1620 	devfs_rules_apply(dmp, de);
1621 
1622 	return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1623 }
1624 
1625 static int
1626 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1627 {
1628 
1629 	return (vnops.fo_truncate(fp, length, cred, td));
1630 }
1631 
1632 /* ARGSUSED */
1633 static int
1634 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td)
1635 {
1636 	struct cdev *dev;
1637 	int error, ioflag, ref, resid;
1638 	struct cdevsw *dsw;
1639 	struct file *fpop;
1640 
1641 	fpop = td->td_fpop;
1642 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1643 	if (error)
1644 		return (error);
1645 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1646 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1647 	if (ioflag & O_DIRECT)
1648 		ioflag |= IO_DIRECT;
1649 	if ((flags & FOF_OFFSET) == 0)
1650 		uio->uio_offset = fp->f_offset;
1651 
1652 	resid = uio->uio_resid;
1653 
1654 	error = dsw->d_write(dev, uio, ioflag);
1655 	if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1656 		vfs_timestamp(&dev->si_ctime);
1657 		dev->si_mtime = dev->si_ctime;
1658 	}
1659 	td->td_fpop = fpop;
1660 	dev_relthread(dev, ref);
1661 
1662 	if ((flags & FOF_OFFSET) == 0)
1663 		fp->f_offset = uio->uio_offset;
1664 	fp->f_nextoff = uio->uio_offset;
1665 	return (error);
1666 }
1667 
1668 dev_t
1669 dev2udev(struct cdev *x)
1670 {
1671 	if (x == NULL)
1672 		return (NODEV);
1673 	return (cdev2priv(x)->cdp_inode);
1674 }
1675 
1676 static struct fileops devfs_ops_f = {
1677 	.fo_read =	devfs_read_f,
1678 	.fo_write =	devfs_write_f,
1679 	.fo_truncate =	devfs_truncate_f,
1680 	.fo_ioctl =	devfs_ioctl_f,
1681 	.fo_poll =	devfs_poll_f,
1682 	.fo_kqfilter =	devfs_kqfilter_f,
1683 	.fo_stat =	devfs_stat_f,
1684 	.fo_close =	devfs_close_f,
1685 	.fo_chmod =	vn_chmod,
1686 	.fo_chown =	vn_chown,
1687 	.fo_flags =	DFLAG_PASSABLE | DFLAG_SEEKABLE
1688 };
1689 
1690 static struct vop_vector devfs_vnodeops = {
1691 	.vop_default =		&default_vnodeops,
1692 
1693 	.vop_access =		devfs_access,
1694 	.vop_getattr =		devfs_getattr,
1695 	.vop_ioctl =		devfs_rioctl,
1696 	.vop_lookup =		devfs_lookup,
1697 	.vop_mknod =		devfs_mknod,
1698 	.vop_pathconf =		devfs_pathconf,
1699 	.vop_read =		devfs_rread,
1700 	.vop_readdir =		devfs_readdir,
1701 	.vop_readlink =		devfs_readlink,
1702 	.vop_reclaim =		devfs_reclaim,
1703 	.vop_remove =		devfs_remove,
1704 	.vop_revoke =		devfs_revoke,
1705 	.vop_setattr =		devfs_setattr,
1706 #ifdef MAC
1707 	.vop_setlabel =		devfs_setlabel,
1708 #endif
1709 	.vop_symlink =		devfs_symlink,
1710 	.vop_vptocnp =		devfs_vptocnp,
1711 };
1712 
1713 static struct vop_vector devfs_specops = {
1714 	.vop_default =		&default_vnodeops,
1715 
1716 	.vop_access =		devfs_access,
1717 	.vop_bmap =		VOP_PANIC,
1718 	.vop_close =		devfs_close,
1719 	.vop_create =		VOP_PANIC,
1720 	.vop_fsync =		devfs_fsync,
1721 	.vop_getattr =		devfs_getattr,
1722 	.vop_link =		VOP_PANIC,
1723 	.vop_mkdir =		VOP_PANIC,
1724 	.vop_mknod =		VOP_PANIC,
1725 	.vop_open =		devfs_open,
1726 	.vop_pathconf =		devfs_pathconf,
1727 	.vop_print =		devfs_print,
1728 	.vop_read =		VOP_PANIC,
1729 	.vop_readdir =		VOP_PANIC,
1730 	.vop_readlink =		VOP_PANIC,
1731 	.vop_reallocblks =	VOP_PANIC,
1732 	.vop_reclaim =		devfs_reclaim,
1733 	.vop_remove =		devfs_remove,
1734 	.vop_rename =		VOP_PANIC,
1735 	.vop_revoke =		devfs_revoke,
1736 	.vop_rmdir =		VOP_PANIC,
1737 	.vop_setattr =		devfs_setattr,
1738 #ifdef MAC
1739 	.vop_setlabel =		devfs_setlabel,
1740 #endif
1741 	.vop_strategy =		VOP_PANIC,
1742 	.vop_symlink =		VOP_PANIC,
1743 	.vop_vptocnp =		devfs_vptocnp,
1744 	.vop_write =		VOP_PANIC,
1745 };
1746 
1747 /*
1748  * Our calling convention to the device drivers used to be that we passed
1749  * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1750  * flags instead since that's what open(), close() and ioctl() takes and
1751  * we don't really want vnode.h in device drivers.
1752  * We solved the source compatibility by redefining some vnode flags to
1753  * be the same as the fcntl ones and by sending down the bitwise OR of
1754  * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
1755  * pulls the rug out under this.
1756  */
1757 CTASSERT(O_NONBLOCK == IO_NDELAY);
1758 CTASSERT(O_FSYNC == IO_SYNC);
1759