xref: /freebsd/sys/fs/devfs/devfs_vnops.c (revision 6186fd1857626de0f7cb1a9e4dff19082f9ebb11)
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 	if (td && vp == td->td_proc->p_session->s_ttyvp) {
549 		oldvp = NULL;
550 		sx_xlock(&proctree_lock);
551 		if (vp == td->td_proc->p_session->s_ttyvp) {
552 			SESS_LOCK(td->td_proc->p_session);
553 			VI_LOCK(vp);
554 			if (count_dev(dev) == 2 &&
555 			    (vp->v_iflag & VI_DOOMED) == 0) {
556 				td->td_proc->p_session->s_ttyvp = NULL;
557 				td->td_proc->p_session->s_ttydp = NULL;
558 				oldvp = vp;
559 			}
560 			VI_UNLOCK(vp);
561 			SESS_UNLOCK(td->td_proc->p_session);
562 		}
563 		sx_xunlock(&proctree_lock);
564 		if (oldvp != NULL)
565 			vrele(oldvp);
566 	}
567 	/*
568 	 * We do not want to really close the device if it
569 	 * is still in use unless we are trying to close it
570 	 * forcibly. Since every use (buffer, vnode, swap, cmap)
571 	 * holds a reference to the vnode, and because we mark
572 	 * any other vnodes that alias this device, when the
573 	 * sum of the reference counts on all the aliased
574 	 * vnodes descends to one, we are on last close.
575 	 */
576 	dsw = dev_refthread(dev, &ref);
577 	if (dsw == NULL)
578 		return (ENXIO);
579 	VI_LOCK(vp);
580 	if (vp->v_iflag & VI_DOOMED) {
581 		/* Forced close. */
582 	} else if (dsw->d_flags & D_TRACKCLOSE) {
583 		/* Keep device updated on status. */
584 	} else if (count_dev(dev) > 1) {
585 		VI_UNLOCK(vp);
586 		dev_relthread(dev, ref);
587 		return (0);
588 	}
589 	vholdl(vp);
590 	VI_UNLOCK(vp);
591 	vp_locked = VOP_ISLOCKED(vp);
592 	VOP_UNLOCK(vp, 0);
593 	KASSERT(dev->si_refcount > 0,
594 	    ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
595 	error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td);
596 	dev_relthread(dev, ref);
597 	vn_lock(vp, vp_locked | LK_RETRY);
598 	vdrop(vp);
599 	return (error);
600 }
601 
602 static int
603 devfs_close_f(struct file *fp, struct thread *td)
604 {
605 	int error;
606 	struct file *fpop;
607 
608 	/*
609 	 * NB: td may be NULL if this descriptor is closed due to
610 	 * garbage collection from a closed UNIX domain socket.
611 	 */
612 	fpop = curthread->td_fpop;
613 	curthread->td_fpop = fp;
614 	error = vnops.fo_close(fp, td);
615 	curthread->td_fpop = fpop;
616 
617 	/*
618 	 * The f_cdevpriv cannot be assigned non-NULL value while we
619 	 * are destroying the file.
620 	 */
621 	if (fp->f_cdevpriv != NULL)
622 		devfs_fpdrop(fp);
623 	return (error);
624 }
625 
626 static int
627 devfs_fsync(struct vop_fsync_args *ap)
628 {
629 	int error;
630 	struct bufobj *bo;
631 	struct devfs_dirent *de;
632 
633 	if (!vn_isdisk(ap->a_vp, &error)) {
634 		bo = &ap->a_vp->v_bufobj;
635 		de = ap->a_vp->v_data;
636 		if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) {
637 			printf("Device %s went missing before all of the data "
638 			    "could be written to it; expect data loss.\n",
639 			    de->de_dirent->d_name);
640 
641 			error = vop_stdfsync(ap);
642 			if (bo->bo_dirty.bv_cnt != 0 || error != 0)
643 				panic("devfs_fsync: vop_stdfsync failed.");
644 		}
645 
646 		return (0);
647 	}
648 
649 	return (vop_stdfsync(ap));
650 }
651 
652 static int
653 devfs_getattr(struct vop_getattr_args *ap)
654 {
655 	struct vnode *vp = ap->a_vp;
656 	struct vattr *vap = ap->a_vap;
657 	int error;
658 	struct devfs_dirent *de;
659 	struct devfs_mount *dmp;
660 	struct cdev *dev;
661 
662 	error = devfs_populate_vp(vp);
663 	if (error != 0)
664 		return (error);
665 
666 	dmp = VFSTODEVFS(vp->v_mount);
667 	sx_xunlock(&dmp->dm_lock);
668 
669 	de = vp->v_data;
670 	KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
671 	if (vp->v_type == VDIR) {
672 		de = de->de_dir;
673 		KASSERT(de != NULL,
674 		    ("Null dir dirent in devfs_getattr vp=%p", vp));
675 	}
676 	vap->va_uid = de->de_uid;
677 	vap->va_gid = de->de_gid;
678 	vap->va_mode = de->de_mode;
679 	if (vp->v_type == VLNK)
680 		vap->va_size = strlen(de->de_symlink);
681 	else if (vp->v_type == VDIR)
682 		vap->va_size = vap->va_bytes = DEV_BSIZE;
683 	else
684 		vap->va_size = 0;
685 	if (vp->v_type != VDIR)
686 		vap->va_bytes = 0;
687 	vap->va_blocksize = DEV_BSIZE;
688 	vap->va_type = vp->v_type;
689 
690 #define fix(aa)							\
691 	do {							\
692 		if ((aa).tv_sec <= 3600) {			\
693 			(aa).tv_sec = boottime.tv_sec;		\
694 			(aa).tv_nsec = boottime.tv_usec * 1000; \
695 		}						\
696 	} while (0)
697 
698 	if (vp->v_type != VCHR)  {
699 		fix(de->de_atime);
700 		vap->va_atime = de->de_atime;
701 		fix(de->de_mtime);
702 		vap->va_mtime = de->de_mtime;
703 		fix(de->de_ctime);
704 		vap->va_ctime = de->de_ctime;
705 	} else {
706 		dev = vp->v_rdev;
707 		fix(dev->si_atime);
708 		vap->va_atime = dev->si_atime;
709 		fix(dev->si_mtime);
710 		vap->va_mtime = dev->si_mtime;
711 		fix(dev->si_ctime);
712 		vap->va_ctime = dev->si_ctime;
713 
714 		vap->va_rdev = cdev2priv(dev)->cdp_inode;
715 	}
716 	vap->va_gen = 0;
717 	vap->va_flags = 0;
718 	vap->va_filerev = 0;
719 	vap->va_nlink = de->de_links;
720 	vap->va_fileid = de->de_inode;
721 
722 	return (error);
723 }
724 
725 /* ARGSUSED */
726 static int
727 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
728 {
729 	struct cdev *dev;
730 	struct cdevsw *dsw;
731 	struct vnode *vp;
732 	struct vnode *vpold;
733 	int error, i, ref;
734 	const char *p;
735 	struct fiodgname_arg *fgn;
736 	struct file *fpop;
737 
738 	fpop = td->td_fpop;
739 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
740 	if (error)
741 		return (error);
742 
743 	if (com == FIODTYPE) {
744 		*(int *)data = dsw->d_flags & D_TYPEMASK;
745 		td->td_fpop = fpop;
746 		dev_relthread(dev, ref);
747 		return (0);
748 	} else if (com == FIODGNAME) {
749 		fgn = data;
750 		p = devtoname(dev);
751 		i = strlen(p) + 1;
752 		if (i > fgn->len)
753 			error = EINVAL;
754 		else
755 			error = copyout(p, fgn->buf, i);
756 		td->td_fpop = fpop;
757 		dev_relthread(dev, ref);
758 		return (error);
759 	}
760 	error = dsw->d_ioctl(dev, com, data, fp->f_flag, td);
761 	td->td_fpop = NULL;
762 	dev_relthread(dev, ref);
763 	if (error == ENOIOCTL)
764 		error = ENOTTY;
765 	if (error == 0 && com == TIOCSCTTY) {
766 		vp = fp->f_vnode;
767 
768 		/* Do nothing if reassigning same control tty */
769 		sx_slock(&proctree_lock);
770 		if (td->td_proc->p_session->s_ttyvp == vp) {
771 			sx_sunlock(&proctree_lock);
772 			return (0);
773 		}
774 
775 		vpold = td->td_proc->p_session->s_ttyvp;
776 		VREF(vp);
777 		SESS_LOCK(td->td_proc->p_session);
778 		td->td_proc->p_session->s_ttyvp = vp;
779 		td->td_proc->p_session->s_ttydp = cdev2priv(dev);
780 		SESS_UNLOCK(td->td_proc->p_session);
781 
782 		sx_sunlock(&proctree_lock);
783 
784 		/* Get rid of reference to old control tty */
785 		if (vpold)
786 			vrele(vpold);
787 	}
788 	return (error);
789 }
790 
791 /* ARGSUSED */
792 static int
793 devfs_kqfilter_f(struct file *fp, struct knote *kn)
794 {
795 	struct cdev *dev;
796 	struct cdevsw *dsw;
797 	int error, ref;
798 	struct file *fpop;
799 	struct thread *td;
800 
801 	td = curthread;
802 	fpop = td->td_fpop;
803 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
804 	if (error)
805 		return (error);
806 	error = dsw->d_kqfilter(dev, kn);
807 	td->td_fpop = fpop;
808 	dev_relthread(dev, ref);
809 	return (error);
810 }
811 
812 static inline int
813 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
814 {
815 	struct cdev_priv *cdp;
816 	struct ucred *dcr;
817 	int error;
818 
819 	cdp = de->de_cdp;
820 	if (cdp == NULL)
821 		return (0);
822 	dcr = cdp->cdp_c.si_cred;
823 	if (dcr == NULL)
824 		return (0);
825 
826 	error = prison_check(td->td_ucred, dcr);
827 	if (error == 0)
828 		return (0);
829 	/* We do, however, allow access to the controlling terminal */
830 	if (!(td->td_proc->p_flag & P_CONTROLT))
831 		return (error);
832 	if (td->td_proc->p_session->s_ttydp == cdp)
833 		return (0);
834 	return (error);
835 }
836 
837 static int
838 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
839 {
840 	struct componentname *cnp;
841 	struct vnode *dvp, **vpp;
842 	struct thread *td;
843 	struct devfs_dirent *de, *dd;
844 	struct devfs_dirent **dde;
845 	struct devfs_mount *dmp;
846 	struct cdev *cdev;
847 	int error, flags, nameiop, dvplocked;
848 	char specname[SPECNAMELEN + 1], *pname;
849 
850 	cnp = ap->a_cnp;
851 	vpp = ap->a_vpp;
852 	dvp = ap->a_dvp;
853 	pname = cnp->cn_nameptr;
854 	td = cnp->cn_thread;
855 	flags = cnp->cn_flags;
856 	nameiop = cnp->cn_nameiop;
857 	dmp = VFSTODEVFS(dvp->v_mount);
858 	dd = dvp->v_data;
859 	*vpp = NULLVP;
860 
861 	if ((flags & ISLASTCN) && nameiop == RENAME)
862 		return (EOPNOTSUPP);
863 
864 	if (dvp->v_type != VDIR)
865 		return (ENOTDIR);
866 
867 	if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
868 		return (EIO);
869 
870 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
871 	if (error)
872 		return (error);
873 
874 	if (cnp->cn_namelen == 1 && *pname == '.') {
875 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
876 			return (EINVAL);
877 		*vpp = dvp;
878 		VREF(dvp);
879 		return (0);
880 	}
881 
882 	if (flags & ISDOTDOT) {
883 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
884 			return (EINVAL);
885 		de = devfs_parent_dirent(dd);
886 		if (de == NULL)
887 			return (ENOENT);
888 		dvplocked = VOP_ISLOCKED(dvp);
889 		VOP_UNLOCK(dvp, 0);
890 		error = devfs_allocv(de, dvp->v_mount,
891 		    cnp->cn_lkflags & LK_TYPE_MASK, vpp);
892 		*dm_unlock = 0;
893 		vn_lock(dvp, dvplocked | LK_RETRY);
894 		return (error);
895 	}
896 
897 	dd = dvp->v_data;
898 	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
899 	while (de == NULL) {	/* While(...) so we can use break */
900 
901 		if (nameiop == DELETE)
902 			return (ENOENT);
903 
904 		/*
905 		 * OK, we didn't have an entry for the name we were asked for
906 		 * so we try to see if anybody can create it on demand.
907 		 */
908 		pname = devfs_fqpn(specname, dmp, dd, cnp);
909 		if (pname == NULL)
910 			break;
911 
912 		cdev = NULL;
913 		DEVFS_DMP_HOLD(dmp);
914 		sx_xunlock(&dmp->dm_lock);
915 		sx_slock(&clone_drain_lock);
916 		EVENTHANDLER_INVOKE(dev_clone,
917 		    td->td_ucred, pname, strlen(pname), &cdev);
918 		sx_sunlock(&clone_drain_lock);
919 
920 		if (cdev == NULL)
921 			sx_xlock(&dmp->dm_lock);
922 		else if (devfs_populate_vp(dvp) != 0) {
923 			*dm_unlock = 0;
924 			sx_xlock(&dmp->dm_lock);
925 			if (DEVFS_DMP_DROP(dmp)) {
926 				sx_xunlock(&dmp->dm_lock);
927 				devfs_unmount_final(dmp);
928 			} else
929 				sx_xunlock(&dmp->dm_lock);
930 			dev_rel(cdev);
931 			return (ENOENT);
932 		}
933 		if (DEVFS_DMP_DROP(dmp)) {
934 			*dm_unlock = 0;
935 			sx_xunlock(&dmp->dm_lock);
936 			devfs_unmount_final(dmp);
937 			if (cdev != NULL)
938 				dev_rel(cdev);
939 			return (ENOENT);
940 		}
941 
942 		if (cdev == NULL)
943 			break;
944 
945 		dev_lock();
946 		dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
947 		if (dde != NULL && *dde != NULL)
948 			de = *dde;
949 		dev_unlock();
950 		dev_rel(cdev);
951 		break;
952 	}
953 
954 	if (de == NULL || de->de_flags & DE_WHITEOUT) {
955 		if ((nameiop == CREATE || nameiop == RENAME) &&
956 		    (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
957 			cnp->cn_flags |= SAVENAME;
958 			return (EJUSTRETURN);
959 		}
960 		return (ENOENT);
961 	}
962 
963 	if (devfs_prison_check(de, td))
964 		return (ENOENT);
965 
966 	if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
967 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
968 		if (error)
969 			return (error);
970 		if (*vpp == dvp) {
971 			VREF(dvp);
972 			*vpp = dvp;
973 			return (0);
974 		}
975 	}
976 	error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK,
977 	    vpp);
978 	*dm_unlock = 0;
979 	return (error);
980 }
981 
982 static int
983 devfs_lookup(struct vop_lookup_args *ap)
984 {
985 	int j;
986 	struct devfs_mount *dmp;
987 	int dm_unlock;
988 
989 	if (devfs_populate_vp(ap->a_dvp) != 0)
990 		return (ENOTDIR);
991 
992 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
993 	dm_unlock = 1;
994 	j = devfs_lookupx(ap, &dm_unlock);
995 	if (dm_unlock == 1)
996 		sx_xunlock(&dmp->dm_lock);
997 	return (j);
998 }
999 
1000 static int
1001 devfs_mknod(struct vop_mknod_args *ap)
1002 {
1003 	struct componentname *cnp;
1004 	struct vnode *dvp, **vpp;
1005 	struct devfs_dirent *dd, *de;
1006 	struct devfs_mount *dmp;
1007 	int error;
1008 
1009 	/*
1010 	 * The only type of node we should be creating here is a
1011 	 * character device, for anything else return EOPNOTSUPP.
1012 	 */
1013 	if (ap->a_vap->va_type != VCHR)
1014 		return (EOPNOTSUPP);
1015 	dvp = ap->a_dvp;
1016 	dmp = VFSTODEVFS(dvp->v_mount);
1017 
1018 	cnp = ap->a_cnp;
1019 	vpp = ap->a_vpp;
1020 	dd = dvp->v_data;
1021 
1022 	error = ENOENT;
1023 	sx_xlock(&dmp->dm_lock);
1024 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1025 		if (cnp->cn_namelen != de->de_dirent->d_namlen)
1026 			continue;
1027 		if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1028 		    de->de_dirent->d_namlen) != 0)
1029 			continue;
1030 		if (de->de_flags & DE_WHITEOUT)
1031 			break;
1032 		goto notfound;
1033 	}
1034 	if (de == NULL)
1035 		goto notfound;
1036 	de->de_flags &= ~DE_WHITEOUT;
1037 	error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1038 	return (error);
1039 notfound:
1040 	sx_xunlock(&dmp->dm_lock);
1041 	return (error);
1042 }
1043 
1044 /* ARGSUSED */
1045 static int
1046 devfs_open(struct vop_open_args *ap)
1047 {
1048 	struct thread *td = ap->a_td;
1049 	struct vnode *vp = ap->a_vp;
1050 	struct cdev *dev = vp->v_rdev;
1051 	struct file *fp = ap->a_fp;
1052 	int error, ref, vlocked;
1053 	struct cdevsw *dsw;
1054 	struct file *fpop;
1055 	struct mtx *mtxp;
1056 
1057 	if (vp->v_type == VBLK)
1058 		return (ENXIO);
1059 
1060 	if (dev == NULL)
1061 		return (ENXIO);
1062 
1063 	/* Make this field valid before any I/O in d_open. */
1064 	if (dev->si_iosize_max == 0)
1065 		dev->si_iosize_max = DFLTPHYS;
1066 
1067 	dsw = dev_refthread(dev, &ref);
1068 	if (dsw == NULL)
1069 		return (ENXIO);
1070 	if (fp == NULL && dsw->d_fdopen != NULL) {
1071 		dev_relthread(dev, ref);
1072 		return (ENXIO);
1073 	}
1074 
1075 	vlocked = VOP_ISLOCKED(vp);
1076 	VOP_UNLOCK(vp, 0);
1077 
1078 	fpop = td->td_fpop;
1079 	td->td_fpop = fp;
1080 	if (fp != NULL) {
1081 		fp->f_data = dev;
1082 		fp->f_vnode = vp;
1083 	}
1084 	if (dsw->d_fdopen != NULL)
1085 		error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1086 	else
1087 		error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1088 	/* cleanup any cdevpriv upon error */
1089 	if (error != 0)
1090 		devfs_clear_cdevpriv();
1091 	td->td_fpop = fpop;
1092 
1093 	vn_lock(vp, vlocked | LK_RETRY);
1094 	dev_relthread(dev, ref);
1095 	if (error != 0) {
1096 		if (error == ERESTART)
1097 			error = EINTR;
1098 		return (error);
1099 	}
1100 
1101 #if 0	/* /dev/console */
1102 	KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1103 #else
1104 	if (fp == NULL)
1105 		return (error);
1106 #endif
1107 	if (fp->f_ops == &badfileops)
1108 		finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1109 	mtxp = mtx_pool_find(mtxpool_sleep, fp);
1110 
1111 	/*
1112 	 * Hint to the dofilewrite() to not force the buffer draining
1113 	 * on the writer to the file.  Most likely, the write would
1114 	 * not need normal buffers.
1115 	 */
1116 	mtx_lock(mtxp);
1117 	fp->f_vnread_flags |= FDEVFS_VNODE;
1118 	mtx_unlock(mtxp);
1119 	return (error);
1120 }
1121 
1122 static int
1123 devfs_pathconf(struct vop_pathconf_args *ap)
1124 {
1125 
1126 	switch (ap->a_name) {
1127 	case _PC_MAC_PRESENT:
1128 #ifdef MAC
1129 		/*
1130 		 * If MAC is enabled, devfs automatically supports
1131 		 * trivial non-persistant label storage.
1132 		 */
1133 		*ap->a_retval = 1;
1134 #else
1135 		*ap->a_retval = 0;
1136 #endif
1137 		return (0);
1138 	default:
1139 		return (vop_stdpathconf(ap));
1140 	}
1141 	/* NOTREACHED */
1142 }
1143 
1144 /* ARGSUSED */
1145 static int
1146 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1147 {
1148 	struct cdev *dev;
1149 	struct cdevsw *dsw;
1150 	int error, ref;
1151 	struct file *fpop;
1152 
1153 	fpop = td->td_fpop;
1154 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1155 	if (error)
1156 		return (poll_no_poll(events));
1157 	error = dsw->d_poll(dev, events, td);
1158 	td->td_fpop = fpop;
1159 	dev_relthread(dev, ref);
1160 	return(error);
1161 }
1162 
1163 /*
1164  * Print out the contents of a special device vnode.
1165  */
1166 static int
1167 devfs_print(struct vop_print_args *ap)
1168 {
1169 
1170 	printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1171 	return (0);
1172 }
1173 
1174 static int
1175 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1176     int flags, struct thread *td)
1177 {
1178 	struct cdev *dev;
1179 	int ioflag, error, ref;
1180 	ssize_t resid;
1181 	struct cdevsw *dsw;
1182 	struct file *fpop;
1183 
1184 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1185 		return (EINVAL);
1186 	fpop = td->td_fpop;
1187 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1188 	if (error)
1189 		return (error);
1190 	resid = uio->uio_resid;
1191 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1192 	if (ioflag & O_DIRECT)
1193 		ioflag |= IO_DIRECT;
1194 
1195 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1196 	error = dsw->d_read(dev, uio, ioflag);
1197 	if (uio->uio_resid != resid || (error == 0 && resid != 0))
1198 		vfs_timestamp(&dev->si_atime);
1199 	td->td_fpop = fpop;
1200 	dev_relthread(dev, ref);
1201 
1202 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1203 	return (error);
1204 }
1205 
1206 static int
1207 devfs_readdir(struct vop_readdir_args *ap)
1208 {
1209 	int error;
1210 	struct uio *uio;
1211 	struct dirent *dp;
1212 	struct devfs_dirent *dd;
1213 	struct devfs_dirent *de;
1214 	struct devfs_mount *dmp;
1215 	off_t off;
1216 	int *tmp_ncookies = NULL;
1217 
1218 	if (ap->a_vp->v_type != VDIR)
1219 		return (ENOTDIR);
1220 
1221 	uio = ap->a_uio;
1222 	if (uio->uio_offset < 0)
1223 		return (EINVAL);
1224 
1225 	/*
1226 	 * XXX: This is a temporary hack to get around this filesystem not
1227 	 * supporting cookies. We store the location of the ncookies pointer
1228 	 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1229 	 * and set the number of cookies to 0. We then set the pointer to
1230 	 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1231 	 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1232 	 * pointer to its original location before returning to the caller.
1233 	 */
1234 	if (ap->a_ncookies != NULL) {
1235 		tmp_ncookies = ap->a_ncookies;
1236 		*ap->a_ncookies = 0;
1237 		ap->a_ncookies = NULL;
1238 	}
1239 
1240 	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1241 	if (devfs_populate_vp(ap->a_vp) != 0) {
1242 		if (tmp_ncookies != NULL)
1243 			ap->a_ncookies = tmp_ncookies;
1244 		return (EIO);
1245 	}
1246 	error = 0;
1247 	de = ap->a_vp->v_data;
1248 	off = 0;
1249 	TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1250 		KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1251 		if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1252 			continue;
1253 		if (devfs_prison_check(dd, uio->uio_td))
1254 			continue;
1255 		if (dd->de_dirent->d_type == DT_DIR)
1256 			de = dd->de_dir;
1257 		else
1258 			de = dd;
1259 		dp = dd->de_dirent;
1260 		if (dp->d_reclen > uio->uio_resid)
1261 			break;
1262 		dp->d_fileno = de->de_inode;
1263 		if (off >= uio->uio_offset) {
1264 			error = vfs_read_dirent(ap, dp, off);
1265 			if (error)
1266 				break;
1267 		}
1268 		off += dp->d_reclen;
1269 	}
1270 	sx_xunlock(&dmp->dm_lock);
1271 	uio->uio_offset = off;
1272 
1273 	/*
1274 	 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1275 	 * place.
1276 	 */
1277 	if (tmp_ncookies != NULL)
1278 		ap->a_ncookies = tmp_ncookies;
1279 
1280 	return (error);
1281 }
1282 
1283 static int
1284 devfs_readlink(struct vop_readlink_args *ap)
1285 {
1286 	struct devfs_dirent *de;
1287 
1288 	de = ap->a_vp->v_data;
1289 	return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1290 }
1291 
1292 static int
1293 devfs_reclaim(struct vop_reclaim_args *ap)
1294 {
1295 	struct vnode *vp = ap->a_vp;
1296 	struct devfs_dirent *de;
1297 	struct cdev *dev;
1298 
1299 	mtx_lock(&devfs_de_interlock);
1300 	de = vp->v_data;
1301 	if (de != NULL) {
1302 		de->de_vnode = NULL;
1303 		vp->v_data = NULL;
1304 	}
1305 	mtx_unlock(&devfs_de_interlock);
1306 
1307 	vnode_destroy_vobject(vp);
1308 
1309 	VI_LOCK(vp);
1310 	dev_lock();
1311 	dev = vp->v_rdev;
1312 	vp->v_rdev = NULL;
1313 
1314 	if (dev == NULL) {
1315 		dev_unlock();
1316 		VI_UNLOCK(vp);
1317 		return (0);
1318 	}
1319 
1320 	dev->si_usecount -= vp->v_usecount;
1321 	dev_unlock();
1322 	VI_UNLOCK(vp);
1323 	dev_rel(dev);
1324 	return (0);
1325 }
1326 
1327 static int
1328 devfs_remove(struct vop_remove_args *ap)
1329 {
1330 	struct vnode *dvp = ap->a_dvp;
1331 	struct vnode *vp = ap->a_vp;
1332 	struct devfs_dirent *dd;
1333 	struct devfs_dirent *de, *de_covered;
1334 	struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1335 
1336 	ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1337 	ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1338 
1339 	sx_xlock(&dmp->dm_lock);
1340 	dd = ap->a_dvp->v_data;
1341 	de = vp->v_data;
1342 	if (de->de_cdp == NULL) {
1343 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1344 		if (de->de_dirent->d_type == DT_LNK) {
1345 			de_covered = devfs_find(dd, de->de_dirent->d_name,
1346 			    de->de_dirent->d_namlen, 0);
1347 			if (de_covered != NULL)
1348 				de_covered->de_flags &= ~DE_COVERED;
1349 		}
1350 		/* We need to unlock dvp because devfs_delete() may lock it. */
1351 		VOP_UNLOCK(vp, 0);
1352 		if (dvp != vp)
1353 			VOP_UNLOCK(dvp, 0);
1354 		devfs_delete(dmp, de, 0);
1355 		sx_xunlock(&dmp->dm_lock);
1356 		if (dvp != vp)
1357 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1358 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1359 	} else {
1360 		de->de_flags |= DE_WHITEOUT;
1361 		sx_xunlock(&dmp->dm_lock);
1362 	}
1363 	return (0);
1364 }
1365 
1366 /*
1367  * Revoke is called on a tty when a terminal session ends.  The vnode
1368  * is orphaned by setting v_op to deadfs so we need to let go of it
1369  * as well so that we create a new one next time around.
1370  *
1371  */
1372 static int
1373 devfs_revoke(struct vop_revoke_args *ap)
1374 {
1375 	struct vnode *vp = ap->a_vp, *vp2;
1376 	struct cdev *dev;
1377 	struct cdev_priv *cdp;
1378 	struct devfs_dirent *de;
1379 	int i;
1380 
1381 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1382 
1383 	dev = vp->v_rdev;
1384 	cdp = cdev2priv(dev);
1385 
1386 	dev_lock();
1387 	cdp->cdp_inuse++;
1388 	dev_unlock();
1389 
1390 	vhold(vp);
1391 	vgone(vp);
1392 	vdrop(vp);
1393 
1394 	VOP_UNLOCK(vp,0);
1395  loop:
1396 	for (;;) {
1397 		mtx_lock(&devfs_de_interlock);
1398 		dev_lock();
1399 		vp2 = NULL;
1400 		for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1401 			de = cdp->cdp_dirents[i];
1402 			if (de == NULL)
1403 				continue;
1404 
1405 			vp2 = de->de_vnode;
1406 			if (vp2 != NULL) {
1407 				dev_unlock();
1408 				VI_LOCK(vp2);
1409 				mtx_unlock(&devfs_de_interlock);
1410 				if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1411 				    curthread))
1412 					goto loop;
1413 				vhold(vp2);
1414 				vgone(vp2);
1415 				vdrop(vp2);
1416 				vput(vp2);
1417 				break;
1418 			}
1419 		}
1420 		if (vp2 != NULL) {
1421 			continue;
1422 		}
1423 		dev_unlock();
1424 		mtx_unlock(&devfs_de_interlock);
1425 		break;
1426 	}
1427 	dev_lock();
1428 	cdp->cdp_inuse--;
1429 	if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1430 		TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1431 		dev_unlock();
1432 		dev_rel(&cdp->cdp_c);
1433 	} else
1434 		dev_unlock();
1435 
1436 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1437 	return (0);
1438 }
1439 
1440 static int
1441 devfs_rioctl(struct vop_ioctl_args *ap)
1442 {
1443 	struct vnode *vp;
1444 	struct devfs_mount *dmp;
1445 	int error;
1446 
1447 	vp = ap->a_vp;
1448 	vn_lock(vp, LK_SHARED | LK_RETRY);
1449 	if (vp->v_iflag & VI_DOOMED) {
1450 		VOP_UNLOCK(vp, 0);
1451 		return (EBADF);
1452 	}
1453 	dmp = VFSTODEVFS(vp->v_mount);
1454 	sx_xlock(&dmp->dm_lock);
1455 	VOP_UNLOCK(vp, 0);
1456 	DEVFS_DMP_HOLD(dmp);
1457 	devfs_populate(dmp);
1458 	if (DEVFS_DMP_DROP(dmp)) {
1459 		sx_xunlock(&dmp->dm_lock);
1460 		devfs_unmount_final(dmp);
1461 		return (ENOENT);
1462 	}
1463 	error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1464 	sx_xunlock(&dmp->dm_lock);
1465 	return (error);
1466 }
1467 
1468 static int
1469 devfs_rread(struct vop_read_args *ap)
1470 {
1471 
1472 	if (ap->a_vp->v_type != VDIR)
1473 		return (EINVAL);
1474 	return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1475 }
1476 
1477 static int
1478 devfs_setattr(struct vop_setattr_args *ap)
1479 {
1480 	struct devfs_dirent *de;
1481 	struct vattr *vap;
1482 	struct vnode *vp;
1483 	struct thread *td;
1484 	int c, error;
1485 	uid_t uid;
1486 	gid_t gid;
1487 
1488 	vap = ap->a_vap;
1489 	vp = ap->a_vp;
1490 	td = curthread;
1491 	if ((vap->va_type != VNON) ||
1492 	    (vap->va_nlink != VNOVAL) ||
1493 	    (vap->va_fsid != VNOVAL) ||
1494 	    (vap->va_fileid != VNOVAL) ||
1495 	    (vap->va_blocksize != VNOVAL) ||
1496 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1497 	    (vap->va_rdev != VNOVAL) ||
1498 	    ((int)vap->va_bytes != VNOVAL) ||
1499 	    (vap->va_gen != VNOVAL)) {
1500 		return (EINVAL);
1501 	}
1502 
1503 	de = vp->v_data;
1504 	if (vp->v_type == VDIR)
1505 		de = de->de_dir;
1506 
1507 	error = c = 0;
1508 	if (vap->va_uid == (uid_t)VNOVAL)
1509 		uid = de->de_uid;
1510 	else
1511 		uid = vap->va_uid;
1512 	if (vap->va_gid == (gid_t)VNOVAL)
1513 		gid = de->de_gid;
1514 	else
1515 		gid = vap->va_gid;
1516 	if (uid != de->de_uid || gid != de->de_gid) {
1517 		if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1518 		    (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1519 			error = priv_check(td, PRIV_VFS_CHOWN);
1520 			if (error)
1521 				return (error);
1522 		}
1523 		de->de_uid = uid;
1524 		de->de_gid = gid;
1525 		c = 1;
1526 	}
1527 
1528 	if (vap->va_mode != (mode_t)VNOVAL) {
1529 		if (ap->a_cred->cr_uid != de->de_uid) {
1530 			error = priv_check(td, PRIV_VFS_ADMIN);
1531 			if (error)
1532 				return (error);
1533 		}
1534 		de->de_mode = vap->va_mode;
1535 		c = 1;
1536 	}
1537 
1538 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1539 		error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1540 		if (error != 0)
1541 			return (error);
1542 		if (vap->va_atime.tv_sec != VNOVAL) {
1543 			if (vp->v_type == VCHR)
1544 				vp->v_rdev->si_atime = vap->va_atime;
1545 			else
1546 				de->de_atime = vap->va_atime;
1547 		}
1548 		if (vap->va_mtime.tv_sec != VNOVAL) {
1549 			if (vp->v_type == VCHR)
1550 				vp->v_rdev->si_mtime = vap->va_mtime;
1551 			else
1552 				de->de_mtime = vap->va_mtime;
1553 		}
1554 		c = 1;
1555 	}
1556 
1557 	if (c) {
1558 		if (vp->v_type == VCHR)
1559 			vfs_timestamp(&vp->v_rdev->si_ctime);
1560 		else
1561 			vfs_timestamp(&de->de_mtime);
1562 	}
1563 	return (0);
1564 }
1565 
1566 #ifdef MAC
1567 static int
1568 devfs_setlabel(struct vop_setlabel_args *ap)
1569 {
1570 	struct vnode *vp;
1571 	struct devfs_dirent *de;
1572 
1573 	vp = ap->a_vp;
1574 	de = vp->v_data;
1575 
1576 	mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1577 	mac_devfs_update(vp->v_mount, de, vp);
1578 
1579 	return (0);
1580 }
1581 #endif
1582 
1583 static int
1584 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1585 {
1586 
1587 	return (vnops.fo_stat(fp, sb, cred, td));
1588 }
1589 
1590 static int
1591 devfs_symlink(struct vop_symlink_args *ap)
1592 {
1593 	int i, error;
1594 	struct devfs_dirent *dd;
1595 	struct devfs_dirent *de, *de_covered, *de_dotdot;
1596 	struct devfs_mount *dmp;
1597 
1598 	error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1599 	if (error)
1600 		return(error);
1601 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1602 	if (devfs_populate_vp(ap->a_dvp) != 0)
1603 		return (ENOENT);
1604 
1605 	dd = ap->a_dvp->v_data;
1606 	de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1607 	de->de_flags = DE_USER;
1608 	de->de_uid = 0;
1609 	de->de_gid = 0;
1610 	de->de_mode = 0755;
1611 	de->de_inode = alloc_unr(devfs_inos);
1612 	de->de_dir = dd;
1613 	de->de_dirent->d_type = DT_LNK;
1614 	i = strlen(ap->a_target) + 1;
1615 	de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1616 	bcopy(ap->a_target, de->de_symlink, i);
1617 #ifdef MAC
1618 	mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1619 #endif
1620 	de_covered = devfs_find(dd, de->de_dirent->d_name,
1621 	    de->de_dirent->d_namlen, 0);
1622 	if (de_covered != NULL) {
1623 		if ((de_covered->de_flags & DE_USER) != 0) {
1624 			devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1625 			sx_xunlock(&dmp->dm_lock);
1626 			return (EEXIST);
1627 		}
1628 		KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1629 		    ("devfs_symlink: entry %p already covered", de_covered));
1630 		de_covered->de_flags |= DE_COVERED;
1631 	}
1632 
1633 	de_dotdot = TAILQ_FIRST(&dd->de_dlist);		/* "." */
1634 	de_dotdot = TAILQ_NEXT(de_dotdot, de_list);	/* ".." */
1635 	TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1636 	devfs_dir_ref_de(dmp, dd);
1637 	devfs_rules_apply(dmp, de);
1638 
1639 	return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1640 }
1641 
1642 static int
1643 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1644 {
1645 
1646 	return (vnops.fo_truncate(fp, length, cred, td));
1647 }
1648 
1649 static int
1650 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1651     int flags, struct thread *td)
1652 {
1653 	struct cdev *dev;
1654 	int error, ioflag, ref;
1655 	ssize_t resid;
1656 	struct cdevsw *dsw;
1657 	struct file *fpop;
1658 
1659 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1660 		return (EINVAL);
1661 	fpop = td->td_fpop;
1662 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1663 	if (error)
1664 		return (error);
1665 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1666 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1667 	if (ioflag & O_DIRECT)
1668 		ioflag |= IO_DIRECT;
1669 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1670 
1671 	resid = uio->uio_resid;
1672 
1673 	error = dsw->d_write(dev, uio, ioflag);
1674 	if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1675 		vfs_timestamp(&dev->si_ctime);
1676 		dev->si_mtime = dev->si_ctime;
1677 	}
1678 	td->td_fpop = fpop;
1679 	dev_relthread(dev, ref);
1680 
1681 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1682 	return (error);
1683 }
1684 
1685 dev_t
1686 dev2udev(struct cdev *x)
1687 {
1688 	if (x == NULL)
1689 		return (NODEV);
1690 	return (cdev2priv(x)->cdp_inode);
1691 }
1692 
1693 static struct fileops devfs_ops_f = {
1694 	.fo_read =	devfs_read_f,
1695 	.fo_write =	devfs_write_f,
1696 	.fo_truncate =	devfs_truncate_f,
1697 	.fo_ioctl =	devfs_ioctl_f,
1698 	.fo_poll =	devfs_poll_f,
1699 	.fo_kqfilter =	devfs_kqfilter_f,
1700 	.fo_stat =	devfs_stat_f,
1701 	.fo_close =	devfs_close_f,
1702 	.fo_chmod =	vn_chmod,
1703 	.fo_chown =	vn_chown,
1704 	.fo_sendfile =	vn_sendfile,
1705 	.fo_seek =	vn_seek,
1706 	.fo_fill_kinfo = vn_fill_kinfo,
1707 	.fo_flags =	DFLAG_PASSABLE | DFLAG_SEEKABLE
1708 };
1709 
1710 static struct vop_vector devfs_vnodeops = {
1711 	.vop_default =		&default_vnodeops,
1712 
1713 	.vop_access =		devfs_access,
1714 	.vop_getattr =		devfs_getattr,
1715 	.vop_ioctl =		devfs_rioctl,
1716 	.vop_lookup =		devfs_lookup,
1717 	.vop_mknod =		devfs_mknod,
1718 	.vop_pathconf =		devfs_pathconf,
1719 	.vop_read =		devfs_rread,
1720 	.vop_readdir =		devfs_readdir,
1721 	.vop_readlink =		devfs_readlink,
1722 	.vop_reclaim =		devfs_reclaim,
1723 	.vop_remove =		devfs_remove,
1724 	.vop_revoke =		devfs_revoke,
1725 	.vop_setattr =		devfs_setattr,
1726 #ifdef MAC
1727 	.vop_setlabel =		devfs_setlabel,
1728 #endif
1729 	.vop_symlink =		devfs_symlink,
1730 	.vop_vptocnp =		devfs_vptocnp,
1731 };
1732 
1733 static struct vop_vector devfs_specops = {
1734 	.vop_default =		&default_vnodeops,
1735 
1736 	.vop_access =		devfs_access,
1737 	.vop_bmap =		VOP_PANIC,
1738 	.vop_close =		devfs_close,
1739 	.vop_create =		VOP_PANIC,
1740 	.vop_fsync =		devfs_fsync,
1741 	.vop_getattr =		devfs_getattr,
1742 	.vop_link =		VOP_PANIC,
1743 	.vop_mkdir =		VOP_PANIC,
1744 	.vop_mknod =		VOP_PANIC,
1745 	.vop_open =		devfs_open,
1746 	.vop_pathconf =		devfs_pathconf,
1747 	.vop_print =		devfs_print,
1748 	.vop_read =		VOP_PANIC,
1749 	.vop_readdir =		VOP_PANIC,
1750 	.vop_readlink =		VOP_PANIC,
1751 	.vop_reallocblks =	VOP_PANIC,
1752 	.vop_reclaim =		devfs_reclaim,
1753 	.vop_remove =		devfs_remove,
1754 	.vop_rename =		VOP_PANIC,
1755 	.vop_revoke =		devfs_revoke,
1756 	.vop_rmdir =		VOP_PANIC,
1757 	.vop_setattr =		devfs_setattr,
1758 #ifdef MAC
1759 	.vop_setlabel =		devfs_setlabel,
1760 #endif
1761 	.vop_strategy =		VOP_PANIC,
1762 	.vop_symlink =		VOP_PANIC,
1763 	.vop_vptocnp =		devfs_vptocnp,
1764 	.vop_write =		VOP_PANIC,
1765 };
1766 
1767 /*
1768  * Our calling convention to the device drivers used to be that we passed
1769  * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1770  * flags instead since that's what open(), close() and ioctl() takes and
1771  * we don't really want vnode.h in device drivers.
1772  * We solved the source compatibility by redefining some vnode flags to
1773  * be the same as the fcntl ones and by sending down the bitwise OR of
1774  * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
1775  * pulls the rug out under this.
1776  */
1777 CTASSERT(O_NONBLOCK == IO_NDELAY);
1778 CTASSERT(O_FSYNC == IO_SYNC);
1779