devfs_vnops.c revision 193511
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: head/sys/fs/devfs/devfs_vnops.c 193511 2009-06-05 14:55:22Z rwatson $
35 */
36
37/*
38 * TODO:
39 *	remove empty directories
40 *	mkdir: want it ?
41 */
42
43#include <sys/param.h>
44#include <sys/systm.h>
45#include <sys/conf.h>
46#include <sys/dirent.h>
47#include <sys/fcntl.h>
48#include <sys/file.h>
49#include <sys/filedesc.h>
50#include <sys/filio.h>
51#include <sys/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
65static struct vop_vector devfs_vnodeops;
66static struct vop_vector devfs_specops;
67static 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
74static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
75
76struct mtx	devfs_de_interlock;
77MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
78struct sx	clone_drain_lock;
79SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
80struct mtx	cdevpriv_mtx;
81MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
82
83static int
84devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp)
85{
86
87	*dswp = devvn_refthread(fp->f_vnode, devp);
88	if (*devp != fp->f_data) {
89		if (*dswp != NULL)
90			dev_relthread(*devp);
91		return (ENXIO);
92	}
93	KASSERT((*devp)->si_refcount > 0,
94	    ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
95	if (*dswp == NULL)
96		return (ENXIO);
97	curthread->td_fpop = fp;
98	return (0);
99}
100
101int
102devfs_get_cdevpriv(void **datap)
103{
104	struct file *fp;
105	struct cdev_privdata *p;
106	int error;
107
108	fp = curthread->td_fpop;
109	if (fp == NULL)
110		return (EBADF);
111	p = fp->f_cdevpriv;
112	if (p != NULL) {
113		error = 0;
114		*datap = p->cdpd_data;
115	} else
116		error = ENOENT;
117	return (error);
118}
119
120int
121devfs_set_cdevpriv(void *priv, cdevpriv_dtr_t priv_dtr)
122{
123	struct file *fp;
124	struct cdev_priv *cdp;
125	struct cdev_privdata *p;
126	int error;
127
128	fp = curthread->td_fpop;
129	if (fp == NULL)
130		return (ENOENT);
131	cdp = cdev2priv((struct cdev *)fp->f_data);
132	p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
133	p->cdpd_data = priv;
134	p->cdpd_dtr = priv_dtr;
135	p->cdpd_fp = fp;
136	mtx_lock(&cdevpriv_mtx);
137	if (fp->f_cdevpriv == NULL) {
138		LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
139		fp->f_cdevpriv = p;
140		mtx_unlock(&cdevpriv_mtx);
141		error = 0;
142	} else {
143		mtx_unlock(&cdevpriv_mtx);
144		free(p, M_CDEVPDATA);
145		error = EBUSY;
146	}
147	return (error);
148}
149
150void
151devfs_destroy_cdevpriv(struct cdev_privdata *p)
152{
153
154	mtx_assert(&cdevpriv_mtx, MA_OWNED);
155	p->cdpd_fp->f_cdevpriv = NULL;
156	LIST_REMOVE(p, cdpd_list);
157	mtx_unlock(&cdevpriv_mtx);
158	(p->cdpd_dtr)(p->cdpd_data);
159	free(p, M_CDEVPDATA);
160}
161
162void
163devfs_fpdrop(struct file *fp)
164{
165	struct cdev_privdata *p;
166
167	mtx_lock(&cdevpriv_mtx);
168	if ((p = fp->f_cdevpriv) == NULL) {
169		mtx_unlock(&cdevpriv_mtx);
170		return;
171	}
172	devfs_destroy_cdevpriv(p);
173}
174
175void
176devfs_clear_cdevpriv(void)
177{
178	struct file *fp;
179
180	fp = curthread->td_fpop;
181	if (fp == NULL)
182		return;
183	devfs_fpdrop(fp);
184}
185
186static int
187devfs_vptocnp(struct vop_vptocnp_args *ap)
188{
189	struct vnode *vp = ap->a_vp;
190	struct vnode **dvp = ap->a_vpp;
191	struct devfs_mount *dmp;
192	char *buf = ap->a_buf;
193	int *buflen = ap->a_buflen;
194	struct devfs_dirent *dd, *de;
195	int i, error;
196
197	dmp = VFSTODEVFS(vp->v_mount);
198	i = *buflen;
199	dd = vp->v_data;
200	error = 0;
201
202	sx_xlock(&dmp->dm_lock);
203
204	if (vp->v_type == VCHR) {
205		i -= strlen(dd->de_cdp->cdp_c.si_name);
206		if (i < 0) {
207			error = ENOMEM;
208			goto finished;
209		}
210		bcopy(dd->de_cdp->cdp_c.si_name, buf + i,
211		    strlen(dd->de_cdp->cdp_c.si_name));
212		de = dd->de_dir;
213	} else if (vp->v_type == VDIR) {
214		if (dd == dmp->dm_rootdir) {
215			*dvp = vp;
216			vhold(*dvp);
217			goto finished;
218		}
219		i -= dd->de_dirent->d_namlen;
220		if (i < 0) {
221			error = ENOMEM;
222			goto finished;
223		}
224		bcopy(dd->de_dirent->d_name, buf + i,
225		    dd->de_dirent->d_namlen);
226		de = dd;
227	} else {
228		error = ENOENT;
229		goto finished;
230	}
231	*buflen = i;
232	de = TAILQ_FIRST(&de->de_dlist);	/* "." */
233	de = TAILQ_NEXT(de, de_list);		/* ".." */
234	de = de->de_dir;
235	mtx_lock(&devfs_de_interlock);
236	*dvp = de->de_vnode;
237	if (*dvp != NULL) {
238		VI_LOCK(*dvp);
239		mtx_unlock(&devfs_de_interlock);
240		vholdl(*dvp);
241		VI_UNLOCK(*dvp);
242	} else {
243		mtx_unlock(&devfs_de_interlock);
244		error = ENOENT;
245	}
246finished:
247	sx_xunlock(&dmp->dm_lock);
248	return (error);
249}
250
251/*
252 * Construct the fully qualified path name relative to the mountpoint
253 */
254static char *
255devfs_fqpn(char *buf, struct vnode *dvp, struct componentname *cnp)
256{
257	int i;
258	struct devfs_dirent *de, *dd;
259	struct devfs_mount *dmp;
260
261	dmp = VFSTODEVFS(dvp->v_mount);
262	dd = dvp->v_data;
263	i = SPECNAMELEN;
264	buf[i] = '\0';
265	i -= cnp->cn_namelen;
266	if (i < 0)
267		 return (NULL);
268	bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
269	de = dd;
270	while (de != dmp->dm_rootdir) {
271		i--;
272		if (i < 0)
273			 return (NULL);
274		buf[i] = '/';
275		i -= de->de_dirent->d_namlen;
276		if (i < 0)
277			 return (NULL);
278		bcopy(de->de_dirent->d_name, buf + i,
279		    de->de_dirent->d_namlen);
280		de = TAILQ_FIRST(&de->de_dlist);	/* "." */
281		de = TAILQ_NEXT(de, de_list);		/* ".." */
282		de = de->de_dir;
283	}
284	return (buf + i);
285}
286
287static int
288devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
289	struct devfs_dirent *de)
290{
291	int not_found;
292
293	not_found = 0;
294	if (de->de_flags & DE_DOOMED)
295		not_found = 1;
296	if (DEVFS_DE_DROP(de)) {
297		KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
298		devfs_dirent_free(de);
299	}
300	if (DEVFS_DMP_DROP(dmp)) {
301		KASSERT(not_found == 1,
302			("DEVFS mount struct freed before dirent"));
303		not_found = 2;
304		sx_xunlock(&dmp->dm_lock);
305		devfs_unmount_final(dmp);
306	}
307	if (not_found == 1 || (drop_dm_lock && not_found != 2))
308		sx_unlock(&dmp->dm_lock);
309	return (not_found);
310}
311
312static void
313devfs_insmntque_dtr(struct vnode *vp, void *arg)
314{
315	struct devfs_dirent *de;
316
317	de = (struct devfs_dirent *)arg;
318	mtx_lock(&devfs_de_interlock);
319	vp->v_data = NULL;
320	de->de_vnode = NULL;
321	mtx_unlock(&devfs_de_interlock);
322	vgone(vp);
323	vput(vp);
324}
325
326/*
327 * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
328 * it on return.
329 */
330int
331devfs_allocv(struct devfs_dirent *de, struct mount *mp, struct vnode **vpp)
332{
333	int error;
334	struct vnode *vp;
335	struct cdev *dev;
336	struct devfs_mount *dmp;
337
338	dmp = VFSTODEVFS(mp);
339	if (de->de_flags & DE_DOOMED) {
340		sx_xunlock(&dmp->dm_lock);
341		return (ENOENT);
342	}
343	DEVFS_DE_HOLD(de);
344	DEVFS_DMP_HOLD(dmp);
345	mtx_lock(&devfs_de_interlock);
346	vp = de->de_vnode;
347	if (vp != NULL) {
348		VI_LOCK(vp);
349		mtx_unlock(&devfs_de_interlock);
350		sx_xunlock(&dmp->dm_lock);
351		error = vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, curthread);
352		sx_xlock(&dmp->dm_lock);
353		if (devfs_allocv_drop_refs(0, dmp, de)) {
354			if (error == 0)
355				vput(vp);
356			return (ENOENT);
357		}
358		else if (error) {
359			sx_xunlock(&dmp->dm_lock);
360			return (error);
361		}
362		sx_xunlock(&dmp->dm_lock);
363		*vpp = vp;
364		return (0);
365	}
366	mtx_unlock(&devfs_de_interlock);
367	if (de->de_dirent->d_type == DT_CHR) {
368		if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
369			devfs_allocv_drop_refs(1, dmp, de);
370			return (ENOENT);
371		}
372		dev = &de->de_cdp->cdp_c;
373	} else {
374		dev = NULL;
375	}
376	error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
377	if (error != 0) {
378		devfs_allocv_drop_refs(1, dmp, de);
379		printf("devfs_allocv: failed to allocate new vnode\n");
380		return (error);
381	}
382
383	if (de->de_dirent->d_type == DT_CHR) {
384		vp->v_type = VCHR;
385		VI_LOCK(vp);
386		dev_lock();
387		dev_refl(dev);
388		/* XXX: v_rdev should be protect by vnode lock */
389		vp->v_rdev = dev;
390		KASSERT(vp->v_usecount == 1,
391		    ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount));
392		dev->si_usecount += vp->v_usecount;
393		dev_unlock();
394		VI_UNLOCK(vp);
395		vp->v_op = &devfs_specops;
396	} else if (de->de_dirent->d_type == DT_DIR) {
397		vp->v_type = VDIR;
398	} else if (de->de_dirent->d_type == DT_LNK) {
399		vp->v_type = VLNK;
400	} else {
401		vp->v_type = VBAD;
402	}
403	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
404	mtx_lock(&devfs_de_interlock);
405	vp->v_data = de;
406	de->de_vnode = vp;
407	mtx_unlock(&devfs_de_interlock);
408	error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
409	if (error != 0) {
410		(void) devfs_allocv_drop_refs(1, dmp, de);
411		return (error);
412	}
413	if (devfs_allocv_drop_refs(0, dmp, de)) {
414		vput(vp);
415		return (ENOENT);
416	}
417#ifdef MAC
418	mac_devfs_vnode_associate(mp, de, vp);
419#endif
420	sx_xunlock(&dmp->dm_lock);
421	*vpp = vp;
422	return (0);
423}
424
425static int
426devfs_access(struct vop_access_args *ap)
427{
428	struct vnode *vp = ap->a_vp;
429	struct devfs_dirent *de;
430	int error;
431
432	de = vp->v_data;
433	if (vp->v_type == VDIR)
434		de = de->de_dir;
435
436	error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
437	    ap->a_accmode, ap->a_cred, NULL);
438	if (!error)
439		return (error);
440	if (error != EACCES)
441		return (error);
442	/* We do, however, allow access to the controlling terminal */
443	if (!(ap->a_td->td_proc->p_flag & P_CONTROLT))
444		return (error);
445	if (ap->a_td->td_proc->p_session->s_ttyvp == de->de_vnode)
446		return (0);
447	return (error);
448}
449
450/* ARGSUSED */
451static int
452devfs_close(struct vop_close_args *ap)
453{
454	struct vnode *vp = ap->a_vp, *oldvp;
455	struct thread *td = ap->a_td;
456	struct cdev *dev = vp->v_rdev;
457	struct cdevsw *dsw;
458	int vp_locked, error;
459
460	/*
461	 * Hack: a tty device that is a controlling terminal
462	 * has a reference from the session structure.
463	 * We cannot easily tell that a character device is
464	 * a controlling terminal, unless it is the closing
465	 * process' controlling terminal.  In that case,
466	 * if the reference count is 2 (this last descriptor
467	 * plus the session), release the reference from the session.
468	 */
469	oldvp = NULL;
470	sx_xlock(&proctree_lock);
471	if (td && vp == td->td_proc->p_session->s_ttyvp) {
472		SESS_LOCK(td->td_proc->p_session);
473		VI_LOCK(vp);
474		if (count_dev(dev) == 2 && (vp->v_iflag & VI_DOOMED) == 0) {
475			td->td_proc->p_session->s_ttyvp = NULL;
476			oldvp = vp;
477		}
478		VI_UNLOCK(vp);
479		SESS_UNLOCK(td->td_proc->p_session);
480	}
481	sx_xunlock(&proctree_lock);
482	if (oldvp != NULL)
483		vrele(oldvp);
484	/*
485	 * We do not want to really close the device if it
486	 * is still in use unless we are trying to close it
487	 * forcibly. Since every use (buffer, vnode, swap, cmap)
488	 * holds a reference to the vnode, and because we mark
489	 * any other vnodes that alias this device, when the
490	 * sum of the reference counts on all the aliased
491	 * vnodes descends to one, we are on last close.
492	 */
493	dsw = dev_refthread(dev);
494	if (dsw == NULL)
495		return (ENXIO);
496	VI_LOCK(vp);
497	if (vp->v_iflag & VI_DOOMED) {
498		/* Forced close. */
499	} else if (dsw->d_flags & D_TRACKCLOSE) {
500		/* Keep device updated on status. */
501	} else if (count_dev(dev) > 1) {
502		VI_UNLOCK(vp);
503		dev_relthread(dev);
504		return (0);
505	}
506	vholdl(vp);
507	VI_UNLOCK(vp);
508	vp_locked = VOP_ISLOCKED(vp);
509	VOP_UNLOCK(vp, 0);
510	KASSERT(dev->si_refcount > 0,
511	    ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
512	error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td);
513	dev_relthread(dev);
514	vn_lock(vp, vp_locked | LK_RETRY);
515	vdrop(vp);
516	return (error);
517}
518
519static int
520devfs_close_f(struct file *fp, struct thread *td)
521{
522	int error;
523	struct file *fpop;
524
525	fpop = td->td_fpop;
526	td->td_fpop = fp;
527	error = vnops.fo_close(fp, td);
528	td->td_fpop = fpop;
529	return (error);
530}
531
532static int
533devfs_fsync(struct vop_fsync_args *ap)
534{
535	int error;
536	struct bufobj *bo;
537	struct devfs_dirent *de;
538
539	if (!vn_isdisk(ap->a_vp, &error)) {
540		bo = &ap->a_vp->v_bufobj;
541		de = ap->a_vp->v_data;
542		if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) {
543			printf("Device %s went missing before all of the data "
544			    "could be written to it; expect data loss.\n",
545			    de->de_dirent->d_name);
546
547			error = vop_stdfsync(ap);
548			if (bo->bo_dirty.bv_cnt != 0 || error != 0)
549				panic("devfs_fsync: vop_stdfsync failed.");
550		}
551
552		return (0);
553	}
554
555	return (vop_stdfsync(ap));
556}
557
558static int
559devfs_getattr(struct vop_getattr_args *ap)
560{
561	struct vnode *vp = ap->a_vp;
562	struct vattr *vap = ap->a_vap;
563	int error = 0;
564	struct devfs_dirent *de;
565	struct cdev *dev;
566
567	de = vp->v_data;
568	KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
569	if (vp->v_type == VDIR) {
570		de = de->de_dir;
571		KASSERT(de != NULL,
572		    ("Null dir dirent in devfs_getattr vp=%p", vp));
573	}
574	vap->va_uid = de->de_uid;
575	vap->va_gid = de->de_gid;
576	vap->va_mode = de->de_mode;
577	if (vp->v_type == VLNK)
578		vap->va_size = strlen(de->de_symlink);
579	else if (vp->v_type == VDIR)
580		vap->va_size = vap->va_bytes = DEV_BSIZE;
581	else
582		vap->va_size = 0;
583	if (vp->v_type != VDIR)
584		vap->va_bytes = 0;
585	vap->va_blocksize = DEV_BSIZE;
586	vap->va_type = vp->v_type;
587
588#define fix(aa)							\
589	do {							\
590		if ((aa).tv_sec <= 3600) {			\
591			(aa).tv_sec = boottime.tv_sec;		\
592			(aa).tv_nsec = boottime.tv_usec * 1000; \
593		}						\
594	} while (0)
595
596	if (vp->v_type != VCHR)  {
597		fix(de->de_atime);
598		vap->va_atime = de->de_atime;
599		fix(de->de_mtime);
600		vap->va_mtime = de->de_mtime;
601		fix(de->de_ctime);
602		vap->va_ctime = de->de_ctime;
603	} else {
604		dev = vp->v_rdev;
605		fix(dev->si_atime);
606		vap->va_atime = dev->si_atime;
607		fix(dev->si_mtime);
608		vap->va_mtime = dev->si_mtime;
609		fix(dev->si_ctime);
610		vap->va_ctime = dev->si_ctime;
611
612		vap->va_rdev = cdev2priv(dev)->cdp_inode;
613	}
614	vap->va_gen = 0;
615	vap->va_flags = 0;
616	vap->va_filerev = 0;
617	vap->va_nlink = de->de_links;
618	vap->va_fileid = de->de_inode;
619
620	return (error);
621}
622
623/* ARGSUSED */
624static int
625devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
626{
627	struct cdev *dev;
628	struct cdevsw *dsw;
629	struct vnode *vp;
630	struct vnode *vpold;
631	int error, i;
632	const char *p;
633	struct fiodgname_arg *fgn;
634	struct file *fpop;
635
636	fpop = td->td_fpop;
637	error = devfs_fp_check(fp, &dev, &dsw);
638	if (error)
639		return (error);
640
641	if (com == FIODTYPE) {
642		*(int *)data = dsw->d_flags & D_TYPEMASK;
643		td->td_fpop = fpop;
644		dev_relthread(dev);
645		return (0);
646	} else if (com == FIODGNAME) {
647		fgn = data;
648		p = devtoname(dev);
649		i = strlen(p) + 1;
650		if (i > fgn->len)
651			error = EINVAL;
652		else
653			error = copyout(p, fgn->buf, i);
654		td->td_fpop = fpop;
655		dev_relthread(dev);
656		return (error);
657	}
658	error = dsw->d_ioctl(dev, com, data, fp->f_flag, td);
659	td->td_fpop = NULL;
660	dev_relthread(dev);
661	if (error == ENOIOCTL)
662		error = ENOTTY;
663	if (error == 0 && com == TIOCSCTTY) {
664		vp = fp->f_vnode;
665
666		/* Do nothing if reassigning same control tty */
667		sx_slock(&proctree_lock);
668		if (td->td_proc->p_session->s_ttyvp == vp) {
669			sx_sunlock(&proctree_lock);
670			return (0);
671		}
672
673		vpold = td->td_proc->p_session->s_ttyvp;
674		VREF(vp);
675		SESS_LOCK(td->td_proc->p_session);
676		td->td_proc->p_session->s_ttyvp = vp;
677		SESS_UNLOCK(td->td_proc->p_session);
678
679		sx_sunlock(&proctree_lock);
680
681		/* Get rid of reference to old control tty */
682		if (vpold)
683			vrele(vpold);
684	}
685	return (error);
686}
687
688/* ARGSUSED */
689static int
690devfs_kqfilter_f(struct file *fp, struct knote *kn)
691{
692	struct cdev *dev;
693	struct cdevsw *dsw;
694	int error;
695	struct file *fpop;
696	struct thread *td;
697
698	td = curthread;
699	fpop = td->td_fpop;
700	error = devfs_fp_check(fp, &dev, &dsw);
701	if (error)
702		return (error);
703	error = dsw->d_kqfilter(dev, kn);
704	td->td_fpop = fpop;
705	dev_relthread(dev);
706	return (error);
707}
708
709static int
710devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
711{
712	struct componentname *cnp;
713	struct vnode *dvp, **vpp;
714	struct thread *td;
715	struct devfs_dirent *de, *dd;
716	struct devfs_dirent **dde;
717	struct devfs_mount *dmp;
718	struct cdev *cdev;
719	int error, flags, nameiop;
720	char specname[SPECNAMELEN + 1], *pname;
721
722	cnp = ap->a_cnp;
723	vpp = ap->a_vpp;
724	dvp = ap->a_dvp;
725	pname = cnp->cn_nameptr;
726	td = cnp->cn_thread;
727	flags = cnp->cn_flags;
728	nameiop = cnp->cn_nameiop;
729	dmp = VFSTODEVFS(dvp->v_mount);
730	dd = dvp->v_data;
731	*vpp = NULLVP;
732
733	if ((flags & ISLASTCN) && nameiop == RENAME)
734		return (EOPNOTSUPP);
735
736	if (dvp->v_type != VDIR)
737		return (ENOTDIR);
738
739	if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
740		return (EIO);
741
742	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
743	if (error)
744		return (error);
745
746	if (cnp->cn_namelen == 1 && *pname == '.') {
747		if ((flags & ISLASTCN) && nameiop != LOOKUP)
748			return (EINVAL);
749		*vpp = dvp;
750		VREF(dvp);
751		return (0);
752	}
753
754	if (flags & ISDOTDOT) {
755		if ((flags & ISLASTCN) && nameiop != LOOKUP)
756			return (EINVAL);
757		VOP_UNLOCK(dvp, 0);
758		de = TAILQ_FIRST(&dd->de_dlist);	/* "." */
759		de = TAILQ_NEXT(de, de_list);		/* ".." */
760		de = de->de_dir;
761		error = devfs_allocv(de, dvp->v_mount, vpp);
762		*dm_unlock = 0;
763		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
764		return (error);
765	}
766
767	DEVFS_DMP_HOLD(dmp);
768	devfs_populate(dmp);
769	if (DEVFS_DMP_DROP(dmp)) {
770		*dm_unlock = 0;
771		sx_xunlock(&dmp->dm_lock);
772		devfs_unmount_final(dmp);
773		return (ENOENT);
774	}
775	dd = dvp->v_data;
776	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen);
777	while (de == NULL) {	/* While(...) so we can use break */
778
779		if (nameiop == DELETE)
780			return (ENOENT);
781
782		/*
783		 * OK, we didn't have an entry for the name we were asked for
784		 * so we try to see if anybody can create it on demand.
785		 */
786		pname = devfs_fqpn(specname, dvp, cnp);
787		if (pname == NULL)
788			break;
789
790		cdev = NULL;
791		DEVFS_DMP_HOLD(dmp);
792		sx_xunlock(&dmp->dm_lock);
793		sx_slock(&clone_drain_lock);
794		EVENTHANDLER_INVOKE(dev_clone,
795		    td->td_ucred, pname, strlen(pname), &cdev);
796		sx_sunlock(&clone_drain_lock);
797		sx_xlock(&dmp->dm_lock);
798		if (DEVFS_DMP_DROP(dmp)) {
799			*dm_unlock = 0;
800			sx_xunlock(&dmp->dm_lock);
801			devfs_unmount_final(dmp);
802			return (ENOENT);
803		}
804		if (cdev == NULL)
805			break;
806
807		DEVFS_DMP_HOLD(dmp);
808		devfs_populate(dmp);
809		if (DEVFS_DMP_DROP(dmp)) {
810			*dm_unlock = 0;
811			sx_xunlock(&dmp->dm_lock);
812			devfs_unmount_final(dmp);
813			return (ENOENT);
814		}
815
816		dev_lock();
817		dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
818		if (dde != NULL && *dde != NULL)
819			de = *dde;
820		dev_unlock();
821		dev_rel(cdev);
822		break;
823	}
824
825	if (de == NULL || de->de_flags & DE_WHITEOUT) {
826		if ((nameiop == CREATE || nameiop == RENAME) &&
827		    (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
828			cnp->cn_flags |= SAVENAME;
829			return (EJUSTRETURN);
830		}
831		return (ENOENT);
832	}
833
834	if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
835		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
836		if (error)
837			return (error);
838		if (*vpp == dvp) {
839			VREF(dvp);
840			*vpp = dvp;
841			return (0);
842		}
843	}
844	error = devfs_allocv(de, dvp->v_mount, vpp);
845	*dm_unlock = 0;
846	return (error);
847}
848
849static int
850devfs_lookup(struct vop_lookup_args *ap)
851{
852	int j;
853	struct devfs_mount *dmp;
854	int dm_unlock;
855
856	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
857	dm_unlock = 1;
858	sx_xlock(&dmp->dm_lock);
859	j = devfs_lookupx(ap, &dm_unlock);
860	if (dm_unlock == 1)
861		sx_xunlock(&dmp->dm_lock);
862	return (j);
863}
864
865static int
866devfs_mknod(struct vop_mknod_args *ap)
867{
868	struct componentname *cnp;
869	struct vnode *dvp, **vpp;
870	struct devfs_dirent *dd, *de;
871	struct devfs_mount *dmp;
872	int error;
873
874	/*
875	 * The only type of node we should be creating here is a
876	 * character device, for anything else return EOPNOTSUPP.
877	 */
878	if (ap->a_vap->va_type != VCHR)
879		return (EOPNOTSUPP);
880	dvp = ap->a_dvp;
881	dmp = VFSTODEVFS(dvp->v_mount);
882
883	cnp = ap->a_cnp;
884	vpp = ap->a_vpp;
885	dd = dvp->v_data;
886
887	error = ENOENT;
888	sx_xlock(&dmp->dm_lock);
889	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
890		if (cnp->cn_namelen != de->de_dirent->d_namlen)
891			continue;
892		if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
893		    de->de_dirent->d_namlen) != 0)
894			continue;
895		if (de->de_flags & DE_WHITEOUT)
896			break;
897		goto notfound;
898	}
899	if (de == NULL)
900		goto notfound;
901	de->de_flags &= ~DE_WHITEOUT;
902	error = devfs_allocv(de, dvp->v_mount, vpp);
903	return (error);
904notfound:
905	sx_xunlock(&dmp->dm_lock);
906	return (error);
907}
908
909/* ARGSUSED */
910static int
911devfs_open(struct vop_open_args *ap)
912{
913	struct thread *td = ap->a_td;
914	struct vnode *vp = ap->a_vp;
915	struct cdev *dev = vp->v_rdev;
916	struct file *fp = ap->a_fp;
917	int error;
918	struct cdevsw *dsw;
919	struct file *fpop;
920
921	if (vp->v_type == VBLK)
922		return (ENXIO);
923
924	if (dev == NULL)
925		return (ENXIO);
926
927	/* Make this field valid before any I/O in d_open. */
928	if (dev->si_iosize_max == 0)
929		dev->si_iosize_max = DFLTPHYS;
930
931	dsw = dev_refthread(dev);
932	if (dsw == NULL)
933		return (ENXIO);
934
935	/* XXX: Special casing of ttys for deadfs.  Probably redundant. */
936	if (dsw->d_flags & D_TTY)
937		vp->v_vflag |= VV_ISTTY;
938
939	VOP_UNLOCK(vp, 0);
940
941	fpop = td->td_fpop;
942	td->td_fpop = fp;
943	if (fp != NULL) {
944		fp->f_data = dev;
945		fp->f_vnode = vp;
946	}
947	if (dsw->d_fdopen != NULL)
948		error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
949	else
950		error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
951	td->td_fpop = fpop;
952
953	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
954
955	dev_relthread(dev);
956
957	if (error)
958		return (error);
959
960#if 0	/* /dev/console */
961	KASSERT(fp != NULL,
962	     ("Could not vnode bypass device on NULL fp"));
963#else
964	if(fp == NULL)
965		return (error);
966#endif
967	if (fp->f_ops == &badfileops)
968		finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
969	return (error);
970}
971
972static int
973devfs_pathconf(struct vop_pathconf_args *ap)
974{
975
976	switch (ap->a_name) {
977	case _PC_MAC_PRESENT:
978#ifdef MAC
979		/*
980		 * If MAC is enabled, devfs automatically supports
981		 * trivial non-persistant label storage.
982		 */
983		*ap->a_retval = 1;
984#else
985		*ap->a_retval = 0;
986#endif
987		return (0);
988	default:
989		return (vop_stdpathconf(ap));
990	}
991	/* NOTREACHED */
992}
993
994/* ARGSUSED */
995static int
996devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
997{
998	struct cdev *dev;
999	struct cdevsw *dsw;
1000	int error;
1001	struct file *fpop;
1002
1003	fpop = td->td_fpop;
1004	error = devfs_fp_check(fp, &dev, &dsw);
1005	if (error)
1006		return (poll_no_poll(events));
1007	error = dsw->d_poll(dev, events, td);
1008	td->td_fpop = fpop;
1009	dev_relthread(dev);
1010	return(error);
1011}
1012
1013/*
1014 * Print out the contents of a special device vnode.
1015 */
1016static int
1017devfs_print(struct vop_print_args *ap)
1018{
1019
1020	printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1021	return (0);
1022}
1023
1024/* ARGSUSED */
1025static int
1026devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td)
1027{
1028	struct cdev *dev;
1029	int ioflag, error, resid;
1030	struct cdevsw *dsw;
1031	struct file *fpop;
1032
1033	fpop = td->td_fpop;
1034	error = devfs_fp_check(fp, &dev, &dsw);
1035	if (error)
1036		return (error);
1037	resid = uio->uio_resid;
1038	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1039	if (ioflag & O_DIRECT)
1040		ioflag |= IO_DIRECT;
1041
1042	if ((flags & FOF_OFFSET) == 0)
1043		uio->uio_offset = fp->f_offset;
1044
1045	error = dsw->d_read(dev, uio, ioflag);
1046	if (uio->uio_resid != resid || (error == 0 && resid != 0))
1047		vfs_timestamp(&dev->si_atime);
1048	td->td_fpop = fpop;
1049	dev_relthread(dev);
1050
1051	if ((flags & FOF_OFFSET) == 0)
1052		fp->f_offset = uio->uio_offset;
1053	fp->f_nextoff = uio->uio_offset;
1054	return (error);
1055}
1056
1057static int
1058devfs_readdir(struct vop_readdir_args *ap)
1059{
1060	int error;
1061	struct uio *uio;
1062	struct dirent *dp;
1063	struct devfs_dirent *dd;
1064	struct devfs_dirent *de;
1065	struct devfs_mount *dmp;
1066	off_t off;
1067	int *tmp_ncookies = NULL;
1068
1069	if (ap->a_vp->v_type != VDIR)
1070		return (ENOTDIR);
1071
1072	uio = ap->a_uio;
1073	if (uio->uio_offset < 0)
1074		return (EINVAL);
1075
1076	/*
1077	 * XXX: This is a temporary hack to get around this filesystem not
1078	 * supporting cookies. We store the location of the ncookies pointer
1079	 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1080	 * and set the number of cookies to 0. We then set the pointer to
1081	 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1082	 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1083	 * pointer to its original location before returning to the caller.
1084	 */
1085	if (ap->a_ncookies != NULL) {
1086		tmp_ncookies = ap->a_ncookies;
1087		*ap->a_ncookies = 0;
1088		ap->a_ncookies = NULL;
1089	}
1090
1091	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1092	sx_xlock(&dmp->dm_lock);
1093	DEVFS_DMP_HOLD(dmp);
1094	devfs_populate(dmp);
1095	if (DEVFS_DMP_DROP(dmp)) {
1096		sx_xunlock(&dmp->dm_lock);
1097		devfs_unmount_final(dmp);
1098		if (tmp_ncookies != NULL)
1099			ap->a_ncookies = tmp_ncookies;
1100		return (EIO);
1101	}
1102	error = 0;
1103	de = ap->a_vp->v_data;
1104	off = 0;
1105	TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1106		KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1107		if (dd->de_flags & DE_WHITEOUT)
1108			continue;
1109		if (dd->de_dirent->d_type == DT_DIR)
1110			de = dd->de_dir;
1111		else
1112			de = dd;
1113		dp = dd->de_dirent;
1114		if (dp->d_reclen > uio->uio_resid)
1115			break;
1116		dp->d_fileno = de->de_inode;
1117		if (off >= uio->uio_offset) {
1118			error = vfs_read_dirent(ap, dp, off);
1119			if (error)
1120				break;
1121		}
1122		off += dp->d_reclen;
1123	}
1124	sx_xunlock(&dmp->dm_lock);
1125	uio->uio_offset = off;
1126
1127	/*
1128	 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1129	 * place.
1130	 */
1131	if (tmp_ncookies != NULL)
1132		ap->a_ncookies = tmp_ncookies;
1133
1134	return (error);
1135}
1136
1137static int
1138devfs_readlink(struct vop_readlink_args *ap)
1139{
1140	struct devfs_dirent *de;
1141
1142	de = ap->a_vp->v_data;
1143	return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1144}
1145
1146static int
1147devfs_reclaim(struct vop_reclaim_args *ap)
1148{
1149	struct vnode *vp = ap->a_vp;
1150	struct devfs_dirent *de;
1151	struct cdev *dev;
1152
1153	mtx_lock(&devfs_de_interlock);
1154	de = vp->v_data;
1155	if (de != NULL) {
1156		de->de_vnode = NULL;
1157		vp->v_data = NULL;
1158	}
1159	mtx_unlock(&devfs_de_interlock);
1160
1161	vnode_destroy_vobject(vp);
1162
1163	VI_LOCK(vp);
1164	dev_lock();
1165	dev = vp->v_rdev;
1166	vp->v_rdev = NULL;
1167
1168	if (dev == NULL) {
1169		dev_unlock();
1170		VI_UNLOCK(vp);
1171		return (0);
1172	}
1173
1174	dev->si_usecount -= vp->v_usecount;
1175	dev_unlock();
1176	VI_UNLOCK(vp);
1177	dev_rel(dev);
1178	return (0);
1179}
1180
1181static int
1182devfs_remove(struct vop_remove_args *ap)
1183{
1184	struct vnode *vp = ap->a_vp;
1185	struct devfs_dirent *dd;
1186	struct devfs_dirent *de;
1187	struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1188
1189	sx_xlock(&dmp->dm_lock);
1190	dd = ap->a_dvp->v_data;
1191	de = vp->v_data;
1192	if (de->de_cdp == NULL) {
1193		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1194		devfs_delete(dmp, de, 1);
1195	} else {
1196		de->de_flags |= DE_WHITEOUT;
1197	}
1198	sx_xunlock(&dmp->dm_lock);
1199	return (0);
1200}
1201
1202/*
1203 * Revoke is called on a tty when a terminal session ends.  The vnode
1204 * is orphaned by setting v_op to deadfs so we need to let go of it
1205 * as well so that we create a new one next time around.
1206 *
1207 */
1208static int
1209devfs_revoke(struct vop_revoke_args *ap)
1210{
1211	struct vnode *vp = ap->a_vp, *vp2;
1212	struct cdev *dev;
1213	struct cdev_priv *cdp;
1214	struct devfs_dirent *de;
1215	int i;
1216
1217	KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1218
1219	dev = vp->v_rdev;
1220	cdp = cdev2priv(dev);
1221
1222	dev_lock();
1223	cdp->cdp_inuse++;
1224	dev_unlock();
1225
1226	vhold(vp);
1227	vgone(vp);
1228	vdrop(vp);
1229
1230	VOP_UNLOCK(vp,0);
1231 loop:
1232	for (;;) {
1233		mtx_lock(&devfs_de_interlock);
1234		dev_lock();
1235		vp2 = NULL;
1236		for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1237			de = cdp->cdp_dirents[i];
1238			if (de == NULL)
1239				continue;
1240
1241			vp2 = de->de_vnode;
1242			if (vp2 != NULL) {
1243				dev_unlock();
1244				VI_LOCK(vp2);
1245				mtx_unlock(&devfs_de_interlock);
1246				if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1247				    curthread))
1248					goto loop;
1249				vhold(vp2);
1250				vgone(vp2);
1251				vdrop(vp2);
1252				vput(vp2);
1253				break;
1254			}
1255		}
1256		if (vp2 != NULL) {
1257			continue;
1258		}
1259		dev_unlock();
1260		mtx_unlock(&devfs_de_interlock);
1261		break;
1262	}
1263	dev_lock();
1264	cdp->cdp_inuse--;
1265	if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1266		TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1267		dev_unlock();
1268		dev_rel(&cdp->cdp_c);
1269	} else
1270		dev_unlock();
1271
1272	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1273	return (0);
1274}
1275
1276static int
1277devfs_rioctl(struct vop_ioctl_args *ap)
1278{
1279	int error;
1280	struct devfs_mount *dmp;
1281
1282	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1283	sx_xlock(&dmp->dm_lock);
1284	DEVFS_DMP_HOLD(dmp);
1285	devfs_populate(dmp);
1286	if (DEVFS_DMP_DROP(dmp)) {
1287		sx_xunlock(&dmp->dm_lock);
1288		devfs_unmount_final(dmp);
1289		return (ENOENT);
1290	}
1291	error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1292	sx_xunlock(&dmp->dm_lock);
1293	return (error);
1294}
1295
1296static int
1297devfs_rread(struct vop_read_args *ap)
1298{
1299
1300	if (ap->a_vp->v_type != VDIR)
1301		return (EINVAL);
1302	return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1303}
1304
1305static int
1306devfs_setattr(struct vop_setattr_args *ap)
1307{
1308	struct devfs_dirent *de;
1309	struct vattr *vap;
1310	struct vnode *vp;
1311	struct thread *td;
1312	int c, error;
1313	uid_t uid;
1314	gid_t gid;
1315
1316	vap = ap->a_vap;
1317	vp = ap->a_vp;
1318	td = curthread;
1319	if ((vap->va_type != VNON) ||
1320	    (vap->va_nlink != VNOVAL) ||
1321	    (vap->va_fsid != VNOVAL) ||
1322	    (vap->va_fileid != VNOVAL) ||
1323	    (vap->va_blocksize != VNOVAL) ||
1324	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1325	    (vap->va_rdev != VNOVAL) ||
1326	    ((int)vap->va_bytes != VNOVAL) ||
1327	    (vap->va_gen != VNOVAL)) {
1328		return (EINVAL);
1329	}
1330
1331	de = vp->v_data;
1332	if (vp->v_type == VDIR)
1333		de = de->de_dir;
1334
1335	error = c = 0;
1336	if (vap->va_uid == (uid_t)VNOVAL)
1337		uid = de->de_uid;
1338	else
1339		uid = vap->va_uid;
1340	if (vap->va_gid == (gid_t)VNOVAL)
1341		gid = de->de_gid;
1342	else
1343		gid = vap->va_gid;
1344	if (uid != de->de_uid || gid != de->de_gid) {
1345		if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1346		    (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1347			error = priv_check(td, PRIV_VFS_CHOWN);
1348			if (error)
1349				return (error);
1350		}
1351		de->de_uid = uid;
1352		de->de_gid = gid;
1353		c = 1;
1354	}
1355
1356	if (vap->va_mode != (mode_t)VNOVAL) {
1357		if (ap->a_cred->cr_uid != de->de_uid) {
1358			error = priv_check(td, PRIV_VFS_ADMIN);
1359			if (error)
1360				return (error);
1361		}
1362		de->de_mode = vap->va_mode;
1363		c = 1;
1364	}
1365
1366	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1367		/* See the comment in ufs_vnops::ufs_setattr(). */
1368		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) &&
1369		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1370		    (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td))))
1371			return (error);
1372		if (vap->va_atime.tv_sec != VNOVAL) {
1373			if (vp->v_type == VCHR)
1374				vp->v_rdev->si_atime = vap->va_atime;
1375			else
1376				de->de_atime = vap->va_atime;
1377		}
1378		if (vap->va_mtime.tv_sec != VNOVAL) {
1379			if (vp->v_type == VCHR)
1380				vp->v_rdev->si_mtime = vap->va_mtime;
1381			else
1382				de->de_mtime = vap->va_mtime;
1383		}
1384		c = 1;
1385	}
1386
1387	if (c) {
1388		if (vp->v_type == VCHR)
1389			vfs_timestamp(&vp->v_rdev->si_ctime);
1390		else
1391			vfs_timestamp(&de->de_mtime);
1392	}
1393	return (0);
1394}
1395
1396#ifdef MAC
1397static int
1398devfs_setlabel(struct vop_setlabel_args *ap)
1399{
1400	struct vnode *vp;
1401	struct devfs_dirent *de;
1402
1403	vp = ap->a_vp;
1404	de = vp->v_data;
1405
1406	mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1407	mac_devfs_update(vp->v_mount, de, vp);
1408
1409	return (0);
1410}
1411#endif
1412
1413static int
1414devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1415{
1416
1417	return (vnops.fo_stat(fp, sb, cred, td));
1418}
1419
1420static int
1421devfs_symlink(struct vop_symlink_args *ap)
1422{
1423	int i, error;
1424	struct devfs_dirent *dd;
1425	struct devfs_dirent *de;
1426	struct devfs_mount *dmp;
1427
1428	error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1429	if (error)
1430		return(error);
1431	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1432	dd = ap->a_dvp->v_data;
1433	de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1434	de->de_uid = 0;
1435	de->de_gid = 0;
1436	de->de_mode = 0755;
1437	de->de_inode = alloc_unr(devfs_inos);
1438	de->de_dirent->d_type = DT_LNK;
1439	i = strlen(ap->a_target) + 1;
1440	de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1441	bcopy(ap->a_target, de->de_symlink, i);
1442	sx_xlock(&dmp->dm_lock);
1443#ifdef MAC
1444	mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1445#endif
1446	TAILQ_INSERT_TAIL(&dd->de_dlist, de, de_list);
1447	return (devfs_allocv(de, ap->a_dvp->v_mount, ap->a_vpp));
1448}
1449
1450static int
1451devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1452{
1453
1454	return (vnops.fo_truncate(fp, length, cred, td));
1455}
1456
1457/* ARGSUSED */
1458static int
1459devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred, int flags, struct thread *td)
1460{
1461	struct cdev *dev;
1462	int error, ioflag, resid;
1463	struct cdevsw *dsw;
1464	struct file *fpop;
1465
1466	fpop = td->td_fpop;
1467	error = devfs_fp_check(fp, &dev, &dsw);
1468	if (error)
1469		return (error);
1470	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1471	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1472	if (ioflag & O_DIRECT)
1473		ioflag |= IO_DIRECT;
1474	if ((flags & FOF_OFFSET) == 0)
1475		uio->uio_offset = fp->f_offset;
1476
1477	resid = uio->uio_resid;
1478
1479	error = dsw->d_write(dev, uio, ioflag);
1480	if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1481		vfs_timestamp(&dev->si_ctime);
1482		dev->si_mtime = dev->si_ctime;
1483	}
1484	td->td_fpop = fpop;
1485	dev_relthread(dev);
1486
1487	if ((flags & FOF_OFFSET) == 0)
1488		fp->f_offset = uio->uio_offset;
1489	fp->f_nextoff = uio->uio_offset;
1490	return (error);
1491}
1492
1493dev_t
1494dev2udev(struct cdev *x)
1495{
1496	if (x == NULL)
1497		return (NODEV);
1498	return (cdev2priv(x)->cdp_inode);
1499}
1500
1501static struct fileops devfs_ops_f = {
1502	.fo_read =	devfs_read_f,
1503	.fo_write =	devfs_write_f,
1504	.fo_truncate =	devfs_truncate_f,
1505	.fo_ioctl =	devfs_ioctl_f,
1506	.fo_poll =	devfs_poll_f,
1507	.fo_kqfilter =	devfs_kqfilter_f,
1508	.fo_stat =	devfs_stat_f,
1509	.fo_close =	devfs_close_f,
1510	.fo_flags =	DFLAG_PASSABLE | DFLAG_SEEKABLE
1511};
1512
1513static struct vop_vector devfs_vnodeops = {
1514	.vop_default =		&default_vnodeops,
1515
1516	.vop_access =		devfs_access,
1517	.vop_getattr =		devfs_getattr,
1518	.vop_ioctl =		devfs_rioctl,
1519	.vop_lookup =		devfs_lookup,
1520	.vop_mknod =		devfs_mknod,
1521	.vop_pathconf =		devfs_pathconf,
1522	.vop_read =		devfs_rread,
1523	.vop_readdir =		devfs_readdir,
1524	.vop_readlink =		devfs_readlink,
1525	.vop_reclaim =		devfs_reclaim,
1526	.vop_remove =		devfs_remove,
1527	.vop_revoke =		devfs_revoke,
1528	.vop_setattr =		devfs_setattr,
1529#ifdef MAC
1530	.vop_setlabel =		devfs_setlabel,
1531#endif
1532	.vop_symlink =		devfs_symlink,
1533	.vop_vptocnp =		devfs_vptocnp,
1534};
1535
1536static struct vop_vector devfs_specops = {
1537	.vop_default =		&default_vnodeops,
1538
1539	.vop_access =		devfs_access,
1540	.vop_bmap =		VOP_PANIC,
1541	.vop_close =		devfs_close,
1542	.vop_create =		VOP_PANIC,
1543	.vop_fsync =		devfs_fsync,
1544	.vop_getattr =		devfs_getattr,
1545	.vop_link =		VOP_PANIC,
1546	.vop_mkdir =		VOP_PANIC,
1547	.vop_mknod =		VOP_PANIC,
1548	.vop_open =		devfs_open,
1549	.vop_pathconf =		devfs_pathconf,
1550	.vop_print =		devfs_print,
1551	.vop_read =		VOP_PANIC,
1552	.vop_readdir =		VOP_PANIC,
1553	.vop_readlink =		VOP_PANIC,
1554	.vop_reallocblks =	VOP_PANIC,
1555	.vop_reclaim =		devfs_reclaim,
1556	.vop_remove =		devfs_remove,
1557	.vop_rename =		VOP_PANIC,
1558	.vop_revoke =		devfs_revoke,
1559	.vop_rmdir =		VOP_PANIC,
1560	.vop_setattr =		devfs_setattr,
1561#ifdef MAC
1562	.vop_setlabel =		devfs_setlabel,
1563#endif
1564	.vop_strategy =		VOP_PANIC,
1565	.vop_symlink =		VOP_PANIC,
1566	.vop_vptocnp =		devfs_vptocnp,
1567	.vop_write =		VOP_PANIC,
1568};
1569
1570/*
1571 * Our calling convention to the device drivers used to be that we passed
1572 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1573 * flags instead since that's what open(), close() and ioctl() takes and
1574 * we don't really want vnode.h in device drivers.
1575 * We solved the source compatibility by redefining some vnode flags to
1576 * be the same as the fcntl ones and by sending down the bitwise OR of
1577 * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
1578 * pulls the rug out under this.
1579 */
1580CTASSERT(O_NONBLOCK == IO_NDELAY);
1581CTASSERT(O_FSYNC == IO_SYNC);
1582