xref: /freebsd-14.2/sys/fs/devfs/devfs_vnops.c (revision 9ccd5394)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2000-2004
5  *	Poul-Henning Kamp.  All rights reserved.
6  * Copyright (c) 1989, 1992-1993, 1995
7  *	The Regents of the University of California.  All rights reserved.
8  *
9  * This code is derived from software donated to Berkeley by
10  * Jan-Simon Pendry.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)kernfs_vnops.c	8.15 (Berkeley) 5/21/95
34  * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vnops.c 1.43
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/eventhandler.h>
47 #include <sys/fcntl.h>
48 #include <sys/file.h>
49 #include <sys/filedesc.h>
50 #include <sys/filio.h>
51 #include <sys/jail.h>
52 #include <sys/kernel.h>
53 #include <sys/limits.h>
54 #include <sys/lock.h>
55 #include <sys/malloc.h>
56 #include <sys/mman.h>
57 #include <sys/mount.h>
58 #include <sys/namei.h>
59 #include <sys/priv.h>
60 #include <sys/proc.h>
61 #include <sys/stat.h>
62 #include <sys/sx.h>
63 #include <sys/sysctl.h>
64 #include <sys/time.h>
65 #include <sys/ttycom.h>
66 #include <sys/unistd.h>
67 #include <sys/vnode.h>
68 
69 static struct vop_vector devfs_vnodeops;
70 static struct vop_vector devfs_specops;
71 static struct fileops devfs_ops_f;
72 
73 #include <fs/devfs/devfs.h>
74 #include <fs/devfs/devfs_int.h>
75 
76 #include <security/mac/mac_framework.h>
77 
78 #include <vm/vm.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_object.h>
81 
82 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
83 
84 struct mtx	devfs_de_interlock;
85 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
86 struct mtx	cdevpriv_mtx;
87 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
88 
89 SYSCTL_DECL(_vfs_devfs);
90 
91 static int devfs_dotimes;
92 SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW,
93     &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision");
94 
95 /*
96  * Update devfs node timestamp.  Note that updates are unlocked and
97  * stat(2) could see partially updated times.
98  */
99 static void
devfs_timestamp(struct timespec * tsp)100 devfs_timestamp(struct timespec *tsp)
101 {
102 	time_t ts;
103 
104 	if (devfs_dotimes) {
105 		vfs_timestamp(tsp);
106 	} else {
107 		ts = time_second;
108 		if (tsp->tv_sec != ts) {
109 			tsp->tv_sec = ts;
110 			tsp->tv_nsec = 0;
111 		}
112 	}
113 }
114 
115 static int
devfs_fp_check(struct file * fp,struct cdev ** devp,struct cdevsw ** dswp,int * ref)116 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
117     int *ref)
118 {
119 	*dswp = devvn_refthread(fp->f_vnode, devp, ref);
120 	if (*dswp == NULL || *devp != fp->f_data) {
121 		if (*dswp != NULL)
122 			dev_relthread(*devp, *ref);
123 		return (ENXIO);
124 	}
125 	KASSERT((*devp)->si_refcount > 0,
126 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
127 	if (*dswp == NULL)
128 		return (ENXIO);
129 	curthread->td_fpop = fp;
130 	return (0);
131 }
132 
133 int
devfs_get_cdevpriv(void ** datap)134 devfs_get_cdevpriv(void **datap)
135 {
136 	struct file *fp;
137 	struct cdev_privdata *p;
138 	int error;
139 
140 	fp = curthread->td_fpop;
141 	if (fp == NULL)
142 		return (EBADF);
143 	p = fp->f_cdevpriv;
144 	if (p != NULL) {
145 		error = 0;
146 		*datap = p->cdpd_data;
147 	} else
148 		error = ENOENT;
149 	return (error);
150 }
151 
152 int
devfs_set_cdevpriv(void * priv,d_priv_dtor_t * priv_dtr)153 devfs_set_cdevpriv(void *priv, d_priv_dtor_t *priv_dtr)
154 {
155 	struct file *fp;
156 	struct cdev_priv *cdp;
157 	struct cdev_privdata *p;
158 	int error;
159 
160 	fp = curthread->td_fpop;
161 	if (fp == NULL)
162 		return (ENOENT);
163 	cdp = cdev2priv((struct cdev *)fp->f_data);
164 	p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
165 	p->cdpd_data = priv;
166 	p->cdpd_dtr = priv_dtr;
167 	p->cdpd_fp = fp;
168 	mtx_lock(&cdevpriv_mtx);
169 	if (fp->f_cdevpriv == NULL) {
170 		LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
171 		fp->f_cdevpriv = p;
172 		mtx_unlock(&cdevpriv_mtx);
173 		error = 0;
174 	} else {
175 		mtx_unlock(&cdevpriv_mtx);
176 		free(p, M_CDEVPDATA);
177 		error = EBUSY;
178 	}
179 	return (error);
180 }
181 
182 int
devfs_foreach_cdevpriv(struct cdev * dev,int (* cb)(void * data,void * arg),void * arg)183 devfs_foreach_cdevpriv(struct cdev *dev, int (*cb)(void *data, void *arg),
184     void *arg)
185 {
186 	struct cdev_priv *cdp;
187 	struct cdev_privdata *p;
188 	int error;
189 
190 	cdp = cdev2priv(dev);
191 	error = 0;
192 	mtx_lock(&cdevpriv_mtx);
193 	LIST_FOREACH(p, &cdp->cdp_fdpriv, cdpd_list) {
194 		error = cb(p->cdpd_data, arg);
195 		if (error != 0)
196 			break;
197 	}
198 	mtx_unlock(&cdevpriv_mtx);
199 	return (error);
200 }
201 
202 void
devfs_destroy_cdevpriv(struct cdev_privdata * p)203 devfs_destroy_cdevpriv(struct cdev_privdata *p)
204 {
205 
206 	mtx_assert(&cdevpriv_mtx, MA_OWNED);
207 	KASSERT(p->cdpd_fp->f_cdevpriv == p,
208 	    ("devfs_destoy_cdevpriv %p != %p", p->cdpd_fp->f_cdevpriv, p));
209 	p->cdpd_fp->f_cdevpriv = NULL;
210 	LIST_REMOVE(p, cdpd_list);
211 	mtx_unlock(&cdevpriv_mtx);
212 	(p->cdpd_dtr)(p->cdpd_data);
213 	free(p, M_CDEVPDATA);
214 }
215 
216 static void
devfs_fpdrop(struct file * fp)217 devfs_fpdrop(struct file *fp)
218 {
219 	struct cdev_privdata *p;
220 
221 	mtx_lock(&cdevpriv_mtx);
222 	if ((p = fp->f_cdevpriv) == NULL) {
223 		mtx_unlock(&cdevpriv_mtx);
224 		return;
225 	}
226 	devfs_destroy_cdevpriv(p);
227 }
228 
229 void
devfs_clear_cdevpriv(void)230 devfs_clear_cdevpriv(void)
231 {
232 	struct file *fp;
233 
234 	fp = curthread->td_fpop;
235 	if (fp == NULL)
236 		return;
237 	devfs_fpdrop(fp);
238 }
239 
240 static void
devfs_usecount_add(struct vnode * vp)241 devfs_usecount_add(struct vnode *vp)
242 {
243 	struct devfs_dirent *de;
244 	struct cdev *dev;
245 
246 	mtx_lock(&devfs_de_interlock);
247 	VI_LOCK(vp);
248 	VNPASS(vp->v_type == VCHR || vp->v_type == VBAD, vp);
249 	if (VN_IS_DOOMED(vp)) {
250 		goto out_unlock;
251 	}
252 
253 	de = vp->v_data;
254 	dev = vp->v_rdev;
255 	MPASS(de != NULL);
256 	MPASS(dev != NULL);
257 	dev->si_usecount++;
258 	de->de_usecount++;
259 out_unlock:
260 	VI_UNLOCK(vp);
261 	mtx_unlock(&devfs_de_interlock);
262 }
263 
264 static void
devfs_usecount_subl(struct vnode * vp)265 devfs_usecount_subl(struct vnode *vp)
266 {
267 	struct devfs_dirent *de;
268 	struct cdev *dev;
269 
270 	mtx_assert(&devfs_de_interlock, MA_OWNED);
271 	ASSERT_VI_LOCKED(vp, __func__);
272 	VNPASS(vp->v_type == VCHR || vp->v_type == VBAD, vp);
273 
274 	de = vp->v_data;
275 	dev = vp->v_rdev;
276 	if (de == NULL)
277 		return;
278 	if (dev == NULL) {
279 		MPASS(de->de_usecount == 0);
280 		return;
281 	}
282 	if (dev->si_usecount < de->de_usecount)
283 		panic("%s: si_usecount underflow for dev %p "
284 		    "(has %ld, dirent has %d)\n",
285 		    __func__, dev, dev->si_usecount, de->de_usecount);
286 	if (VN_IS_DOOMED(vp)) {
287 		dev->si_usecount -= de->de_usecount;
288 		de->de_usecount = 0;
289 	} else {
290 		if (de->de_usecount == 0)
291 			panic("%s: de_usecount underflow for dev %p\n",
292 			    __func__, dev);
293 		dev->si_usecount--;
294 		de->de_usecount--;
295 	}
296 }
297 
298 static void
devfs_usecount_sub(struct vnode * vp)299 devfs_usecount_sub(struct vnode *vp)
300 {
301 
302 	mtx_lock(&devfs_de_interlock);
303 	VI_LOCK(vp);
304 	devfs_usecount_subl(vp);
305 	VI_UNLOCK(vp);
306 	mtx_unlock(&devfs_de_interlock);
307 }
308 
309 static int
devfs_usecountl(struct vnode * vp)310 devfs_usecountl(struct vnode *vp)
311 {
312 
313 	VNPASS(vp->v_type == VCHR, vp);
314 	mtx_assert(&devfs_de_interlock, MA_OWNED);
315 	ASSERT_VI_LOCKED(vp, __func__);
316 	return (vp->v_rdev->si_usecount);
317 }
318 
319 int
devfs_usecount(struct vnode * vp)320 devfs_usecount(struct vnode *vp)
321 {
322 	int count;
323 
324 	VNPASS(vp->v_type == VCHR, vp);
325 	mtx_lock(&devfs_de_interlock);
326 	VI_LOCK(vp);
327 	count = devfs_usecountl(vp);
328 	VI_UNLOCK(vp);
329 	mtx_unlock(&devfs_de_interlock);
330 	return (count);
331 }
332 
333 void
devfs_ctty_ref(struct vnode * vp)334 devfs_ctty_ref(struct vnode *vp)
335 {
336 
337 	vrefact(vp);
338 	devfs_usecount_add(vp);
339 }
340 
341 void
devfs_ctty_unref(struct vnode * vp)342 devfs_ctty_unref(struct vnode *vp)
343 {
344 
345 	devfs_usecount_sub(vp);
346 	vrele(vp);
347 }
348 
349 /*
350  * On success devfs_populate_vp() returns with dmp->dm_lock held.
351  */
352 static int
devfs_populate_vp(struct vnode * vp)353 devfs_populate_vp(struct vnode *vp)
354 {
355 	struct devfs_dirent *de;
356 	struct devfs_mount *dmp;
357 	int locked;
358 
359 	ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
360 
361 	dmp = VFSTODEVFS(vp->v_mount);
362 	if (!devfs_populate_needed(dmp)) {
363 		sx_xlock(&dmp->dm_lock);
364 		goto out_nopopulate;
365 	}
366 
367 	locked = VOP_ISLOCKED(vp);
368 
369 	sx_xlock(&dmp->dm_lock);
370 	DEVFS_DMP_HOLD(dmp);
371 
372 	/* Can't call devfs_populate() with the vnode lock held. */
373 	VOP_UNLOCK(vp);
374 	devfs_populate(dmp);
375 
376 	sx_xunlock(&dmp->dm_lock);
377 	vn_lock(vp, locked | LK_RETRY);
378 	sx_xlock(&dmp->dm_lock);
379 	if (DEVFS_DMP_DROP(dmp)) {
380 		sx_xunlock(&dmp->dm_lock);
381 		devfs_unmount_final(dmp);
382 		return (ERESTART);
383 	}
384 out_nopopulate:
385 	if (VN_IS_DOOMED(vp)) {
386 		sx_xunlock(&dmp->dm_lock);
387 		return (ERESTART);
388 	}
389 	de = vp->v_data;
390 	KASSERT(de != NULL,
391 	    ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
392 	if ((de->de_flags & DE_DOOMED) != 0) {
393 		sx_xunlock(&dmp->dm_lock);
394 		return (ERESTART);
395 	}
396 
397 	return (0);
398 }
399 
400 static int
devfs_vptocnp(struct vop_vptocnp_args * ap)401 devfs_vptocnp(struct vop_vptocnp_args *ap)
402 {
403 	struct vnode *vp = ap->a_vp;
404 	struct vnode **dvp = ap->a_vpp;
405 	struct devfs_mount *dmp;
406 	char *buf = ap->a_buf;
407 	size_t *buflen = ap->a_buflen;
408 	struct devfs_dirent *dd, *de;
409 	int i, error;
410 
411 	dmp = VFSTODEVFS(vp->v_mount);
412 
413 	error = devfs_populate_vp(vp);
414 	if (error != 0)
415 		return (error);
416 
417 	if (vp->v_type != VCHR && vp->v_type != VDIR) {
418 		error = ENOENT;
419 		goto finished;
420 	}
421 
422 	dd = vp->v_data;
423 	if (vp->v_type == VDIR && dd == dmp->dm_rootdir) {
424 		*dvp = vp;
425 		vref(*dvp);
426 		goto finished;
427 	}
428 
429 	i = *buflen;
430 	i -= dd->de_dirent->d_namlen;
431 	if (i < 0) {
432 		error = ENOMEM;
433 		goto finished;
434 	}
435 	bcopy(dd->de_dirent->d_name, buf + i, dd->de_dirent->d_namlen);
436 	*buflen = i;
437 	de = devfs_parent_dirent(dd);
438 	if (de == NULL) {
439 		error = ENOENT;
440 		goto finished;
441 	}
442 	mtx_lock(&devfs_de_interlock);
443 	*dvp = de->de_vnode;
444 	if (*dvp != NULL) {
445 		VI_LOCK(*dvp);
446 		mtx_unlock(&devfs_de_interlock);
447 		vholdl(*dvp);
448 		VI_UNLOCK(*dvp);
449 		vref(*dvp);
450 		vdrop(*dvp);
451 	} else {
452 		mtx_unlock(&devfs_de_interlock);
453 		error = ENOENT;
454 	}
455 finished:
456 	sx_xunlock(&dmp->dm_lock);
457 	return (error);
458 }
459 
460 /*
461  * Construct the fully qualified path name relative to the mountpoint.
462  * If a NULL cnp is provided, no '/' is appended to the resulting path.
463  */
464 char *
devfs_fqpn(char * buf,struct devfs_mount * dmp,struct devfs_dirent * dd,struct componentname * cnp)465 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
466     struct componentname *cnp)
467 {
468 	int i;
469 	struct devfs_dirent *de;
470 
471 	sx_assert(&dmp->dm_lock, SA_LOCKED);
472 
473 	i = SPECNAMELEN;
474 	buf[i] = '\0';
475 	if (cnp != NULL)
476 		i -= cnp->cn_namelen;
477 	if (i < 0)
478 		 return (NULL);
479 	if (cnp != NULL)
480 		bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
481 	de = dd;
482 	while (de != dmp->dm_rootdir) {
483 		if (cnp != NULL || i < SPECNAMELEN) {
484 			i--;
485 			if (i < 0)
486 				 return (NULL);
487 			buf[i] = '/';
488 		}
489 		i -= de->de_dirent->d_namlen;
490 		if (i < 0)
491 			 return (NULL);
492 		bcopy(de->de_dirent->d_name, buf + i,
493 		    de->de_dirent->d_namlen);
494 		de = devfs_parent_dirent(de);
495 		if (de == NULL)
496 			return (NULL);
497 	}
498 	return (buf + i);
499 }
500 
501 static int
devfs_allocv_drop_refs(int drop_dm_lock,struct devfs_mount * dmp,struct devfs_dirent * de)502 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
503 	struct devfs_dirent *de)
504 {
505 	int not_found;
506 
507 	not_found = 0;
508 	if (de->de_flags & DE_DOOMED)
509 		not_found = 1;
510 	if (DEVFS_DE_DROP(de)) {
511 		KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
512 		devfs_dirent_free(de);
513 	}
514 	if (DEVFS_DMP_DROP(dmp)) {
515 		KASSERT(not_found == 1,
516 			("DEVFS mount struct freed before dirent"));
517 		not_found = 2;
518 		sx_xunlock(&dmp->dm_lock);
519 		devfs_unmount_final(dmp);
520 	}
521 	if (not_found == 1 || (drop_dm_lock && not_found != 2))
522 		sx_unlock(&dmp->dm_lock);
523 	return (not_found);
524 }
525 
526 /*
527  * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
528  * it on return.
529  */
530 int
devfs_allocv(struct devfs_dirent * de,struct mount * mp,int lockmode,struct vnode ** vpp)531 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
532     struct vnode **vpp)
533 {
534 	int error;
535 	struct vnode *vp;
536 	struct cdev *dev;
537 	struct devfs_mount *dmp;
538 	struct cdevsw *dsw;
539 	enum vgetstate vs;
540 
541 	dmp = VFSTODEVFS(mp);
542 	if (de->de_flags & DE_DOOMED) {
543 		sx_xunlock(&dmp->dm_lock);
544 		return (ENOENT);
545 	}
546 loop:
547 	DEVFS_DE_HOLD(de);
548 	DEVFS_DMP_HOLD(dmp);
549 	mtx_lock(&devfs_de_interlock);
550 	vp = de->de_vnode;
551 	if (vp != NULL) {
552 		vs = vget_prep(vp);
553 		mtx_unlock(&devfs_de_interlock);
554 		sx_xunlock(&dmp->dm_lock);
555 		vget_finish(vp, lockmode | LK_RETRY, vs);
556 		sx_xlock(&dmp->dm_lock);
557 		if (devfs_allocv_drop_refs(0, dmp, de)) {
558 			vput(vp);
559 			return (ENOENT);
560 		} else if (VN_IS_DOOMED(vp)) {
561 			mtx_lock(&devfs_de_interlock);
562 			if (de->de_vnode == vp) {
563 				de->de_vnode = NULL;
564 				vp->v_data = NULL;
565 			}
566 			mtx_unlock(&devfs_de_interlock);
567 			vput(vp);
568 			goto loop;
569 		}
570 		sx_xunlock(&dmp->dm_lock);
571 		*vpp = vp;
572 		return (0);
573 	}
574 	mtx_unlock(&devfs_de_interlock);
575 	if (de->de_dirent->d_type == DT_CHR) {
576 		if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
577 			devfs_allocv_drop_refs(1, dmp, de);
578 			return (ENOENT);
579 		}
580 		dev = &de->de_cdp->cdp_c;
581 	} else {
582 		dev = NULL;
583 	}
584 	error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
585 	if (error != 0) {
586 		devfs_allocv_drop_refs(1, dmp, de);
587 		printf("devfs_allocv: failed to allocate new vnode\n");
588 		return (error);
589 	}
590 
591 	if (de->de_dirent->d_type == DT_CHR) {
592 		vp->v_type = VCHR;
593 		VI_LOCK(vp);
594 		dev_lock();
595 		dev_refl(dev);
596 		/* XXX: v_rdev should be protect by vnode lock */
597 		vp->v_rdev = dev;
598 		VNPASS(vp->v_usecount == 1, vp);
599 		/* Special casing of ttys for deadfs.  Probably redundant. */
600 		dsw = dev->si_devsw;
601 		if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
602 			vp->v_vflag |= VV_ISTTY;
603 		dev_unlock();
604 		VI_UNLOCK(vp);
605 		if ((dev->si_flags & SI_ETERNAL) != 0)
606 			vp->v_vflag |= VV_ETERNALDEV;
607 		vp->v_op = &devfs_specops;
608 	} else if (de->de_dirent->d_type == DT_DIR) {
609 		vp->v_type = VDIR;
610 	} else if (de->de_dirent->d_type == DT_LNK) {
611 		vp->v_type = VLNK;
612 	} else {
613 		vp->v_type = VBAD;
614 	}
615 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
616 	VN_LOCK_ASHARE(vp);
617 	mtx_lock(&devfs_de_interlock);
618 	vp->v_data = de;
619 	de->de_vnode = vp;
620 	mtx_unlock(&devfs_de_interlock);
621 	error = insmntque1(vp, mp);
622 	if (error != 0) {
623 		mtx_lock(&devfs_de_interlock);
624 		vp->v_data = NULL;
625 		de->de_vnode = NULL;
626 		mtx_unlock(&devfs_de_interlock);
627 		vgone(vp);
628 		vput(vp);
629 		(void) devfs_allocv_drop_refs(1, dmp, de);
630 		return (error);
631 	}
632 	if (devfs_allocv_drop_refs(0, dmp, de)) {
633 		vgone(vp);
634 		vput(vp);
635 		return (ENOENT);
636 	}
637 #ifdef MAC
638 	mac_devfs_vnode_associate(mp, de, vp);
639 #endif
640 	sx_xunlock(&dmp->dm_lock);
641 	vn_set_state(vp, VSTATE_CONSTRUCTED);
642 	*vpp = vp;
643 	return (0);
644 }
645 
646 static int
devfs_access(struct vop_access_args * ap)647 devfs_access(struct vop_access_args *ap)
648 {
649 	struct vnode *vp = ap->a_vp;
650 	struct devfs_dirent *de;
651 	struct proc *p;
652 	int error;
653 
654 	de = vp->v_data;
655 	if (vp->v_type == VDIR)
656 		de = de->de_dir;
657 
658 	error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
659 	    ap->a_accmode, ap->a_cred);
660 	if (error == 0)
661 		return (0);
662 	if (error != EACCES)
663 		return (error);
664 	p = ap->a_td->td_proc;
665 	/* We do, however, allow access to the controlling terminal */
666 	PROC_LOCK(p);
667 	if (!(p->p_flag & P_CONTROLT)) {
668 		PROC_UNLOCK(p);
669 		return (error);
670 	}
671 	if (p->p_session->s_ttydp == de->de_cdp)
672 		error = 0;
673 	PROC_UNLOCK(p);
674 	return (error);
675 }
676 
677 _Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0,
678     "devfs-only flag reuse failed");
679 
680 static int
devfs_close(struct vop_close_args * ap)681 devfs_close(struct vop_close_args *ap)
682 {
683 	struct vnode *vp = ap->a_vp, *oldvp;
684 	struct thread *td = ap->a_td;
685 	struct proc *p;
686 	struct cdev *dev = vp->v_rdev;
687 	struct cdevsw *dsw;
688 	struct devfs_dirent *de = vp->v_data;
689 	int dflags, error, ref, vp_locked;
690 
691 	/*
692 	 * XXX: Don't call d_close() if we were called because of
693 	 * XXX: insmntque() failure.
694 	 */
695 	if (vp->v_data == NULL)
696 		return (0);
697 
698 	/*
699 	 * Hack: a tty device that is a controlling terminal
700 	 * has a reference from the session structure.
701 	 * We cannot easily tell that a character device is
702 	 * a controlling terminal, unless it is the closing
703 	 * process' controlling terminal.  In that case,
704 	 * if the reference count is 2 (this last descriptor
705 	 * plus the session), release the reference from the session.
706 	 */
707 	if (de->de_usecount == 2 && td != NULL) {
708 		p = td->td_proc;
709 		PROC_LOCK(p);
710 		if (vp == p->p_session->s_ttyvp) {
711 			PROC_UNLOCK(p);
712 			oldvp = NULL;
713 			sx_xlock(&proctree_lock);
714 			if (vp == p->p_session->s_ttyvp) {
715 				SESS_LOCK(p->p_session);
716 				mtx_lock(&devfs_de_interlock);
717 				VI_LOCK(vp);
718 				if (devfs_usecountl(vp) == 2 && !VN_IS_DOOMED(vp)) {
719 					p->p_session->s_ttyvp = NULL;
720 					p->p_session->s_ttydp = NULL;
721 					oldvp = vp;
722 				}
723 				VI_UNLOCK(vp);
724 				mtx_unlock(&devfs_de_interlock);
725 				SESS_UNLOCK(p->p_session);
726 			}
727 			sx_xunlock(&proctree_lock);
728 			if (oldvp != NULL)
729 				devfs_ctty_unref(oldvp);
730 		} else
731 			PROC_UNLOCK(p);
732 	}
733 	/*
734 	 * We do not want to really close the device if it
735 	 * is still in use unless we are trying to close it
736 	 * forcibly. Since every use (buffer, vnode, swap, cmap)
737 	 * holds a reference to the vnode, and because we mark
738 	 * any other vnodes that alias this device, when the
739 	 * sum of the reference counts on all the aliased
740 	 * vnodes descends to one, we are on last close.
741 	 */
742 	dsw = dev_refthread(dev, &ref);
743 	if (dsw == NULL)
744 		return (ENXIO);
745 	dflags = 0;
746 	mtx_lock(&devfs_de_interlock);
747 	VI_LOCK(vp);
748 	if (devfs_usecountl(vp) == 1)
749 		dflags |= FLASTCLOSE;
750 	devfs_usecount_subl(vp);
751 	mtx_unlock(&devfs_de_interlock);
752 	if (VN_IS_DOOMED(vp)) {
753 		/* Forced close. */
754 		dflags |= FREVOKE | FNONBLOCK;
755 	} else if (dsw->d_flags & D_TRACKCLOSE) {
756 		/* Keep device updated on status. */
757 	} else if ((dflags & FLASTCLOSE) == 0) {
758 		VI_UNLOCK(vp);
759 		dev_relthread(dev, ref);
760 		return (0);
761 	}
762 	vholdnz(vp);
763 	VI_UNLOCK(vp);
764 	vp_locked = VOP_ISLOCKED(vp);
765 	VOP_UNLOCK(vp);
766 	KASSERT(dev->si_refcount > 0,
767 	    ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
768 	error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td);
769 	dev_relthread(dev, ref);
770 	vn_lock(vp, vp_locked | LK_RETRY);
771 	vdrop(vp);
772 	return (error);
773 }
774 
775 static int
devfs_close_f(struct file * fp,struct thread * td)776 devfs_close_f(struct file *fp, struct thread *td)
777 {
778 	int error;
779 	struct file *fpop;
780 
781 	/*
782 	 * NB: td may be NULL if this descriptor is closed due to
783 	 * garbage collection from a closed UNIX domain socket.
784 	 */
785 	fpop = curthread->td_fpop;
786 	curthread->td_fpop = fp;
787 	error = vnops.fo_close(fp, td);
788 	curthread->td_fpop = fpop;
789 
790 	/*
791 	 * The f_cdevpriv cannot be assigned non-NULL value while we
792 	 * are destroying the file.
793 	 */
794 	if (fp->f_cdevpriv != NULL)
795 		devfs_fpdrop(fp);
796 	return (error);
797 }
798 
799 static int
devfs_getattr(struct vop_getattr_args * ap)800 devfs_getattr(struct vop_getattr_args *ap)
801 {
802 	struct vnode *vp = ap->a_vp;
803 	struct vattr *vap = ap->a_vap;
804 	struct devfs_dirent *de;
805 	struct devfs_mount *dmp;
806 	struct cdev *dev;
807 	struct timeval boottime;
808 	int error;
809 
810 	error = devfs_populate_vp(vp);
811 	if (error != 0)
812 		return (error);
813 
814 	dmp = VFSTODEVFS(vp->v_mount);
815 	sx_xunlock(&dmp->dm_lock);
816 
817 	de = vp->v_data;
818 	KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
819 	if (vp->v_type == VDIR) {
820 		de = de->de_dir;
821 		KASSERT(de != NULL,
822 		    ("Null dir dirent in devfs_getattr vp=%p", vp));
823 	}
824 	vap->va_uid = de->de_uid;
825 	vap->va_gid = de->de_gid;
826 	vap->va_mode = de->de_mode;
827 	if (vp->v_type == VLNK)
828 		vap->va_size = strlen(de->de_symlink);
829 	else if (vp->v_type == VDIR)
830 		vap->va_size = vap->va_bytes = DEV_BSIZE;
831 	else
832 		vap->va_size = 0;
833 	if (vp->v_type != VDIR)
834 		vap->va_bytes = 0;
835 	vap->va_blocksize = DEV_BSIZE;
836 	vap->va_type = vp->v_type;
837 
838 	getboottime(&boottime);
839 #define fix(aa)							\
840 	do {							\
841 		if ((aa).tv_sec <= 3600) {			\
842 			(aa).tv_sec = boottime.tv_sec;		\
843 			(aa).tv_nsec = boottime.tv_usec * 1000; \
844 		}						\
845 	} while (0)
846 
847 	if (vp->v_type != VCHR)  {
848 		fix(de->de_atime);
849 		vap->va_atime = de->de_atime;
850 		fix(de->de_mtime);
851 		vap->va_mtime = de->de_mtime;
852 		fix(de->de_ctime);
853 		vap->va_ctime = de->de_ctime;
854 	} else {
855 		dev = vp->v_rdev;
856 		fix(dev->si_atime);
857 		vap->va_atime = dev->si_atime;
858 		fix(dev->si_mtime);
859 		vap->va_mtime = dev->si_mtime;
860 		fix(dev->si_ctime);
861 		vap->va_ctime = dev->si_ctime;
862 
863 		vap->va_rdev = cdev2priv(dev)->cdp_inode;
864 	}
865 	vap->va_gen = 0;
866 	vap->va_flags = 0;
867 	vap->va_filerev = 0;
868 	vap->va_nlink = de->de_links;
869 	vap->va_fileid = de->de_inode;
870 
871 	return (error);
872 }
873 
874 /* ARGSUSED */
875 static int
devfs_ioctl_f(struct file * fp,u_long com,void * data,struct ucred * cred,struct thread * td)876 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
877 {
878 	struct file *fpop;
879 	int error;
880 
881 	fpop = td->td_fpop;
882 	td->td_fpop = fp;
883 	error = vnops.fo_ioctl(fp, com, data, cred, td);
884 	td->td_fpop = fpop;
885 	return (error);
886 }
887 
888 void *
fiodgname_buf_get_ptr(void * fgnp,u_long com)889 fiodgname_buf_get_ptr(void *fgnp, u_long com)
890 {
891 	union {
892 		struct fiodgname_arg	fgn;
893 #ifdef COMPAT_FREEBSD32
894 		struct fiodgname_arg32	fgn32;
895 #endif
896 	} *fgnup;
897 
898 	fgnup = fgnp;
899 	switch (com) {
900 	case FIODGNAME:
901 		return (fgnup->fgn.buf);
902 #ifdef COMPAT_FREEBSD32
903 	case FIODGNAME_32:
904 		return ((void *)(uintptr_t)fgnup->fgn32.buf);
905 #endif
906 	default:
907 		panic("Unhandled ioctl command %ld", com);
908 	}
909 }
910 
911 static int
devfs_ioctl(struct vop_ioctl_args * ap)912 devfs_ioctl(struct vop_ioctl_args *ap)
913 {
914 	struct fiodgname_arg *fgn;
915 	struct vnode *vpold, *vp;
916 	struct cdevsw *dsw;
917 	struct thread *td;
918 	struct session *sess;
919 	struct cdev *dev;
920 	int error, ref, i;
921 	const char *p;
922 	u_long com;
923 
924 	vp = ap->a_vp;
925 	com = ap->a_command;
926 	td = ap->a_td;
927 
928 	dsw = devvn_refthread(vp, &dev, &ref);
929 	if (dsw == NULL)
930 		return (ENXIO);
931 	KASSERT(dev->si_refcount > 0,
932 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(dev)));
933 
934 	switch (com) {
935 	case FIODTYPE:
936 		*(int *)ap->a_data = dsw->d_flags & D_TYPEMASK;
937 		error = 0;
938 		break;
939 	case FIODGNAME:
940 #ifdef	COMPAT_FREEBSD32
941 	case FIODGNAME_32:
942 #endif
943 		fgn = ap->a_data;
944 		p = devtoname(dev);
945 		i = strlen(p) + 1;
946 		if (i > fgn->len)
947 			error = EINVAL;
948 		else
949 			error = copyout(p, fiodgname_buf_get_ptr(fgn, com), i);
950 		break;
951 	default:
952 		error = dsw->d_ioctl(dev, com, ap->a_data, ap->a_fflag, td);
953 	}
954 
955 	dev_relthread(dev, ref);
956 	if (error == ENOIOCTL)
957 		error = ENOTTY;
958 
959 	if (error == 0 && com == TIOCSCTTY) {
960 		/*
961 		 * Do nothing if reassigning same control tty, or if the
962 		 * control tty has already disappeared.  If it disappeared,
963 		 * it's because we were racing with TIOCNOTTY.  TIOCNOTTY
964 		 * already took care of releasing the old vnode and we have
965 		 * nothing left to do.
966 		 */
967 		sx_slock(&proctree_lock);
968 		sess = td->td_proc->p_session;
969 		if (sess->s_ttyvp == vp || sess->s_ttyp == NULL) {
970 			sx_sunlock(&proctree_lock);
971 			return (0);
972 		}
973 
974 		devfs_ctty_ref(vp);
975 		SESS_LOCK(sess);
976 		vpold = sess->s_ttyvp;
977 		sess->s_ttyvp = vp;
978 		sess->s_ttydp = cdev2priv(dev);
979 		SESS_UNLOCK(sess);
980 
981 		sx_sunlock(&proctree_lock);
982 
983 		/* Get rid of reference to old control tty */
984 		if (vpold)
985 			devfs_ctty_unref(vpold);
986 	}
987 	return (error);
988 }
989 
990 /* ARGSUSED */
991 static int
devfs_kqfilter_f(struct file * fp,struct knote * kn)992 devfs_kqfilter_f(struct file *fp, struct knote *kn)
993 {
994 	struct cdev *dev;
995 	struct cdevsw *dsw;
996 	int error, ref;
997 	struct file *fpop;
998 	struct thread *td;
999 
1000 	td = curthread;
1001 	fpop = td->td_fpop;
1002 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1003 	if (error)
1004 		return (error);
1005 	error = dsw->d_kqfilter(dev, kn);
1006 	td->td_fpop = fpop;
1007 	dev_relthread(dev, ref);
1008 	return (error);
1009 }
1010 
1011 static inline int
devfs_prison_check(struct devfs_dirent * de,struct thread * td)1012 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
1013 {
1014 	struct cdev_priv *cdp;
1015 	struct ucred *dcr;
1016 	struct proc *p;
1017 	int error;
1018 
1019 	cdp = de->de_cdp;
1020 	if (cdp == NULL)
1021 		return (0);
1022 	dcr = cdp->cdp_c.si_cred;
1023 	if (dcr == NULL)
1024 		return (0);
1025 
1026 	error = prison_check(td->td_ucred, dcr);
1027 	if (error == 0)
1028 		return (0);
1029 	/* We do, however, allow access to the controlling terminal */
1030 	p = td->td_proc;
1031 	PROC_LOCK(p);
1032 	if (!(p->p_flag & P_CONTROLT)) {
1033 		PROC_UNLOCK(p);
1034 		return (error);
1035 	}
1036 	if (p->p_session->s_ttydp == cdp)
1037 		error = 0;
1038 	PROC_UNLOCK(p);
1039 	return (error);
1040 }
1041 
1042 static int
devfs_lookupx(struct vop_lookup_args * ap,int * dm_unlock)1043 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
1044 {
1045 	struct componentname *cnp;
1046 	struct vnode *dvp, **vpp;
1047 	struct thread *td;
1048 	struct devfs_dirent *de, *dd;
1049 	struct devfs_dirent **dde;
1050 	struct devfs_mount *dmp;
1051 	struct mount *mp;
1052 	struct cdev *cdev;
1053 	int error, flags, nameiop, dvplocked;
1054 	char specname[SPECNAMELEN + 1], *pname;
1055 
1056 	td = curthread;
1057 	cnp = ap->a_cnp;
1058 	vpp = ap->a_vpp;
1059 	dvp = ap->a_dvp;
1060 	pname = cnp->cn_nameptr;
1061 	flags = cnp->cn_flags;
1062 	nameiop = cnp->cn_nameiop;
1063 	mp = dvp->v_mount;
1064 	dmp = VFSTODEVFS(mp);
1065 	dd = dvp->v_data;
1066 	*vpp = NULLVP;
1067 
1068 	if ((flags & ISLASTCN) && nameiop == RENAME)
1069 		return (EOPNOTSUPP);
1070 
1071 	if (dvp->v_type != VDIR)
1072 		return (ENOTDIR);
1073 
1074 	if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
1075 		return (EIO);
1076 
1077 	error = vn_dir_check_exec(dvp, cnp);
1078 	if (error != 0)
1079 		return (error);
1080 
1081 	if (cnp->cn_namelen == 1 && *pname == '.') {
1082 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
1083 			return (EINVAL);
1084 		*vpp = dvp;
1085 		VREF(dvp);
1086 		return (0);
1087 	}
1088 
1089 	if (flags & ISDOTDOT) {
1090 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
1091 			return (EINVAL);
1092 		de = devfs_parent_dirent(dd);
1093 		if (de == NULL)
1094 			return (ENOENT);
1095 		dvplocked = VOP_ISLOCKED(dvp);
1096 		VOP_UNLOCK(dvp);
1097 		error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK,
1098 		    vpp);
1099 		*dm_unlock = 0;
1100 		vn_lock(dvp, dvplocked | LK_RETRY);
1101 		return (error);
1102 	}
1103 
1104 	dd = dvp->v_data;
1105 	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
1106 	while (de == NULL) {	/* While(...) so we can use break */
1107 
1108 		if (nameiop == DELETE)
1109 			return (ENOENT);
1110 
1111 		/*
1112 		 * OK, we didn't have an entry for the name we were asked for
1113 		 * so we try to see if anybody can create it on demand.
1114 		 */
1115 		pname = devfs_fqpn(specname, dmp, dd, cnp);
1116 		if (pname == NULL)
1117 			break;
1118 
1119 		cdev = NULL;
1120 		DEVFS_DMP_HOLD(dmp);
1121 		sx_xunlock(&dmp->dm_lock);
1122 		EVENTHANDLER_INVOKE(dev_clone,
1123 		    td->td_ucred, pname, strlen(pname), &cdev);
1124 
1125 		if (cdev == NULL)
1126 			sx_xlock(&dmp->dm_lock);
1127 		else if (devfs_populate_vp(dvp) != 0) {
1128 			*dm_unlock = 0;
1129 			sx_xlock(&dmp->dm_lock);
1130 			if (DEVFS_DMP_DROP(dmp)) {
1131 				sx_xunlock(&dmp->dm_lock);
1132 				devfs_unmount_final(dmp);
1133 			} else
1134 				sx_xunlock(&dmp->dm_lock);
1135 			dev_rel(cdev);
1136 			return (ENOENT);
1137 		}
1138 		if (DEVFS_DMP_DROP(dmp)) {
1139 			*dm_unlock = 0;
1140 			sx_xunlock(&dmp->dm_lock);
1141 			devfs_unmount_final(dmp);
1142 			if (cdev != NULL)
1143 				dev_rel(cdev);
1144 			return (ENOENT);
1145 		}
1146 
1147 		if (cdev == NULL)
1148 			break;
1149 
1150 		dev_lock();
1151 		dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
1152 		if (dde != NULL && *dde != NULL)
1153 			de = *dde;
1154 		dev_unlock();
1155 		dev_rel(cdev);
1156 		break;
1157 	}
1158 
1159 	if (de == NULL || de->de_flags & DE_WHITEOUT) {
1160 		if ((nameiop == CREATE || nameiop == RENAME) &&
1161 		    (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
1162 			return (EJUSTRETURN);
1163 		}
1164 		return (ENOENT);
1165 	}
1166 
1167 	if (devfs_prison_check(de, td))
1168 		return (ENOENT);
1169 
1170 	if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
1171 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
1172 		if (error)
1173 			return (error);
1174 		if (*vpp == dvp) {
1175 			VREF(dvp);
1176 			*vpp = dvp;
1177 			return (0);
1178 		}
1179 	}
1180 	error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp);
1181 	*dm_unlock = 0;
1182 	return (error);
1183 }
1184 
1185 static int
devfs_lookup(struct vop_lookup_args * ap)1186 devfs_lookup(struct vop_lookup_args *ap)
1187 {
1188 	int j;
1189 	struct devfs_mount *dmp;
1190 	int dm_unlock;
1191 
1192 	if (devfs_populate_vp(ap->a_dvp) != 0)
1193 		return (ENOTDIR);
1194 
1195 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1196 	dm_unlock = 1;
1197 	j = devfs_lookupx(ap, &dm_unlock);
1198 	if (dm_unlock == 1)
1199 		sx_xunlock(&dmp->dm_lock);
1200 	return (j);
1201 }
1202 
1203 static int
devfs_mknod(struct vop_mknod_args * ap)1204 devfs_mknod(struct vop_mknod_args *ap)
1205 {
1206 	struct componentname *cnp;
1207 	struct vnode *dvp, **vpp;
1208 	struct devfs_dirent *dd, *de;
1209 	struct devfs_mount *dmp;
1210 	int error;
1211 
1212 	/*
1213 	 * The only type of node we should be creating here is a
1214 	 * character device, for anything else return EOPNOTSUPP.
1215 	 */
1216 	if (ap->a_vap->va_type != VCHR)
1217 		return (EOPNOTSUPP);
1218 	dvp = ap->a_dvp;
1219 	dmp = VFSTODEVFS(dvp->v_mount);
1220 
1221 	cnp = ap->a_cnp;
1222 	vpp = ap->a_vpp;
1223 	dd = dvp->v_data;
1224 
1225 	error = ENOENT;
1226 	sx_xlock(&dmp->dm_lock);
1227 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1228 		if (cnp->cn_namelen != de->de_dirent->d_namlen)
1229 			continue;
1230 		if (de->de_dirent->d_type == DT_CHR &&
1231 		    (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
1232 			continue;
1233 		if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1234 		    de->de_dirent->d_namlen) != 0)
1235 			continue;
1236 		if (de->de_flags & DE_WHITEOUT)
1237 			break;
1238 		goto notfound;
1239 	}
1240 	if (de == NULL)
1241 		goto notfound;
1242 	de->de_flags &= ~DE_WHITEOUT;
1243 	error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1244 	return (error);
1245 notfound:
1246 	sx_xunlock(&dmp->dm_lock);
1247 	return (error);
1248 }
1249 
1250 /* ARGSUSED */
1251 static int
devfs_open(struct vop_open_args * ap)1252 devfs_open(struct vop_open_args *ap)
1253 {
1254 	struct thread *td = ap->a_td;
1255 	struct vnode *vp = ap->a_vp;
1256 	struct cdev *dev = vp->v_rdev;
1257 	struct file *fp = ap->a_fp;
1258 	int error, ref, vlocked;
1259 	struct cdevsw *dsw;
1260 	struct file *fpop;
1261 
1262 	if (vp->v_type == VBLK)
1263 		return (ENXIO);
1264 
1265 	if (dev == NULL)
1266 		return (ENXIO);
1267 
1268 	/* Make this field valid before any I/O in d_open. */
1269 	if (dev->si_iosize_max == 0)
1270 		dev->si_iosize_max = DFLTPHYS;
1271 
1272 	dsw = dev_refthread(dev, &ref);
1273 	if (dsw == NULL)
1274 		return (ENXIO);
1275 	if (fp == NULL && dsw->d_fdopen != NULL) {
1276 		dev_relthread(dev, ref);
1277 		return (ENXIO);
1278 	}
1279 
1280 	if (vp->v_type == VCHR)
1281 		devfs_usecount_add(vp);
1282 
1283 	vlocked = VOP_ISLOCKED(vp);
1284 	VOP_UNLOCK(vp);
1285 
1286 	fpop = td->td_fpop;
1287 	td->td_fpop = fp;
1288 	if (fp != NULL) {
1289 		fp->f_data = dev;
1290 		fp->f_vnode = vp;
1291 	}
1292 	if (dsw->d_fdopen != NULL)
1293 		error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1294 	else
1295 		error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1296 	/* Clean up any cdevpriv upon error. */
1297 	if (error != 0)
1298 		devfs_clear_cdevpriv();
1299 	td->td_fpop = fpop;
1300 
1301 	vn_lock(vp, vlocked | LK_RETRY);
1302 	if (error != 0 && vp->v_type == VCHR)
1303 		devfs_usecount_sub(vp);
1304 
1305 	dev_relthread(dev, ref);
1306 	if (error != 0) {
1307 		if (error == ERESTART)
1308 			error = EINTR;
1309 		return (error);
1310 	}
1311 
1312 #if 0	/* /dev/console */
1313 	KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1314 #else
1315 	if (fp == NULL)
1316 		return (error);
1317 #endif
1318 	if (fp->f_ops == &badfileops)
1319 		finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1320 	return (error);
1321 }
1322 
1323 static int
devfs_pathconf(struct vop_pathconf_args * ap)1324 devfs_pathconf(struct vop_pathconf_args *ap)
1325 {
1326 
1327 	switch (ap->a_name) {
1328 	case _PC_FILESIZEBITS:
1329 		*ap->a_retval = 64;
1330 		return (0);
1331 	case _PC_NAME_MAX:
1332 		*ap->a_retval = NAME_MAX;
1333 		return (0);
1334 	case _PC_LINK_MAX:
1335 		*ap->a_retval = INT_MAX;
1336 		return (0);
1337 	case _PC_SYMLINK_MAX:
1338 		*ap->a_retval = MAXPATHLEN;
1339 		return (0);
1340 	case _PC_MAX_CANON:
1341 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1342 			*ap->a_retval = MAX_CANON;
1343 			return (0);
1344 		}
1345 		return (EINVAL);
1346 	case _PC_MAX_INPUT:
1347 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1348 			*ap->a_retval = MAX_INPUT;
1349 			return (0);
1350 		}
1351 		return (EINVAL);
1352 	case _PC_VDISABLE:
1353 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1354 			*ap->a_retval = _POSIX_VDISABLE;
1355 			return (0);
1356 		}
1357 		return (EINVAL);
1358 	case _PC_MAC_PRESENT:
1359 #ifdef MAC
1360 		/*
1361 		 * If MAC is enabled, devfs automatically supports
1362 		 * trivial non-persistent label storage.
1363 		 */
1364 		*ap->a_retval = 1;
1365 #else
1366 		*ap->a_retval = 0;
1367 #endif
1368 		return (0);
1369 	case _PC_CHOWN_RESTRICTED:
1370 		*ap->a_retval = 1;
1371 		return (0);
1372 	default:
1373 		return (vop_stdpathconf(ap));
1374 	}
1375 	/* NOTREACHED */
1376 }
1377 
1378 /* ARGSUSED */
1379 static int
devfs_poll_f(struct file * fp,int events,struct ucred * cred,struct thread * td)1380 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1381 {
1382 	struct cdev *dev;
1383 	struct cdevsw *dsw;
1384 	int error, ref;
1385 	struct file *fpop;
1386 
1387 	fpop = td->td_fpop;
1388 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1389 	if (error != 0) {
1390 		error = vnops.fo_poll(fp, events, cred, td);
1391 		return (error);
1392 	}
1393 	error = dsw->d_poll(dev, events, td);
1394 	td->td_fpop = fpop;
1395 	dev_relthread(dev, ref);
1396 	return(error);
1397 }
1398 
1399 /*
1400  * Print out the contents of a special device vnode.
1401  */
1402 static int
devfs_print(struct vop_print_args * ap)1403 devfs_print(struct vop_print_args *ap)
1404 {
1405 
1406 	printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1407 	return (0);
1408 }
1409 
1410 static int
devfs_read_f(struct file * fp,struct uio * uio,struct ucred * cred,int flags,struct thread * td)1411 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1412     int flags, struct thread *td)
1413 {
1414 	struct cdev *dev;
1415 	int ioflag, error, ref;
1416 	ssize_t resid;
1417 	struct cdevsw *dsw;
1418 	struct file *fpop;
1419 
1420 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1421 		return (EINVAL);
1422 	fpop = td->td_fpop;
1423 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1424 	if (error != 0) {
1425 		error = vnops.fo_read(fp, uio, cred, flags, td);
1426 		return (error);
1427 	}
1428 	resid = uio->uio_resid;
1429 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1430 	if (ioflag & O_DIRECT)
1431 		ioflag |= IO_DIRECT;
1432 
1433 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1434 	error = dsw->d_read(dev, uio, ioflag);
1435 	if (uio->uio_resid != resid || (error == 0 && resid != 0))
1436 		devfs_timestamp(&dev->si_atime);
1437 	td->td_fpop = fpop;
1438 	dev_relthread(dev, ref);
1439 
1440 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_R);
1441 	return (error);
1442 }
1443 
1444 static int
devfs_readdir(struct vop_readdir_args * ap)1445 devfs_readdir(struct vop_readdir_args *ap)
1446 {
1447 	int error;
1448 	struct uio *uio;
1449 	struct dirent *dp;
1450 	struct devfs_dirent *dd;
1451 	struct devfs_dirent *de;
1452 	struct devfs_mount *dmp;
1453 	off_t off;
1454 	int *tmp_ncookies = NULL;
1455 
1456 	if (ap->a_vp->v_type != VDIR)
1457 		return (ENOTDIR);
1458 
1459 	uio = ap->a_uio;
1460 	if (uio->uio_offset < 0)
1461 		return (EINVAL);
1462 
1463 	/*
1464 	 * XXX: This is a temporary hack to get around this filesystem not
1465 	 * supporting cookies. We store the location of the ncookies pointer
1466 	 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1467 	 * and set the number of cookies to 0. We then set the pointer to
1468 	 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1469 	 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1470 	 * pointer to its original location before returning to the caller.
1471 	 */
1472 	if (ap->a_ncookies != NULL) {
1473 		tmp_ncookies = ap->a_ncookies;
1474 		*ap->a_ncookies = 0;
1475 		ap->a_ncookies = NULL;
1476 	}
1477 
1478 	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1479 	if (devfs_populate_vp(ap->a_vp) != 0) {
1480 		if (tmp_ncookies != NULL)
1481 			ap->a_ncookies = tmp_ncookies;
1482 		return (EIO);
1483 	}
1484 	error = 0;
1485 	de = ap->a_vp->v_data;
1486 	off = 0;
1487 	TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1488 		KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1489 		if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1490 			continue;
1491 		if (devfs_prison_check(dd, uio->uio_td))
1492 			continue;
1493 		if (dd->de_dirent->d_type == DT_DIR)
1494 			de = dd->de_dir;
1495 		else
1496 			de = dd;
1497 		dp = dd->de_dirent;
1498 		MPASS(dp->d_reclen == GENERIC_DIRSIZ(dp));
1499 		if (dp->d_reclen > uio->uio_resid)
1500 			break;
1501 		dp->d_fileno = de->de_inode;
1502 		/* NOTE: d_off is the offset for the *next* entry. */
1503 		dp->d_off = off + dp->d_reclen;
1504 		if (off >= uio->uio_offset) {
1505 			error = vfs_read_dirent(ap, dp, off);
1506 			if (error)
1507 				break;
1508 		}
1509 		off += dp->d_reclen;
1510 	}
1511 	sx_xunlock(&dmp->dm_lock);
1512 	uio->uio_offset = off;
1513 
1514 	/*
1515 	 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1516 	 * place.
1517 	 */
1518 	if (tmp_ncookies != NULL)
1519 		ap->a_ncookies = tmp_ncookies;
1520 
1521 	return (error);
1522 }
1523 
1524 static int
devfs_readlink(struct vop_readlink_args * ap)1525 devfs_readlink(struct vop_readlink_args *ap)
1526 {
1527 	struct devfs_dirent *de;
1528 
1529 	de = ap->a_vp->v_data;
1530 	return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1531 }
1532 
1533 static void
devfs_reclaiml(struct vnode * vp)1534 devfs_reclaiml(struct vnode *vp)
1535 {
1536 	struct devfs_dirent *de;
1537 
1538 	mtx_assert(&devfs_de_interlock, MA_OWNED);
1539 	de = vp->v_data;
1540 	if (de != NULL) {
1541 		MPASS(de->de_usecount == 0);
1542 		de->de_vnode = NULL;
1543 		vp->v_data = NULL;
1544 	}
1545 }
1546 
1547 static int
devfs_reclaim(struct vop_reclaim_args * ap)1548 devfs_reclaim(struct vop_reclaim_args *ap)
1549 {
1550 	struct vnode *vp;
1551 
1552 	vp = ap->a_vp;
1553 	mtx_lock(&devfs_de_interlock);
1554 	devfs_reclaiml(vp);
1555 	mtx_unlock(&devfs_de_interlock);
1556 	return (0);
1557 }
1558 
1559 static int
devfs_reclaim_vchr(struct vop_reclaim_args * ap)1560 devfs_reclaim_vchr(struct vop_reclaim_args *ap)
1561 {
1562 	struct vnode *vp;
1563 	struct cdev *dev;
1564 
1565 	vp = ap->a_vp;
1566 	MPASS(vp->v_type == VCHR);
1567 
1568 	mtx_lock(&devfs_de_interlock);
1569 	VI_LOCK(vp);
1570 	devfs_usecount_subl(vp);
1571 	devfs_reclaiml(vp);
1572 	mtx_unlock(&devfs_de_interlock);
1573 	dev_lock();
1574 	dev = vp->v_rdev;
1575 	vp->v_rdev = NULL;
1576 	dev_unlock();
1577 	VI_UNLOCK(vp);
1578 	if (dev != NULL)
1579 		dev_rel(dev);
1580 	return (0);
1581 }
1582 
1583 static int
devfs_remove(struct vop_remove_args * ap)1584 devfs_remove(struct vop_remove_args *ap)
1585 {
1586 	struct vnode *dvp = ap->a_dvp;
1587 	struct vnode *vp = ap->a_vp;
1588 	struct devfs_dirent *dd;
1589 	struct devfs_dirent *de, *de_covered;
1590 	struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1591 
1592 	ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1593 	ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1594 
1595 	sx_xlock(&dmp->dm_lock);
1596 	dd = ap->a_dvp->v_data;
1597 	de = vp->v_data;
1598 	if (de->de_cdp == NULL) {
1599 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1600 		if (de->de_dirent->d_type == DT_LNK) {
1601 			de_covered = devfs_find(dd, de->de_dirent->d_name,
1602 			    de->de_dirent->d_namlen, 0);
1603 			if (de_covered != NULL)
1604 				de_covered->de_flags &= ~DE_COVERED;
1605 		}
1606 		/* We need to unlock dvp because devfs_delete() may lock it. */
1607 		VOP_UNLOCK(vp);
1608 		if (dvp != vp)
1609 			VOP_UNLOCK(dvp);
1610 		devfs_delete(dmp, de, 0);
1611 		sx_xunlock(&dmp->dm_lock);
1612 		if (dvp != vp)
1613 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1614 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1615 	} else {
1616 		de->de_flags |= DE_WHITEOUT;
1617 		sx_xunlock(&dmp->dm_lock);
1618 	}
1619 	return (0);
1620 }
1621 
1622 /*
1623  * Revoke is called on a tty when a terminal session ends.  The vnode
1624  * is orphaned by setting v_op to deadfs so we need to let go of it
1625  * as well so that we create a new one next time around.
1626  *
1627  */
1628 static int
devfs_revoke(struct vop_revoke_args * ap)1629 devfs_revoke(struct vop_revoke_args *ap)
1630 {
1631 	struct vnode *vp = ap->a_vp, *vp2;
1632 	struct cdev *dev;
1633 	struct cdev_priv *cdp;
1634 	struct devfs_dirent *de;
1635 	enum vgetstate vs;
1636 	u_int i;
1637 
1638 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1639 
1640 	dev = vp->v_rdev;
1641 	cdp = cdev2priv(dev);
1642 
1643 	dev_lock();
1644 	cdp->cdp_inuse++;
1645 	dev_unlock();
1646 
1647 	vhold(vp);
1648 	vgone(vp);
1649 	vdrop(vp);
1650 
1651 	VOP_UNLOCK(vp);
1652  loop:
1653 	for (;;) {
1654 		mtx_lock(&devfs_de_interlock);
1655 		dev_lock();
1656 		vp2 = NULL;
1657 		for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1658 			de = cdp->cdp_dirents[i];
1659 			if (de == NULL)
1660 				continue;
1661 
1662 			vp2 = de->de_vnode;
1663 			if (vp2 != NULL) {
1664 				dev_unlock();
1665 				vs = vget_prep(vp2);
1666 				mtx_unlock(&devfs_de_interlock);
1667 				if (vget_finish(vp2, LK_EXCLUSIVE, vs) != 0)
1668 					goto loop;
1669 				vhold(vp2);
1670 				vgone(vp2);
1671 				vdrop(vp2);
1672 				vput(vp2);
1673 				break;
1674 			}
1675 		}
1676 		if (vp2 != NULL) {
1677 			continue;
1678 		}
1679 		dev_unlock();
1680 		mtx_unlock(&devfs_de_interlock);
1681 		break;
1682 	}
1683 	dev_lock();
1684 	cdp->cdp_inuse--;
1685 	if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1686 		KASSERT((cdp->cdp_flags & CDP_ON_ACTIVE_LIST) != 0,
1687 		    ("%s: cdp %p (%s) not on active list",
1688 		    __func__, cdp, dev->si_name));
1689 		cdp->cdp_flags &= ~CDP_ON_ACTIVE_LIST;
1690 		TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1691 		dev_unlock();
1692 		dev_rel(&cdp->cdp_c);
1693 	} else
1694 		dev_unlock();
1695 
1696 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1697 	return (0);
1698 }
1699 
1700 static int
devfs_rioctl(struct vop_ioctl_args * ap)1701 devfs_rioctl(struct vop_ioctl_args *ap)
1702 {
1703 	struct vnode *vp;
1704 	struct devfs_mount *dmp;
1705 	int error;
1706 
1707 	vp = ap->a_vp;
1708 	vn_lock(vp, LK_SHARED | LK_RETRY);
1709 	if (VN_IS_DOOMED(vp)) {
1710 		VOP_UNLOCK(vp);
1711 		return (EBADF);
1712 	}
1713 	dmp = VFSTODEVFS(vp->v_mount);
1714 	sx_xlock(&dmp->dm_lock);
1715 	VOP_UNLOCK(vp);
1716 	DEVFS_DMP_HOLD(dmp);
1717 	devfs_populate(dmp);
1718 	if (DEVFS_DMP_DROP(dmp)) {
1719 		sx_xunlock(&dmp->dm_lock);
1720 		devfs_unmount_final(dmp);
1721 		return (ENOENT);
1722 	}
1723 	error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1724 	sx_xunlock(&dmp->dm_lock);
1725 	return (error);
1726 }
1727 
1728 static int
devfs_rread(struct vop_read_args * ap)1729 devfs_rread(struct vop_read_args *ap)
1730 {
1731 
1732 	if (ap->a_vp->v_type != VDIR)
1733 		return (EINVAL);
1734 	return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1735 }
1736 
1737 static int
devfs_setattr(struct vop_setattr_args * ap)1738 devfs_setattr(struct vop_setattr_args *ap)
1739 {
1740 	struct devfs_dirent *de;
1741 	struct vattr *vap;
1742 	struct vnode *vp;
1743 	struct thread *td;
1744 	int c, error;
1745 	uid_t uid;
1746 	gid_t gid;
1747 
1748 	vap = ap->a_vap;
1749 	vp = ap->a_vp;
1750 	td = curthread;
1751 	if ((vap->va_type != VNON) ||
1752 	    (vap->va_nlink != VNOVAL) ||
1753 	    (vap->va_fsid != VNOVAL) ||
1754 	    (vap->va_fileid != VNOVAL) ||
1755 	    (vap->va_blocksize != VNOVAL) ||
1756 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1757 	    (vap->va_rdev != VNOVAL) ||
1758 	    ((int)vap->va_bytes != VNOVAL) ||
1759 	    (vap->va_gen != VNOVAL)) {
1760 		return (EINVAL);
1761 	}
1762 
1763 	error = devfs_populate_vp(vp);
1764 	if (error != 0)
1765 		return (error);
1766 
1767 	de = vp->v_data;
1768 	if (vp->v_type == VDIR)
1769 		de = de->de_dir;
1770 
1771 	c = 0;
1772 	if (vap->va_uid == (uid_t)VNOVAL)
1773 		uid = de->de_uid;
1774 	else
1775 		uid = vap->va_uid;
1776 	if (vap->va_gid == (gid_t)VNOVAL)
1777 		gid = de->de_gid;
1778 	else
1779 		gid = vap->va_gid;
1780 	if (uid != de->de_uid || gid != de->de_gid) {
1781 		if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1782 		    (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1783 			error = priv_check(td, PRIV_VFS_CHOWN);
1784 			if (error != 0)
1785 				goto ret;
1786 		}
1787 		de->de_uid = uid;
1788 		de->de_gid = gid;
1789 		c = 1;
1790 	}
1791 
1792 	if (vap->va_mode != (mode_t)VNOVAL) {
1793 		if (ap->a_cred->cr_uid != de->de_uid) {
1794 			error = priv_check(td, PRIV_VFS_ADMIN);
1795 			if (error != 0)
1796 				goto ret;
1797 		}
1798 		de->de_mode = vap->va_mode;
1799 		c = 1;
1800 	}
1801 
1802 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1803 		error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1804 		if (error != 0)
1805 			goto ret;
1806 		if (vap->va_atime.tv_sec != VNOVAL) {
1807 			if (vp->v_type == VCHR)
1808 				vp->v_rdev->si_atime = vap->va_atime;
1809 			else
1810 				de->de_atime = vap->va_atime;
1811 		}
1812 		if (vap->va_mtime.tv_sec != VNOVAL) {
1813 			if (vp->v_type == VCHR)
1814 				vp->v_rdev->si_mtime = vap->va_mtime;
1815 			else
1816 				de->de_mtime = vap->va_mtime;
1817 		}
1818 		c = 1;
1819 	}
1820 
1821 	if (c) {
1822 		if (vp->v_type == VCHR)
1823 			vfs_timestamp(&vp->v_rdev->si_ctime);
1824 		else
1825 			vfs_timestamp(&de->de_mtime);
1826 	}
1827 
1828 ret:
1829 	sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock);
1830 	return (error);
1831 }
1832 
1833 #ifdef MAC
1834 static int
devfs_setlabel(struct vop_setlabel_args * ap)1835 devfs_setlabel(struct vop_setlabel_args *ap)
1836 {
1837 	struct vnode *vp;
1838 	struct devfs_dirent *de;
1839 
1840 	vp = ap->a_vp;
1841 	de = vp->v_data;
1842 
1843 	mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1844 	mac_devfs_update(vp->v_mount, de, vp);
1845 
1846 	return (0);
1847 }
1848 #endif
1849 
1850 static int
devfs_stat_f(struct file * fp,struct stat * sb,struct ucred * cred)1851 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred)
1852 {
1853 
1854 	return (vnops.fo_stat(fp, sb, cred));
1855 }
1856 
1857 static int
devfs_symlink(struct vop_symlink_args * ap)1858 devfs_symlink(struct vop_symlink_args *ap)
1859 {
1860 	int i, error;
1861 	struct devfs_dirent *dd;
1862 	struct devfs_dirent *de, *de_covered, *de_dotdot;
1863 	struct devfs_mount *dmp;
1864 
1865 	error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1866 	if (error)
1867 		return(error);
1868 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1869 	if (devfs_populate_vp(ap->a_dvp) != 0)
1870 		return (ENOENT);
1871 
1872 	dd = ap->a_dvp->v_data;
1873 	de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1874 	de->de_flags = DE_USER;
1875 	de->de_uid = 0;
1876 	de->de_gid = 0;
1877 	de->de_mode = 0755;
1878 	de->de_inode = alloc_unr(devfs_inos);
1879 	de->de_dir = dd;
1880 	de->de_dirent->d_type = DT_LNK;
1881 	i = strlen(ap->a_target) + 1;
1882 	de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1883 	bcopy(ap->a_target, de->de_symlink, i);
1884 #ifdef MAC
1885 	mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1886 #endif
1887 	de_covered = devfs_find(dd, de->de_dirent->d_name,
1888 	    de->de_dirent->d_namlen, 0);
1889 	if (de_covered != NULL) {
1890 		if ((de_covered->de_flags & DE_USER) != 0) {
1891 			devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1892 			sx_xunlock(&dmp->dm_lock);
1893 			return (EEXIST);
1894 		}
1895 		KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1896 		    ("devfs_symlink: entry %p already covered", de_covered));
1897 		de_covered->de_flags |= DE_COVERED;
1898 	}
1899 
1900 	de_dotdot = TAILQ_FIRST(&dd->de_dlist);		/* "." */
1901 	de_dotdot = TAILQ_NEXT(de_dotdot, de_list);	/* ".." */
1902 	TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1903 	devfs_dir_ref_de(dmp, dd);
1904 	devfs_rules_apply(dmp, de);
1905 
1906 	return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1907 }
1908 
1909 static int
devfs_truncate_f(struct file * fp,off_t length,struct ucred * cred,struct thread * td)1910 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1911 {
1912 
1913 	return (vnops.fo_truncate(fp, length, cred, td));
1914 }
1915 
1916 static int
devfs_write_f(struct file * fp,struct uio * uio,struct ucred * cred,int flags,struct thread * td)1917 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1918     int flags, struct thread *td)
1919 {
1920 	struct cdev *dev;
1921 	int error, ioflag, ref;
1922 	ssize_t resid;
1923 	struct cdevsw *dsw;
1924 	struct file *fpop;
1925 
1926 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1927 		return (EINVAL);
1928 	fpop = td->td_fpop;
1929 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1930 	if (error != 0) {
1931 		error = vnops.fo_write(fp, uio, cred, flags, td);
1932 		return (error);
1933 	}
1934 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1935 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1936 	if (ioflag & O_DIRECT)
1937 		ioflag |= IO_DIRECT;
1938 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1939 
1940 	resid = uio->uio_resid;
1941 
1942 	error = dsw->d_write(dev, uio, ioflag);
1943 	if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1944 		devfs_timestamp(&dev->si_ctime);
1945 		dev->si_mtime = dev->si_ctime;
1946 	}
1947 	td->td_fpop = fpop;
1948 	dev_relthread(dev, ref);
1949 
1950 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_W);
1951 	return (error);
1952 }
1953 
1954 static int
devfs_mmap_f(struct file * fp,vm_map_t map,vm_offset_t * addr,vm_size_t size,vm_prot_t prot,vm_prot_t cap_maxprot,int flags,vm_ooffset_t foff,struct thread * td)1955 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1956     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1957     struct thread *td)
1958 {
1959 	struct cdev *dev;
1960 	struct cdevsw *dsw;
1961 	struct mount *mp;
1962 	struct vnode *vp;
1963 	struct file *fpop;
1964 	vm_object_t object;
1965 	vm_prot_t maxprot;
1966 	int error, ref;
1967 
1968 	vp = fp->f_vnode;
1969 
1970 	/*
1971 	 * Ensure that file and memory protections are
1972 	 * compatible.
1973 	 */
1974 	mp = vp->v_mount;
1975 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1976 		maxprot = VM_PROT_NONE;
1977 		if ((prot & VM_PROT_EXECUTE) != 0)
1978 			return (EACCES);
1979 	} else
1980 		maxprot = VM_PROT_EXECUTE;
1981 	if ((fp->f_flag & FREAD) != 0)
1982 		maxprot |= VM_PROT_READ;
1983 	else if ((prot & VM_PROT_READ) != 0)
1984 		return (EACCES);
1985 
1986 	/*
1987 	 * If we are sharing potential changes via MAP_SHARED and we
1988 	 * are trying to get write permission although we opened it
1989 	 * without asking for it, bail out.
1990 	 *
1991 	 * Note that most character devices always share mappings.
1992 	 * The one exception is that D_MMAP_ANON devices
1993 	 * (i.e. /dev/zero) permit private writable mappings.
1994 	 *
1995 	 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests
1996 	 * as well as updating maxprot to permit writing for
1997 	 * D_MMAP_ANON devices rather than doing that here.
1998 	 */
1999 	if ((flags & MAP_SHARED) != 0) {
2000 		if ((fp->f_flag & FWRITE) != 0)
2001 			maxprot |= VM_PROT_WRITE;
2002 		else if ((prot & VM_PROT_WRITE) != 0)
2003 			return (EACCES);
2004 	}
2005 	maxprot &= cap_maxprot;
2006 
2007 	fpop = td->td_fpop;
2008 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
2009 	if (error != 0)
2010 		return (error);
2011 
2012 	error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff,
2013 	    &object);
2014 	td->td_fpop = fpop;
2015 	dev_relthread(dev, ref);
2016 	if (error != 0)
2017 		return (error);
2018 
2019 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2020 	    foff, FALSE, td);
2021 	if (error != 0)
2022 		vm_object_deallocate(object);
2023 	return (error);
2024 }
2025 
2026 dev_t
dev2udev(struct cdev * x)2027 dev2udev(struct cdev *x)
2028 {
2029 	if (x == NULL)
2030 		return (NODEV);
2031 	return (cdev2priv(x)->cdp_inode);
2032 }
2033 
2034 static int
devfs_cmp_f(struct file * fp1,struct file * fp2,struct thread * td)2035 devfs_cmp_f(struct file *fp1, struct file *fp2, struct thread *td)
2036 {
2037 	if (fp2->f_type != DTYPE_VNODE || fp2->f_ops != &devfs_ops_f)
2038 		return (3);
2039 	return (kcmp_cmp((uintptr_t)fp1->f_data, (uintptr_t)fp2->f_data));
2040 }
2041 
2042 static struct fileops devfs_ops_f = {
2043 	.fo_read =	devfs_read_f,
2044 	.fo_write =	devfs_write_f,
2045 	.fo_truncate =	devfs_truncate_f,
2046 	.fo_ioctl =	devfs_ioctl_f,
2047 	.fo_poll =	devfs_poll_f,
2048 	.fo_kqfilter =	devfs_kqfilter_f,
2049 	.fo_stat =	devfs_stat_f,
2050 	.fo_close =	devfs_close_f,
2051 	.fo_chmod =	vn_chmod,
2052 	.fo_chown =	vn_chown,
2053 	.fo_sendfile =	vn_sendfile,
2054 	.fo_seek =	vn_seek,
2055 	.fo_fill_kinfo = vn_fill_kinfo,
2056 	.fo_mmap =	devfs_mmap_f,
2057 	.fo_cmp =	devfs_cmp_f,
2058 	.fo_flags =	DFLAG_PASSABLE | DFLAG_SEEKABLE
2059 };
2060 
2061 /* Vops for non-CHR vnodes in /dev. */
2062 static struct vop_vector devfs_vnodeops = {
2063 	.vop_default =		&default_vnodeops,
2064 
2065 	.vop_access =		devfs_access,
2066 	.vop_getattr =		devfs_getattr,
2067 	.vop_ioctl =		devfs_rioctl,
2068 	.vop_lookup =		devfs_lookup,
2069 	.vop_mknod =		devfs_mknod,
2070 	.vop_pathconf =		devfs_pathconf,
2071 	.vop_read =		devfs_rread,
2072 	.vop_readdir =		devfs_readdir,
2073 	.vop_readlink =		devfs_readlink,
2074 	.vop_reclaim =		devfs_reclaim,
2075 	.vop_remove =		devfs_remove,
2076 	.vop_revoke =		devfs_revoke,
2077 	.vop_setattr =		devfs_setattr,
2078 #ifdef MAC
2079 	.vop_setlabel =		devfs_setlabel,
2080 #endif
2081 	.vop_symlink =		devfs_symlink,
2082 	.vop_vptocnp =		devfs_vptocnp,
2083 	.vop_lock1 =		vop_lock,
2084 	.vop_unlock =		vop_unlock,
2085 	.vop_islocked =		vop_islocked,
2086 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
2087 };
2088 VFS_VOP_VECTOR_REGISTER(devfs_vnodeops);
2089 
2090 /* Vops for VCHR vnodes in /dev. */
2091 static struct vop_vector devfs_specops = {
2092 	.vop_default =		&default_vnodeops,
2093 
2094 	.vop_access =		devfs_access,
2095 	.vop_bmap =		VOP_PANIC,
2096 	.vop_close =		devfs_close,
2097 	.vop_create =		VOP_PANIC,
2098 	.vop_fsync =		vop_stdfsync,
2099 	.vop_getattr =		devfs_getattr,
2100 	.vop_ioctl =		devfs_ioctl,
2101 	.vop_link =		VOP_PANIC,
2102 	.vop_mkdir =		VOP_PANIC,
2103 	.vop_mknod =		VOP_PANIC,
2104 	.vop_open =		devfs_open,
2105 	.vop_pathconf =		devfs_pathconf,
2106 	.vop_poll =		dead_poll,
2107 	.vop_print =		devfs_print,
2108 	.vop_read =		dead_read,
2109 	.vop_readdir =		VOP_PANIC,
2110 	.vop_readlink =		VOP_PANIC,
2111 	.vop_reallocblks =	VOP_PANIC,
2112 	.vop_reclaim =		devfs_reclaim_vchr,
2113 	.vop_remove =		devfs_remove,
2114 	.vop_rename =		VOP_PANIC,
2115 	.vop_revoke =		devfs_revoke,
2116 	.vop_rmdir =		VOP_PANIC,
2117 	.vop_setattr =		devfs_setattr,
2118 #ifdef MAC
2119 	.vop_setlabel =		devfs_setlabel,
2120 #endif
2121 	.vop_strategy =		VOP_PANIC,
2122 	.vop_symlink =		VOP_PANIC,
2123 	.vop_vptocnp =		devfs_vptocnp,
2124 	.vop_write =		dead_write,
2125 	.vop_lock1 =		vop_lock,
2126 	.vop_unlock =		vop_unlock,
2127 	.vop_islocked =		vop_islocked,
2128 	.vop_add_writecount =	vop_stdadd_writecount_nomsync,
2129 };
2130 VFS_VOP_VECTOR_REGISTER(devfs_specops);
2131 
2132 /*
2133  * Our calling convention to the device drivers used to be that we passed
2134  * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
2135  * flags instead since that's what open(), close() and ioctl() takes and
2136  * we don't really want vnode.h in device drivers.
2137  * We solved the source compatibility by redefining some vnode flags to
2138  * be the same as the fcntl ones and by sending down the bitwise OR of
2139  * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
2140  * pulls the rug out under this.
2141  */
2142 CTASSERT(O_NONBLOCK == IO_NDELAY);
2143 CTASSERT(O_FSYNC == IO_SYNC);
2144