xref: /freebsd-13-stable/sys/fs/devfs/devfs_vnops.c (revision 7c23465541f328acab2192c870be31324ce436d9)
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 sx	clone_drain_lock;
87 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
88 struct mtx	cdevpriv_mtx;
89 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
90 
91 SYSCTL_DECL(_vfs_devfs);
92 
93 static int devfs_dotimes;
94 SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW,
95     &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision");
96 
97 /*
98  * Update devfs node timestamp.  Note that updates are unlocked and
99  * stat(2) could see partially updated times.
100  */
101 static void
devfs_timestamp(struct timespec * tsp)102 devfs_timestamp(struct timespec *tsp)
103 {
104 	time_t ts;
105 
106 	if (devfs_dotimes) {
107 		vfs_timestamp(tsp);
108 	} else {
109 		ts = time_second;
110 		if (tsp->tv_sec != ts) {
111 			tsp->tv_sec = ts;
112 			tsp->tv_nsec = 0;
113 		}
114 	}
115 }
116 
117 static int
devfs_fp_check(struct file * fp,struct cdev ** devp,struct cdevsw ** dswp,int * ref)118 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
119     int *ref)
120 {
121 
122 	*dswp = devvn_refthread(fp->f_vnode, devp, ref);
123 	if (*devp != fp->f_data) {
124 		if (*dswp != NULL)
125 			dev_relthread(*devp, *ref);
126 		return (ENXIO);
127 	}
128 	KASSERT((*devp)->si_refcount > 0,
129 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
130 	if (*dswp == NULL)
131 		return (ENXIO);
132 	curthread->td_fpop = fp;
133 	return (0);
134 }
135 
136 int
devfs_get_cdevpriv(void ** datap)137 devfs_get_cdevpriv(void **datap)
138 {
139 	struct file *fp;
140 	struct cdev_privdata *p;
141 	int error;
142 
143 	fp = curthread->td_fpop;
144 	if (fp == NULL)
145 		return (EBADF);
146 	p = fp->f_cdevpriv;
147 	if (p != NULL) {
148 		error = 0;
149 		*datap = p->cdpd_data;
150 	} else
151 		error = ENOENT;
152 	return (error);
153 }
154 
155 int
devfs_set_cdevpriv(void * priv,d_priv_dtor_t * priv_dtr)156 devfs_set_cdevpriv(void *priv, d_priv_dtor_t *priv_dtr)
157 {
158 	struct file *fp;
159 	struct cdev_priv *cdp;
160 	struct cdev_privdata *p;
161 	int error;
162 
163 	fp = curthread->td_fpop;
164 	if (fp == NULL)
165 		return (ENOENT);
166 	cdp = cdev2priv((struct cdev *)fp->f_data);
167 	p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
168 	p->cdpd_data = priv;
169 	p->cdpd_dtr = priv_dtr;
170 	p->cdpd_fp = fp;
171 	mtx_lock(&cdevpriv_mtx);
172 	if (fp->f_cdevpriv == NULL) {
173 		LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
174 		fp->f_cdevpriv = p;
175 		mtx_unlock(&cdevpriv_mtx);
176 		error = 0;
177 	} else {
178 		mtx_unlock(&cdevpriv_mtx);
179 		free(p, M_CDEVPDATA);
180 		error = EBUSY;
181 	}
182 	return (error);
183 }
184 
185 int
devfs_foreach_cdevpriv(struct cdev * dev,int (* cb)(void * data,void * arg),void * arg)186 devfs_foreach_cdevpriv(struct cdev *dev, int (*cb)(void *data, void *arg),
187     void *arg)
188 {
189 	struct cdev_priv *cdp;
190 	struct cdev_privdata *p;
191 	int error;
192 
193 	cdp = cdev2priv(dev);
194 	error = 0;
195 	mtx_lock(&cdevpriv_mtx);
196 	LIST_FOREACH(p, &cdp->cdp_fdpriv, cdpd_list) {
197 		error = cb(p->cdpd_data, arg);
198 		if (error != 0)
199 			break;
200 	}
201 	mtx_unlock(&cdevpriv_mtx);
202 	return (error);
203 }
204 
205 void
devfs_destroy_cdevpriv(struct cdev_privdata * p)206 devfs_destroy_cdevpriv(struct cdev_privdata *p)
207 {
208 
209 	mtx_assert(&cdevpriv_mtx, MA_OWNED);
210 	KASSERT(p->cdpd_fp->f_cdevpriv == p,
211 	    ("devfs_destoy_cdevpriv %p != %p", p->cdpd_fp->f_cdevpriv, p));
212 	p->cdpd_fp->f_cdevpriv = NULL;
213 	LIST_REMOVE(p, cdpd_list);
214 	mtx_unlock(&cdevpriv_mtx);
215 	(p->cdpd_dtr)(p->cdpd_data);
216 	free(p, M_CDEVPDATA);
217 }
218 
219 static void
devfs_fpdrop(struct file * fp)220 devfs_fpdrop(struct file *fp)
221 {
222 	struct cdev_privdata *p;
223 
224 	mtx_lock(&cdevpriv_mtx);
225 	if ((p = fp->f_cdevpriv) == NULL) {
226 		mtx_unlock(&cdevpriv_mtx);
227 		return;
228 	}
229 	devfs_destroy_cdevpriv(p);
230 }
231 
232 void
devfs_clear_cdevpriv(void)233 devfs_clear_cdevpriv(void)
234 {
235 	struct file *fp;
236 
237 	fp = curthread->td_fpop;
238 	if (fp == NULL)
239 		return;
240 	devfs_fpdrop(fp);
241 }
242 
243 static void
devfs_usecount_add(struct vnode * vp)244 devfs_usecount_add(struct vnode *vp)
245 {
246 	struct devfs_dirent *de;
247 	struct cdev *dev;
248 
249 	mtx_lock(&devfs_de_interlock);
250 	VI_LOCK(vp);
251 	VNPASS(vp->v_type == VCHR || vp->v_type == VBAD, vp);
252 	if (VN_IS_DOOMED(vp)) {
253 		goto out_unlock;
254 	}
255 
256 	de = vp->v_data;
257 	dev = vp->v_rdev;
258 	MPASS(de != NULL);
259 	MPASS(dev != NULL);
260 	dev->si_usecount++;
261 	de->de_usecount++;
262 out_unlock:
263 	VI_UNLOCK(vp);
264 	mtx_unlock(&devfs_de_interlock);
265 }
266 
267 static void
devfs_usecount_subl(struct vnode * vp)268 devfs_usecount_subl(struct vnode *vp)
269 {
270 	struct devfs_dirent *de;
271 	struct cdev *dev;
272 
273 	mtx_assert(&devfs_de_interlock, MA_OWNED);
274 	ASSERT_VI_LOCKED(vp, __func__);
275 	VNPASS(vp->v_type == VCHR || vp->v_type == VBAD, vp);
276 
277 	de = vp->v_data;
278 	dev = vp->v_rdev;
279 	if (de == NULL)
280 		return;
281 	if (dev == NULL) {
282 		MPASS(de->de_usecount == 0);
283 		return;
284 	}
285 	if (dev->si_usecount < de->de_usecount)
286 		panic("%s: si_usecount underflow for dev %p "
287 		    "(has %ld, dirent has %d)\n",
288 		    __func__, dev, dev->si_usecount, de->de_usecount);
289 	if (VN_IS_DOOMED(vp)) {
290 		dev->si_usecount -= de->de_usecount;
291 		de->de_usecount = 0;
292 	} else {
293 		if (de->de_usecount == 0)
294 			panic("%s: de_usecount underflow for dev %p\n",
295 			    __func__, dev);
296 		dev->si_usecount--;
297 		de->de_usecount--;
298 	}
299 }
300 
301 static void
devfs_usecount_sub(struct vnode * vp)302 devfs_usecount_sub(struct vnode *vp)
303 {
304 
305 	mtx_lock(&devfs_de_interlock);
306 	VI_LOCK(vp);
307 	devfs_usecount_subl(vp);
308 	VI_UNLOCK(vp);
309 	mtx_unlock(&devfs_de_interlock);
310 }
311 
312 static int
devfs_usecountl(struct vnode * vp)313 devfs_usecountl(struct vnode *vp)
314 {
315 
316 	VNPASS(vp->v_type == VCHR, vp);
317 	mtx_assert(&devfs_de_interlock, MA_OWNED);
318 	ASSERT_VI_LOCKED(vp, __func__);
319 	return (vp->v_rdev->si_usecount);
320 }
321 
322 int
devfs_usecount(struct vnode * vp)323 devfs_usecount(struct vnode *vp)
324 {
325 	int count;
326 
327 	VNPASS(vp->v_type == VCHR, vp);
328 	mtx_lock(&devfs_de_interlock);
329 	VI_LOCK(vp);
330 	count = devfs_usecountl(vp);
331 	VI_UNLOCK(vp);
332 	mtx_unlock(&devfs_de_interlock);
333 	return (count);
334 }
335 
336 void
devfs_ctty_ref(struct vnode * vp)337 devfs_ctty_ref(struct vnode *vp)
338 {
339 
340 	vrefact(vp);
341 	devfs_usecount_add(vp);
342 }
343 
344 void
devfs_ctty_unref(struct vnode * vp)345 devfs_ctty_unref(struct vnode *vp)
346 {
347 
348 	devfs_usecount_sub(vp);
349 	vrele(vp);
350 }
351 
352 /*
353  * On success devfs_populate_vp() returns with dmp->dm_lock held.
354  */
355 static int
devfs_populate_vp(struct vnode * vp)356 devfs_populate_vp(struct vnode *vp)
357 {
358 	struct devfs_dirent *de;
359 	struct devfs_mount *dmp;
360 	int locked;
361 
362 	ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
363 
364 	dmp = VFSTODEVFS(vp->v_mount);
365 	if (!devfs_populate_needed(dmp)) {
366 		sx_xlock(&dmp->dm_lock);
367 		goto out_nopopulate;
368 	}
369 
370 	locked = VOP_ISLOCKED(vp);
371 
372 	sx_xlock(&dmp->dm_lock);
373 	DEVFS_DMP_HOLD(dmp);
374 
375 	/* Can't call devfs_populate() with the vnode lock held. */
376 	VOP_UNLOCK(vp);
377 	devfs_populate(dmp);
378 
379 	sx_xunlock(&dmp->dm_lock);
380 	vn_lock(vp, locked | LK_RETRY);
381 	sx_xlock(&dmp->dm_lock);
382 	if (DEVFS_DMP_DROP(dmp)) {
383 		sx_xunlock(&dmp->dm_lock);
384 		devfs_unmount_final(dmp);
385 		return (ERESTART);
386 	}
387 out_nopopulate:
388 	if (VN_IS_DOOMED(vp)) {
389 		sx_xunlock(&dmp->dm_lock);
390 		return (ERESTART);
391 	}
392 	de = vp->v_data;
393 	KASSERT(de != NULL,
394 	    ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
395 	if ((de->de_flags & DE_DOOMED) != 0) {
396 		sx_xunlock(&dmp->dm_lock);
397 		return (ERESTART);
398 	}
399 
400 	return (0);
401 }
402 
403 static int
devfs_vptocnp(struct vop_vptocnp_args * ap)404 devfs_vptocnp(struct vop_vptocnp_args *ap)
405 {
406 	struct vnode *vp = ap->a_vp;
407 	struct vnode **dvp = ap->a_vpp;
408 	struct devfs_mount *dmp;
409 	char *buf = ap->a_buf;
410 	size_t *buflen = ap->a_buflen;
411 	struct devfs_dirent *dd, *de;
412 	int i, error;
413 
414 	dmp = VFSTODEVFS(vp->v_mount);
415 
416 	error = devfs_populate_vp(vp);
417 	if (error != 0)
418 		return (error);
419 
420 	if (vp->v_type != VCHR && vp->v_type != VDIR) {
421 		error = ENOENT;
422 		goto finished;
423 	}
424 
425 	dd = vp->v_data;
426 	if (vp->v_type == VDIR && dd == dmp->dm_rootdir) {
427 		*dvp = vp;
428 		vref(*dvp);
429 		goto finished;
430 	}
431 
432 	i = *buflen;
433 	i -= dd->de_dirent->d_namlen;
434 	if (i < 0) {
435 		error = ENOMEM;
436 		goto finished;
437 	}
438 	bcopy(dd->de_dirent->d_name, buf + i, dd->de_dirent->d_namlen);
439 	*buflen = i;
440 	de = devfs_parent_dirent(dd);
441 	if (de == NULL) {
442 		error = ENOENT;
443 		goto finished;
444 	}
445 	mtx_lock(&devfs_de_interlock);
446 	*dvp = de->de_vnode;
447 	if (*dvp != NULL) {
448 		VI_LOCK(*dvp);
449 		mtx_unlock(&devfs_de_interlock);
450 		vholdl(*dvp);
451 		VI_UNLOCK(*dvp);
452 		vref(*dvp);
453 		vdrop(*dvp);
454 	} else {
455 		mtx_unlock(&devfs_de_interlock);
456 		error = ENOENT;
457 	}
458 finished:
459 	sx_xunlock(&dmp->dm_lock);
460 	return (error);
461 }
462 
463 /*
464  * Construct the fully qualified path name relative to the mountpoint.
465  * If a NULL cnp is provided, no '/' is appended to the resulting path.
466  */
467 char *
devfs_fqpn(char * buf,struct devfs_mount * dmp,struct devfs_dirent * dd,struct componentname * cnp)468 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
469     struct componentname *cnp)
470 {
471 	int i;
472 	struct devfs_dirent *de;
473 
474 	sx_assert(&dmp->dm_lock, SA_LOCKED);
475 
476 	i = SPECNAMELEN;
477 	buf[i] = '\0';
478 	if (cnp != NULL)
479 		i -= cnp->cn_namelen;
480 	if (i < 0)
481 		 return (NULL);
482 	if (cnp != NULL)
483 		bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
484 	de = dd;
485 	while (de != dmp->dm_rootdir) {
486 		if (cnp != NULL || i < SPECNAMELEN) {
487 			i--;
488 			if (i < 0)
489 				 return (NULL);
490 			buf[i] = '/';
491 		}
492 		i -= de->de_dirent->d_namlen;
493 		if (i < 0)
494 			 return (NULL);
495 		bcopy(de->de_dirent->d_name, buf + i,
496 		    de->de_dirent->d_namlen);
497 		de = devfs_parent_dirent(de);
498 		if (de == NULL)
499 			return (NULL);
500 	}
501 	return (buf + i);
502 }
503 
504 static int
devfs_allocv_drop_refs(int drop_dm_lock,struct devfs_mount * dmp,struct devfs_dirent * de)505 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
506 	struct devfs_dirent *de)
507 {
508 	int not_found;
509 
510 	not_found = 0;
511 	if (de->de_flags & DE_DOOMED)
512 		not_found = 1;
513 	if (DEVFS_DE_DROP(de)) {
514 		KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
515 		devfs_dirent_free(de);
516 	}
517 	if (DEVFS_DMP_DROP(dmp)) {
518 		KASSERT(not_found == 1,
519 			("DEVFS mount struct freed before dirent"));
520 		not_found = 2;
521 		sx_xunlock(&dmp->dm_lock);
522 		devfs_unmount_final(dmp);
523 	}
524 	if (not_found == 1 || (drop_dm_lock && not_found != 2))
525 		sx_unlock(&dmp->dm_lock);
526 	return (not_found);
527 }
528 
529 static void
devfs_insmntque_dtr(struct vnode * vp,void * arg)530 devfs_insmntque_dtr(struct vnode *vp, void *arg)
531 {
532 	struct devfs_dirent *de;
533 
534 	de = (struct devfs_dirent *)arg;
535 	mtx_lock(&devfs_de_interlock);
536 	vp->v_data = NULL;
537 	de->de_vnode = NULL;
538 	mtx_unlock(&devfs_de_interlock);
539 	vgone(vp);
540 	vput(vp);
541 }
542 
543 /*
544  * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
545  * it on return.
546  */
547 int
devfs_allocv(struct devfs_dirent * de,struct mount * mp,int lockmode,struct vnode ** vpp)548 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
549     struct vnode **vpp)
550 {
551 	int error;
552 	struct vnode *vp;
553 	struct cdev *dev;
554 	struct devfs_mount *dmp;
555 	struct cdevsw *dsw;
556 	enum vgetstate vs;
557 
558 	dmp = VFSTODEVFS(mp);
559 	if (de->de_flags & DE_DOOMED) {
560 		sx_xunlock(&dmp->dm_lock);
561 		return (ENOENT);
562 	}
563 loop:
564 	DEVFS_DE_HOLD(de);
565 	DEVFS_DMP_HOLD(dmp);
566 	mtx_lock(&devfs_de_interlock);
567 	vp = de->de_vnode;
568 	if (vp != NULL) {
569 		vs = vget_prep(vp);
570 		mtx_unlock(&devfs_de_interlock);
571 		sx_xunlock(&dmp->dm_lock);
572 		vget_finish(vp, lockmode | LK_RETRY, vs);
573 		sx_xlock(&dmp->dm_lock);
574 		if (devfs_allocv_drop_refs(0, dmp, de)) {
575 			vput(vp);
576 			return (ENOENT);
577 		} else if (VN_IS_DOOMED(vp)) {
578 			mtx_lock(&devfs_de_interlock);
579 			if (de->de_vnode == vp) {
580 				de->de_vnode = NULL;
581 				vp->v_data = NULL;
582 			}
583 			mtx_unlock(&devfs_de_interlock);
584 			vput(vp);
585 			goto loop;
586 		}
587 		sx_xunlock(&dmp->dm_lock);
588 		*vpp = vp;
589 		return (0);
590 	}
591 	mtx_unlock(&devfs_de_interlock);
592 	if (de->de_dirent->d_type == DT_CHR) {
593 		if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
594 			devfs_allocv_drop_refs(1, dmp, de);
595 			return (ENOENT);
596 		}
597 		dev = &de->de_cdp->cdp_c;
598 	} else {
599 		dev = NULL;
600 	}
601 	error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
602 	if (error != 0) {
603 		devfs_allocv_drop_refs(1, dmp, de);
604 		printf("devfs_allocv: failed to allocate new vnode\n");
605 		return (error);
606 	}
607 
608 	if (de->de_dirent->d_type == DT_CHR) {
609 		vp->v_type = VCHR;
610 		VI_LOCK(vp);
611 		dev_lock();
612 		dev_refl(dev);
613 		/* XXX: v_rdev should be protect by vnode lock */
614 		vp->v_rdev = dev;
615 		VNPASS(vp->v_usecount == 1, vp);
616 		/* Special casing of ttys for deadfs.  Probably redundant. */
617 		dsw = dev->si_devsw;
618 		if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
619 			vp->v_vflag |= VV_ISTTY;
620 		dev_unlock();
621 		VI_UNLOCK(vp);
622 		if ((dev->si_flags & SI_ETERNAL) != 0)
623 			vp->v_vflag |= VV_ETERNALDEV;
624 		vp->v_op = &devfs_specops;
625 	} else if (de->de_dirent->d_type == DT_DIR) {
626 		vp->v_type = VDIR;
627 	} else if (de->de_dirent->d_type == DT_LNK) {
628 		vp->v_type = VLNK;
629 	} else {
630 		vp->v_type = VBAD;
631 	}
632 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
633 	VN_LOCK_ASHARE(vp);
634 	mtx_lock(&devfs_de_interlock);
635 	vp->v_data = de;
636 	de->de_vnode = vp;
637 	mtx_unlock(&devfs_de_interlock);
638 	error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
639 	if (error != 0) {
640 		(void) devfs_allocv_drop_refs(1, dmp, de);
641 		return (error);
642 	}
643 	if (devfs_allocv_drop_refs(0, dmp, de)) {
644 		vput(vp);
645 		return (ENOENT);
646 	}
647 #ifdef MAC
648 	mac_devfs_vnode_associate(mp, de, vp);
649 #endif
650 	sx_xunlock(&dmp->dm_lock);
651 	*vpp = vp;
652 	return (0);
653 }
654 
655 static int
devfs_access(struct vop_access_args * ap)656 devfs_access(struct vop_access_args *ap)
657 {
658 	struct vnode *vp = ap->a_vp;
659 	struct devfs_dirent *de;
660 	struct proc *p;
661 	int error;
662 
663 	de = vp->v_data;
664 	if (vp->v_type == VDIR)
665 		de = de->de_dir;
666 
667 	error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
668 	    ap->a_accmode, ap->a_cred);
669 	if (error == 0)
670 		return (0);
671 	if (error != EACCES)
672 		return (error);
673 	p = ap->a_td->td_proc;
674 	/* We do, however, allow access to the controlling terminal */
675 	PROC_LOCK(p);
676 	if (!(p->p_flag & P_CONTROLT)) {
677 		PROC_UNLOCK(p);
678 		return (error);
679 	}
680 	if (p->p_session->s_ttydp == de->de_cdp)
681 		error = 0;
682 	PROC_UNLOCK(p);
683 	return (error);
684 }
685 
686 _Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0,
687     "devfs-only flag reuse failed");
688 
689 static int
devfs_close(struct vop_close_args * ap)690 devfs_close(struct vop_close_args *ap)
691 {
692 	struct vnode *vp = ap->a_vp, *oldvp;
693 	struct thread *td = ap->a_td;
694 	struct proc *p;
695 	struct cdev *dev = vp->v_rdev;
696 	struct cdevsw *dsw;
697 	struct devfs_dirent *de = vp->v_data;
698 	int dflags, error, ref, vp_locked;
699 
700 	/*
701 	 * XXX: Don't call d_close() if we were called because of
702 	 * XXX: insmntque1() failure.
703 	 */
704 	if (vp->v_data == NULL)
705 		return (0);
706 
707 	/*
708 	 * Hack: a tty device that is a controlling terminal
709 	 * has a reference from the session structure.
710 	 * We cannot easily tell that a character device is
711 	 * a controlling terminal, unless it is the closing
712 	 * process' controlling terminal.  In that case,
713 	 * if the reference count is 2 (this last descriptor
714 	 * plus the session), release the reference from the session.
715 	 */
716 	if (de->de_usecount == 2 && td != NULL) {
717 		p = td->td_proc;
718 		PROC_LOCK(p);
719 		if (vp == p->p_session->s_ttyvp) {
720 			PROC_UNLOCK(p);
721 			oldvp = NULL;
722 			sx_xlock(&proctree_lock);
723 			if (vp == p->p_session->s_ttyvp) {
724 				SESS_LOCK(p->p_session);
725 				mtx_lock(&devfs_de_interlock);
726 				VI_LOCK(vp);
727 				if (devfs_usecountl(vp) == 2 && !VN_IS_DOOMED(vp)) {
728 					p->p_session->s_ttyvp = NULL;
729 					p->p_session->s_ttydp = NULL;
730 					oldvp = vp;
731 				}
732 				VI_UNLOCK(vp);
733 				mtx_unlock(&devfs_de_interlock);
734 				SESS_UNLOCK(p->p_session);
735 			}
736 			sx_xunlock(&proctree_lock);
737 			if (oldvp != NULL)
738 				devfs_ctty_unref(oldvp);
739 		} else
740 			PROC_UNLOCK(p);
741 	}
742 	/*
743 	 * We do not want to really close the device if it
744 	 * is still in use unless we are trying to close it
745 	 * forcibly. Since every use (buffer, vnode, swap, cmap)
746 	 * holds a reference to the vnode, and because we mark
747 	 * any other vnodes that alias this device, when the
748 	 * sum of the reference counts on all the aliased
749 	 * vnodes descends to one, we are on last close.
750 	 */
751 	dsw = dev_refthread(dev, &ref);
752 	if (dsw == NULL)
753 		return (ENXIO);
754 	dflags = 0;
755 	mtx_lock(&devfs_de_interlock);
756 	VI_LOCK(vp);
757 	if (devfs_usecountl(vp) == 1)
758 		dflags |= FLASTCLOSE;
759 	devfs_usecount_subl(vp);
760 	mtx_unlock(&devfs_de_interlock);
761 	if (VN_IS_DOOMED(vp)) {
762 		/* Forced close. */
763 		dflags |= FREVOKE | FNONBLOCK;
764 	} else if (dsw->d_flags & D_TRACKCLOSE) {
765 		/* Keep device updated on status. */
766 	} else if ((dflags & FLASTCLOSE) == 0) {
767 		VI_UNLOCK(vp);
768 		dev_relthread(dev, ref);
769 		return (0);
770 	}
771 	vholdnz(vp);
772 	VI_UNLOCK(vp);
773 	vp_locked = VOP_ISLOCKED(vp);
774 	VOP_UNLOCK(vp);
775 	KASSERT(dev->si_refcount > 0,
776 	    ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
777 	error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td);
778 	dev_relthread(dev, ref);
779 	vn_lock(vp, vp_locked | LK_RETRY);
780 	vdrop(vp);
781 	return (error);
782 }
783 
784 static int
devfs_close_f(struct file * fp,struct thread * td)785 devfs_close_f(struct file *fp, struct thread *td)
786 {
787 	int error;
788 	struct file *fpop;
789 
790 	/*
791 	 * NB: td may be NULL if this descriptor is closed due to
792 	 * garbage collection from a closed UNIX domain socket.
793 	 */
794 	fpop = curthread->td_fpop;
795 	curthread->td_fpop = fp;
796 	error = vnops.fo_close(fp, td);
797 	curthread->td_fpop = fpop;
798 
799 	/*
800 	 * The f_cdevpriv cannot be assigned non-NULL value while we
801 	 * are destroying the file.
802 	 */
803 	if (fp->f_cdevpriv != NULL)
804 		devfs_fpdrop(fp);
805 	return (error);
806 }
807 
808 static int
devfs_getattr(struct vop_getattr_args * ap)809 devfs_getattr(struct vop_getattr_args *ap)
810 {
811 	struct vnode *vp = ap->a_vp;
812 	struct vattr *vap = ap->a_vap;
813 	struct devfs_dirent *de;
814 	struct devfs_mount *dmp;
815 	struct cdev *dev;
816 	struct timeval boottime;
817 	int error;
818 
819 	error = devfs_populate_vp(vp);
820 	if (error != 0)
821 		return (error);
822 
823 	dmp = VFSTODEVFS(vp->v_mount);
824 	sx_xunlock(&dmp->dm_lock);
825 
826 	de = vp->v_data;
827 	KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
828 	if (vp->v_type == VDIR) {
829 		de = de->de_dir;
830 		KASSERT(de != NULL,
831 		    ("Null dir dirent in devfs_getattr vp=%p", vp));
832 	}
833 	vap->va_uid = de->de_uid;
834 	vap->va_gid = de->de_gid;
835 	vap->va_mode = de->de_mode;
836 	if (vp->v_type == VLNK)
837 		vap->va_size = strlen(de->de_symlink);
838 	else if (vp->v_type == VDIR)
839 		vap->va_size = vap->va_bytes = DEV_BSIZE;
840 	else
841 		vap->va_size = 0;
842 	if (vp->v_type != VDIR)
843 		vap->va_bytes = 0;
844 	vap->va_blocksize = DEV_BSIZE;
845 	vap->va_type = vp->v_type;
846 
847 	getboottime(&boottime);
848 #define fix(aa)							\
849 	do {							\
850 		if ((aa).tv_sec <= 3600) {			\
851 			(aa).tv_sec = boottime.tv_sec;		\
852 			(aa).tv_nsec = boottime.tv_usec * 1000; \
853 		}						\
854 	} while (0)
855 
856 	if (vp->v_type != VCHR)  {
857 		fix(de->de_atime);
858 		vap->va_atime = de->de_atime;
859 		fix(de->de_mtime);
860 		vap->va_mtime = de->de_mtime;
861 		fix(de->de_ctime);
862 		vap->va_ctime = de->de_ctime;
863 	} else {
864 		dev = vp->v_rdev;
865 		fix(dev->si_atime);
866 		vap->va_atime = dev->si_atime;
867 		fix(dev->si_mtime);
868 		vap->va_mtime = dev->si_mtime;
869 		fix(dev->si_ctime);
870 		vap->va_ctime = dev->si_ctime;
871 
872 		vap->va_rdev = cdev2priv(dev)->cdp_inode;
873 	}
874 	vap->va_gen = 0;
875 	vap->va_flags = 0;
876 	vap->va_filerev = 0;
877 	vap->va_nlink = de->de_links;
878 	vap->va_fileid = de->de_inode;
879 
880 	return (error);
881 }
882 
883 /* ARGSUSED */
884 static int
devfs_ioctl_f(struct file * fp,u_long com,void * data,struct ucred * cred,struct thread * td)885 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
886 {
887 	struct file *fpop;
888 	int error;
889 
890 	fpop = td->td_fpop;
891 	td->td_fpop = fp;
892 	error = vnops.fo_ioctl(fp, com, data, cred, td);
893 	td->td_fpop = fpop;
894 	return (error);
895 }
896 
897 void *
fiodgname_buf_get_ptr(void * fgnp,u_long com)898 fiodgname_buf_get_ptr(void *fgnp, u_long com)
899 {
900 	union {
901 		struct fiodgname_arg	fgn;
902 #ifdef COMPAT_FREEBSD32
903 		struct fiodgname_arg32	fgn32;
904 #endif
905 	} *fgnup;
906 
907 	fgnup = fgnp;
908 	switch (com) {
909 	case FIODGNAME:
910 		return (fgnup->fgn.buf);
911 #ifdef COMPAT_FREEBSD32
912 	case FIODGNAME_32:
913 		return ((void *)(uintptr_t)fgnup->fgn32.buf);
914 #endif
915 	default:
916 		panic("Unhandled ioctl command %ld", com);
917 	}
918 }
919 
920 static int
devfs_ioctl(struct vop_ioctl_args * ap)921 devfs_ioctl(struct vop_ioctl_args *ap)
922 {
923 	struct fiodgname_arg *fgn;
924 	struct vnode *vpold, *vp;
925 	struct cdevsw *dsw;
926 	struct thread *td;
927 	struct session *sess;
928 	struct cdev *dev;
929 	int error, ref, i;
930 	const char *p;
931 	u_long com;
932 
933 	vp = ap->a_vp;
934 	com = ap->a_command;
935 	td = ap->a_td;
936 
937 	dsw = devvn_refthread(vp, &dev, &ref);
938 	if (dsw == NULL)
939 		return (ENXIO);
940 	KASSERT(dev->si_refcount > 0,
941 	    ("devfs: un-referenced struct cdev *(%s)", devtoname(dev)));
942 
943 	switch (com) {
944 	case FIODTYPE:
945 		*(int *)ap->a_data = dsw->d_flags & D_TYPEMASK;
946 		error = 0;
947 		break;
948 	case FIODGNAME:
949 #ifdef	COMPAT_FREEBSD32
950 	case FIODGNAME_32:
951 #endif
952 		fgn = ap->a_data;
953 		p = devtoname(dev);
954 		i = strlen(p) + 1;
955 		if (i > fgn->len)
956 			error = EINVAL;
957 		else
958 			error = copyout(p, fiodgname_buf_get_ptr(fgn, com), i);
959 		break;
960 	default:
961 		error = dsw->d_ioctl(dev, com, ap->a_data, ap->a_fflag, td);
962 	}
963 
964 	dev_relthread(dev, ref);
965 	if (error == ENOIOCTL)
966 		error = ENOTTY;
967 
968 	if (error == 0 && com == TIOCSCTTY) {
969 		/*
970 		 * Do nothing if reassigning same control tty, or if the
971 		 * control tty has already disappeared.  If it disappeared,
972 		 * it's because we were racing with TIOCNOTTY.  TIOCNOTTY
973 		 * already took care of releasing the old vnode and we have
974 		 * nothing left to do.
975 		 */
976 		sx_slock(&proctree_lock);
977 		sess = td->td_proc->p_session;
978 		if (sess->s_ttyvp == vp || sess->s_ttyp == NULL) {
979 			sx_sunlock(&proctree_lock);
980 			return (0);
981 		}
982 
983 		devfs_ctty_ref(vp);
984 		SESS_LOCK(sess);
985 		vpold = sess->s_ttyvp;
986 		sess->s_ttyvp = vp;
987 		sess->s_ttydp = cdev2priv(dev);
988 		SESS_UNLOCK(sess);
989 
990 		sx_sunlock(&proctree_lock);
991 
992 		/* Get rid of reference to old control tty */
993 		if (vpold)
994 			devfs_ctty_unref(vpold);
995 	}
996 	return (error);
997 }
998 
999 /* ARGSUSED */
1000 static int
devfs_kqfilter_f(struct file * fp,struct knote * kn)1001 devfs_kqfilter_f(struct file *fp, struct knote *kn)
1002 {
1003 	struct cdev *dev;
1004 	struct cdevsw *dsw;
1005 	int error, ref;
1006 	struct file *fpop;
1007 	struct thread *td;
1008 
1009 	td = curthread;
1010 	fpop = td->td_fpop;
1011 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1012 	if (error)
1013 		return (error);
1014 	error = dsw->d_kqfilter(dev, kn);
1015 	td->td_fpop = fpop;
1016 	dev_relthread(dev, ref);
1017 	return (error);
1018 }
1019 
1020 static inline int
devfs_prison_check(struct devfs_dirent * de,struct thread * td)1021 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
1022 {
1023 	struct cdev_priv *cdp;
1024 	struct ucred *dcr;
1025 	struct proc *p;
1026 	int error;
1027 
1028 	cdp = de->de_cdp;
1029 	if (cdp == NULL)
1030 		return (0);
1031 	dcr = cdp->cdp_c.si_cred;
1032 	if (dcr == NULL)
1033 		return (0);
1034 
1035 	error = prison_check(td->td_ucred, dcr);
1036 	if (error == 0)
1037 		return (0);
1038 	/* We do, however, allow access to the controlling terminal */
1039 	p = td->td_proc;
1040 	PROC_LOCK(p);
1041 	if (!(p->p_flag & P_CONTROLT)) {
1042 		PROC_UNLOCK(p);
1043 		return (error);
1044 	}
1045 	if (p->p_session->s_ttydp == cdp)
1046 		error = 0;
1047 	PROC_UNLOCK(p);
1048 	return (error);
1049 }
1050 
1051 static int
devfs_lookupx(struct vop_lookup_args * ap,int * dm_unlock)1052 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
1053 {
1054 	struct componentname *cnp;
1055 	struct vnode *dvp, **vpp;
1056 	struct thread *td;
1057 	struct devfs_dirent *de, *dd;
1058 	struct devfs_dirent **dde;
1059 	struct devfs_mount *dmp;
1060 	struct mount *mp;
1061 	struct cdev *cdev;
1062 	int error, flags, nameiop, dvplocked;
1063 	char specname[SPECNAMELEN + 1], *pname;
1064 
1065 	cnp = ap->a_cnp;
1066 	vpp = ap->a_vpp;
1067 	dvp = ap->a_dvp;
1068 	pname = cnp->cn_nameptr;
1069 	td = cnp->cn_thread;
1070 	flags = cnp->cn_flags;
1071 	nameiop = cnp->cn_nameiop;
1072 	mp = dvp->v_mount;
1073 	dmp = VFSTODEVFS(mp);
1074 	dd = dvp->v_data;
1075 	*vpp = NULLVP;
1076 
1077 	if ((flags & ISLASTCN) && nameiop == RENAME)
1078 		return (EOPNOTSUPP);
1079 
1080 	if (dvp->v_type != VDIR)
1081 		return (ENOTDIR);
1082 
1083 	if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
1084 		return (EIO);
1085 
1086 	error = vn_dir_check_exec(dvp, cnp);
1087 	if (error != 0)
1088 		return (error);
1089 
1090 	if (cnp->cn_namelen == 1 && *pname == '.') {
1091 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
1092 			return (EINVAL);
1093 		*vpp = dvp;
1094 		VREF(dvp);
1095 		return (0);
1096 	}
1097 
1098 	if (flags & ISDOTDOT) {
1099 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
1100 			return (EINVAL);
1101 		de = devfs_parent_dirent(dd);
1102 		if (de == NULL)
1103 			return (ENOENT);
1104 		dvplocked = VOP_ISLOCKED(dvp);
1105 		VOP_UNLOCK(dvp);
1106 		error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK,
1107 		    vpp);
1108 		*dm_unlock = 0;
1109 		vn_lock(dvp, dvplocked | LK_RETRY);
1110 		return (error);
1111 	}
1112 
1113 	dd = dvp->v_data;
1114 	de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
1115 	while (de == NULL) {	/* While(...) so we can use break */
1116 
1117 		if (nameiop == DELETE)
1118 			return (ENOENT);
1119 
1120 		/*
1121 		 * OK, we didn't have an entry for the name we were asked for
1122 		 * so we try to see if anybody can create it on demand.
1123 		 */
1124 		pname = devfs_fqpn(specname, dmp, dd, cnp);
1125 		if (pname == NULL)
1126 			break;
1127 
1128 		cdev = NULL;
1129 		DEVFS_DMP_HOLD(dmp);
1130 		sx_xunlock(&dmp->dm_lock);
1131 		sx_slock(&clone_drain_lock);
1132 		EVENTHANDLER_INVOKE(dev_clone,
1133 		    td->td_ucred, pname, strlen(pname), &cdev);
1134 		sx_sunlock(&clone_drain_lock);
1135 
1136 		if (cdev == NULL)
1137 			sx_xlock(&dmp->dm_lock);
1138 		else if (devfs_populate_vp(dvp) != 0) {
1139 			*dm_unlock = 0;
1140 			sx_xlock(&dmp->dm_lock);
1141 			if (DEVFS_DMP_DROP(dmp)) {
1142 				sx_xunlock(&dmp->dm_lock);
1143 				devfs_unmount_final(dmp);
1144 			} else
1145 				sx_xunlock(&dmp->dm_lock);
1146 			dev_rel(cdev);
1147 			return (ENOENT);
1148 		}
1149 		if (DEVFS_DMP_DROP(dmp)) {
1150 			*dm_unlock = 0;
1151 			sx_xunlock(&dmp->dm_lock);
1152 			devfs_unmount_final(dmp);
1153 			if (cdev != NULL)
1154 				dev_rel(cdev);
1155 			return (ENOENT);
1156 		}
1157 
1158 		if (cdev == NULL)
1159 			break;
1160 
1161 		dev_lock();
1162 		dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
1163 		if (dde != NULL && *dde != NULL)
1164 			de = *dde;
1165 		dev_unlock();
1166 		dev_rel(cdev);
1167 		break;
1168 	}
1169 
1170 	if (de == NULL || de->de_flags & DE_WHITEOUT) {
1171 		if ((nameiop == CREATE || nameiop == RENAME) &&
1172 		    (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
1173 			cnp->cn_flags |= SAVENAME;
1174 			return (EJUSTRETURN);
1175 		}
1176 		return (ENOENT);
1177 	}
1178 
1179 	if (devfs_prison_check(de, td))
1180 		return (ENOENT);
1181 
1182 	if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
1183 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
1184 		if (error)
1185 			return (error);
1186 		if (*vpp == dvp) {
1187 			VREF(dvp);
1188 			*vpp = dvp;
1189 			return (0);
1190 		}
1191 	}
1192 	error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp);
1193 	*dm_unlock = 0;
1194 	return (error);
1195 }
1196 
1197 static int
devfs_lookup(struct vop_lookup_args * ap)1198 devfs_lookup(struct vop_lookup_args *ap)
1199 {
1200 	int j;
1201 	struct devfs_mount *dmp;
1202 	int dm_unlock;
1203 
1204 	if (devfs_populate_vp(ap->a_dvp) != 0)
1205 		return (ENOTDIR);
1206 
1207 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1208 	dm_unlock = 1;
1209 	j = devfs_lookupx(ap, &dm_unlock);
1210 	if (dm_unlock == 1)
1211 		sx_xunlock(&dmp->dm_lock);
1212 	return (j);
1213 }
1214 
1215 static int
devfs_mknod(struct vop_mknod_args * ap)1216 devfs_mknod(struct vop_mknod_args *ap)
1217 {
1218 	struct componentname *cnp;
1219 	struct vnode *dvp, **vpp;
1220 	struct devfs_dirent *dd, *de;
1221 	struct devfs_mount *dmp;
1222 	int error;
1223 
1224 	/*
1225 	 * The only type of node we should be creating here is a
1226 	 * character device, for anything else return EOPNOTSUPP.
1227 	 */
1228 	if (ap->a_vap->va_type != VCHR)
1229 		return (EOPNOTSUPP);
1230 	dvp = ap->a_dvp;
1231 	dmp = VFSTODEVFS(dvp->v_mount);
1232 
1233 	cnp = ap->a_cnp;
1234 	vpp = ap->a_vpp;
1235 	dd = dvp->v_data;
1236 
1237 	error = ENOENT;
1238 	sx_xlock(&dmp->dm_lock);
1239 	TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1240 		if (cnp->cn_namelen != de->de_dirent->d_namlen)
1241 			continue;
1242 		if (de->de_dirent->d_type == DT_CHR &&
1243 		    (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
1244 			continue;
1245 		if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1246 		    de->de_dirent->d_namlen) != 0)
1247 			continue;
1248 		if (de->de_flags & DE_WHITEOUT)
1249 			break;
1250 		goto notfound;
1251 	}
1252 	if (de == NULL)
1253 		goto notfound;
1254 	de->de_flags &= ~DE_WHITEOUT;
1255 	error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1256 	return (error);
1257 notfound:
1258 	sx_xunlock(&dmp->dm_lock);
1259 	return (error);
1260 }
1261 
1262 /* ARGSUSED */
1263 static int
devfs_open(struct vop_open_args * ap)1264 devfs_open(struct vop_open_args *ap)
1265 {
1266 	struct thread *td = ap->a_td;
1267 	struct vnode *vp = ap->a_vp;
1268 	struct cdev *dev = vp->v_rdev;
1269 	struct file *fp = ap->a_fp;
1270 	int error, ref, vlocked;
1271 	struct cdevsw *dsw;
1272 	struct file *fpop;
1273 
1274 	if (vp->v_type == VBLK)
1275 		return (ENXIO);
1276 
1277 	if (dev == NULL)
1278 		return (ENXIO);
1279 
1280 	/* Make this field valid before any I/O in d_open. */
1281 	if (dev->si_iosize_max == 0)
1282 		dev->si_iosize_max = DFLTPHYS;
1283 
1284 	dsw = dev_refthread(dev, &ref);
1285 	if (dsw == NULL)
1286 		return (ENXIO);
1287 	if (fp == NULL && dsw->d_fdopen != NULL) {
1288 		dev_relthread(dev, ref);
1289 		return (ENXIO);
1290 	}
1291 
1292 	if (vp->v_type == VCHR)
1293 		devfs_usecount_add(vp);
1294 
1295 	vlocked = VOP_ISLOCKED(vp);
1296 	VOP_UNLOCK(vp);
1297 
1298 	fpop = td->td_fpop;
1299 	td->td_fpop = fp;
1300 	if (fp != NULL) {
1301 		fp->f_data = dev;
1302 		fp->f_vnode = vp;
1303 	}
1304 	if (dsw->d_fdopen != NULL)
1305 		error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1306 	else
1307 		error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1308 	/* Clean up any cdevpriv upon error. */
1309 	if (error != 0)
1310 		devfs_clear_cdevpriv();
1311 	td->td_fpop = fpop;
1312 
1313 	vn_lock(vp, vlocked | LK_RETRY);
1314 	if (error != 0 && vp->v_type == VCHR)
1315 		devfs_usecount_sub(vp);
1316 
1317 	dev_relthread(dev, ref);
1318 	if (error != 0) {
1319 		if (error == ERESTART)
1320 			error = EINTR;
1321 		return (error);
1322 	}
1323 
1324 #if 0	/* /dev/console */
1325 	KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1326 #else
1327 	if (fp == NULL)
1328 		return (error);
1329 #endif
1330 	if (fp->f_ops == &badfileops)
1331 		finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1332 	return (error);
1333 }
1334 
1335 static int
devfs_pathconf(struct vop_pathconf_args * ap)1336 devfs_pathconf(struct vop_pathconf_args *ap)
1337 {
1338 
1339 	switch (ap->a_name) {
1340 	case _PC_FILESIZEBITS:
1341 		*ap->a_retval = 64;
1342 		return (0);
1343 	case _PC_NAME_MAX:
1344 		*ap->a_retval = NAME_MAX;
1345 		return (0);
1346 	case _PC_LINK_MAX:
1347 		*ap->a_retval = INT_MAX;
1348 		return (0);
1349 	case _PC_SYMLINK_MAX:
1350 		*ap->a_retval = MAXPATHLEN;
1351 		return (0);
1352 	case _PC_MAX_CANON:
1353 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1354 			*ap->a_retval = MAX_CANON;
1355 			return (0);
1356 		}
1357 		return (EINVAL);
1358 	case _PC_MAX_INPUT:
1359 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1360 			*ap->a_retval = MAX_INPUT;
1361 			return (0);
1362 		}
1363 		return (EINVAL);
1364 	case _PC_VDISABLE:
1365 		if (ap->a_vp->v_vflag & VV_ISTTY) {
1366 			*ap->a_retval = _POSIX_VDISABLE;
1367 			return (0);
1368 		}
1369 		return (EINVAL);
1370 	case _PC_MAC_PRESENT:
1371 #ifdef MAC
1372 		/*
1373 		 * If MAC is enabled, devfs automatically supports
1374 		 * trivial non-persistent label storage.
1375 		 */
1376 		*ap->a_retval = 1;
1377 #else
1378 		*ap->a_retval = 0;
1379 #endif
1380 		return (0);
1381 	case _PC_CHOWN_RESTRICTED:
1382 		*ap->a_retval = 1;
1383 		return (0);
1384 	default:
1385 		return (vop_stdpathconf(ap));
1386 	}
1387 	/* NOTREACHED */
1388 }
1389 
1390 /* ARGSUSED */
1391 static int
devfs_poll_f(struct file * fp,int events,struct ucred * cred,struct thread * td)1392 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1393 {
1394 	struct cdev *dev;
1395 	struct cdevsw *dsw;
1396 	int error, ref;
1397 	struct file *fpop;
1398 
1399 	fpop = td->td_fpop;
1400 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1401 	if (error != 0) {
1402 		error = vnops.fo_poll(fp, events, cred, td);
1403 		return (error);
1404 	}
1405 	error = dsw->d_poll(dev, events, td);
1406 	td->td_fpop = fpop;
1407 	dev_relthread(dev, ref);
1408 	return(error);
1409 }
1410 
1411 /*
1412  * Print out the contents of a special device vnode.
1413  */
1414 static int
devfs_print(struct vop_print_args * ap)1415 devfs_print(struct vop_print_args *ap)
1416 {
1417 
1418 	printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1419 	return (0);
1420 }
1421 
1422 static int
devfs_read_f(struct file * fp,struct uio * uio,struct ucred * cred,int flags,struct thread * td)1423 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1424     int flags, struct thread *td)
1425 {
1426 	struct cdev *dev;
1427 	int ioflag, error, ref;
1428 	ssize_t resid;
1429 	struct cdevsw *dsw;
1430 	struct file *fpop;
1431 
1432 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1433 		return (EINVAL);
1434 	fpop = td->td_fpop;
1435 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1436 	if (error != 0) {
1437 		error = vnops.fo_read(fp, uio, cred, flags, td);
1438 		return (error);
1439 	}
1440 	resid = uio->uio_resid;
1441 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1442 	if (ioflag & O_DIRECT)
1443 		ioflag |= IO_DIRECT;
1444 
1445 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1446 	error = dsw->d_read(dev, uio, ioflag);
1447 	if (uio->uio_resid != resid || (error == 0 && resid != 0))
1448 		devfs_timestamp(&dev->si_atime);
1449 	td->td_fpop = fpop;
1450 	dev_relthread(dev, ref);
1451 
1452 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_R);
1453 	return (error);
1454 }
1455 
1456 static int
devfs_readdir(struct vop_readdir_args * ap)1457 devfs_readdir(struct vop_readdir_args *ap)
1458 {
1459 	int error;
1460 	struct uio *uio;
1461 	struct dirent *dp;
1462 	struct devfs_dirent *dd;
1463 	struct devfs_dirent *de;
1464 	struct devfs_mount *dmp;
1465 	off_t off;
1466 	int *tmp_ncookies = NULL;
1467 
1468 	if (ap->a_vp->v_type != VDIR)
1469 		return (ENOTDIR);
1470 
1471 	uio = ap->a_uio;
1472 	if (uio->uio_offset < 0)
1473 		return (EINVAL);
1474 
1475 	/*
1476 	 * XXX: This is a temporary hack to get around this filesystem not
1477 	 * supporting cookies. We store the location of the ncookies pointer
1478 	 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1479 	 * and set the number of cookies to 0. We then set the pointer to
1480 	 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1481 	 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1482 	 * pointer to its original location before returning to the caller.
1483 	 */
1484 	if (ap->a_ncookies != NULL) {
1485 		tmp_ncookies = ap->a_ncookies;
1486 		*ap->a_ncookies = 0;
1487 		ap->a_ncookies = NULL;
1488 	}
1489 
1490 	dmp = VFSTODEVFS(ap->a_vp->v_mount);
1491 	if (devfs_populate_vp(ap->a_vp) != 0) {
1492 		if (tmp_ncookies != NULL)
1493 			ap->a_ncookies = tmp_ncookies;
1494 		return (EIO);
1495 	}
1496 	error = 0;
1497 	de = ap->a_vp->v_data;
1498 	off = 0;
1499 	TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1500 		KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1501 		if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1502 			continue;
1503 		if (devfs_prison_check(dd, uio->uio_td))
1504 			continue;
1505 		if (dd->de_dirent->d_type == DT_DIR)
1506 			de = dd->de_dir;
1507 		else
1508 			de = dd;
1509 		dp = dd->de_dirent;
1510 		MPASS(dp->d_reclen == GENERIC_DIRSIZ(dp));
1511 		if (dp->d_reclen > uio->uio_resid)
1512 			break;
1513 		dp->d_fileno = de->de_inode;
1514 		/* NOTE: d_off is the offset for the *next* entry. */
1515 		dp->d_off = off + dp->d_reclen;
1516 		if (off >= uio->uio_offset) {
1517 			error = vfs_read_dirent(ap, dp, off);
1518 			if (error)
1519 				break;
1520 		}
1521 		off += dp->d_reclen;
1522 	}
1523 	sx_xunlock(&dmp->dm_lock);
1524 	uio->uio_offset = off;
1525 
1526 	/*
1527 	 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1528 	 * place.
1529 	 */
1530 	if (tmp_ncookies != NULL)
1531 		ap->a_ncookies = tmp_ncookies;
1532 
1533 	return (error);
1534 }
1535 
1536 static int
devfs_readlink(struct vop_readlink_args * ap)1537 devfs_readlink(struct vop_readlink_args *ap)
1538 {
1539 	struct devfs_dirent *de;
1540 
1541 	de = ap->a_vp->v_data;
1542 	return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1543 }
1544 
1545 static void
devfs_reclaiml(struct vnode * vp)1546 devfs_reclaiml(struct vnode *vp)
1547 {
1548 	struct devfs_dirent *de;
1549 
1550 	mtx_assert(&devfs_de_interlock, MA_OWNED);
1551 	de = vp->v_data;
1552 	if (de != NULL) {
1553 		MPASS(de->de_usecount == 0);
1554 		de->de_vnode = NULL;
1555 		vp->v_data = NULL;
1556 	}
1557 }
1558 
1559 static int
devfs_reclaim(struct vop_reclaim_args * ap)1560 devfs_reclaim(struct vop_reclaim_args *ap)
1561 {
1562 	struct vnode *vp;
1563 
1564 	vp = ap->a_vp;
1565 	mtx_lock(&devfs_de_interlock);
1566 	devfs_reclaiml(vp);
1567 	mtx_unlock(&devfs_de_interlock);
1568 	return (0);
1569 }
1570 
1571 static int
devfs_reclaim_vchr(struct vop_reclaim_args * ap)1572 devfs_reclaim_vchr(struct vop_reclaim_args *ap)
1573 {
1574 	struct vnode *vp;
1575 	struct cdev *dev;
1576 
1577 	vp = ap->a_vp;
1578 	MPASS(vp->v_type == VCHR);
1579 
1580 	mtx_lock(&devfs_de_interlock);
1581 	VI_LOCK(vp);
1582 	devfs_usecount_subl(vp);
1583 	devfs_reclaiml(vp);
1584 	mtx_unlock(&devfs_de_interlock);
1585 	dev_lock();
1586 	dev = vp->v_rdev;
1587 	vp->v_rdev = NULL;
1588 	dev_unlock();
1589 	VI_UNLOCK(vp);
1590 	if (dev != NULL)
1591 		dev_rel(dev);
1592 	return (0);
1593 }
1594 
1595 static int
devfs_remove(struct vop_remove_args * ap)1596 devfs_remove(struct vop_remove_args *ap)
1597 {
1598 	struct vnode *dvp = ap->a_dvp;
1599 	struct vnode *vp = ap->a_vp;
1600 	struct devfs_dirent *dd;
1601 	struct devfs_dirent *de, *de_covered;
1602 	struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1603 
1604 	ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1605 	ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1606 
1607 	sx_xlock(&dmp->dm_lock);
1608 	dd = ap->a_dvp->v_data;
1609 	de = vp->v_data;
1610 	if (de->de_cdp == NULL) {
1611 		TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1612 		if (de->de_dirent->d_type == DT_LNK) {
1613 			de_covered = devfs_find(dd, de->de_dirent->d_name,
1614 			    de->de_dirent->d_namlen, 0);
1615 			if (de_covered != NULL)
1616 				de_covered->de_flags &= ~DE_COVERED;
1617 		}
1618 		/* We need to unlock dvp because devfs_delete() may lock it. */
1619 		VOP_UNLOCK(vp);
1620 		if (dvp != vp)
1621 			VOP_UNLOCK(dvp);
1622 		devfs_delete(dmp, de, 0);
1623 		sx_xunlock(&dmp->dm_lock);
1624 		if (dvp != vp)
1625 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1626 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1627 	} else {
1628 		de->de_flags |= DE_WHITEOUT;
1629 		sx_xunlock(&dmp->dm_lock);
1630 	}
1631 	return (0);
1632 }
1633 
1634 /*
1635  * Revoke is called on a tty when a terminal session ends.  The vnode
1636  * is orphaned by setting v_op to deadfs so we need to let go of it
1637  * as well so that we create a new one next time around.
1638  *
1639  */
1640 static int
devfs_revoke(struct vop_revoke_args * ap)1641 devfs_revoke(struct vop_revoke_args *ap)
1642 {
1643 	struct vnode *vp = ap->a_vp, *vp2;
1644 	struct cdev *dev;
1645 	struct cdev_priv *cdp;
1646 	struct devfs_dirent *de;
1647 	enum vgetstate vs;
1648 	u_int i;
1649 
1650 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1651 
1652 	dev = vp->v_rdev;
1653 	cdp = cdev2priv(dev);
1654 
1655 	dev_lock();
1656 	cdp->cdp_inuse++;
1657 	dev_unlock();
1658 
1659 	vhold(vp);
1660 	vgone(vp);
1661 	vdrop(vp);
1662 
1663 	VOP_UNLOCK(vp);
1664  loop:
1665 	for (;;) {
1666 		mtx_lock(&devfs_de_interlock);
1667 		dev_lock();
1668 		vp2 = NULL;
1669 		for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1670 			de = cdp->cdp_dirents[i];
1671 			if (de == NULL)
1672 				continue;
1673 
1674 			vp2 = de->de_vnode;
1675 			if (vp2 != NULL) {
1676 				dev_unlock();
1677 				vs = vget_prep(vp2);
1678 				mtx_unlock(&devfs_de_interlock);
1679 				if (vget_finish(vp2, LK_EXCLUSIVE, vs) != 0)
1680 					goto loop;
1681 				vhold(vp2);
1682 				vgone(vp2);
1683 				vdrop(vp2);
1684 				vput(vp2);
1685 				break;
1686 			}
1687 		}
1688 		if (vp2 != NULL) {
1689 			continue;
1690 		}
1691 		dev_unlock();
1692 		mtx_unlock(&devfs_de_interlock);
1693 		break;
1694 	}
1695 	dev_lock();
1696 	cdp->cdp_inuse--;
1697 	if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1698 		KASSERT((cdp->cdp_flags & CDP_ON_ACTIVE_LIST) != 0,
1699 		    ("%s: cdp %p (%s) not on active list",
1700 		    __func__, cdp, dev->si_name));
1701 		cdp->cdp_flags &= ~CDP_ON_ACTIVE_LIST;
1702 		TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1703 		dev_unlock();
1704 		dev_rel(&cdp->cdp_c);
1705 	} else
1706 		dev_unlock();
1707 
1708 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1709 	return (0);
1710 }
1711 
1712 static int
devfs_rioctl(struct vop_ioctl_args * ap)1713 devfs_rioctl(struct vop_ioctl_args *ap)
1714 {
1715 	struct vnode *vp;
1716 	struct devfs_mount *dmp;
1717 	int error;
1718 
1719 	vp = ap->a_vp;
1720 	vn_lock(vp, LK_SHARED | LK_RETRY);
1721 	if (VN_IS_DOOMED(vp)) {
1722 		VOP_UNLOCK(vp);
1723 		return (EBADF);
1724 	}
1725 	dmp = VFSTODEVFS(vp->v_mount);
1726 	sx_xlock(&dmp->dm_lock);
1727 	VOP_UNLOCK(vp);
1728 	DEVFS_DMP_HOLD(dmp);
1729 	devfs_populate(dmp);
1730 	if (DEVFS_DMP_DROP(dmp)) {
1731 		sx_xunlock(&dmp->dm_lock);
1732 		devfs_unmount_final(dmp);
1733 		return (ENOENT);
1734 	}
1735 	error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1736 	sx_xunlock(&dmp->dm_lock);
1737 	return (error);
1738 }
1739 
1740 static int
devfs_rread(struct vop_read_args * ap)1741 devfs_rread(struct vop_read_args *ap)
1742 {
1743 
1744 	if (ap->a_vp->v_type != VDIR)
1745 		return (EINVAL);
1746 	return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1747 }
1748 
1749 static int
devfs_setattr(struct vop_setattr_args * ap)1750 devfs_setattr(struct vop_setattr_args *ap)
1751 {
1752 	struct devfs_dirent *de;
1753 	struct vattr *vap;
1754 	struct vnode *vp;
1755 	struct thread *td;
1756 	int c, error;
1757 	uid_t uid;
1758 	gid_t gid;
1759 
1760 	vap = ap->a_vap;
1761 	vp = ap->a_vp;
1762 	td = curthread;
1763 	if ((vap->va_type != VNON) ||
1764 	    (vap->va_nlink != VNOVAL) ||
1765 	    (vap->va_fsid != VNOVAL) ||
1766 	    (vap->va_fileid != VNOVAL) ||
1767 	    (vap->va_blocksize != VNOVAL) ||
1768 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1769 	    (vap->va_rdev != VNOVAL) ||
1770 	    ((int)vap->va_bytes != VNOVAL) ||
1771 	    (vap->va_gen != VNOVAL)) {
1772 		return (EINVAL);
1773 	}
1774 
1775 	error = devfs_populate_vp(vp);
1776 	if (error != 0)
1777 		return (error);
1778 
1779 	de = vp->v_data;
1780 	if (vp->v_type == VDIR)
1781 		de = de->de_dir;
1782 
1783 	c = 0;
1784 	if (vap->va_uid == (uid_t)VNOVAL)
1785 		uid = de->de_uid;
1786 	else
1787 		uid = vap->va_uid;
1788 	if (vap->va_gid == (gid_t)VNOVAL)
1789 		gid = de->de_gid;
1790 	else
1791 		gid = vap->va_gid;
1792 	if (uid != de->de_uid || gid != de->de_gid) {
1793 		if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1794 		    (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1795 			error = priv_check(td, PRIV_VFS_CHOWN);
1796 			if (error != 0)
1797 				goto ret;
1798 		}
1799 		de->de_uid = uid;
1800 		de->de_gid = gid;
1801 		c = 1;
1802 	}
1803 
1804 	if (vap->va_mode != (mode_t)VNOVAL) {
1805 		if (ap->a_cred->cr_uid != de->de_uid) {
1806 			error = priv_check(td, PRIV_VFS_ADMIN);
1807 			if (error != 0)
1808 				goto ret;
1809 		}
1810 		de->de_mode = vap->va_mode;
1811 		c = 1;
1812 	}
1813 
1814 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1815 		error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1816 		if (error != 0)
1817 			goto ret;
1818 		if (vap->va_atime.tv_sec != VNOVAL) {
1819 			if (vp->v_type == VCHR)
1820 				vp->v_rdev->si_atime = vap->va_atime;
1821 			else
1822 				de->de_atime = vap->va_atime;
1823 		}
1824 		if (vap->va_mtime.tv_sec != VNOVAL) {
1825 			if (vp->v_type == VCHR)
1826 				vp->v_rdev->si_mtime = vap->va_mtime;
1827 			else
1828 				de->de_mtime = vap->va_mtime;
1829 		}
1830 		c = 1;
1831 	}
1832 
1833 	if (c) {
1834 		if (vp->v_type == VCHR)
1835 			vfs_timestamp(&vp->v_rdev->si_ctime);
1836 		else
1837 			vfs_timestamp(&de->de_mtime);
1838 	}
1839 
1840 ret:
1841 	sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock);
1842 	return (error);
1843 }
1844 
1845 #ifdef MAC
1846 static int
devfs_setlabel(struct vop_setlabel_args * ap)1847 devfs_setlabel(struct vop_setlabel_args *ap)
1848 {
1849 	struct vnode *vp;
1850 	struct devfs_dirent *de;
1851 
1852 	vp = ap->a_vp;
1853 	de = vp->v_data;
1854 
1855 	mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1856 	mac_devfs_update(vp->v_mount, de, vp);
1857 
1858 	return (0);
1859 }
1860 #endif
1861 
1862 static int
devfs_stat_f(struct file * fp,struct stat * sb,struct ucred * cred,struct thread * td)1863 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1864 {
1865 
1866 	return (vnops.fo_stat(fp, sb, cred, td));
1867 }
1868 
1869 static int
devfs_symlink(struct vop_symlink_args * ap)1870 devfs_symlink(struct vop_symlink_args *ap)
1871 {
1872 	int i, error;
1873 	struct devfs_dirent *dd;
1874 	struct devfs_dirent *de, *de_covered, *de_dotdot;
1875 	struct devfs_mount *dmp;
1876 
1877 	error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1878 	if (error)
1879 		return(error);
1880 	dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1881 	if (devfs_populate_vp(ap->a_dvp) != 0)
1882 		return (ENOENT);
1883 
1884 	dd = ap->a_dvp->v_data;
1885 	de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1886 	de->de_flags = DE_USER;
1887 	de->de_uid = 0;
1888 	de->de_gid = 0;
1889 	de->de_mode = 0755;
1890 	de->de_inode = alloc_unr(devfs_inos);
1891 	de->de_dir = dd;
1892 	de->de_dirent->d_type = DT_LNK;
1893 	i = strlen(ap->a_target) + 1;
1894 	de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1895 	bcopy(ap->a_target, de->de_symlink, i);
1896 #ifdef MAC
1897 	mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1898 #endif
1899 	de_covered = devfs_find(dd, de->de_dirent->d_name,
1900 	    de->de_dirent->d_namlen, 0);
1901 	if (de_covered != NULL) {
1902 		if ((de_covered->de_flags & DE_USER) != 0) {
1903 			devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1904 			sx_xunlock(&dmp->dm_lock);
1905 			return (EEXIST);
1906 		}
1907 		KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1908 		    ("devfs_symlink: entry %p already covered", de_covered));
1909 		de_covered->de_flags |= DE_COVERED;
1910 	}
1911 
1912 	de_dotdot = TAILQ_FIRST(&dd->de_dlist);		/* "." */
1913 	de_dotdot = TAILQ_NEXT(de_dotdot, de_list);	/* ".." */
1914 	TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1915 	devfs_dir_ref_de(dmp, dd);
1916 	devfs_rules_apply(dmp, de);
1917 
1918 	return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1919 }
1920 
1921 static int
devfs_truncate_f(struct file * fp,off_t length,struct ucred * cred,struct thread * td)1922 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1923 {
1924 
1925 	return (vnops.fo_truncate(fp, length, cred, td));
1926 }
1927 
1928 static int
devfs_write_f(struct file * fp,struct uio * uio,struct ucred * cred,int flags,struct thread * td)1929 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1930     int flags, struct thread *td)
1931 {
1932 	struct cdev *dev;
1933 	int error, ioflag, ref;
1934 	ssize_t resid;
1935 	struct cdevsw *dsw;
1936 	struct file *fpop;
1937 
1938 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1939 		return (EINVAL);
1940 	fpop = td->td_fpop;
1941 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
1942 	if (error != 0) {
1943 		error = vnops.fo_write(fp, uio, cred, flags, td);
1944 		return (error);
1945 	}
1946 	KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1947 	ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1948 	if (ioflag & O_DIRECT)
1949 		ioflag |= IO_DIRECT;
1950 	foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1951 
1952 	resid = uio->uio_resid;
1953 
1954 	error = dsw->d_write(dev, uio, ioflag);
1955 	if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1956 		devfs_timestamp(&dev->si_ctime);
1957 		dev->si_mtime = dev->si_ctime;
1958 	}
1959 	td->td_fpop = fpop;
1960 	dev_relthread(dev, ref);
1961 
1962 	foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF_W);
1963 	return (error);
1964 }
1965 
1966 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)1967 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1968     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1969     struct thread *td)
1970 {
1971 	struct cdev *dev;
1972 	struct cdevsw *dsw;
1973 	struct mount *mp;
1974 	struct vnode *vp;
1975 	struct file *fpop;
1976 	vm_object_t object;
1977 	vm_prot_t maxprot;
1978 	int error, ref;
1979 
1980 	vp = fp->f_vnode;
1981 
1982 	/*
1983 	 * Ensure that file and memory protections are
1984 	 * compatible.
1985 	 */
1986 	mp = vp->v_mount;
1987 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1988 		maxprot = VM_PROT_NONE;
1989 		if ((prot & VM_PROT_EXECUTE) != 0)
1990 			return (EACCES);
1991 	} else
1992 		maxprot = VM_PROT_EXECUTE;
1993 	if ((fp->f_flag & FREAD) != 0)
1994 		maxprot |= VM_PROT_READ;
1995 	else if ((prot & VM_PROT_READ) != 0)
1996 		return (EACCES);
1997 
1998 	/*
1999 	 * If we are sharing potential changes via MAP_SHARED and we
2000 	 * are trying to get write permission although we opened it
2001 	 * without asking for it, bail out.
2002 	 *
2003 	 * Note that most character devices always share mappings.
2004 	 * The one exception is that D_MMAP_ANON devices
2005 	 * (i.e. /dev/zero) permit private writable mappings.
2006 	 *
2007 	 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests
2008 	 * as well as updating maxprot to permit writing for
2009 	 * D_MMAP_ANON devices rather than doing that here.
2010 	 */
2011 	if ((flags & MAP_SHARED) != 0) {
2012 		if ((fp->f_flag & FWRITE) != 0)
2013 			maxprot |= VM_PROT_WRITE;
2014 		else if ((prot & VM_PROT_WRITE) != 0)
2015 			return (EACCES);
2016 	}
2017 	maxprot &= cap_maxprot;
2018 
2019 	fpop = td->td_fpop;
2020 	error = devfs_fp_check(fp, &dev, &dsw, &ref);
2021 	if (error != 0)
2022 		return (error);
2023 
2024 	error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff,
2025 	    &object);
2026 	td->td_fpop = fpop;
2027 	dev_relthread(dev, ref);
2028 	if (error != 0)
2029 		return (error);
2030 
2031 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2032 	    foff, FALSE, td);
2033 	if (error != 0)
2034 		vm_object_deallocate(object);
2035 	return (error);
2036 }
2037 
2038 dev_t
dev2udev(struct cdev * x)2039 dev2udev(struct cdev *x)
2040 {
2041 	if (x == NULL)
2042 		return (NODEV);
2043 	return (cdev2priv(x)->cdp_inode);
2044 }
2045 
2046 static int
devfs_cmp_f(struct file * fp1,struct file * fp2,struct thread * td)2047 devfs_cmp_f(struct file *fp1, struct file *fp2, struct thread *td)
2048 {
2049 	if (fp2->f_type != DTYPE_VNODE || fp2->f_ops != &devfs_ops_f)
2050 		return (3);
2051 	return (kcmp_cmp((uintptr_t)fp1->f_data, (uintptr_t)fp2->f_data));
2052 }
2053 
2054 static struct fileops devfs_ops_f = {
2055 	.fo_read =	devfs_read_f,
2056 	.fo_write =	devfs_write_f,
2057 	.fo_truncate =	devfs_truncate_f,
2058 	.fo_ioctl =	devfs_ioctl_f,
2059 	.fo_poll =	devfs_poll_f,
2060 	.fo_kqfilter =	devfs_kqfilter_f,
2061 	.fo_stat =	devfs_stat_f,
2062 	.fo_close =	devfs_close_f,
2063 	.fo_chmod =	vn_chmod,
2064 	.fo_chown =	vn_chown,
2065 	.fo_sendfile =	vn_sendfile,
2066 	.fo_seek =	vn_seek,
2067 	.fo_fill_kinfo = vn_fill_kinfo,
2068 	.fo_mmap =	devfs_mmap_f,
2069 	.fo_cmp =	devfs_cmp_f,
2070 	.fo_flags =	DFLAG_PASSABLE | DFLAG_SEEKABLE
2071 };
2072 
2073 /* Vops for non-CHR vnodes in /dev. */
2074 static struct vop_vector devfs_vnodeops = {
2075 	.vop_default =		&default_vnodeops,
2076 
2077 	.vop_access =		devfs_access,
2078 	.vop_getattr =		devfs_getattr,
2079 	.vop_ioctl =		devfs_rioctl,
2080 	.vop_lookup =		devfs_lookup,
2081 	.vop_mknod =		devfs_mknod,
2082 	.vop_pathconf =		devfs_pathconf,
2083 	.vop_read =		devfs_rread,
2084 	.vop_readdir =		devfs_readdir,
2085 	.vop_readlink =		devfs_readlink,
2086 	.vop_reclaim =		devfs_reclaim,
2087 	.vop_remove =		devfs_remove,
2088 	.vop_revoke =		devfs_revoke,
2089 	.vop_setattr =		devfs_setattr,
2090 #ifdef MAC
2091 	.vop_setlabel =		devfs_setlabel,
2092 #endif
2093 	.vop_symlink =		devfs_symlink,
2094 	.vop_vptocnp =		devfs_vptocnp,
2095 	.vop_lock1 =		vop_lock,
2096 	.vop_unlock =		vop_unlock,
2097 	.vop_islocked =		vop_islocked,
2098 };
2099 VFS_VOP_VECTOR_REGISTER(devfs_vnodeops);
2100 
2101 /* Vops for VCHR vnodes in /dev. */
2102 static struct vop_vector devfs_specops = {
2103 	.vop_default =		&default_vnodeops,
2104 
2105 	.vop_access =		devfs_access,
2106 	.vop_bmap =		VOP_PANIC,
2107 	.vop_close =		devfs_close,
2108 	.vop_create =		VOP_PANIC,
2109 	.vop_fsync =		vop_stdfsync,
2110 	.vop_getattr =		devfs_getattr,
2111 	.vop_ioctl =		devfs_ioctl,
2112 	.vop_link =		VOP_PANIC,
2113 	.vop_mkdir =		VOP_PANIC,
2114 	.vop_mknod =		VOP_PANIC,
2115 	.vop_open =		devfs_open,
2116 	.vop_pathconf =		devfs_pathconf,
2117 	.vop_poll =		dead_poll,
2118 	.vop_print =		devfs_print,
2119 	.vop_read =		dead_read,
2120 	.vop_readdir =		VOP_PANIC,
2121 	.vop_readlink =		VOP_PANIC,
2122 	.vop_reallocblks =	VOP_PANIC,
2123 	.vop_reclaim =		devfs_reclaim_vchr,
2124 	.vop_remove =		devfs_remove,
2125 	.vop_rename =		VOP_PANIC,
2126 	.vop_revoke =		devfs_revoke,
2127 	.vop_rmdir =		VOP_PANIC,
2128 	.vop_setattr =		devfs_setattr,
2129 #ifdef MAC
2130 	.vop_setlabel =		devfs_setlabel,
2131 #endif
2132 	.vop_strategy =		VOP_PANIC,
2133 	.vop_symlink =		VOP_PANIC,
2134 	.vop_vptocnp =		devfs_vptocnp,
2135 	.vop_write =		dead_write,
2136 	.vop_lock1 =		vop_lock,
2137 	.vop_unlock =		vop_unlock,
2138 	.vop_islocked =		vop_islocked,
2139 };
2140 VFS_VOP_VECTOR_REGISTER(devfs_specops);
2141 
2142 /*
2143  * Our calling convention to the device drivers used to be that we passed
2144  * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
2145  * flags instead since that's what open(), close() and ioctl() takes and
2146  * we don't really want vnode.h in device drivers.
2147  * We solved the source compatibility by redefining some vnode flags to
2148  * be the same as the fcntl ones and by sending down the bitwise OR of
2149  * the respective fcntl/vnode flags.  These CTASSERTS make sure nobody
2150  * pulls the rug out under this.
2151  */
2152 CTASSERT(O_NONBLOCK == IO_NDELAY);
2153 CTASSERT(O_FSYNC == IO_SYNC);
2154