1 /*-
2 * Copyright (c) 2000-2004
3 * Poul-Henning Kamp. All rights reserved.
4 * Copyright (c) 1989, 1992-1993, 1995
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software donated to Berkeley by
8 * Jan-Simon Pendry.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * @(#)kernfs_vnops.c 8.15 (Berkeley) 5/21/95
32 * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vnops.c 1.43
33 *
34 * $FreeBSD$
35 */
36
37 /*
38 * TODO:
39 * mkdir: want it ?
40 */
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/conf.h>
45 #include <sys/dirent.h>
46 #include <sys/fcntl.h>
47 #include <sys/file.h>
48 #include <sys/filedesc.h>
49 #include <sys/filio.h>
50 #include <sys/jail.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/mman.h>
55 #include <sys/mount.h>
56 #include <sys/namei.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/stat.h>
60 #include <sys/sx.h>
61 #include <sys/sysctl.h>
62 #include <sys/time.h>
63 #include <sys/ttycom.h>
64 #include <sys/unistd.h>
65 #include <sys/vnode.h>
66
67 static struct vop_vector devfs_vnodeops;
68 static struct vop_vector devfs_specops;
69 static struct fileops devfs_ops_f;
70
71 #include <fs/devfs/devfs.h>
72 #include <fs/devfs/devfs_int.h>
73
74 #include <security/mac/mac_framework.h>
75
76 #include <vm/vm.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_object.h>
79
80 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
81
82 struct mtx devfs_de_interlock;
83 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
84 struct sx clone_drain_lock;
85 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
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
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
116 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
117 int *ref)
118 {
119
120 *dswp = devvn_refthread(fp->f_vnode, devp, ref);
121 if (*devp != fp->f_data) {
122 if (*dswp != NULL)
123 dev_relthread(*devp, *ref);
124 return (ENXIO);
125 }
126 KASSERT((*devp)->si_refcount > 0,
127 ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
128 if (*dswp == NULL)
129 return (ENXIO);
130 curthread->td_fpop = fp;
131 return (0);
132 }
133
134 int
135 devfs_get_cdevpriv(void **datap)
136 {
137 struct file *fp;
138 struct cdev_privdata *p;
139 int error;
140
141 fp = curthread->td_fpop;
142 if (fp == NULL)
143 return (EBADF);
144 p = fp->f_cdevpriv;
145 if (p != NULL) {
146 error = 0;
147 *datap = p->cdpd_data;
148 } else
149 error = ENOENT;
150 return (error);
151 }
152
153 int
154 devfs_set_cdevpriv(void *priv, d_priv_dtor_t *priv_dtr)
155 {
156 struct file *fp;
157 struct cdev_priv *cdp;
158 struct cdev_privdata *p;
159 int error;
160
161 fp = curthread->td_fpop;
162 if (fp == NULL)
163 return (ENOENT);
164 cdp = cdev2priv((struct cdev *)fp->f_data);
165 p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
166 p->cdpd_data = priv;
167 p->cdpd_dtr = priv_dtr;
168 p->cdpd_fp = fp;
169 mtx_lock(&cdevpriv_mtx);
170 if (fp->f_cdevpriv == NULL) {
171 LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
172 fp->f_cdevpriv = p;
173 mtx_unlock(&cdevpriv_mtx);
174 error = 0;
175 } else {
176 mtx_unlock(&cdevpriv_mtx);
177 free(p, M_CDEVPDATA);
178 error = EBUSY;
179 }
180 return (error);
181 }
182
183 void
184 devfs_destroy_cdevpriv(struct cdev_privdata *p)
185 {
186
187 mtx_assert(&cdevpriv_mtx, MA_OWNED);
188 KASSERT(p->cdpd_fp->f_cdevpriv == p,
189 ("devfs_destoy_cdevpriv %p != %p", p->cdpd_fp->f_cdevpriv, p));
190 p->cdpd_fp->f_cdevpriv = NULL;
191 LIST_REMOVE(p, cdpd_list);
192 mtx_unlock(&cdevpriv_mtx);
193 (p->cdpd_dtr)(p->cdpd_data);
194 free(p, M_CDEVPDATA);
195 }
196
197 static void
198 devfs_fpdrop(struct file *fp)
199 {
200 struct cdev_privdata *p;
201
202 mtx_lock(&cdevpriv_mtx);
203 if ((p = fp->f_cdevpriv) == NULL) {
204 mtx_unlock(&cdevpriv_mtx);
205 return;
206 }
207 devfs_destroy_cdevpriv(p);
208 }
209
210 void
211 devfs_clear_cdevpriv(void)
212 {
213 struct file *fp;
214
215 fp = curthread->td_fpop;
216 if (fp == NULL)
217 return;
218 devfs_fpdrop(fp);
219 }
220
221 /*
222 * On success devfs_populate_vp() returns with dmp->dm_lock held.
223 */
224 static int
225 devfs_populate_vp(struct vnode *vp)
226 {
227 struct devfs_dirent *de;
228 struct devfs_mount *dmp;
229 int locked;
230
231 ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
232
233 dmp = VFSTODEVFS(vp->v_mount);
234 locked = VOP_ISLOCKED(vp);
235
236 sx_xlock(&dmp->dm_lock);
237 DEVFS_DMP_HOLD(dmp);
238
239 /* Can't call devfs_populate() with the vnode lock held. */
240 VOP_UNLOCK(vp, 0);
241 devfs_populate(dmp);
242
243 sx_xunlock(&dmp->dm_lock);
244 vn_lock(vp, locked | LK_RETRY);
245 sx_xlock(&dmp->dm_lock);
246 if (DEVFS_DMP_DROP(dmp)) {
247 sx_xunlock(&dmp->dm_lock);
248 devfs_unmount_final(dmp);
249 return (ERESTART);
250 }
251 if ((vp->v_iflag & VI_DOOMED) != 0) {
252 sx_xunlock(&dmp->dm_lock);
253 return (ERESTART);
254 }
255 de = vp->v_data;
256 KASSERT(de != NULL,
257 ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
258 if ((de->de_flags & DE_DOOMED) != 0) {
259 sx_xunlock(&dmp->dm_lock);
260 return (ERESTART);
261 }
262
263 return (0);
264 }
265
266 static int
267 devfs_vptocnp(struct vop_vptocnp_args *ap)
268 {
269 struct vnode *vp = ap->a_vp;
270 struct vnode **dvp = ap->a_vpp;
271 struct devfs_mount *dmp;
272 char *buf = ap->a_buf;
273 int *buflen = ap->a_buflen;
274 struct devfs_dirent *dd, *de;
275 int i, error;
276
277 dmp = VFSTODEVFS(vp->v_mount);
278
279 error = devfs_populate_vp(vp);
280 if (error != 0)
281 return (error);
282
283 if (vp->v_type != VCHR && vp->v_type != VDIR) {
284 error = ENOENT;
285 goto finished;
286 }
287
288 dd = vp->v_data;
289 if (vp->v_type == VDIR && dd == dmp->dm_rootdir) {
290 *dvp = vp;
291 vref(*dvp);
292 goto finished;
293 }
294
295 i = *buflen;
296 i -= dd->de_dirent->d_namlen;
297 if (i < 0) {
298 error = ENOMEM;
299 goto finished;
300 }
301 bcopy(dd->de_dirent->d_name, buf + i, dd->de_dirent->d_namlen);
302 *buflen = i;
303 de = devfs_parent_dirent(dd);
304 if (de == NULL) {
305 error = ENOENT;
306 goto finished;
307 }
308 mtx_lock(&devfs_de_interlock);
309 *dvp = de->de_vnode;
310 if (*dvp != NULL) {
311 VI_LOCK(*dvp);
312 mtx_unlock(&devfs_de_interlock);
313 vholdl(*dvp);
314 VI_UNLOCK(*dvp);
315 vref(*dvp);
316 vdrop(*dvp);
317 } else {
318 mtx_unlock(&devfs_de_interlock);
319 error = ENOENT;
320 }
321 finished:
322 sx_xunlock(&dmp->dm_lock);
323 return (error);
324 }
325
326 /*
327 * Construct the fully qualified path name relative to the mountpoint.
328 * If a NULL cnp is provided, no '/' is appended to the resulting path.
329 */
330 char *
331 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
332 struct componentname *cnp)
333 {
334 int i;
335 struct devfs_dirent *de;
336
337 sx_assert(&dmp->dm_lock, SA_LOCKED);
338
339 i = SPECNAMELEN;
340 buf[i] = '\0';
341 if (cnp != NULL)
342 i -= cnp->cn_namelen;
343 if (i < 0)
344 return (NULL);
345 if (cnp != NULL)
346 bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
347 de = dd;
348 while (de != dmp->dm_rootdir) {
349 if (cnp != NULL || i < SPECNAMELEN) {
350 i--;
351 if (i < 0)
352 return (NULL);
353 buf[i] = '/';
354 }
355 i -= de->de_dirent->d_namlen;
356 if (i < 0)
357 return (NULL);
358 bcopy(de->de_dirent->d_name, buf + i,
359 de->de_dirent->d_namlen);
360 de = devfs_parent_dirent(de);
361 if (de == NULL)
362 return (NULL);
363 }
364 return (buf + i);
365 }
366
367 static int
368 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
369 struct devfs_dirent *de)
370 {
371 int not_found;
372
373 not_found = 0;
374 if (de->de_flags & DE_DOOMED)
375 not_found = 1;
376 if (DEVFS_DE_DROP(de)) {
377 KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
378 devfs_dirent_free(de);
379 }
380 if (DEVFS_DMP_DROP(dmp)) {
381 KASSERT(not_found == 1,
382 ("DEVFS mount struct freed before dirent"));
383 not_found = 2;
384 sx_xunlock(&dmp->dm_lock);
385 devfs_unmount_final(dmp);
386 }
387 if (not_found == 1 || (drop_dm_lock && not_found != 2))
388 sx_unlock(&dmp->dm_lock);
389 return (not_found);
390 }
391
392 static void
393 devfs_insmntque_dtr(struct vnode *vp, void *arg)
394 {
395 struct devfs_dirent *de;
396
397 de = (struct devfs_dirent *)arg;
398 mtx_lock(&devfs_de_interlock);
399 vp->v_data = NULL;
400 de->de_vnode = NULL;
401 mtx_unlock(&devfs_de_interlock);
402 vgone(vp);
403 vput(vp);
404 }
405
406 /*
407 * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
408 * it on return.
409 */
410 int
411 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
412 struct vnode **vpp)
413 {
414 int error;
415 struct vnode *vp;
416 struct cdev *dev;
417 struct devfs_mount *dmp;
418 struct cdevsw *dsw;
419
420 dmp = VFSTODEVFS(mp);
421 if (de->de_flags & DE_DOOMED) {
422 sx_xunlock(&dmp->dm_lock);
423 return (ENOENT);
424 }
425 loop:
426 DEVFS_DE_HOLD(de);
427 DEVFS_DMP_HOLD(dmp);
428 mtx_lock(&devfs_de_interlock);
429 vp = de->de_vnode;
430 if (vp != NULL) {
431 VI_LOCK(vp);
432 mtx_unlock(&devfs_de_interlock);
433 sx_xunlock(&dmp->dm_lock);
434 vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread);
435 sx_xlock(&dmp->dm_lock);
436 if (devfs_allocv_drop_refs(0, dmp, de)) {
437 vput(vp);
438 return (ENOENT);
439 }
440 else if ((vp->v_iflag & VI_DOOMED) != 0) {
441 mtx_lock(&devfs_de_interlock);
442 if (de->de_vnode == vp) {
443 de->de_vnode = NULL;
444 vp->v_data = NULL;
445 }
446 mtx_unlock(&devfs_de_interlock);
447 vput(vp);
448 goto loop;
449 }
450 sx_xunlock(&dmp->dm_lock);
451 *vpp = vp;
452 return (0);
453 }
454 mtx_unlock(&devfs_de_interlock);
455 if (de->de_dirent->d_type == DT_CHR) {
456 if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
457 devfs_allocv_drop_refs(1, dmp, de);
458 return (ENOENT);
459 }
460 dev = &de->de_cdp->cdp_c;
461 } else {
462 dev = NULL;
463 }
464 error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
465 if (error != 0) {
466 devfs_allocv_drop_refs(1, dmp, de);
467 printf("devfs_allocv: failed to allocate new vnode\n");
468 return (error);
469 }
470
471 if (de->de_dirent->d_type == DT_CHR) {
472 vp->v_type = VCHR;
473 VI_LOCK(vp);
474 dev_lock();
475 dev_refl(dev);
476 /* XXX: v_rdev should be protect by vnode lock */
477 vp->v_rdev = dev;
478 KASSERT(vp->v_usecount == 1,
479 ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount));
480 dev->si_usecount += vp->v_usecount;
481 /* Special casing of ttys for deadfs. Probably redundant. */
482 dsw = dev->si_devsw;
483 if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
484 vp->v_vflag |= VV_ISTTY;
485 dev_unlock();
486 VI_UNLOCK(vp);
487 if ((dev->si_flags & SI_ETERNAL) != 0)
488 vp->v_vflag |= VV_ETERNALDEV;
489 vp->v_op = &devfs_specops;
490 } else if (de->de_dirent->d_type == DT_DIR) {
491 vp->v_type = VDIR;
492 } else if (de->de_dirent->d_type == DT_LNK) {
493 vp->v_type = VLNK;
494 } else {
495 vp->v_type = VBAD;
496 }
497 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
498 VN_LOCK_ASHARE(vp);
499 mtx_lock(&devfs_de_interlock);
500 vp->v_data = de;
501 de->de_vnode = vp;
502 mtx_unlock(&devfs_de_interlock);
503 error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
504 if (error != 0) {
505 (void) devfs_allocv_drop_refs(1, dmp, de);
506 return (error);
507 }
508 if (devfs_allocv_drop_refs(0, dmp, de)) {
509 vput(vp);
510 return (ENOENT);
511 }
512 #ifdef MAC
513 mac_devfs_vnode_associate(mp, de, vp);
514 #endif
515 sx_xunlock(&dmp->dm_lock);
516 *vpp = vp;
517 return (0);
518 }
519
520 static int
521 devfs_access(struct vop_access_args *ap)
522 {
523 struct vnode *vp = ap->a_vp;
524 struct devfs_dirent *de;
525 struct proc *p;
526 int error;
527
528 de = vp->v_data;
529 if (vp->v_type == VDIR)
530 de = de->de_dir;
531
532 error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
533 ap->a_accmode, ap->a_cred, NULL);
534 if (error == 0)
535 return (0);
536 if (error != EACCES)
537 return (error);
538 p = ap->a_td->td_proc;
539 /* We do, however, allow access to the controlling terminal */
540 PROC_LOCK(p);
541 if (!(p->p_flag & P_CONTROLT)) {
542 PROC_UNLOCK(p);
543 return (error);
544 }
545 if (p->p_session->s_ttydp == de->de_cdp)
546 error = 0;
547 PROC_UNLOCK(p);
548 return (error);
549 }
550
551 _Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0,
552 "devfs-only flag reuse failed");
553
554 static int
555 devfs_close(struct vop_close_args *ap)
556 {
557 struct vnode *vp = ap->a_vp, *oldvp;
558 struct thread *td = ap->a_td;
559 struct proc *p;
560 struct cdev *dev = vp->v_rdev;
561 struct cdevsw *dsw;
562 int dflags, error, ref, vp_locked;
563
564 /*
565 * XXX: Don't call d_close() if we were called because of
566 * XXX: insmntque1() failure.
567 */
568 if (vp->v_data == NULL)
569 return (0);
570
571 /*
572 * Hack: a tty device that is a controlling terminal
573 * has a reference from the session structure.
574 * We cannot easily tell that a character device is
575 * a controlling terminal, unless it is the closing
576 * process' controlling terminal. In that case,
577 * if the reference count is 2 (this last descriptor
578 * plus the session), release the reference from the session.
579 */
580 if (td != NULL) {
581 p = td->td_proc;
582 PROC_LOCK(p);
583 if (vp == p->p_session->s_ttyvp) {
584 PROC_UNLOCK(p);
585 oldvp = NULL;
586 sx_xlock(&proctree_lock);
587 if (vp == p->p_session->s_ttyvp) {
588 SESS_LOCK(p->p_session);
589 VI_LOCK(vp);
590 if (count_dev(dev) == 2 &&
591 (vp->v_iflag & VI_DOOMED) == 0) {
592 p->p_session->s_ttyvp = NULL;
593 p->p_session->s_ttydp = NULL;
594 oldvp = vp;
595 }
596 VI_UNLOCK(vp);
597 SESS_UNLOCK(p->p_session);
598 }
599 sx_xunlock(&proctree_lock);
600 if (oldvp != NULL)
601 vrele(oldvp);
602 } else
603 PROC_UNLOCK(p);
604 }
605 /*
606 * We do not want to really close the device if it
607 * is still in use unless we are trying to close it
608 * forcibly. Since every use (buffer, vnode, swap, cmap)
609 * holds a reference to the vnode, and because we mark
610 * any other vnodes that alias this device, when the
611 * sum of the reference counts on all the aliased
612 * vnodes descends to one, we are on last close.
613 */
614 dsw = dev_refthread(dev, &ref);
615 if (dsw == NULL)
616 return (ENXIO);
617 dflags = 0;
618 VI_LOCK(vp);
619 if (vp->v_iflag & VI_DOOMED) {
620 /* Forced close. */
621 dflags |= FREVOKE | FNONBLOCK;
622 } else if (dsw->d_flags & D_TRACKCLOSE) {
623 /* Keep device updated on status. */
624 } else if (count_dev(dev) > 1) {
625 VI_UNLOCK(vp);
626 dev_relthread(dev, ref);
627 return (0);
628 }
629 if (count_dev(dev) == 1)
630 dflags |= FLASTCLOSE;
631 vholdl(vp);
632 VI_UNLOCK(vp);
633 vp_locked = VOP_ISLOCKED(vp);
634 VOP_UNLOCK(vp, 0);
635 KASSERT(dev->si_refcount > 0,
636 ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
637 error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td);
638 dev_relthread(dev, ref);
639 vn_lock(vp, vp_locked | LK_RETRY);
640 vdrop(vp);
641 return (error);
642 }
643
644 static int
645 devfs_close_f(struct file *fp, struct thread *td)
646 {
647 int error;
648 struct file *fpop;
649
650 /*
651 * NB: td may be NULL if this descriptor is closed due to
652 * garbage collection from a closed UNIX domain socket.
653 */
654 fpop = curthread->td_fpop;
655 curthread->td_fpop = fp;
656 error = vnops.fo_close(fp, td);
657 curthread->td_fpop = fpop;
658
659 /*
660 * The f_cdevpriv cannot be assigned non-NULL value while we
661 * are destroying the file.
662 */
663 if (fp->f_cdevpriv != NULL)
664 devfs_fpdrop(fp);
665 return (error);
666 }
667
668 static int
669 devfs_getattr(struct vop_getattr_args *ap)
670 {
671 struct vnode *vp = ap->a_vp;
672 struct vattr *vap = ap->a_vap;
673 struct devfs_dirent *de;
674 struct devfs_mount *dmp;
675 struct cdev *dev;
676 struct timeval boottime;
677 int error;
678
679 error = devfs_populate_vp(vp);
680 if (error != 0)
681 return (error);
682
683 dmp = VFSTODEVFS(vp->v_mount);
684 sx_xunlock(&dmp->dm_lock);
685
686 de = vp->v_data;
687 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
688 if (vp->v_type == VDIR) {
689 de = de->de_dir;
690 KASSERT(de != NULL,
691 ("Null dir dirent in devfs_getattr vp=%p", vp));
692 }
693 vap->va_uid = de->de_uid;
694 vap->va_gid = de->de_gid;
695 vap->va_mode = de->de_mode;
696 if (vp->v_type == VLNK)
697 vap->va_size = strlen(de->de_symlink);
698 else if (vp->v_type == VDIR)
699 vap->va_size = vap->va_bytes = DEV_BSIZE;
700 else
701 vap->va_size = 0;
702 if (vp->v_type != VDIR)
703 vap->va_bytes = 0;
704 vap->va_blocksize = DEV_BSIZE;
705 vap->va_type = vp->v_type;
706
707 getboottime(&boottime);
708 #define fix(aa) \
709 do { \
710 if ((aa).tv_sec <= 3600) { \
711 (aa).tv_sec = boottime.tv_sec; \
712 (aa).tv_nsec = boottime.tv_usec * 1000; \
713 } \
714 } while (0)
715
716 if (vp->v_type != VCHR) {
717 fix(de->de_atime);
718 vap->va_atime = de->de_atime;
719 fix(de->de_mtime);
720 vap->va_mtime = de->de_mtime;
721 fix(de->de_ctime);
722 vap->va_ctime = de->de_ctime;
723 } else {
724 dev = vp->v_rdev;
725 fix(dev->si_atime);
726 vap->va_atime = dev->si_atime;
727 fix(dev->si_mtime);
728 vap->va_mtime = dev->si_mtime;
729 fix(dev->si_ctime);
730 vap->va_ctime = dev->si_ctime;
731
732 vap->va_rdev = cdev2priv(dev)->cdp_inode;
733 }
734 vap->va_gen = 0;
735 vap->va_flags = 0;
736 vap->va_filerev = 0;
737 vap->va_nlink = de->de_links;
738 vap->va_fileid = de->de_inode;
739
740 return (error);
741 }
742
743 /* ARGSUSED */
744 static int
745 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
746 {
747 struct cdev *dev;
748 struct cdevsw *dsw;
749 struct vnode *vp;
750 struct vnode *vpold;
751 int error, i, ref;
752 const char *p;
753 struct fiodgname_arg *fgn;
754 struct file *fpop;
755
756 fpop = td->td_fpop;
757 error = devfs_fp_check(fp, &dev, &dsw, &ref);
758 if (error != 0) {
759 error = vnops.fo_ioctl(fp, com, data, cred, td);
760 return (error);
761 }
762
763 if (com == FIODTYPE) {
764 *(int *)data = dsw->d_flags & D_TYPEMASK;
765 td->td_fpop = fpop;
766 dev_relthread(dev, ref);
767 return (0);
768 } else if (com == FIODGNAME) {
769 fgn = data;
770 p = devtoname(dev);
771 i = strlen(p) + 1;
772 if (i > fgn->len)
773 error = EINVAL;
774 else
775 error = copyout(p, fgn->buf, i);
776 td->td_fpop = fpop;
777 dev_relthread(dev, ref);
778 return (error);
779 }
780 error = dsw->d_ioctl(dev, com, data, fp->f_flag, td);
781 td->td_fpop = NULL;
782 dev_relthread(dev, ref);
783 if (error == ENOIOCTL)
784 error = ENOTTY;
785 if (error == 0 && com == TIOCSCTTY) {
786 vp = fp->f_vnode;
787
788 /*
789 * Do nothing if reassigning same control tty, or if the
790 * control tty has already disappeared. If it disappeared,
791 * it's because we were racing with TIOCNOTTY. TIOCNOTTY
792 * already took care of releasing the old vnode and we have
793 * nothing left to do.
794 */
795 sx_slock(&proctree_lock);
796 if (td->td_proc->p_session->s_ttyvp == vp ||
797 td->td_proc->p_session->s_ttyp == NULL) {
798 sx_sunlock(&proctree_lock);
799 return (0);
800 }
801
802 vpold = td->td_proc->p_session->s_ttyvp;
803 VREF(vp);
804 SESS_LOCK(td->td_proc->p_session);
805 td->td_proc->p_session->s_ttyvp = vp;
806 td->td_proc->p_session->s_ttydp = cdev2priv(dev);
807 SESS_UNLOCK(td->td_proc->p_session);
808
809 sx_sunlock(&proctree_lock);
810
811 /* Get rid of reference to old control tty */
812 if (vpold)
813 vrele(vpold);
814 }
815 return (error);
816 }
817
818 /* ARGSUSED */
819 static int
820 devfs_kqfilter_f(struct file *fp, struct knote *kn)
821 {
822 struct cdev *dev;
823 struct cdevsw *dsw;
824 int error, ref;
825 struct file *fpop;
826 struct thread *td;
827
828 td = curthread;
829 fpop = td->td_fpop;
830 error = devfs_fp_check(fp, &dev, &dsw, &ref);
831 if (error)
832 return (error);
833 error = dsw->d_kqfilter(dev, kn);
834 td->td_fpop = fpop;
835 dev_relthread(dev, ref);
836 return (error);
837 }
838
839 static inline int
840 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
841 {
842 struct cdev_priv *cdp;
843 struct ucred *dcr;
844 struct proc *p;
845 int error;
846
847 cdp = de->de_cdp;
848 if (cdp == NULL)
849 return (0);
850 dcr = cdp->cdp_c.si_cred;
851 if (dcr == NULL)
852 return (0);
853
854 error = prison_check(td->td_ucred, dcr);
855 if (error == 0)
856 return (0);
857 /* We do, however, allow access to the controlling terminal */
858 p = td->td_proc;
859 PROC_LOCK(p);
860 if (!(p->p_flag & P_CONTROLT)) {
861 PROC_UNLOCK(p);
862 return (error);
863 }
864 if (p->p_session->s_ttydp == cdp)
865 error = 0;
866 PROC_UNLOCK(p);
867 return (error);
868 }
869
870 static int
871 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
872 {
873 struct componentname *cnp;
874 struct vnode *dvp, **vpp;
875 struct thread *td;
876 struct devfs_dirent *de, *dd;
877 struct devfs_dirent **dde;
878 struct devfs_mount *dmp;
879 struct mount *mp;
880 struct cdev *cdev;
881 int error, flags, nameiop, dvplocked;
882 char specname[SPECNAMELEN + 1], *pname;
883
884 cnp = ap->a_cnp;
885 vpp = ap->a_vpp;
886 dvp = ap->a_dvp;
887 pname = cnp->cn_nameptr;
888 td = cnp->cn_thread;
889 flags = cnp->cn_flags;
890 nameiop = cnp->cn_nameiop;
891 mp = dvp->v_mount;
892 dmp = VFSTODEVFS(mp);
893 dd = dvp->v_data;
894 *vpp = NULLVP;
895
896 if ((flags & ISLASTCN) && nameiop == RENAME)
897 return (EOPNOTSUPP);
898
899 if (dvp->v_type != VDIR)
900 return (ENOTDIR);
901
902 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
903 return (EIO);
904
905 error = vn_dir_check_exec(dvp, cnp);
906 if (error != 0)
907 return (error);
908
909 if (cnp->cn_namelen == 1 && *pname == '.') {
910 if ((flags & ISLASTCN) && nameiop != LOOKUP)
911 return (EINVAL);
912 *vpp = dvp;
913 VREF(dvp);
914 return (0);
915 }
916
917 if (flags & ISDOTDOT) {
918 if ((flags & ISLASTCN) && nameiop != LOOKUP)
919 return (EINVAL);
920 de = devfs_parent_dirent(dd);
921 if (de == NULL)
922 return (ENOENT);
923 dvplocked = VOP_ISLOCKED(dvp);
924 VOP_UNLOCK(dvp, 0);
925 error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK,
926 vpp);
927 *dm_unlock = 0;
928 vn_lock(dvp, dvplocked | LK_RETRY);
929 return (error);
930 }
931
932 dd = dvp->v_data;
933 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
934 while (de == NULL) { /* While(...) so we can use break */
935
936 if (nameiop == DELETE)
937 return (ENOENT);
938
939 /*
940 * OK, we didn't have an entry for the name we were asked for
941 * so we try to see if anybody can create it on demand.
942 */
943 pname = devfs_fqpn(specname, dmp, dd, cnp);
944 if (pname == NULL)
945 break;
946
947 cdev = NULL;
948 DEVFS_DMP_HOLD(dmp);
949 sx_xunlock(&dmp->dm_lock);
950 sx_slock(&clone_drain_lock);
951 EVENTHANDLER_INVOKE(dev_clone,
952 td->td_ucred, pname, strlen(pname), &cdev);
953 sx_sunlock(&clone_drain_lock);
954
955 if (cdev == NULL)
956 sx_xlock(&dmp->dm_lock);
957 else if (devfs_populate_vp(dvp) != 0) {
958 *dm_unlock = 0;
959 sx_xlock(&dmp->dm_lock);
960 if (DEVFS_DMP_DROP(dmp)) {
961 sx_xunlock(&dmp->dm_lock);
962 devfs_unmount_final(dmp);
963 } else
964 sx_xunlock(&dmp->dm_lock);
965 dev_rel(cdev);
966 return (ENOENT);
967 }
968 if (DEVFS_DMP_DROP(dmp)) {
969 *dm_unlock = 0;
970 sx_xunlock(&dmp->dm_lock);
971 devfs_unmount_final(dmp);
972 if (cdev != NULL)
973 dev_rel(cdev);
974 return (ENOENT);
975 }
976
977 if (cdev == NULL)
978 break;
979
980 dev_lock();
981 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
982 if (dde != NULL && *dde != NULL)
983 de = *dde;
984 dev_unlock();
985 dev_rel(cdev);
986 break;
987 }
988
989 if (de == NULL || de->de_flags & DE_WHITEOUT) {
990 if ((nameiop == CREATE || nameiop == RENAME) &&
991 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
992 cnp->cn_flags |= SAVENAME;
993 return (EJUSTRETURN);
994 }
995 return (ENOENT);
996 }
997
998 if (devfs_prison_check(de, td))
999 return (ENOENT);
1000
1001 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
1002 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
1003 if (error)
1004 return (error);
1005 if (*vpp == dvp) {
1006 VREF(dvp);
1007 *vpp = dvp;
1008 return (0);
1009 }
1010 }
1011 error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp);
1012 *dm_unlock = 0;
1013 return (error);
1014 }
1015
1016 static int
1017 devfs_lookup(struct vop_lookup_args *ap)
1018 {
1019 int j;
1020 struct devfs_mount *dmp;
1021 int dm_unlock;
1022
1023 if (devfs_populate_vp(ap->a_dvp) != 0)
1024 return (ENOTDIR);
1025
1026 dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1027 dm_unlock = 1;
1028 j = devfs_lookupx(ap, &dm_unlock);
1029 if (dm_unlock == 1)
1030 sx_xunlock(&dmp->dm_lock);
1031 return (j);
1032 }
1033
1034 static int
1035 devfs_mknod(struct vop_mknod_args *ap)
1036 {
1037 struct componentname *cnp;
1038 struct vnode *dvp, **vpp;
1039 struct devfs_dirent *dd, *de;
1040 struct devfs_mount *dmp;
1041 int error;
1042
1043 /*
1044 * The only type of node we should be creating here is a
1045 * character device, for anything else return EOPNOTSUPP.
1046 */
1047 if (ap->a_vap->va_type != VCHR)
1048 return (EOPNOTSUPP);
1049 dvp = ap->a_dvp;
1050 dmp = VFSTODEVFS(dvp->v_mount);
1051
1052 cnp = ap->a_cnp;
1053 vpp = ap->a_vpp;
1054 dd = dvp->v_data;
1055
1056 error = ENOENT;
1057 sx_xlock(&dmp->dm_lock);
1058 TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1059 if (cnp->cn_namelen != de->de_dirent->d_namlen)
1060 continue;
1061 if (de->de_dirent->d_type == DT_CHR &&
1062 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
1063 continue;
1064 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1065 de->de_dirent->d_namlen) != 0)
1066 continue;
1067 if (de->de_flags & DE_WHITEOUT)
1068 break;
1069 goto notfound;
1070 }
1071 if (de == NULL)
1072 goto notfound;
1073 de->de_flags &= ~DE_WHITEOUT;
1074 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1075 return (error);
1076 notfound:
1077 sx_xunlock(&dmp->dm_lock);
1078 return (error);
1079 }
1080
1081 /* ARGSUSED */
1082 static int
1083 devfs_open(struct vop_open_args *ap)
1084 {
1085 struct thread *td = ap->a_td;
1086 struct vnode *vp = ap->a_vp;
1087 struct cdev *dev = vp->v_rdev;
1088 struct file *fp = ap->a_fp;
1089 int error, ref, vlocked;
1090 struct cdevsw *dsw;
1091 struct file *fpop;
1092 struct mtx *mtxp;
1093
1094 if (vp->v_type == VBLK)
1095 return (ENXIO);
1096
1097 if (dev == NULL)
1098 return (ENXIO);
1099
1100 /* Make this field valid before any I/O in d_open. */
1101 if (dev->si_iosize_max == 0)
1102 dev->si_iosize_max = DFLTPHYS;
1103
1104 dsw = dev_refthread(dev, &ref);
1105 if (dsw == NULL)
1106 return (ENXIO);
1107 if (fp == NULL && dsw->d_fdopen != NULL) {
1108 dev_relthread(dev, ref);
1109 return (ENXIO);
1110 }
1111
1112 vlocked = VOP_ISLOCKED(vp);
1113 VOP_UNLOCK(vp, 0);
1114
1115 fpop = td->td_fpop;
1116 td->td_fpop = fp;
1117 if (fp != NULL) {
1118 fp->f_data = dev;
1119 fp->f_vnode = vp;
1120 }
1121 if (dsw->d_fdopen != NULL)
1122 error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1123 else
1124 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1125 /* Clean up any cdevpriv upon error. */
1126 if (error != 0)
1127 devfs_clear_cdevpriv();
1128 td->td_fpop = fpop;
1129
1130 vn_lock(vp, vlocked | LK_RETRY);
1131 dev_relthread(dev, ref);
1132 if (error != 0) {
1133 if (error == ERESTART)
1134 error = EINTR;
1135 return (error);
1136 }
1137
1138 #if 0 /* /dev/console */
1139 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1140 #else
1141 if (fp == NULL)
1142 return (error);
1143 #endif
1144 if (fp->f_ops == &badfileops)
1145 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1146 mtxp = mtx_pool_find(mtxpool_sleep, fp);
1147
1148 /*
1149 * Hint to the dofilewrite() to not force the buffer draining
1150 * on the writer to the file. Most likely, the write would
1151 * not need normal buffers.
1152 */
1153 mtx_lock(mtxp);
1154 fp->f_vnread_flags |= FDEVFS_VNODE;
1155 mtx_unlock(mtxp);
1156 return (error);
1157 }
1158
1159 static int
1160 devfs_pathconf(struct vop_pathconf_args *ap)
1161 {
1162
1163 switch (ap->a_name) {
1164 case _PC_FILESIZEBITS:
1165 *ap->a_retval = 64;
1166 return (0);
1167 case _PC_NAME_MAX:
1168 *ap->a_retval = NAME_MAX;
1169 return (0);
1170 case _PC_LINK_MAX:
1171 *ap->a_retval = LINK_MAX;
1172 return (0);
1173 case _PC_SYMLINK_MAX:
1174 *ap->a_retval = MAXPATHLEN;
1175 return (0);
1176 case _PC_MAX_CANON:
1177 if (ap->a_vp->v_vflag & VV_ISTTY) {
1178 *ap->a_retval = MAX_CANON;
1179 return (0);
1180 }
1181 return (EINVAL);
1182 case _PC_MAX_INPUT:
1183 if (ap->a_vp->v_vflag & VV_ISTTY) {
1184 *ap->a_retval = MAX_INPUT;
1185 return (0);
1186 }
1187 return (EINVAL);
1188 case _PC_VDISABLE:
1189 if (ap->a_vp->v_vflag & VV_ISTTY) {
1190 *ap->a_retval = _POSIX_VDISABLE;
1191 return (0);
1192 }
1193 return (EINVAL);
1194 case _PC_MAC_PRESENT:
1195 #ifdef MAC
1196 /*
1197 * If MAC is enabled, devfs automatically supports
1198 * trivial non-persistant label storage.
1199 */
1200 *ap->a_retval = 1;
1201 #else
1202 *ap->a_retval = 0;
1203 #endif
1204 return (0);
1205 case _PC_CHOWN_RESTRICTED:
1206 *ap->a_retval = 1;
1207 return (0);
1208 default:
1209 return (vop_stdpathconf(ap));
1210 }
1211 /* NOTREACHED */
1212 }
1213
1214 /* ARGSUSED */
1215 static int
1216 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1217 {
1218 struct cdev *dev;
1219 struct cdevsw *dsw;
1220 int error, ref;
1221 struct file *fpop;
1222
1223 fpop = td->td_fpop;
1224 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1225 if (error != 0) {
1226 error = vnops.fo_poll(fp, events, cred, td);
1227 return (error);
1228 }
1229 error = dsw->d_poll(dev, events, td);
1230 td->td_fpop = fpop;
1231 dev_relthread(dev, ref);
1232 return(error);
1233 }
1234
1235 /*
1236 * Print out the contents of a special device vnode.
1237 */
1238 static int
1239 devfs_print(struct vop_print_args *ap)
1240 {
1241
1242 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1243 return (0);
1244 }
1245
1246 static int
1247 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1248 int flags, struct thread *td)
1249 {
1250 struct cdev *dev;
1251 int ioflag, error, ref;
1252 ssize_t resid;
1253 struct cdevsw *dsw;
1254 struct file *fpop;
1255
1256 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1257 return (EINVAL);
1258 fpop = td->td_fpop;
1259 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1260 if (error != 0) {
1261 error = vnops.fo_read(fp, uio, cred, flags, td);
1262 return (error);
1263 }
1264 resid = uio->uio_resid;
1265 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1266 if (ioflag & O_DIRECT)
1267 ioflag |= IO_DIRECT;
1268
1269 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1270 error = dsw->d_read(dev, uio, ioflag);
1271 if (uio->uio_resid != resid || (error == 0 && resid != 0))
1272 devfs_timestamp(&dev->si_atime);
1273 td->td_fpop = fpop;
1274 dev_relthread(dev, ref);
1275
1276 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1277 return (error);
1278 }
1279
1280 static int
1281 devfs_readdir(struct vop_readdir_args *ap)
1282 {
1283 int error;
1284 struct uio *uio;
1285 struct dirent *dp;
1286 struct devfs_dirent *dd;
1287 struct devfs_dirent *de;
1288 struct devfs_mount *dmp;
1289 off_t off;
1290 int *tmp_ncookies = NULL;
1291
1292 if (ap->a_vp->v_type != VDIR)
1293 return (ENOTDIR);
1294
1295 uio = ap->a_uio;
1296 if (uio->uio_offset < 0)
1297 return (EINVAL);
1298
1299 /*
1300 * XXX: This is a temporary hack to get around this filesystem not
1301 * supporting cookies. We store the location of the ncookies pointer
1302 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1303 * and set the number of cookies to 0. We then set the pointer to
1304 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1305 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1306 * pointer to its original location before returning to the caller.
1307 */
1308 if (ap->a_ncookies != NULL) {
1309 tmp_ncookies = ap->a_ncookies;
1310 *ap->a_ncookies = 0;
1311 ap->a_ncookies = NULL;
1312 }
1313
1314 dmp = VFSTODEVFS(ap->a_vp->v_mount);
1315 if (devfs_populate_vp(ap->a_vp) != 0) {
1316 if (tmp_ncookies != NULL)
1317 ap->a_ncookies = tmp_ncookies;
1318 return (EIO);
1319 }
1320 error = 0;
1321 de = ap->a_vp->v_data;
1322 off = 0;
1323 TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1324 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1325 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1326 continue;
1327 if (devfs_prison_check(dd, uio->uio_td))
1328 continue;
1329 if (dd->de_dirent->d_type == DT_DIR)
1330 de = dd->de_dir;
1331 else
1332 de = dd;
1333 dp = dd->de_dirent;
1334 if (dp->d_reclen > uio->uio_resid)
1335 break;
1336 dp->d_fileno = de->de_inode;
1337 if (off >= uio->uio_offset) {
1338 error = vfs_read_dirent(ap, dp, off);
1339 if (error)
1340 break;
1341 }
1342 off += dp->d_reclen;
1343 }
1344 sx_xunlock(&dmp->dm_lock);
1345 uio->uio_offset = off;
1346
1347 /*
1348 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1349 * place.
1350 */
1351 if (tmp_ncookies != NULL)
1352 ap->a_ncookies = tmp_ncookies;
1353
1354 return (error);
1355 }
1356
1357 static int
1358 devfs_readlink(struct vop_readlink_args *ap)
1359 {
1360 struct devfs_dirent *de;
1361
1362 de = ap->a_vp->v_data;
1363 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1364 }
1365
1366 static int
1367 devfs_reclaim(struct vop_reclaim_args *ap)
1368 {
1369 struct vnode *vp;
1370 struct devfs_dirent *de;
1371
1372 vp = ap->a_vp;
1373 mtx_lock(&devfs_de_interlock);
1374 de = vp->v_data;
1375 if (de != NULL) {
1376 de->de_vnode = NULL;
1377 vp->v_data = NULL;
1378 }
1379 mtx_unlock(&devfs_de_interlock);
1380 vnode_destroy_vobject(vp);
1381 return (0);
1382 }
1383
1384 static int
1385 devfs_reclaim_vchr(struct vop_reclaim_args *ap)
1386 {
1387 struct vnode *vp;
1388 struct cdev *dev;
1389
1390 vp = ap->a_vp;
1391 MPASS(vp->v_type == VCHR);
1392
1393 devfs_reclaim(ap);
1394
1395 VI_LOCK(vp);
1396 dev_lock();
1397 dev = vp->v_rdev;
1398 vp->v_rdev = NULL;
1399 if (dev != NULL)
1400 dev->si_usecount -= vp->v_usecount;
1401 dev_unlock();
1402 VI_UNLOCK(vp);
1403 if (dev != NULL)
1404 dev_rel(dev);
1405 return (0);
1406 }
1407
1408 static int
1409 devfs_remove(struct vop_remove_args *ap)
1410 {
1411 struct vnode *dvp = ap->a_dvp;
1412 struct vnode *vp = ap->a_vp;
1413 struct devfs_dirent *dd;
1414 struct devfs_dirent *de, *de_covered;
1415 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1416
1417 ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1418 ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1419
1420 sx_xlock(&dmp->dm_lock);
1421 dd = ap->a_dvp->v_data;
1422 de = vp->v_data;
1423 if (de->de_cdp == NULL) {
1424 TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1425 if (de->de_dirent->d_type == DT_LNK) {
1426 de_covered = devfs_find(dd, de->de_dirent->d_name,
1427 de->de_dirent->d_namlen, 0);
1428 if (de_covered != NULL)
1429 de_covered->de_flags &= ~DE_COVERED;
1430 }
1431 /* We need to unlock dvp because devfs_delete() may lock it. */
1432 VOP_UNLOCK(vp, 0);
1433 if (dvp != vp)
1434 VOP_UNLOCK(dvp, 0);
1435 devfs_delete(dmp, de, 0);
1436 sx_xunlock(&dmp->dm_lock);
1437 if (dvp != vp)
1438 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1439 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1440 } else {
1441 de->de_flags |= DE_WHITEOUT;
1442 sx_xunlock(&dmp->dm_lock);
1443 }
1444 return (0);
1445 }
1446
1447 /*
1448 * Revoke is called on a tty when a terminal session ends. The vnode
1449 * is orphaned by setting v_op to deadfs so we need to let go of it
1450 * as well so that we create a new one next time around.
1451 *
1452 */
1453 static int
1454 devfs_revoke(struct vop_revoke_args *ap)
1455 {
1456 struct vnode *vp = ap->a_vp, *vp2;
1457 struct cdev *dev;
1458 struct cdev_priv *cdp;
1459 struct devfs_dirent *de;
1460 u_int i;
1461
1462 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1463
1464 dev = vp->v_rdev;
1465 cdp = cdev2priv(dev);
1466
1467 dev_lock();
1468 cdp->cdp_inuse++;
1469 dev_unlock();
1470
1471 vhold(vp);
1472 vgone(vp);
1473 vdrop(vp);
1474
1475 VOP_UNLOCK(vp,0);
1476 loop:
1477 for (;;) {
1478 mtx_lock(&devfs_de_interlock);
1479 dev_lock();
1480 vp2 = NULL;
1481 for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1482 de = cdp->cdp_dirents[i];
1483 if (de == NULL)
1484 continue;
1485
1486 vp2 = de->de_vnode;
1487 if (vp2 != NULL) {
1488 dev_unlock();
1489 VI_LOCK(vp2);
1490 mtx_unlock(&devfs_de_interlock);
1491 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1492 curthread))
1493 goto loop;
1494 vhold(vp2);
1495 vgone(vp2);
1496 vdrop(vp2);
1497 vput(vp2);
1498 break;
1499 }
1500 }
1501 if (vp2 != NULL) {
1502 continue;
1503 }
1504 dev_unlock();
1505 mtx_unlock(&devfs_de_interlock);
1506 break;
1507 }
1508 dev_lock();
1509 cdp->cdp_inuse--;
1510 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1511 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1512 dev_unlock();
1513 dev_rel(&cdp->cdp_c);
1514 } else
1515 dev_unlock();
1516
1517 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1518 return (0);
1519 }
1520
1521 static int
1522 devfs_rioctl(struct vop_ioctl_args *ap)
1523 {
1524 struct vnode *vp;
1525 struct devfs_mount *dmp;
1526 int error;
1527
1528 vp = ap->a_vp;
1529 vn_lock(vp, LK_SHARED | LK_RETRY);
1530 if (vp->v_iflag & VI_DOOMED) {
1531 VOP_UNLOCK(vp, 0);
1532 return (EBADF);
1533 }
1534 dmp = VFSTODEVFS(vp->v_mount);
1535 sx_xlock(&dmp->dm_lock);
1536 VOP_UNLOCK(vp, 0);
1537 DEVFS_DMP_HOLD(dmp);
1538 devfs_populate(dmp);
1539 if (DEVFS_DMP_DROP(dmp)) {
1540 sx_xunlock(&dmp->dm_lock);
1541 devfs_unmount_final(dmp);
1542 return (ENOENT);
1543 }
1544 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1545 sx_xunlock(&dmp->dm_lock);
1546 return (error);
1547 }
1548
1549 static int
1550 devfs_rread(struct vop_read_args *ap)
1551 {
1552
1553 if (ap->a_vp->v_type != VDIR)
1554 return (EINVAL);
1555 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1556 }
1557
1558 static int
1559 devfs_setattr(struct vop_setattr_args *ap)
1560 {
1561 struct devfs_dirent *de;
1562 struct vattr *vap;
1563 struct vnode *vp;
1564 struct thread *td;
1565 int c, error;
1566 uid_t uid;
1567 gid_t gid;
1568
1569 vap = ap->a_vap;
1570 vp = ap->a_vp;
1571 td = curthread;
1572 if ((vap->va_type != VNON) ||
1573 (vap->va_nlink != VNOVAL) ||
1574 (vap->va_fsid != VNOVAL) ||
1575 (vap->va_fileid != VNOVAL) ||
1576 (vap->va_blocksize != VNOVAL) ||
1577 (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1578 (vap->va_rdev != VNOVAL) ||
1579 ((int)vap->va_bytes != VNOVAL) ||
1580 (vap->va_gen != VNOVAL)) {
1581 return (EINVAL);
1582 }
1583
1584 error = devfs_populate_vp(vp);
1585 if (error != 0)
1586 return (error);
1587
1588 de = vp->v_data;
1589 if (vp->v_type == VDIR)
1590 de = de->de_dir;
1591
1592 c = 0;
1593 if (vap->va_uid == (uid_t)VNOVAL)
1594 uid = de->de_uid;
1595 else
1596 uid = vap->va_uid;
1597 if (vap->va_gid == (gid_t)VNOVAL)
1598 gid = de->de_gid;
1599 else
1600 gid = vap->va_gid;
1601 if (uid != de->de_uid || gid != de->de_gid) {
1602 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1603 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1604 error = priv_check(td, PRIV_VFS_CHOWN);
1605 if (error != 0)
1606 goto ret;
1607 }
1608 de->de_uid = uid;
1609 de->de_gid = gid;
1610 c = 1;
1611 }
1612
1613 if (vap->va_mode != (mode_t)VNOVAL) {
1614 if (ap->a_cred->cr_uid != de->de_uid) {
1615 error = priv_check(td, PRIV_VFS_ADMIN);
1616 if (error != 0)
1617 goto ret;
1618 }
1619 de->de_mode = vap->va_mode;
1620 c = 1;
1621 }
1622
1623 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1624 error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1625 if (error != 0)
1626 goto ret;
1627 if (vap->va_atime.tv_sec != VNOVAL) {
1628 if (vp->v_type == VCHR)
1629 vp->v_rdev->si_atime = vap->va_atime;
1630 else
1631 de->de_atime = vap->va_atime;
1632 }
1633 if (vap->va_mtime.tv_sec != VNOVAL) {
1634 if (vp->v_type == VCHR)
1635 vp->v_rdev->si_mtime = vap->va_mtime;
1636 else
1637 de->de_mtime = vap->va_mtime;
1638 }
1639 c = 1;
1640 }
1641
1642 if (c) {
1643 if (vp->v_type == VCHR)
1644 vfs_timestamp(&vp->v_rdev->si_ctime);
1645 else
1646 vfs_timestamp(&de->de_mtime);
1647 }
1648
1649 ret:
1650 sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock);
1651 return (error);
1652 }
1653
1654 #ifdef MAC
1655 static int
1656 devfs_setlabel(struct vop_setlabel_args *ap)
1657 {
1658 struct vnode *vp;
1659 struct devfs_dirent *de;
1660
1661 vp = ap->a_vp;
1662 de = vp->v_data;
1663
1664 mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1665 mac_devfs_update(vp->v_mount, de, vp);
1666
1667 return (0);
1668 }
1669 #endif
1670
1671 static int
1672 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1673 {
1674
1675 return (vnops.fo_stat(fp, sb, cred, td));
1676 }
1677
1678 static int
1679 devfs_symlink(struct vop_symlink_args *ap)
1680 {
1681 int i, error;
1682 struct devfs_dirent *dd;
1683 struct devfs_dirent *de, *de_covered, *de_dotdot;
1684 struct devfs_mount *dmp;
1685
1686 error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1687 if (error)
1688 return(error);
1689 dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1690 if (devfs_populate_vp(ap->a_dvp) != 0)
1691 return (ENOENT);
1692
1693 dd = ap->a_dvp->v_data;
1694 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1695 de->de_flags = DE_USER;
1696 de->de_uid = 0;
1697 de->de_gid = 0;
1698 de->de_mode = 0755;
1699 de->de_inode = alloc_unr(devfs_inos);
1700 de->de_dir = dd;
1701 de->de_dirent->d_type = DT_LNK;
1702 i = strlen(ap->a_target) + 1;
1703 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1704 bcopy(ap->a_target, de->de_symlink, i);
1705 #ifdef MAC
1706 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1707 #endif
1708 de_covered = devfs_find(dd, de->de_dirent->d_name,
1709 de->de_dirent->d_namlen, 0);
1710 if (de_covered != NULL) {
1711 if ((de_covered->de_flags & DE_USER) != 0) {
1712 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1713 sx_xunlock(&dmp->dm_lock);
1714 return (EEXIST);
1715 }
1716 KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1717 ("devfs_symlink: entry %p already covered", de_covered));
1718 de_covered->de_flags |= DE_COVERED;
1719 }
1720
1721 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */
1722 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */
1723 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1724 devfs_dir_ref_de(dmp, dd);
1725 devfs_rules_apply(dmp, de);
1726
1727 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1728 }
1729
1730 static int
1731 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1732 {
1733
1734 return (vnops.fo_truncate(fp, length, cred, td));
1735 }
1736
1737 static int
1738 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1739 int flags, struct thread *td)
1740 {
1741 struct cdev *dev;
1742 int error, ioflag, ref;
1743 ssize_t resid;
1744 struct cdevsw *dsw;
1745 struct file *fpop;
1746
1747 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1748 return (EINVAL);
1749 fpop = td->td_fpop;
1750 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1751 if (error != 0) {
1752 error = vnops.fo_write(fp, uio, cred, flags, td);
1753 return (error);
1754 }
1755 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1756 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1757 if (ioflag & O_DIRECT)
1758 ioflag |= IO_DIRECT;
1759 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1760
1761 resid = uio->uio_resid;
1762
1763 error = dsw->d_write(dev, uio, ioflag);
1764 if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1765 devfs_timestamp(&dev->si_ctime);
1766 dev->si_mtime = dev->si_ctime;
1767 }
1768 td->td_fpop = fpop;
1769 dev_relthread(dev, ref);
1770
1771 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1772 return (error);
1773 }
1774
1775 static int
1776 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1777 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1778 struct thread *td)
1779 {
1780 struct cdev *dev;
1781 struct cdevsw *dsw;
1782 struct mount *mp;
1783 struct vnode *vp;
1784 struct file *fpop;
1785 vm_object_t object;
1786 vm_prot_t maxprot;
1787 int error, ref;
1788
1789 vp = fp->f_vnode;
1790
1791 /*
1792 * Ensure that file and memory protections are
1793 * compatible.
1794 */
1795 mp = vp->v_mount;
1796 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1797 maxprot = VM_PROT_NONE;
1798 if ((prot & VM_PROT_EXECUTE) != 0)
1799 return (EACCES);
1800 } else
1801 maxprot = VM_PROT_EXECUTE;
1802 if ((fp->f_flag & FREAD) != 0)
1803 maxprot |= VM_PROT_READ;
1804 else if ((prot & VM_PROT_READ) != 0)
1805 return (EACCES);
1806
1807 /*
1808 * If we are sharing potential changes via MAP_SHARED and we
1809 * are trying to get write permission although we opened it
1810 * without asking for it, bail out.
1811 *
1812 * Note that most character devices always share mappings.
1813 * The one exception is that D_MMAP_ANON devices
1814 * (i.e. /dev/zero) permit private writable mappings.
1815 *
1816 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests
1817 * as well as updating maxprot to permit writing for
1818 * D_MMAP_ANON devices rather than doing that here.
1819 */
1820 if ((flags & MAP_SHARED) != 0) {
1821 if ((fp->f_flag & FWRITE) != 0)
1822 maxprot |= VM_PROT_WRITE;
1823 else if ((prot & VM_PROT_WRITE) != 0)
1824 return (EACCES);
1825 }
1826 maxprot &= cap_maxprot;
1827
1828 fpop = td->td_fpop;
1829 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1830 if (error != 0)
1831 return (error);
1832
1833 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff,
1834 &object);
1835 td->td_fpop = fpop;
1836 dev_relthread(dev, ref);
1837 if (error != 0)
1838 return (error);
1839
1840 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1841 foff, FALSE, td);
1842 if (error != 0)
1843 vm_object_deallocate(object);
1844 return (error);
1845 }
1846
1847 dev_t
1848 dev2udev(struct cdev *x)
1849 {
1850 if (x == NULL)
1851 return (NODEV);
1852 return (cdev2priv(x)->cdp_inode);
1853 }
1854
1855 static struct fileops devfs_ops_f = {
1856 .fo_read = devfs_read_f,
1857 .fo_write = devfs_write_f,
1858 .fo_truncate = devfs_truncate_f,
1859 .fo_ioctl = devfs_ioctl_f,
1860 .fo_poll = devfs_poll_f,
1861 .fo_kqfilter = devfs_kqfilter_f,
1862 .fo_stat = devfs_stat_f,
1863 .fo_close = devfs_close_f,
1864 .fo_chmod = vn_chmod,
1865 .fo_chown = vn_chown,
1866 .fo_sendfile = vn_sendfile,
1867 .fo_seek = vn_seek,
1868 .fo_fill_kinfo = vn_fill_kinfo,
1869 .fo_mmap = devfs_mmap_f,
1870 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
1871 };
1872
1873 static struct vop_vector devfs_vnodeops = {
1874 .vop_default = &default_vnodeops,
1875
1876 .vop_access = devfs_access,
1877 .vop_getattr = devfs_getattr,
1878 .vop_ioctl = devfs_rioctl,
1879 .vop_lookup = devfs_lookup,
1880 .vop_mknod = devfs_mknod,
1881 .vop_pathconf = devfs_pathconf,
1882 .vop_read = devfs_rread,
1883 .vop_readdir = devfs_readdir,
1884 .vop_readlink = devfs_readlink,
1885 .vop_reclaim = devfs_reclaim,
1886 .vop_remove = devfs_remove,
1887 .vop_revoke = devfs_revoke,
1888 .vop_setattr = devfs_setattr,
1889 #ifdef MAC
1890 .vop_setlabel = devfs_setlabel,
1891 #endif
1892 .vop_symlink = devfs_symlink,
1893 .vop_vptocnp = devfs_vptocnp,
1894 };
1895
1896 static struct vop_vector devfs_specops = {
1897 .vop_default = &default_vnodeops,
1898
1899 .vop_access = devfs_access,
1900 .vop_bmap = VOP_PANIC,
1901 .vop_close = devfs_close,
1902 .vop_create = VOP_PANIC,
1903 .vop_fsync = vop_stdfsync,
1904 .vop_getattr = devfs_getattr,
1905 .vop_link = VOP_PANIC,
1906 .vop_mkdir = VOP_PANIC,
1907 .vop_mknod = VOP_PANIC,
1908 .vop_open = devfs_open,
1909 .vop_pathconf = devfs_pathconf,
1910 .vop_poll = dead_poll,
1911 .vop_print = devfs_print,
1912 .vop_read = dead_read,
1913 .vop_readdir = VOP_PANIC,
1914 .vop_readlink = VOP_PANIC,
1915 .vop_reallocblks = VOP_PANIC,
1916 .vop_reclaim = devfs_reclaim_vchr,
1917 .vop_remove = devfs_remove,
1918 .vop_rename = VOP_PANIC,
1919 .vop_revoke = devfs_revoke,
1920 .vop_rmdir = VOP_PANIC,
1921 .vop_setattr = devfs_setattr,
1922 #ifdef MAC
1923 .vop_setlabel = devfs_setlabel,
1924 #endif
1925 .vop_strategy = VOP_PANIC,
1926 .vop_symlink = VOP_PANIC,
1927 .vop_vptocnp = devfs_vptocnp,
1928 .vop_write = dead_write,
1929 };
1930
1931 /*
1932 * Our calling convention to the device drivers used to be that we passed
1933 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1934 * flags instead since that's what open(), close() and ioctl() takes and
1935 * we don't really want vnode.h in device drivers.
1936 * We solved the source compatibility by redefining some vnode flags to
1937 * be the same as the fcntl ones and by sending down the bitwise OR of
1938 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody
1939 * pulls the rug out under this.
1940 */
1941 CTASSERT(O_NONBLOCK == IO_NDELAY);
1942 CTASSERT(O_FSYNC == IO_SYNC);
Cache object: 341873c5b1b159e7c2aee5cb1d677a33
|