1 /*-
2 * Copyright (c) 2000-2004
3 * Poul-Henning Kamp. All rights reserved.
4 * Copyright (c) 1989, 1992-1993, 1995
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software donated to Berkeley by
8 * Jan-Simon Pendry.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * @(#)kernfs_vnops.c 8.15 (Berkeley) 5/21/95
32 * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vnops.c 1.43
33 *
34 * $FreeBSD$
35 */
36
37 /*
38 * TODO:
39 * mkdir: want it ?
40 */
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/conf.h>
45 #include <sys/dirent.h>
46 #include <sys/fcntl.h>
47 #include <sys/file.h>
48 #include <sys/filedesc.h>
49 #include <sys/filio.h>
50 #include <sys/jail.h>
51 #include <sys/kernel.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/mount.h>
55 #include <sys/namei.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/stat.h>
59 #include <sys/sx.h>
60 #include <sys/sysctl.h>
61 #include <sys/time.h>
62 #include <sys/ttycom.h>
63 #include <sys/unistd.h>
64 #include <sys/vnode.h>
65
66 static struct vop_vector devfs_vnodeops;
67 static struct fileops devfs_ops_f;
68
69 #include <fs/devfs/devfs.h>
70 #include <fs/devfs/devfs_int.h>
71
72 #include <security/mac/mac_framework.h>
73
74 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
75
76 struct mtx devfs_de_interlock;
77 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
78 struct sx clone_drain_lock;
79 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
80 struct mtx cdevpriv_mtx;
81 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
82
83 SYSCTL_DECL(_vfs_devfs);
84
85 static int devfs_dotimes;
86 SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW,
87 &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision");
88
89 /*
90 * Update devfs node timestamp. Note that updates are unlocked and
91 * stat(2) could see partially updated times.
92 */
93 static void
devfs_timestamp(struct timespec * tsp)94 devfs_timestamp(struct timespec *tsp)
95 {
96 time_t ts;
97
98 if (devfs_dotimes) {
99 vfs_timestamp(tsp);
100 } else {
101 ts = time_second;
102 if (tsp->tv_sec != ts) {
103 tsp->tv_sec = ts;
104 tsp->tv_nsec = 0;
105 }
106 }
107 }
108
109 static int
devfs_fp_check(struct file * fp,struct cdev ** devp,struct cdevsw ** dswp,int * ref)110 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
111 int *ref)
112 {
113
114 *dswp = devvn_refthread(fp->f_vnode, devp, ref);
115 if (*devp != fp->f_data) {
116 if (*dswp != NULL)
117 dev_relthread(*devp, *ref);
118 return (ENXIO);
119 }
120 KASSERT((*devp)->si_refcount > 0,
121 ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
122 if (*dswp == NULL)
123 return (ENXIO);
124 curthread->td_fpop = fp;
125 return (0);
126 }
127
128 int
devfs_get_cdevpriv(void ** datap)129 devfs_get_cdevpriv(void **datap)
130 {
131 struct file *fp;
132 struct cdev_privdata *p;
133 int error;
134
135 fp = curthread->td_fpop;
136 if (fp == NULL)
137 return (EBADF);
138 p = fp->f_cdevpriv;
139 if (p != NULL) {
140 error = 0;
141 *datap = p->cdpd_data;
142 } else
143 error = ENOENT;
144 return (error);
145 }
146
147 int
devfs_set_cdevpriv(void * priv,cdevpriv_dtr_t priv_dtr)148 devfs_set_cdevpriv(void *priv, cdevpriv_dtr_t priv_dtr)
149 {
150 struct file *fp;
151 struct cdev_priv *cdp;
152 struct cdev_privdata *p;
153 int error;
154
155 fp = curthread->td_fpop;
156 if (fp == NULL)
157 return (ENOENT);
158 cdp = cdev2priv((struct cdev *)fp->f_data);
159 p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
160 p->cdpd_data = priv;
161 p->cdpd_dtr = priv_dtr;
162 p->cdpd_fp = fp;
163 mtx_lock(&cdevpriv_mtx);
164 if (fp->f_cdevpriv == NULL) {
165 LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
166 fp->f_cdevpriv = p;
167 mtx_unlock(&cdevpriv_mtx);
168 error = 0;
169 } else {
170 mtx_unlock(&cdevpriv_mtx);
171 free(p, M_CDEVPDATA);
172 error = EBUSY;
173 }
174 return (error);
175 }
176
177 void
devfs_destroy_cdevpriv(struct cdev_privdata * p)178 devfs_destroy_cdevpriv(struct cdev_privdata *p)
179 {
180
181 mtx_assert(&cdevpriv_mtx, MA_OWNED);
182 p->cdpd_fp->f_cdevpriv = NULL;
183 LIST_REMOVE(p, cdpd_list);
184 mtx_unlock(&cdevpriv_mtx);
185 (p->cdpd_dtr)(p->cdpd_data);
186 free(p, M_CDEVPDATA);
187 }
188
189 void
devfs_fpdrop(struct file * fp)190 devfs_fpdrop(struct file *fp)
191 {
192 struct cdev_privdata *p;
193
194 mtx_lock(&cdevpriv_mtx);
195 if ((p = fp->f_cdevpriv) == NULL) {
196 mtx_unlock(&cdevpriv_mtx);
197 return;
198 }
199 devfs_destroy_cdevpriv(p);
200 }
201
202 void
devfs_clear_cdevpriv(void)203 devfs_clear_cdevpriv(void)
204 {
205 struct file *fp;
206
207 fp = curthread->td_fpop;
208 if (fp == NULL)
209 return;
210 devfs_fpdrop(fp);
211 }
212
213 /*
214 * On success devfs_populate_vp() returns with dmp->dm_lock held.
215 */
216 static int
devfs_populate_vp(struct vnode * vp)217 devfs_populate_vp(struct vnode *vp)
218 {
219 struct devfs_dirent *de;
220 struct devfs_mount *dmp;
221 int locked;
222
223 ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
224
225 dmp = VFSTODEVFS(vp->v_mount);
226 locked = VOP_ISLOCKED(vp);
227
228 sx_xlock(&dmp->dm_lock);
229 DEVFS_DMP_HOLD(dmp);
230
231 /* Can't call devfs_populate() with the vnode lock held. */
232 VOP_UNLOCK(vp, 0);
233 devfs_populate(dmp);
234
235 sx_xunlock(&dmp->dm_lock);
236 vn_lock(vp, locked | LK_RETRY);
237 sx_xlock(&dmp->dm_lock);
238 if (DEVFS_DMP_DROP(dmp)) {
239 sx_xunlock(&dmp->dm_lock);
240 devfs_unmount_final(dmp);
241 return (EBADF);
242 }
243 if ((vp->v_iflag & VI_DOOMED) != 0) {
244 sx_xunlock(&dmp->dm_lock);
245 return (EBADF);
246 }
247 de = vp->v_data;
248 KASSERT(de != NULL,
249 ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
250 if ((de->de_flags & DE_DOOMED) != 0) {
251 sx_xunlock(&dmp->dm_lock);
252 return (EBADF);
253 }
254
255 return (0);
256 }
257
258 static int
devfs_vptocnp(struct vop_vptocnp_args * ap)259 devfs_vptocnp(struct vop_vptocnp_args *ap)
260 {
261 struct vnode *vp = ap->a_vp;
262 struct vnode **dvp = ap->a_vpp;
263 struct devfs_mount *dmp;
264 char *buf = ap->a_buf;
265 int *buflen = ap->a_buflen;
266 struct devfs_dirent *dd, *de;
267 int i, error;
268
269 dmp = VFSTODEVFS(vp->v_mount);
270
271 error = devfs_populate_vp(vp);
272 if (error != 0)
273 return (error);
274
275 i = *buflen;
276 dd = vp->v_data;
277
278 if (vp->v_type == VCHR) {
279 i -= strlen(dd->de_cdp->cdp_c.si_name);
280 if (i < 0) {
281 error = ENOMEM;
282 goto finished;
283 }
284 bcopy(dd->de_cdp->cdp_c.si_name, buf + i,
285 strlen(dd->de_cdp->cdp_c.si_name));
286 de = dd->de_dir;
287 } else if (vp->v_type == VDIR) {
288 if (dd == dmp->dm_rootdir) {
289 *dvp = vp;
290 vref(*dvp);
291 goto finished;
292 }
293 i -= dd->de_dirent->d_namlen;
294 if (i < 0) {
295 error = ENOMEM;
296 goto finished;
297 }
298 bcopy(dd->de_dirent->d_name, buf + i,
299 dd->de_dirent->d_namlen);
300 de = dd;
301 } else {
302 error = ENOENT;
303 goto finished;
304 }
305 *buflen = i;
306 de = devfs_parent_dirent(de);
307 if (de == NULL) {
308 error = ENOENT;
309 goto finished;
310 }
311 mtx_lock(&devfs_de_interlock);
312 *dvp = de->de_vnode;
313 if (*dvp != NULL) {
314 VI_LOCK(*dvp);
315 mtx_unlock(&devfs_de_interlock);
316 vholdl(*dvp);
317 VI_UNLOCK(*dvp);
318 vref(*dvp);
319 vdrop(*dvp);
320 } else {
321 mtx_unlock(&devfs_de_interlock);
322 error = ENOENT;
323 }
324 finished:
325 sx_xunlock(&dmp->dm_lock);
326 return (error);
327 }
328
329 /*
330 * Construct the fully qualified path name relative to the mountpoint.
331 * If a NULL cnp is provided, no '/' is appended to the resulting path.
332 */
333 char *
devfs_fqpn(char * buf,struct devfs_mount * dmp,struct devfs_dirent * dd,struct componentname * cnp)334 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
335 struct componentname *cnp)
336 {
337 int i;
338 struct devfs_dirent *de;
339
340 sx_assert(&dmp->dm_lock, SA_LOCKED);
341
342 i = SPECNAMELEN;
343 buf[i] = '\0';
344 if (cnp != NULL)
345 i -= cnp->cn_namelen;
346 if (i < 0)
347 return (NULL);
348 if (cnp != NULL)
349 bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
350 de = dd;
351 while (de != dmp->dm_rootdir) {
352 if (cnp != NULL || i < SPECNAMELEN) {
353 i--;
354 if (i < 0)
355 return (NULL);
356 buf[i] = '/';
357 }
358 i -= de->de_dirent->d_namlen;
359 if (i < 0)
360 return (NULL);
361 bcopy(de->de_dirent->d_name, buf + i,
362 de->de_dirent->d_namlen);
363 de = devfs_parent_dirent(de);
364 if (de == NULL)
365 return (NULL);
366 }
367 return (buf + i);
368 }
369
370 static int
devfs_allocv_drop_refs(int drop_dm_lock,struct devfs_mount * dmp,struct devfs_dirent * de)371 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
372 struct devfs_dirent *de)
373 {
374 int not_found;
375
376 not_found = 0;
377 if (de->de_flags & DE_DOOMED)
378 not_found = 1;
379 if (DEVFS_DE_DROP(de)) {
380 KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
381 devfs_dirent_free(de);
382 }
383 if (DEVFS_DMP_DROP(dmp)) {
384 KASSERT(not_found == 1,
385 ("DEVFS mount struct freed before dirent"));
386 not_found = 2;
387 sx_xunlock(&dmp->dm_lock);
388 devfs_unmount_final(dmp);
389 }
390 if (not_found == 1 || (drop_dm_lock && not_found != 2))
391 sx_unlock(&dmp->dm_lock);
392 return (not_found);
393 }
394
395 static void
devfs_insmntque_dtr(struct vnode * vp,void * arg)396 devfs_insmntque_dtr(struct vnode *vp, void *arg)
397 {
398 struct devfs_dirent *de;
399
400 de = (struct devfs_dirent *)arg;
401 mtx_lock(&devfs_de_interlock);
402 vp->v_data = NULL;
403 de->de_vnode = NULL;
404 mtx_unlock(&devfs_de_interlock);
405 vgone(vp);
406 vput(vp);
407 }
408
409 /*
410 * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
411 * it on return.
412 */
413 int
devfs_allocv(struct devfs_dirent * de,struct mount * mp,int lockmode,struct vnode ** vpp)414 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
415 struct vnode **vpp)
416 {
417 int error;
418 struct vnode *vp;
419 struct cdev *dev;
420 struct devfs_mount *dmp;
421 struct cdevsw *dsw;
422
423 dmp = VFSTODEVFS(mp);
424 if (de->de_flags & DE_DOOMED) {
425 sx_xunlock(&dmp->dm_lock);
426 return (ENOENT);
427 }
428 loop:
429 DEVFS_DE_HOLD(de);
430 DEVFS_DMP_HOLD(dmp);
431 mtx_lock(&devfs_de_interlock);
432 vp = de->de_vnode;
433 if (vp != NULL) {
434 VI_LOCK(vp);
435 mtx_unlock(&devfs_de_interlock);
436 sx_xunlock(&dmp->dm_lock);
437 vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread);
438 sx_xlock(&dmp->dm_lock);
439 if (devfs_allocv_drop_refs(0, dmp, de)) {
440 vput(vp);
441 return (ENOENT);
442 }
443 else if ((vp->v_iflag & VI_DOOMED) != 0) {
444 mtx_lock(&devfs_de_interlock);
445 if (de->de_vnode == vp) {
446 de->de_vnode = NULL;
447 vp->v_data = NULL;
448 }
449 mtx_unlock(&devfs_de_interlock);
450 vput(vp);
451 goto loop;
452 }
453 sx_xunlock(&dmp->dm_lock);
454 *vpp = vp;
455 return (0);
456 }
457 mtx_unlock(&devfs_de_interlock);
458 if (de->de_dirent->d_type == DT_CHR) {
459 if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
460 devfs_allocv_drop_refs(1, dmp, de);
461 return (ENOENT);
462 }
463 dev = &de->de_cdp->cdp_c;
464 } else {
465 dev = NULL;
466 }
467 error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
468 if (error != 0) {
469 devfs_allocv_drop_refs(1, dmp, de);
470 printf("devfs_allocv: failed to allocate new vnode\n");
471 return (error);
472 }
473
474 if (de->de_dirent->d_type == DT_CHR) {
475 vp->v_type = VCHR;
476 VI_LOCK(vp);
477 dev_lock();
478 dev_refl(dev);
479 /* XXX: v_rdev should be protect by vnode lock */
480 vp->v_rdev = dev;
481 KASSERT(vp->v_usecount == 1,
482 ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount));
483 dev->si_usecount += vp->v_usecount;
484 /* Special casing of ttys for deadfs. Probably redundant. */
485 dsw = dev->si_devsw;
486 if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
487 vp->v_vflag |= VV_ISTTY;
488 dev_unlock();
489 VI_UNLOCK(vp);
490 if ((dev->si_flags & SI_ETERNAL) != 0)
491 vp->v_vflag |= VV_ETERNALDEV;
492 vp->v_op = &devfs_specops;
493 } else if (de->de_dirent->d_type == DT_DIR) {
494 vp->v_type = VDIR;
495 } else if (de->de_dirent->d_type == DT_LNK) {
496 vp->v_type = VLNK;
497 } else {
498 vp->v_type = VBAD;
499 }
500 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
501 VN_LOCK_ASHARE(vp);
502 mtx_lock(&devfs_de_interlock);
503 vp->v_data = de;
504 de->de_vnode = vp;
505 mtx_unlock(&devfs_de_interlock);
506 error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
507 if (error != 0) {
508 (void) devfs_allocv_drop_refs(1, dmp, de);
509 return (error);
510 }
511 if (devfs_allocv_drop_refs(0, dmp, de)) {
512 vput(vp);
513 return (ENOENT);
514 }
515 #ifdef MAC
516 mac_devfs_vnode_associate(mp, de, vp);
517 #endif
518 sx_xunlock(&dmp->dm_lock);
519 *vpp = vp;
520 return (0);
521 }
522
523 static int
devfs_access(struct vop_access_args * ap)524 devfs_access(struct vop_access_args *ap)
525 {
526 struct vnode *vp = ap->a_vp;
527 struct devfs_dirent *de;
528 int error;
529
530 de = vp->v_data;
531 if (vp->v_type == VDIR)
532 de = de->de_dir;
533
534 error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
535 ap->a_accmode, ap->a_cred, NULL);
536 if (error == 0)
537 return (0);
538 if (error != EACCES)
539 return (error);
540 /* We do, however, allow access to the controlling terminal */
541 if (!(ap->a_td->td_proc->p_flag & P_CONTROLT))
542 return (error);
543 if (ap->a_td->td_proc->p_session->s_ttydp == de->de_cdp)
544 return (0);
545 return (error);
546 }
547
548 /* ARGSUSED */
549 static int
devfs_close(struct vop_close_args * ap)550 devfs_close(struct vop_close_args *ap)
551 {
552 struct vnode *vp = ap->a_vp, *oldvp;
553 struct thread *td = ap->a_td;
554 struct cdev *dev = vp->v_rdev;
555 struct cdevsw *dsw;
556 int vp_locked, error, ref;
557
558 /*
559 * XXX: Don't call d_close() if we were called because of
560 * XXX: insmntque1() failure.
561 */
562 if (vp->v_data == NULL)
563 return (0);
564
565 /*
566 * Hack: a tty device that is a controlling terminal
567 * has a reference from the session structure.
568 * We cannot easily tell that a character device is
569 * a controlling terminal, unless it is the closing
570 * process' controlling terminal. In that case,
571 * if the reference count is 2 (this last descriptor
572 * plus the session), release the reference from the session.
573 */
574 oldvp = NULL;
575 sx_xlock(&proctree_lock);
576 if (td && vp == td->td_proc->p_session->s_ttyvp) {
577 SESS_LOCK(td->td_proc->p_session);
578 VI_LOCK(vp);
579 if (count_dev(dev) == 2 && (vp->v_iflag & VI_DOOMED) == 0) {
580 td->td_proc->p_session->s_ttyvp = NULL;
581 td->td_proc->p_session->s_ttydp = NULL;
582 oldvp = vp;
583 }
584 VI_UNLOCK(vp);
585 SESS_UNLOCK(td->td_proc->p_session);
586 }
587 sx_xunlock(&proctree_lock);
588 if (oldvp != NULL)
589 vrele(oldvp);
590 /*
591 * We do not want to really close the device if it
592 * is still in use unless we are trying to close it
593 * forcibly. Since every use (buffer, vnode, swap, cmap)
594 * holds a reference to the vnode, and because we mark
595 * any other vnodes that alias this device, when the
596 * sum of the reference counts on all the aliased
597 * vnodes descends to one, we are on last close.
598 */
599 dsw = dev_refthread(dev, &ref);
600 if (dsw == NULL)
601 return (ENXIO);
602 VI_LOCK(vp);
603 if (vp->v_iflag & VI_DOOMED) {
604 /* Forced close. */
605 } else if (dsw->d_flags & D_TRACKCLOSE) {
606 /* Keep device updated on status. */
607 } else if (count_dev(dev) > 1) {
608 VI_UNLOCK(vp);
609 dev_relthread(dev, ref);
610 return (0);
611 }
612 vholdl(vp);
613 VI_UNLOCK(vp);
614 vp_locked = VOP_ISLOCKED(vp);
615 VOP_UNLOCK(vp, 0);
616 KASSERT(dev->si_refcount > 0,
617 ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
618 error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td);
619 dev_relthread(dev, ref);
620 vn_lock(vp, vp_locked | LK_RETRY);
621 vdrop(vp);
622 return (error);
623 }
624
625 static int
devfs_close_f(struct file * fp,struct thread * td)626 devfs_close_f(struct file *fp, struct thread *td)
627 {
628 int error;
629 struct file *fpop;
630
631 /*
632 * NB: td may be NULL if this descriptor is closed due to
633 * garbage collection from a closed UNIX domain socket.
634 */
635 fpop = curthread->td_fpop;
636 curthread->td_fpop = fp;
637 error = vnops.fo_close(fp, td);
638 curthread->td_fpop = fpop;
639
640 /*
641 * The f_cdevpriv cannot be assigned non-NULL value while we
642 * are destroying the file.
643 */
644 if (fp->f_cdevpriv != NULL)
645 devfs_fpdrop(fp);
646 return (error);
647 }
648
649 static int
devfs_fsync(struct vop_fsync_args * ap)650 devfs_fsync(struct vop_fsync_args *ap)
651 {
652 int error;
653 struct bufobj *bo;
654 struct devfs_dirent *de;
655
656 if (!vn_isdisk(ap->a_vp, &error)) {
657 bo = &ap->a_vp->v_bufobj;
658 de = ap->a_vp->v_data;
659 if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) {
660 printf("Device %s went missing before all of the data "
661 "could be written to it; expect data loss.\n",
662 de->de_dirent->d_name);
663
664 error = vop_stdfsync(ap);
665 if (bo->bo_dirty.bv_cnt != 0 || error != 0)
666 panic("devfs_fsync: vop_stdfsync failed.");
667 }
668
669 return (0);
670 }
671
672 return (vop_stdfsync(ap));
673 }
674
675 static int
devfs_getattr(struct vop_getattr_args * ap)676 devfs_getattr(struct vop_getattr_args *ap)
677 {
678 struct vnode *vp = ap->a_vp;
679 struct vattr *vap = ap->a_vap;
680 int error;
681 struct devfs_dirent *de;
682 struct devfs_mount *dmp;
683 struct cdev *dev;
684
685 error = devfs_populate_vp(vp);
686 if (error != 0)
687 return (error);
688
689 dmp = VFSTODEVFS(vp->v_mount);
690 sx_xunlock(&dmp->dm_lock);
691
692 de = vp->v_data;
693 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
694 if (vp->v_type == VDIR) {
695 de = de->de_dir;
696 KASSERT(de != NULL,
697 ("Null dir dirent in devfs_getattr vp=%p", vp));
698 }
699 vap->va_uid = de->de_uid;
700 vap->va_gid = de->de_gid;
701 vap->va_mode = de->de_mode;
702 if (vp->v_type == VLNK)
703 vap->va_size = strlen(de->de_symlink);
704 else if (vp->v_type == VDIR)
705 vap->va_size = vap->va_bytes = DEV_BSIZE;
706 else
707 vap->va_size = 0;
708 if (vp->v_type != VDIR)
709 vap->va_bytes = 0;
710 vap->va_blocksize = DEV_BSIZE;
711 vap->va_type = vp->v_type;
712
713 #define fix(aa) \
714 do { \
715 if ((aa).tv_sec <= 3600) { \
716 (aa).tv_sec = boottime.tv_sec; \
717 (aa).tv_nsec = boottime.tv_usec * 1000; \
718 } \
719 } while (0)
720
721 if (vp->v_type != VCHR) {
722 fix(de->de_atime);
723 vap->va_atime = de->de_atime;
724 fix(de->de_mtime);
725 vap->va_mtime = de->de_mtime;
726 fix(de->de_ctime);
727 vap->va_ctime = de->de_ctime;
728 } else {
729 dev = vp->v_rdev;
730 fix(dev->si_atime);
731 vap->va_atime = dev->si_atime;
732 fix(dev->si_mtime);
733 vap->va_mtime = dev->si_mtime;
734 fix(dev->si_ctime);
735 vap->va_ctime = dev->si_ctime;
736
737 vap->va_rdev = cdev2priv(dev)->cdp_inode;
738 }
739 vap->va_gen = 0;
740 vap->va_flags = 0;
741 vap->va_filerev = 0;
742 vap->va_nlink = de->de_links;
743 vap->va_fileid = de->de_inode;
744
745 return (error);
746 }
747
748 /* ARGSUSED */
749 static int
devfs_ioctl_f(struct file * fp,u_long com,void * data,struct ucred * cred,struct thread * td)750 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
751 {
752 struct cdev *dev;
753 struct cdevsw *dsw;
754 struct vnode *vp;
755 struct vnode *vpold;
756 int error, i, ref;
757 const char *p;
758 struct fiodgname_arg *fgn;
759 struct file *fpop;
760
761 fpop = td->td_fpop;
762 error = devfs_fp_check(fp, &dev, &dsw, &ref);
763 if (error != 0) {
764 error = vnops.fo_ioctl(fp, com, data, cred, td);
765 return (error);
766 }
767
768 if (com == FIODTYPE) {
769 *(int *)data = dsw->d_flags & D_TYPEMASK;
770 td->td_fpop = fpop;
771 dev_relthread(dev, ref);
772 return (0);
773 } else if (com == FIODGNAME) {
774 fgn = data;
775 p = devtoname(dev);
776 i = strlen(p) + 1;
777 if (i > fgn->len)
778 error = EINVAL;
779 else
780 error = copyout(p, fgn->buf, i);
781 td->td_fpop = fpop;
782 dev_relthread(dev, ref);
783 return (error);
784 }
785 error = dsw->d_ioctl(dev, com, data, fp->f_flag, td);
786 td->td_fpop = NULL;
787 dev_relthread(dev, ref);
788 if (error == ENOIOCTL)
789 error = ENOTTY;
790 if (error == 0 && com == TIOCSCTTY) {
791 vp = fp->f_vnode;
792
793 /* Do nothing if reassigning same control tty */
794 sx_slock(&proctree_lock);
795 if (td->td_proc->p_session->s_ttyvp == vp) {
796 sx_sunlock(&proctree_lock);
797 return (0);
798 }
799
800 vpold = td->td_proc->p_session->s_ttyvp;
801 VREF(vp);
802 SESS_LOCK(td->td_proc->p_session);
803 td->td_proc->p_session->s_ttyvp = vp;
804 td->td_proc->p_session->s_ttydp = cdev2priv(dev);
805 SESS_UNLOCK(td->td_proc->p_session);
806
807 sx_sunlock(&proctree_lock);
808
809 /* Get rid of reference to old control tty */
810 if (vpold)
811 vrele(vpold);
812 }
813 return (error);
814 }
815
816 /* ARGSUSED */
817 static int
devfs_kqfilter_f(struct file * fp,struct knote * kn)818 devfs_kqfilter_f(struct file *fp, struct knote *kn)
819 {
820 struct cdev *dev;
821 struct cdevsw *dsw;
822 int error, ref;
823 struct file *fpop;
824 struct thread *td;
825
826 td = curthread;
827 fpop = td->td_fpop;
828 error = devfs_fp_check(fp, &dev, &dsw, &ref);
829 if (error)
830 return (error);
831 error = dsw->d_kqfilter(dev, kn);
832 td->td_fpop = fpop;
833 dev_relthread(dev, ref);
834 return (error);
835 }
836
837 static inline int
devfs_prison_check(struct devfs_dirent * de,struct thread * td)838 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
839 {
840 struct cdev_priv *cdp;
841 struct ucred *dcr;
842 int error;
843
844 cdp = de->de_cdp;
845 if (cdp == NULL)
846 return (0);
847 dcr = cdp->cdp_c.si_cred;
848 if (dcr == NULL)
849 return (0);
850
851 error = prison_check(td->td_ucred, dcr);
852 if (error == 0)
853 return (0);
854 /* We do, however, allow access to the controlling terminal */
855 if (!(td->td_proc->p_flag & P_CONTROLT))
856 return (error);
857 if (td->td_proc->p_session->s_ttydp == cdp)
858 return (0);
859 return (error);
860 }
861
862 static int
devfs_lookupx(struct vop_lookup_args * ap,int * dm_unlock)863 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
864 {
865 struct componentname *cnp;
866 struct vnode *dvp, **vpp;
867 struct thread *td;
868 struct devfs_dirent *de, *dd;
869 struct devfs_dirent **dde;
870 struct devfs_mount *dmp;
871 struct cdev *cdev;
872 int error, flags, nameiop, dvplocked;
873 char specname[SPECNAMELEN + 1], *pname;
874
875 cnp = ap->a_cnp;
876 vpp = ap->a_vpp;
877 dvp = ap->a_dvp;
878 pname = cnp->cn_nameptr;
879 td = cnp->cn_thread;
880 flags = cnp->cn_flags;
881 nameiop = cnp->cn_nameiop;
882 dmp = VFSTODEVFS(dvp->v_mount);
883 dd = dvp->v_data;
884 *vpp = NULLVP;
885
886 if ((flags & ISLASTCN) && nameiop == RENAME)
887 return (EOPNOTSUPP);
888
889 if (dvp->v_type != VDIR)
890 return (ENOTDIR);
891
892 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
893 return (EIO);
894
895 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
896 if (error)
897 return (error);
898
899 if (cnp->cn_namelen == 1 && *pname == '.') {
900 if ((flags & ISLASTCN) && nameiop != LOOKUP)
901 return (EINVAL);
902 *vpp = dvp;
903 VREF(dvp);
904 return (0);
905 }
906
907 if (flags & ISDOTDOT) {
908 if ((flags & ISLASTCN) && nameiop != LOOKUP)
909 return (EINVAL);
910 de = devfs_parent_dirent(dd);
911 if (de == NULL)
912 return (ENOENT);
913 dvplocked = VOP_ISLOCKED(dvp);
914 VOP_UNLOCK(dvp, 0);
915 error = devfs_allocv(de, dvp->v_mount,
916 cnp->cn_lkflags & LK_TYPE_MASK, vpp);
917 *dm_unlock = 0;
918 vn_lock(dvp, dvplocked | LK_RETRY);
919 return (error);
920 }
921
922 dd = dvp->v_data;
923 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
924 while (de == NULL) { /* While(...) so we can use break */
925
926 if (nameiop == DELETE)
927 return (ENOENT);
928
929 /*
930 * OK, we didn't have an entry for the name we were asked for
931 * so we try to see if anybody can create it on demand.
932 */
933 pname = devfs_fqpn(specname, dmp, dd, cnp);
934 if (pname == NULL)
935 break;
936
937 cdev = NULL;
938 DEVFS_DMP_HOLD(dmp);
939 sx_xunlock(&dmp->dm_lock);
940 sx_slock(&clone_drain_lock);
941 EVENTHANDLER_INVOKE(dev_clone,
942 td->td_ucred, pname, strlen(pname), &cdev);
943 sx_sunlock(&clone_drain_lock);
944
945 if (cdev == NULL)
946 sx_xlock(&dmp->dm_lock);
947 else if (devfs_populate_vp(dvp) != 0) {
948 *dm_unlock = 0;
949 sx_xlock(&dmp->dm_lock);
950 if (DEVFS_DMP_DROP(dmp)) {
951 sx_xunlock(&dmp->dm_lock);
952 devfs_unmount_final(dmp);
953 } else
954 sx_xunlock(&dmp->dm_lock);
955 dev_rel(cdev);
956 return (ENOENT);
957 }
958 if (DEVFS_DMP_DROP(dmp)) {
959 *dm_unlock = 0;
960 sx_xunlock(&dmp->dm_lock);
961 devfs_unmount_final(dmp);
962 if (cdev != NULL)
963 dev_rel(cdev);
964 return (ENOENT);
965 }
966
967 if (cdev == NULL)
968 break;
969
970 dev_lock();
971 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
972 if (dde != NULL && *dde != NULL)
973 de = *dde;
974 dev_unlock();
975 dev_rel(cdev);
976 break;
977 }
978
979 if (de == NULL || de->de_flags & DE_WHITEOUT) {
980 if ((nameiop == CREATE || nameiop == RENAME) &&
981 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
982 cnp->cn_flags |= SAVENAME;
983 return (EJUSTRETURN);
984 }
985 return (ENOENT);
986 }
987
988 if (devfs_prison_check(de, td))
989 return (ENOENT);
990
991 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
992 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
993 if (error)
994 return (error);
995 if (*vpp == dvp) {
996 VREF(dvp);
997 *vpp = dvp;
998 return (0);
999 }
1000 }
1001 error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK,
1002 vpp);
1003 *dm_unlock = 0;
1004 return (error);
1005 }
1006
1007 static int
devfs_lookup(struct vop_lookup_args * ap)1008 devfs_lookup(struct vop_lookup_args *ap)
1009 {
1010 int j;
1011 struct devfs_mount *dmp;
1012 int dm_unlock;
1013
1014 if (devfs_populate_vp(ap->a_dvp) != 0)
1015 return (ENOTDIR);
1016
1017 dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1018 dm_unlock = 1;
1019 j = devfs_lookupx(ap, &dm_unlock);
1020 if (dm_unlock == 1)
1021 sx_xunlock(&dmp->dm_lock);
1022 return (j);
1023 }
1024
1025 static int
devfs_mknod(struct vop_mknod_args * ap)1026 devfs_mknod(struct vop_mknod_args *ap)
1027 {
1028 struct componentname *cnp;
1029 struct vnode *dvp, **vpp;
1030 struct devfs_dirent *dd, *de;
1031 struct devfs_mount *dmp;
1032 int error;
1033
1034 /*
1035 * The only type of node we should be creating here is a
1036 * character device, for anything else return EOPNOTSUPP.
1037 */
1038 if (ap->a_vap->va_type != VCHR)
1039 return (EOPNOTSUPP);
1040 dvp = ap->a_dvp;
1041 dmp = VFSTODEVFS(dvp->v_mount);
1042
1043 cnp = ap->a_cnp;
1044 vpp = ap->a_vpp;
1045 dd = dvp->v_data;
1046
1047 error = ENOENT;
1048 sx_xlock(&dmp->dm_lock);
1049 TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1050 if (cnp->cn_namelen != de->de_dirent->d_namlen)
1051 continue;
1052 if (de->de_dirent->d_type == DT_CHR &&
1053 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
1054 continue;
1055 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1056 de->de_dirent->d_namlen) != 0)
1057 continue;
1058 if (de->de_flags & DE_WHITEOUT)
1059 break;
1060 goto notfound;
1061 }
1062 if (de == NULL)
1063 goto notfound;
1064 de->de_flags &= ~DE_WHITEOUT;
1065 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1066 return (error);
1067 notfound:
1068 sx_xunlock(&dmp->dm_lock);
1069 return (error);
1070 }
1071
1072 /* ARGSUSED */
1073 static int
devfs_open(struct vop_open_args * ap)1074 devfs_open(struct vop_open_args *ap)
1075 {
1076 struct thread *td = ap->a_td;
1077 struct vnode *vp = ap->a_vp;
1078 struct cdev *dev = vp->v_rdev;
1079 struct file *fp = ap->a_fp;
1080 int error, ref, vlocked;
1081 struct cdevsw *dsw;
1082 struct file *fpop;
1083 struct mtx *mtxp;
1084
1085 if (vp->v_type == VBLK)
1086 return (ENXIO);
1087
1088 if (dev == NULL)
1089 return (ENXIO);
1090
1091 /* Make this field valid before any I/O in d_open. */
1092 if (dev->si_iosize_max == 0)
1093 dev->si_iosize_max = DFLTPHYS;
1094
1095 dsw = dev_refthread(dev, &ref);
1096 if (dsw == NULL)
1097 return (ENXIO);
1098 if (fp == NULL && dsw->d_fdopen != NULL) {
1099 dev_relthread(dev, ref);
1100 return (ENXIO);
1101 }
1102
1103 vlocked = VOP_ISLOCKED(vp);
1104 VOP_UNLOCK(vp, 0);
1105
1106 fpop = td->td_fpop;
1107 td->td_fpop = fp;
1108 if (fp != NULL) {
1109 fp->f_data = dev;
1110 fp->f_vnode = vp;
1111 }
1112 if (dsw->d_fdopen != NULL)
1113 error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1114 else
1115 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1116 /* cleanup any cdevpriv upon error */
1117 if (error != 0)
1118 devfs_clear_cdevpriv();
1119 td->td_fpop = fpop;
1120
1121 vn_lock(vp, vlocked | LK_RETRY);
1122 dev_relthread(dev, ref);
1123 if (error != 0) {
1124 if (error == ERESTART)
1125 error = EINTR;
1126 return (error);
1127 }
1128
1129 #if 0 /* /dev/console */
1130 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1131 #else
1132 if (fp == NULL)
1133 return (error);
1134 #endif
1135 if (fp->f_ops == &badfileops)
1136 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1137 mtxp = mtx_pool_find(mtxpool_sleep, fp);
1138
1139 /*
1140 * Hint to the dofilewrite() to not force the buffer draining
1141 * on the writer to the file. Most likely, the write would
1142 * not need normal buffers.
1143 */
1144 mtx_lock(mtxp);
1145 fp->f_vnread_flags |= FDEVFS_VNODE;
1146 mtx_unlock(mtxp);
1147 return (error);
1148 }
1149
1150 static int
devfs_pathconf(struct vop_pathconf_args * ap)1151 devfs_pathconf(struct vop_pathconf_args *ap)
1152 {
1153
1154 switch (ap->a_name) {
1155 case _PC_MAC_PRESENT:
1156 #ifdef MAC
1157 /*
1158 * If MAC is enabled, devfs automatically supports
1159 * trivial non-persistant label storage.
1160 */
1161 *ap->a_retval = 1;
1162 #else
1163 *ap->a_retval = 0;
1164 #endif
1165 return (0);
1166 default:
1167 return (vop_stdpathconf(ap));
1168 }
1169 /* NOTREACHED */
1170 }
1171
1172 /* ARGSUSED */
1173 static int
devfs_poll_f(struct file * fp,int events,struct ucred * cred,struct thread * td)1174 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1175 {
1176 struct cdev *dev;
1177 struct cdevsw *dsw;
1178 int error, ref;
1179 struct file *fpop;
1180
1181 fpop = td->td_fpop;
1182 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1183 if (error != 0) {
1184 error = vnops.fo_poll(fp, events, cred, td);
1185 return (error);
1186 }
1187 error = dsw->d_poll(dev, events, td);
1188 td->td_fpop = fpop;
1189 dev_relthread(dev, ref);
1190 return(error);
1191 }
1192
1193 /*
1194 * Print out the contents of a special device vnode.
1195 */
1196 static int
devfs_print(struct vop_print_args * ap)1197 devfs_print(struct vop_print_args *ap)
1198 {
1199
1200 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1201 return (0);
1202 }
1203
1204 static int
devfs_read_f(struct file * fp,struct uio * uio,struct ucred * cred,int flags,struct thread * td)1205 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1206 int flags, struct thread *td)
1207 {
1208 struct cdev *dev;
1209 int ioflag, error, ref;
1210 ssize_t resid;
1211 struct cdevsw *dsw;
1212 struct file *fpop;
1213
1214 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1215 return (EINVAL);
1216 fpop = td->td_fpop;
1217 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1218 if (error != 0) {
1219 error = vnops.fo_read(fp, uio, cred, flags, td);
1220 return (error);
1221 }
1222 resid = uio->uio_resid;
1223 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1224 if (ioflag & O_DIRECT)
1225 ioflag |= IO_DIRECT;
1226
1227 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1228 error = dsw->d_read(dev, uio, ioflag);
1229 if (uio->uio_resid != resid || (error == 0 && resid != 0))
1230 devfs_timestamp(&dev->si_atime);
1231 td->td_fpop = fpop;
1232 dev_relthread(dev, ref);
1233
1234 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1235 return (error);
1236 }
1237
1238 static int
devfs_readdir(struct vop_readdir_args * ap)1239 devfs_readdir(struct vop_readdir_args *ap)
1240 {
1241 int error;
1242 struct uio *uio;
1243 struct dirent *dp;
1244 struct devfs_dirent *dd;
1245 struct devfs_dirent *de;
1246 struct devfs_mount *dmp;
1247 off_t off;
1248 int *tmp_ncookies = NULL;
1249
1250 if (ap->a_vp->v_type != VDIR)
1251 return (ENOTDIR);
1252
1253 uio = ap->a_uio;
1254 if (uio->uio_offset < 0)
1255 return (EINVAL);
1256
1257 /*
1258 * XXX: This is a temporary hack to get around this filesystem not
1259 * supporting cookies. We store the location of the ncookies pointer
1260 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1261 * and set the number of cookies to 0. We then set the pointer to
1262 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1263 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1264 * pointer to its original location before returning to the caller.
1265 */
1266 if (ap->a_ncookies != NULL) {
1267 tmp_ncookies = ap->a_ncookies;
1268 *ap->a_ncookies = 0;
1269 ap->a_ncookies = NULL;
1270 }
1271
1272 dmp = VFSTODEVFS(ap->a_vp->v_mount);
1273 if (devfs_populate_vp(ap->a_vp) != 0) {
1274 if (tmp_ncookies != NULL)
1275 ap->a_ncookies = tmp_ncookies;
1276 return (EIO);
1277 }
1278 error = 0;
1279 de = ap->a_vp->v_data;
1280 off = 0;
1281 TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1282 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1283 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1284 continue;
1285 if (devfs_prison_check(dd, uio->uio_td))
1286 continue;
1287 if (dd->de_dirent->d_type == DT_DIR)
1288 de = dd->de_dir;
1289 else
1290 de = dd;
1291 dp = dd->de_dirent;
1292 if (dp->d_reclen > uio->uio_resid)
1293 break;
1294 dp->d_fileno = de->de_inode;
1295 if (off >= uio->uio_offset) {
1296 error = vfs_read_dirent(ap, dp, off);
1297 if (error)
1298 break;
1299 }
1300 off += dp->d_reclen;
1301 }
1302 sx_xunlock(&dmp->dm_lock);
1303 uio->uio_offset = off;
1304
1305 /*
1306 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1307 * place.
1308 */
1309 if (tmp_ncookies != NULL)
1310 ap->a_ncookies = tmp_ncookies;
1311
1312 return (error);
1313 }
1314
1315 static int
devfs_readlink(struct vop_readlink_args * ap)1316 devfs_readlink(struct vop_readlink_args *ap)
1317 {
1318 struct devfs_dirent *de;
1319 struct cdev_priv *cdp;
1320
1321 de = ap->a_vp->v_data;
1322 cdp = de->de_cdp;
1323
1324 if (cdp != NULL && (cdp->cdp_c.si_flags & SI_ALIAS) != 0) {
1325 struct devfs_mount *dmp;
1326 struct prison *pr;
1327 char *mp;
1328 int mp_len;
1329 int pr_path_len;
1330 int err;
1331
1332 /*
1333 * For device aliases, construct an absolute symlink (to
1334 * shorten its length and avoid the ugliness of a relative
1335 * link) by prepending the fully qualified path to the root
1336 * of this devfs. For a non-jailed process, the devfs root
1337 * is our mount point. For a jailed process, we must remove
1338 * any jail prefix in our mount point so that our response
1339 * matches the user process's world view.
1340 */
1341 dmp = VFSTODEVFS(ap->a_vp->v_mount);
1342 mp = dmp->dm_mount->mnt_stat.f_mntonname;
1343 mp_len = strlen(mp);
1344
1345 pr = ap->a_cred->cr_prison;
1346 pr_path_len = strlen(pr->pr_path);
1347
1348 if (strncmp(pr->pr_path, mp, pr_path_len) == 0
1349 && mp[pr_path_len] == '/') {
1350 mp += pr_path_len;
1351 mp_len -= pr_path_len;
1352 }
1353
1354 err = uiomove(mp, mp_len, ap->a_uio);
1355 if (err != 0)
1356 return (err);
1357
1358 /*
1359 * Devfs cannot be the root file system, so its
1360 * mount point must always be terminated by a '/'.
1361 */
1362 err = uiomove("/", 1, ap->a_uio);
1363 if (err != 0)
1364 return (err);
1365 }
1366 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1367 }
1368
1369 static int
devfs_reclaim(struct vop_reclaim_args * ap)1370 devfs_reclaim(struct vop_reclaim_args *ap)
1371 {
1372 struct vnode *vp = ap->a_vp;
1373 struct devfs_dirent *de;
1374 struct cdev *dev;
1375
1376 mtx_lock(&devfs_de_interlock);
1377 de = vp->v_data;
1378 if (de != NULL) {
1379 de->de_vnode = NULL;
1380 vp->v_data = NULL;
1381 }
1382 mtx_unlock(&devfs_de_interlock);
1383
1384 vnode_destroy_vobject(vp);
1385
1386 VI_LOCK(vp);
1387 dev_lock();
1388 dev = vp->v_rdev;
1389 vp->v_rdev = NULL;
1390
1391 if (dev == NULL) {
1392 dev_unlock();
1393 VI_UNLOCK(vp);
1394 return (0);
1395 }
1396
1397 dev->si_usecount -= vp->v_usecount;
1398 dev_unlock();
1399 VI_UNLOCK(vp);
1400 dev_rel(dev);
1401 return (0);
1402 }
1403
1404 static int
devfs_remove(struct vop_remove_args * ap)1405 devfs_remove(struct vop_remove_args *ap)
1406 {
1407 struct vnode *dvp = ap->a_dvp;
1408 struct vnode *vp = ap->a_vp;
1409 struct devfs_dirent *dd;
1410 struct devfs_dirent *de, *de_covered;
1411 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1412
1413 ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1414 ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1415
1416 sx_xlock(&dmp->dm_lock);
1417 dd = ap->a_dvp->v_data;
1418 de = vp->v_data;
1419 if (de->de_cdp == NULL) {
1420 TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1421 if (de->de_dirent->d_type == DT_LNK) {
1422 de_covered = devfs_find(dd, de->de_dirent->d_name,
1423 de->de_dirent->d_namlen, 0);
1424 if (de_covered != NULL)
1425 de_covered->de_flags &= ~DE_COVERED;
1426 }
1427 /* We need to unlock dvp because devfs_delete() may lock it. */
1428 VOP_UNLOCK(vp, 0);
1429 if (dvp != vp)
1430 VOP_UNLOCK(dvp, 0);
1431 devfs_delete(dmp, de, 0);
1432 sx_xunlock(&dmp->dm_lock);
1433 if (dvp != vp)
1434 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1435 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1436 } else {
1437 de->de_flags |= DE_WHITEOUT;
1438 sx_xunlock(&dmp->dm_lock);
1439 }
1440 return (0);
1441 }
1442
1443 /*
1444 * Revoke is called on a tty when a terminal session ends. The vnode
1445 * is orphaned by setting v_op to deadfs so we need to let go of it
1446 * as well so that we create a new one next time around.
1447 *
1448 */
1449 static int
devfs_revoke(struct vop_revoke_args * ap)1450 devfs_revoke(struct vop_revoke_args *ap)
1451 {
1452 struct vnode *vp = ap->a_vp, *vp2;
1453 struct cdev *dev;
1454 struct cdev_priv *cdp;
1455 struct devfs_dirent *de;
1456 int i;
1457
1458 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1459
1460 dev = vp->v_rdev;
1461 cdp = cdev2priv(dev);
1462
1463 dev_lock();
1464 cdp->cdp_inuse++;
1465 dev_unlock();
1466
1467 vhold(vp);
1468 vgone(vp);
1469 vdrop(vp);
1470
1471 VOP_UNLOCK(vp,0);
1472 loop:
1473 for (;;) {
1474 mtx_lock(&devfs_de_interlock);
1475 dev_lock();
1476 vp2 = NULL;
1477 for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1478 de = cdp->cdp_dirents[i];
1479 if (de == NULL)
1480 continue;
1481
1482 vp2 = de->de_vnode;
1483 if (vp2 != NULL) {
1484 dev_unlock();
1485 VI_LOCK(vp2);
1486 mtx_unlock(&devfs_de_interlock);
1487 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1488 curthread))
1489 goto loop;
1490 vhold(vp2);
1491 vgone(vp2);
1492 vdrop(vp2);
1493 vput(vp2);
1494 break;
1495 }
1496 }
1497 if (vp2 != NULL) {
1498 continue;
1499 }
1500 dev_unlock();
1501 mtx_unlock(&devfs_de_interlock);
1502 break;
1503 }
1504 dev_lock();
1505 cdp->cdp_inuse--;
1506 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1507 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1508 dev_unlock();
1509 dev_rel(&cdp->cdp_c);
1510 } else
1511 dev_unlock();
1512
1513 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1514 return (0);
1515 }
1516
1517 static int
devfs_rioctl(struct vop_ioctl_args * ap)1518 devfs_rioctl(struct vop_ioctl_args *ap)
1519 {
1520 struct vnode *vp;
1521 struct devfs_mount *dmp;
1522 int error;
1523
1524 vp = ap->a_vp;
1525 vn_lock(vp, LK_SHARED | LK_RETRY);
1526 if (vp->v_iflag & VI_DOOMED) {
1527 VOP_UNLOCK(vp, 0);
1528 return (EBADF);
1529 }
1530 dmp = VFSTODEVFS(vp->v_mount);
1531 sx_xlock(&dmp->dm_lock);
1532 VOP_UNLOCK(vp, 0);
1533 DEVFS_DMP_HOLD(dmp);
1534 devfs_populate(dmp);
1535 if (DEVFS_DMP_DROP(dmp)) {
1536 sx_xunlock(&dmp->dm_lock);
1537 devfs_unmount_final(dmp);
1538 return (ENOENT);
1539 }
1540 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1541 sx_xunlock(&dmp->dm_lock);
1542 return (error);
1543 }
1544
1545 static int
devfs_rread(struct vop_read_args * ap)1546 devfs_rread(struct vop_read_args *ap)
1547 {
1548
1549 if (ap->a_vp->v_type != VDIR)
1550 return (EINVAL);
1551 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1552 }
1553
1554 static int
devfs_setattr(struct vop_setattr_args * ap)1555 devfs_setattr(struct vop_setattr_args *ap)
1556 {
1557 struct devfs_dirent *de;
1558 struct vattr *vap;
1559 struct vnode *vp;
1560 struct thread *td;
1561 int c, error;
1562 uid_t uid;
1563 gid_t gid;
1564
1565 vap = ap->a_vap;
1566 vp = ap->a_vp;
1567 td = curthread;
1568 if ((vap->va_type != VNON) ||
1569 (vap->va_nlink != VNOVAL) ||
1570 (vap->va_fsid != VNOVAL) ||
1571 (vap->va_fileid != VNOVAL) ||
1572 (vap->va_blocksize != VNOVAL) ||
1573 (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1574 (vap->va_rdev != VNOVAL) ||
1575 ((int)vap->va_bytes != VNOVAL) ||
1576 (vap->va_gen != VNOVAL)) {
1577 return (EINVAL);
1578 }
1579
1580 de = vp->v_data;
1581 if (vp->v_type == VDIR)
1582 de = de->de_dir;
1583
1584 error = c = 0;
1585 if (vap->va_uid == (uid_t)VNOVAL)
1586 uid = de->de_uid;
1587 else
1588 uid = vap->va_uid;
1589 if (vap->va_gid == (gid_t)VNOVAL)
1590 gid = de->de_gid;
1591 else
1592 gid = vap->va_gid;
1593 if (uid != de->de_uid || gid != de->de_gid) {
1594 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1595 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1596 error = priv_check(td, PRIV_VFS_CHOWN);
1597 if (error)
1598 return (error);
1599 }
1600 de->de_uid = uid;
1601 de->de_gid = gid;
1602 c = 1;
1603 }
1604
1605 if (vap->va_mode != (mode_t)VNOVAL) {
1606 if (ap->a_cred->cr_uid != de->de_uid) {
1607 error = priv_check(td, PRIV_VFS_ADMIN);
1608 if (error)
1609 return (error);
1610 }
1611 de->de_mode = vap->va_mode;
1612 c = 1;
1613 }
1614
1615 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1616 error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1617 if (error != 0)
1618 return (error);
1619 if (vap->va_atime.tv_sec != VNOVAL) {
1620 if (vp->v_type == VCHR)
1621 vp->v_rdev->si_atime = vap->va_atime;
1622 else
1623 de->de_atime = vap->va_atime;
1624 }
1625 if (vap->va_mtime.tv_sec != VNOVAL) {
1626 if (vp->v_type == VCHR)
1627 vp->v_rdev->si_mtime = vap->va_mtime;
1628 else
1629 de->de_mtime = vap->va_mtime;
1630 }
1631 c = 1;
1632 }
1633
1634 if (c) {
1635 if (vp->v_type == VCHR)
1636 vfs_timestamp(&vp->v_rdev->si_ctime);
1637 else
1638 vfs_timestamp(&de->de_mtime);
1639 }
1640 return (0);
1641 }
1642
1643 #ifdef MAC
1644 static int
devfs_setlabel(struct vop_setlabel_args * ap)1645 devfs_setlabel(struct vop_setlabel_args *ap)
1646 {
1647 struct vnode *vp;
1648 struct devfs_dirent *de;
1649
1650 vp = ap->a_vp;
1651 de = vp->v_data;
1652
1653 mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1654 mac_devfs_update(vp->v_mount, de, vp);
1655
1656 return (0);
1657 }
1658 #endif
1659
1660 static int
devfs_stat_f(struct file * fp,struct stat * sb,struct ucred * cred,struct thread * td)1661 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1662 {
1663
1664 return (vnops.fo_stat(fp, sb, cred, td));
1665 }
1666
1667 static int
devfs_symlink(struct vop_symlink_args * ap)1668 devfs_symlink(struct vop_symlink_args *ap)
1669 {
1670 int i, error;
1671 struct devfs_dirent *dd;
1672 struct devfs_dirent *de, *de_covered, *de_dotdot;
1673 struct devfs_mount *dmp;
1674
1675 error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1676 if (error)
1677 return(error);
1678 dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1679 if (devfs_populate_vp(ap->a_dvp) != 0)
1680 return (ENOENT);
1681
1682 dd = ap->a_dvp->v_data;
1683 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1684 de->de_flags = DE_USER;
1685 de->de_uid = 0;
1686 de->de_gid = 0;
1687 de->de_mode = 0755;
1688 de->de_inode = alloc_unr(devfs_inos);
1689 de->de_dir = dd;
1690 de->de_dirent->d_type = DT_LNK;
1691 i = strlen(ap->a_target) + 1;
1692 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1693 bcopy(ap->a_target, de->de_symlink, i);
1694 #ifdef MAC
1695 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1696 #endif
1697 de_covered = devfs_find(dd, de->de_dirent->d_name,
1698 de->de_dirent->d_namlen, 0);
1699 if (de_covered != NULL) {
1700 if ((de_covered->de_flags & DE_USER) != 0) {
1701 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1702 sx_xunlock(&dmp->dm_lock);
1703 return (EEXIST);
1704 }
1705 KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1706 ("devfs_symlink: entry %p already covered", de_covered));
1707 de_covered->de_flags |= DE_COVERED;
1708 }
1709
1710 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */
1711 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */
1712 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1713 devfs_dir_ref_de(dmp, dd);
1714 devfs_rules_apply(dmp, de);
1715
1716 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1717 }
1718
1719 static int
devfs_truncate_f(struct file * fp,off_t length,struct ucred * cred,struct thread * td)1720 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1721 {
1722
1723 return (vnops.fo_truncate(fp, length, cred, td));
1724 }
1725
1726 static int
devfs_write_f(struct file * fp,struct uio * uio,struct ucred * cred,int flags,struct thread * td)1727 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1728 int flags, struct thread *td)
1729 {
1730 struct cdev *dev;
1731 int error, ioflag, ref;
1732 ssize_t resid;
1733 struct cdevsw *dsw;
1734 struct file *fpop;
1735
1736 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1737 return (EINVAL);
1738 fpop = td->td_fpop;
1739 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1740 if (error != 0) {
1741 error = vnops.fo_write(fp, uio, cred, flags, td);
1742 return (error);
1743 }
1744 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1745 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1746 if (ioflag & O_DIRECT)
1747 ioflag |= IO_DIRECT;
1748 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1749
1750 resid = uio->uio_resid;
1751
1752 error = dsw->d_write(dev, uio, ioflag);
1753 if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1754 devfs_timestamp(&dev->si_ctime);
1755 dev->si_mtime = dev->si_ctime;
1756 }
1757 td->td_fpop = fpop;
1758 dev_relthread(dev, ref);
1759
1760 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1761 return (error);
1762 }
1763
1764 dev_t
dev2udev(struct cdev * x)1765 dev2udev(struct cdev *x)
1766 {
1767 if (x == NULL)
1768 return (NODEV);
1769 return (cdev2priv(x)->cdp_inode);
1770 }
1771
1772 static struct fileops devfs_ops_f = {
1773 .fo_read = devfs_read_f,
1774 .fo_write = devfs_write_f,
1775 .fo_truncate = devfs_truncate_f,
1776 .fo_ioctl = devfs_ioctl_f,
1777 .fo_poll = devfs_poll_f,
1778 .fo_kqfilter = devfs_kqfilter_f,
1779 .fo_stat = devfs_stat_f,
1780 .fo_close = devfs_close_f,
1781 .fo_chmod = vn_chmod,
1782 .fo_chown = vn_chown,
1783 .fo_sendfile = vn_sendfile,
1784 .fo_seek = vn_seek,
1785 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
1786 };
1787
1788 static struct vop_vector devfs_vnodeops = {
1789 .vop_default = &default_vnodeops,
1790
1791 .vop_access = devfs_access,
1792 .vop_getattr = devfs_getattr,
1793 .vop_ioctl = devfs_rioctl,
1794 .vop_lookup = devfs_lookup,
1795 .vop_mknod = devfs_mknod,
1796 .vop_pathconf = devfs_pathconf,
1797 .vop_read = devfs_rread,
1798 .vop_readdir = devfs_readdir,
1799 .vop_readlink = devfs_readlink,
1800 .vop_reclaim = devfs_reclaim,
1801 .vop_remove = devfs_remove,
1802 .vop_revoke = devfs_revoke,
1803 .vop_setattr = devfs_setattr,
1804 #ifdef MAC
1805 .vop_setlabel = devfs_setlabel,
1806 #endif
1807 .vop_symlink = devfs_symlink,
1808 .vop_vptocnp = devfs_vptocnp,
1809 };
1810
1811 struct vop_vector devfs_specops = {
1812 .vop_default = &default_vnodeops,
1813
1814 .vop_access = devfs_access,
1815 .vop_bmap = VOP_PANIC,
1816 .vop_close = devfs_close,
1817 .vop_create = VOP_PANIC,
1818 .vop_fsync = devfs_fsync,
1819 .vop_getattr = devfs_getattr,
1820 .vop_link = VOP_PANIC,
1821 .vop_mkdir = VOP_PANIC,
1822 .vop_mknod = VOP_PANIC,
1823 .vop_open = devfs_open,
1824 .vop_pathconf = devfs_pathconf,
1825 .vop_poll = dead_poll,
1826 .vop_print = devfs_print,
1827 .vop_read = dead_read,
1828 .vop_readdir = VOP_PANIC,
1829 .vop_readlink = VOP_PANIC,
1830 .vop_reallocblks = VOP_PANIC,
1831 .vop_reclaim = devfs_reclaim,
1832 .vop_remove = devfs_remove,
1833 .vop_rename = VOP_PANIC,
1834 .vop_revoke = devfs_revoke,
1835 .vop_rmdir = VOP_PANIC,
1836 .vop_setattr = devfs_setattr,
1837 #ifdef MAC
1838 .vop_setlabel = devfs_setlabel,
1839 #endif
1840 .vop_strategy = VOP_PANIC,
1841 .vop_symlink = VOP_PANIC,
1842 .vop_vptocnp = devfs_vptocnp,
1843 .vop_write = dead_write,
1844 };
1845
1846 /*
1847 * Our calling convention to the device drivers used to be that we passed
1848 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1849 * flags instead since that's what open(), close() and ioctl() takes and
1850 * we don't really want vnode.h in device drivers.
1851 * We solved the source compatibility by redefining some vnode flags to
1852 * be the same as the fcntl ones and by sending down the bitwise OR of
1853 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody
1854 * pulls the rug out under this.
1855 */
1856 CTASSERT(O_NONBLOCK == IO_NDELAY);
1857 CTASSERT(O_FSYNC == IO_SYNC);
1858