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: releng/10.0/sys/fs/devfs/devfs_vnops.c 257122 2013-10-25 16:33:24Z kib $
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/time.h>
61 #include <sys/ttycom.h>
62 #include <sys/unistd.h>
63 #include <sys/vnode.h>
64
65 static struct vop_vector devfs_vnodeops;
66 static struct vop_vector devfs_specops;
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 static int
84 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
85 int *ref)
86 {
87
88 *dswp = devvn_refthread(fp->f_vnode, devp, ref);
89 if (*devp != fp->f_data) {
90 if (*dswp != NULL)
91 dev_relthread(*devp, *ref);
92 return (ENXIO);
93 }
94 KASSERT((*devp)->si_refcount > 0,
95 ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
96 if (*dswp == NULL)
97 return (ENXIO);
98 curthread->td_fpop = fp;
99 return (0);
100 }
101
102 int
103 devfs_get_cdevpriv(void **datap)
104 {
105 struct file *fp;
106 struct cdev_privdata *p;
107 int error;
108
109 fp = curthread->td_fpop;
110 if (fp == NULL)
111 return (EBADF);
112 p = fp->f_cdevpriv;
113 if (p != NULL) {
114 error = 0;
115 *datap = p->cdpd_data;
116 } else
117 error = ENOENT;
118 return (error);
119 }
120
121 int
122 devfs_set_cdevpriv(void *priv, cdevpriv_dtr_t priv_dtr)
123 {
124 struct file *fp;
125 struct cdev_priv *cdp;
126 struct cdev_privdata *p;
127 int error;
128
129 fp = curthread->td_fpop;
130 if (fp == NULL)
131 return (ENOENT);
132 cdp = cdev2priv((struct cdev *)fp->f_data);
133 p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
134 p->cdpd_data = priv;
135 p->cdpd_dtr = priv_dtr;
136 p->cdpd_fp = fp;
137 mtx_lock(&cdevpriv_mtx);
138 if (fp->f_cdevpriv == NULL) {
139 LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
140 fp->f_cdevpriv = p;
141 mtx_unlock(&cdevpriv_mtx);
142 error = 0;
143 } else {
144 mtx_unlock(&cdevpriv_mtx);
145 free(p, M_CDEVPDATA);
146 error = EBUSY;
147 }
148 return (error);
149 }
150
151 void
152 devfs_destroy_cdevpriv(struct cdev_privdata *p)
153 {
154
155 mtx_assert(&cdevpriv_mtx, MA_OWNED);
156 p->cdpd_fp->f_cdevpriv = NULL;
157 LIST_REMOVE(p, cdpd_list);
158 mtx_unlock(&cdevpriv_mtx);
159 (p->cdpd_dtr)(p->cdpd_data);
160 free(p, M_CDEVPDATA);
161 }
162
163 void
164 devfs_fpdrop(struct file *fp)
165 {
166 struct cdev_privdata *p;
167
168 mtx_lock(&cdevpriv_mtx);
169 if ((p = fp->f_cdevpriv) == NULL) {
170 mtx_unlock(&cdevpriv_mtx);
171 return;
172 }
173 devfs_destroy_cdevpriv(p);
174 }
175
176 void
177 devfs_clear_cdevpriv(void)
178 {
179 struct file *fp;
180
181 fp = curthread->td_fpop;
182 if (fp == NULL)
183 return;
184 devfs_fpdrop(fp);
185 }
186
187 /*
188 * On success devfs_populate_vp() returns with dmp->dm_lock held.
189 */
190 static int
191 devfs_populate_vp(struct vnode *vp)
192 {
193 struct devfs_dirent *de;
194 struct devfs_mount *dmp;
195 int locked;
196
197 ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
198
199 dmp = VFSTODEVFS(vp->v_mount);
200 locked = VOP_ISLOCKED(vp);
201
202 sx_xlock(&dmp->dm_lock);
203 DEVFS_DMP_HOLD(dmp);
204
205 /* Can't call devfs_populate() with the vnode lock held. */
206 VOP_UNLOCK(vp, 0);
207 devfs_populate(dmp);
208
209 sx_xunlock(&dmp->dm_lock);
210 vn_lock(vp, locked | LK_RETRY);
211 sx_xlock(&dmp->dm_lock);
212 if (DEVFS_DMP_DROP(dmp)) {
213 sx_xunlock(&dmp->dm_lock);
214 devfs_unmount_final(dmp);
215 return (EBADF);
216 }
217 if ((vp->v_iflag & VI_DOOMED) != 0) {
218 sx_xunlock(&dmp->dm_lock);
219 return (EBADF);
220 }
221 de = vp->v_data;
222 KASSERT(de != NULL,
223 ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
224 if ((de->de_flags & DE_DOOMED) != 0) {
225 sx_xunlock(&dmp->dm_lock);
226 return (EBADF);
227 }
228
229 return (0);
230 }
231
232 static int
233 devfs_vptocnp(struct vop_vptocnp_args *ap)
234 {
235 struct vnode *vp = ap->a_vp;
236 struct vnode **dvp = ap->a_vpp;
237 struct devfs_mount *dmp;
238 char *buf = ap->a_buf;
239 int *buflen = ap->a_buflen;
240 struct devfs_dirent *dd, *de;
241 int i, error;
242
243 dmp = VFSTODEVFS(vp->v_mount);
244
245 error = devfs_populate_vp(vp);
246 if (error != 0)
247 return (error);
248
249 i = *buflen;
250 dd = vp->v_data;
251
252 if (vp->v_type == VCHR) {
253 i -= strlen(dd->de_cdp->cdp_c.si_name);
254 if (i < 0) {
255 error = ENOMEM;
256 goto finished;
257 }
258 bcopy(dd->de_cdp->cdp_c.si_name, buf + i,
259 strlen(dd->de_cdp->cdp_c.si_name));
260 de = dd->de_dir;
261 } else if (vp->v_type == VDIR) {
262 if (dd == dmp->dm_rootdir) {
263 *dvp = vp;
264 vref(*dvp);
265 goto finished;
266 }
267 i -= dd->de_dirent->d_namlen;
268 if (i < 0) {
269 error = ENOMEM;
270 goto finished;
271 }
272 bcopy(dd->de_dirent->d_name, buf + i,
273 dd->de_dirent->d_namlen);
274 de = dd;
275 } else {
276 error = ENOENT;
277 goto finished;
278 }
279 *buflen = i;
280 de = devfs_parent_dirent(de);
281 if (de == NULL) {
282 error = ENOENT;
283 goto finished;
284 }
285 mtx_lock(&devfs_de_interlock);
286 *dvp = de->de_vnode;
287 if (*dvp != NULL) {
288 VI_LOCK(*dvp);
289 mtx_unlock(&devfs_de_interlock);
290 vholdl(*dvp);
291 VI_UNLOCK(*dvp);
292 vref(*dvp);
293 vdrop(*dvp);
294 } else {
295 mtx_unlock(&devfs_de_interlock);
296 error = ENOENT;
297 }
298 finished:
299 sx_xunlock(&dmp->dm_lock);
300 return (error);
301 }
302
303 /*
304 * Construct the fully qualified path name relative to the mountpoint.
305 * If a NULL cnp is provided, no '/' is appended to the resulting path.
306 */
307 char *
308 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
309 struct componentname *cnp)
310 {
311 int i;
312 struct devfs_dirent *de;
313
314 sx_assert(&dmp->dm_lock, SA_LOCKED);
315
316 i = SPECNAMELEN;
317 buf[i] = '\0';
318 if (cnp != NULL)
319 i -= cnp->cn_namelen;
320 if (i < 0)
321 return (NULL);
322 if (cnp != NULL)
323 bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
324 de = dd;
325 while (de != dmp->dm_rootdir) {
326 if (cnp != NULL || i < SPECNAMELEN) {
327 i--;
328 if (i < 0)
329 return (NULL);
330 buf[i] = '/';
331 }
332 i -= de->de_dirent->d_namlen;
333 if (i < 0)
334 return (NULL);
335 bcopy(de->de_dirent->d_name, buf + i,
336 de->de_dirent->d_namlen);
337 de = devfs_parent_dirent(de);
338 if (de == NULL)
339 return (NULL);
340 }
341 return (buf + i);
342 }
343
344 static int
345 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
346 struct devfs_dirent *de)
347 {
348 int not_found;
349
350 not_found = 0;
351 if (de->de_flags & DE_DOOMED)
352 not_found = 1;
353 if (DEVFS_DE_DROP(de)) {
354 KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
355 devfs_dirent_free(de);
356 }
357 if (DEVFS_DMP_DROP(dmp)) {
358 KASSERT(not_found == 1,
359 ("DEVFS mount struct freed before dirent"));
360 not_found = 2;
361 sx_xunlock(&dmp->dm_lock);
362 devfs_unmount_final(dmp);
363 }
364 if (not_found == 1 || (drop_dm_lock && not_found != 2))
365 sx_unlock(&dmp->dm_lock);
366 return (not_found);
367 }
368
369 static void
370 devfs_insmntque_dtr(struct vnode *vp, void *arg)
371 {
372 struct devfs_dirent *de;
373
374 de = (struct devfs_dirent *)arg;
375 mtx_lock(&devfs_de_interlock);
376 vp->v_data = NULL;
377 de->de_vnode = NULL;
378 mtx_unlock(&devfs_de_interlock);
379 vgone(vp);
380 vput(vp);
381 }
382
383 /*
384 * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
385 * it on return.
386 */
387 int
388 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
389 struct vnode **vpp)
390 {
391 int error;
392 struct vnode *vp;
393 struct cdev *dev;
394 struct devfs_mount *dmp;
395 struct cdevsw *dsw;
396
397 dmp = VFSTODEVFS(mp);
398 if (de->de_flags & DE_DOOMED) {
399 sx_xunlock(&dmp->dm_lock);
400 return (ENOENT);
401 }
402 loop:
403 DEVFS_DE_HOLD(de);
404 DEVFS_DMP_HOLD(dmp);
405 mtx_lock(&devfs_de_interlock);
406 vp = de->de_vnode;
407 if (vp != NULL) {
408 VI_LOCK(vp);
409 mtx_unlock(&devfs_de_interlock);
410 sx_xunlock(&dmp->dm_lock);
411 vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread);
412 sx_xlock(&dmp->dm_lock);
413 if (devfs_allocv_drop_refs(0, dmp, de)) {
414 vput(vp);
415 return (ENOENT);
416 }
417 else if ((vp->v_iflag & VI_DOOMED) != 0) {
418 mtx_lock(&devfs_de_interlock);
419 if (de->de_vnode == vp) {
420 de->de_vnode = NULL;
421 vp->v_data = NULL;
422 }
423 mtx_unlock(&devfs_de_interlock);
424 vput(vp);
425 goto loop;
426 }
427 sx_xunlock(&dmp->dm_lock);
428 *vpp = vp;
429 return (0);
430 }
431 mtx_unlock(&devfs_de_interlock);
432 if (de->de_dirent->d_type == DT_CHR) {
433 if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
434 devfs_allocv_drop_refs(1, dmp, de);
435 return (ENOENT);
436 }
437 dev = &de->de_cdp->cdp_c;
438 } else {
439 dev = NULL;
440 }
441 error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
442 if (error != 0) {
443 devfs_allocv_drop_refs(1, dmp, de);
444 printf("devfs_allocv: failed to allocate new vnode\n");
445 return (error);
446 }
447
448 if (de->de_dirent->d_type == DT_CHR) {
449 vp->v_type = VCHR;
450 VI_LOCK(vp);
451 dev_lock();
452 dev_refl(dev);
453 /* XXX: v_rdev should be protect by vnode lock */
454 vp->v_rdev = dev;
455 KASSERT(vp->v_usecount == 1,
456 ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount));
457 dev->si_usecount += vp->v_usecount;
458 /* Special casing of ttys for deadfs. Probably redundant. */
459 dsw = dev->si_devsw;
460 if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
461 vp->v_vflag |= VV_ISTTY;
462 dev_unlock();
463 VI_UNLOCK(vp);
464 if ((dev->si_flags & SI_ETERNAL) != 0)
465 vp->v_vflag |= VV_ETERNALDEV;
466 vp->v_op = &devfs_specops;
467 } else if (de->de_dirent->d_type == DT_DIR) {
468 vp->v_type = VDIR;
469 } else if (de->de_dirent->d_type == DT_LNK) {
470 vp->v_type = VLNK;
471 } else {
472 vp->v_type = VBAD;
473 }
474 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
475 VN_LOCK_ASHARE(vp);
476 mtx_lock(&devfs_de_interlock);
477 vp->v_data = de;
478 de->de_vnode = vp;
479 mtx_unlock(&devfs_de_interlock);
480 error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
481 if (error != 0) {
482 (void) devfs_allocv_drop_refs(1, dmp, de);
483 return (error);
484 }
485 if (devfs_allocv_drop_refs(0, dmp, de)) {
486 vput(vp);
487 return (ENOENT);
488 }
489 #ifdef MAC
490 mac_devfs_vnode_associate(mp, de, vp);
491 #endif
492 sx_xunlock(&dmp->dm_lock);
493 *vpp = vp;
494 return (0);
495 }
496
497 static int
498 devfs_access(struct vop_access_args *ap)
499 {
500 struct vnode *vp = ap->a_vp;
501 struct devfs_dirent *de;
502 int error;
503
504 de = vp->v_data;
505 if (vp->v_type == VDIR)
506 de = de->de_dir;
507
508 error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
509 ap->a_accmode, ap->a_cred, NULL);
510 if (error == 0)
511 return (0);
512 if (error != EACCES)
513 return (error);
514 /* We do, however, allow access to the controlling terminal */
515 if (!(ap->a_td->td_proc->p_flag & P_CONTROLT))
516 return (error);
517 if (ap->a_td->td_proc->p_session->s_ttydp == de->de_cdp)
518 return (0);
519 return (error);
520 }
521
522 /* ARGSUSED */
523 static int
524 devfs_close(struct vop_close_args *ap)
525 {
526 struct vnode *vp = ap->a_vp, *oldvp;
527 struct thread *td = ap->a_td;
528 struct cdev *dev = vp->v_rdev;
529 struct cdevsw *dsw;
530 int vp_locked, error, ref;
531
532 /*
533 * XXX: Don't call d_close() if we were called because of
534 * XXX: insmntque1() failure.
535 */
536 if (vp->v_data == NULL)
537 return (0);
538
539 /*
540 * Hack: a tty device that is a controlling terminal
541 * has a reference from the session structure.
542 * We cannot easily tell that a character device is
543 * a controlling terminal, unless it is the closing
544 * process' controlling terminal. In that case,
545 * if the reference count is 2 (this last descriptor
546 * plus the session), release the reference from the session.
547 */
548 oldvp = NULL;
549 sx_xlock(&proctree_lock);
550 if (td && vp == td->td_proc->p_session->s_ttyvp) {
551 SESS_LOCK(td->td_proc->p_session);
552 VI_LOCK(vp);
553 if (count_dev(dev) == 2 && (vp->v_iflag & VI_DOOMED) == 0) {
554 td->td_proc->p_session->s_ttyvp = NULL;
555 td->td_proc->p_session->s_ttydp = NULL;
556 oldvp = vp;
557 }
558 VI_UNLOCK(vp);
559 SESS_UNLOCK(td->td_proc->p_session);
560 }
561 sx_xunlock(&proctree_lock);
562 if (oldvp != NULL)
563 vrele(oldvp);
564 /*
565 * We do not want to really close the device if it
566 * is still in use unless we are trying to close it
567 * forcibly. Since every use (buffer, vnode, swap, cmap)
568 * holds a reference to the vnode, and because we mark
569 * any other vnodes that alias this device, when the
570 * sum of the reference counts on all the aliased
571 * vnodes descends to one, we are on last close.
572 */
573 dsw = dev_refthread(dev, &ref);
574 if (dsw == NULL)
575 return (ENXIO);
576 VI_LOCK(vp);
577 if (vp->v_iflag & VI_DOOMED) {
578 /* Forced close. */
579 } else if (dsw->d_flags & D_TRACKCLOSE) {
580 /* Keep device updated on status. */
581 } else if (count_dev(dev) > 1) {
582 VI_UNLOCK(vp);
583 dev_relthread(dev, ref);
584 return (0);
585 }
586 vholdl(vp);
587 VI_UNLOCK(vp);
588 vp_locked = VOP_ISLOCKED(vp);
589 VOP_UNLOCK(vp, 0);
590 KASSERT(dev->si_refcount > 0,
591 ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
592 error = dsw->d_close(dev, ap->a_fflag, S_IFCHR, td);
593 dev_relthread(dev, ref);
594 vn_lock(vp, vp_locked | LK_RETRY);
595 vdrop(vp);
596 return (error);
597 }
598
599 static int
600 devfs_close_f(struct file *fp, struct thread *td)
601 {
602 int error;
603 struct file *fpop;
604
605 /*
606 * NB: td may be NULL if this descriptor is closed due to
607 * garbage collection from a closed UNIX domain socket.
608 */
609 fpop = curthread->td_fpop;
610 curthread->td_fpop = fp;
611 error = vnops.fo_close(fp, td);
612 curthread->td_fpop = fpop;
613
614 /*
615 * The f_cdevpriv cannot be assigned non-NULL value while we
616 * are destroying the file.
617 */
618 if (fp->f_cdevpriv != NULL)
619 devfs_fpdrop(fp);
620 return (error);
621 }
622
623 static int
624 devfs_fsync(struct vop_fsync_args *ap)
625 {
626 int error;
627 struct bufobj *bo;
628 struct devfs_dirent *de;
629
630 if (!vn_isdisk(ap->a_vp, &error)) {
631 bo = &ap->a_vp->v_bufobj;
632 de = ap->a_vp->v_data;
633 if (error == ENXIO && bo->bo_dirty.bv_cnt > 0) {
634 printf("Device %s went missing before all of the data "
635 "could be written to it; expect data loss.\n",
636 de->de_dirent->d_name);
637
638 error = vop_stdfsync(ap);
639 if (bo->bo_dirty.bv_cnt != 0 || error != 0)
640 panic("devfs_fsync: vop_stdfsync failed.");
641 }
642
643 return (0);
644 }
645
646 return (vop_stdfsync(ap));
647 }
648
649 static int
650 devfs_getattr(struct vop_getattr_args *ap)
651 {
652 struct vnode *vp = ap->a_vp;
653 struct vattr *vap = ap->a_vap;
654 int error;
655 struct devfs_dirent *de;
656 struct devfs_mount *dmp;
657 struct cdev *dev;
658
659 error = devfs_populate_vp(vp);
660 if (error != 0)
661 return (error);
662
663 dmp = VFSTODEVFS(vp->v_mount);
664 sx_xunlock(&dmp->dm_lock);
665
666 de = vp->v_data;
667 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
668 if (vp->v_type == VDIR) {
669 de = de->de_dir;
670 KASSERT(de != NULL,
671 ("Null dir dirent in devfs_getattr vp=%p", vp));
672 }
673 vap->va_uid = de->de_uid;
674 vap->va_gid = de->de_gid;
675 vap->va_mode = de->de_mode;
676 if (vp->v_type == VLNK)
677 vap->va_size = strlen(de->de_symlink);
678 else if (vp->v_type == VDIR)
679 vap->va_size = vap->va_bytes = DEV_BSIZE;
680 else
681 vap->va_size = 0;
682 if (vp->v_type != VDIR)
683 vap->va_bytes = 0;
684 vap->va_blocksize = DEV_BSIZE;
685 vap->va_type = vp->v_type;
686
687 #define fix(aa) \
688 do { \
689 if ((aa).tv_sec <= 3600) { \
690 (aa).tv_sec = boottime.tv_sec; \
691 (aa).tv_nsec = boottime.tv_usec * 1000; \
692 } \
693 } while (0)
694
695 if (vp->v_type != VCHR) {
696 fix(de->de_atime);
697 vap->va_atime = de->de_atime;
698 fix(de->de_mtime);
699 vap->va_mtime = de->de_mtime;
700 fix(de->de_ctime);
701 vap->va_ctime = de->de_ctime;
702 } else {
703 dev = vp->v_rdev;
704 fix(dev->si_atime);
705 vap->va_atime = dev->si_atime;
706 fix(dev->si_mtime);
707 vap->va_mtime = dev->si_mtime;
708 fix(dev->si_ctime);
709 vap->va_ctime = dev->si_ctime;
710
711 vap->va_rdev = cdev2priv(dev)->cdp_inode;
712 }
713 vap->va_gen = 0;
714 vap->va_flags = 0;
715 vap->va_filerev = 0;
716 vap->va_nlink = de->de_links;
717 vap->va_fileid = de->de_inode;
718
719 return (error);
720 }
721
722 /* ARGSUSED */
723 static int
724 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
725 {
726 struct cdev *dev;
727 struct cdevsw *dsw;
728 struct vnode *vp;
729 struct vnode *vpold;
730 int error, i, ref;
731 const char *p;
732 struct fiodgname_arg *fgn;
733 struct file *fpop;
734
735 fpop = td->td_fpop;
736 error = devfs_fp_check(fp, &dev, &dsw, &ref);
737 if (error)
738 return (error);
739
740 if (com == FIODTYPE) {
741 *(int *)data = dsw->d_flags & D_TYPEMASK;
742 td->td_fpop = fpop;
743 dev_relthread(dev, ref);
744 return (0);
745 } else if (com == FIODGNAME) {
746 fgn = data;
747 p = devtoname(dev);
748 i = strlen(p) + 1;
749 if (i > fgn->len)
750 error = EINVAL;
751 else
752 error = copyout(p, fgn->buf, i);
753 td->td_fpop = fpop;
754 dev_relthread(dev, ref);
755 return (error);
756 }
757 error = dsw->d_ioctl(dev, com, data, fp->f_flag, td);
758 td->td_fpop = NULL;
759 dev_relthread(dev, ref);
760 if (error == ENOIOCTL)
761 error = ENOTTY;
762 if (error == 0 && com == TIOCSCTTY) {
763 vp = fp->f_vnode;
764
765 /* Do nothing if reassigning same control tty */
766 sx_slock(&proctree_lock);
767 if (td->td_proc->p_session->s_ttyvp == vp) {
768 sx_sunlock(&proctree_lock);
769 return (0);
770 }
771
772 vpold = td->td_proc->p_session->s_ttyvp;
773 VREF(vp);
774 SESS_LOCK(td->td_proc->p_session);
775 td->td_proc->p_session->s_ttyvp = vp;
776 td->td_proc->p_session->s_ttydp = cdev2priv(dev);
777 SESS_UNLOCK(td->td_proc->p_session);
778
779 sx_sunlock(&proctree_lock);
780
781 /* Get rid of reference to old control tty */
782 if (vpold)
783 vrele(vpold);
784 }
785 return (error);
786 }
787
788 /* ARGSUSED */
789 static int
790 devfs_kqfilter_f(struct file *fp, struct knote *kn)
791 {
792 struct cdev *dev;
793 struct cdevsw *dsw;
794 int error, ref;
795 struct file *fpop;
796 struct thread *td;
797
798 td = curthread;
799 fpop = td->td_fpop;
800 error = devfs_fp_check(fp, &dev, &dsw, &ref);
801 if (error)
802 return (error);
803 error = dsw->d_kqfilter(dev, kn);
804 td->td_fpop = fpop;
805 dev_relthread(dev, ref);
806 return (error);
807 }
808
809 static inline int
810 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
811 {
812 struct cdev_priv *cdp;
813 struct ucred *dcr;
814 int error;
815
816 cdp = de->de_cdp;
817 if (cdp == NULL)
818 return (0);
819 dcr = cdp->cdp_c.si_cred;
820 if (dcr == NULL)
821 return (0);
822
823 error = prison_check(td->td_ucred, dcr);
824 if (error == 0)
825 return (0);
826 /* We do, however, allow access to the controlling terminal */
827 if (!(td->td_proc->p_flag & P_CONTROLT))
828 return (error);
829 if (td->td_proc->p_session->s_ttydp == cdp)
830 return (0);
831 return (error);
832 }
833
834 static int
835 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
836 {
837 struct componentname *cnp;
838 struct vnode *dvp, **vpp;
839 struct thread *td;
840 struct devfs_dirent *de, *dd;
841 struct devfs_dirent **dde;
842 struct devfs_mount *dmp;
843 struct cdev *cdev;
844 int error, flags, nameiop, dvplocked;
845 char specname[SPECNAMELEN + 1], *pname;
846
847 cnp = ap->a_cnp;
848 vpp = ap->a_vpp;
849 dvp = ap->a_dvp;
850 pname = cnp->cn_nameptr;
851 td = cnp->cn_thread;
852 flags = cnp->cn_flags;
853 nameiop = cnp->cn_nameiop;
854 dmp = VFSTODEVFS(dvp->v_mount);
855 dd = dvp->v_data;
856 *vpp = NULLVP;
857
858 if ((flags & ISLASTCN) && nameiop == RENAME)
859 return (EOPNOTSUPP);
860
861 if (dvp->v_type != VDIR)
862 return (ENOTDIR);
863
864 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
865 return (EIO);
866
867 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
868 if (error)
869 return (error);
870
871 if (cnp->cn_namelen == 1 && *pname == '.') {
872 if ((flags & ISLASTCN) && nameiop != LOOKUP)
873 return (EINVAL);
874 *vpp = dvp;
875 VREF(dvp);
876 return (0);
877 }
878
879 if (flags & ISDOTDOT) {
880 if ((flags & ISLASTCN) && nameiop != LOOKUP)
881 return (EINVAL);
882 de = devfs_parent_dirent(dd);
883 if (de == NULL)
884 return (ENOENT);
885 dvplocked = VOP_ISLOCKED(dvp);
886 VOP_UNLOCK(dvp, 0);
887 error = devfs_allocv(de, dvp->v_mount,
888 cnp->cn_lkflags & LK_TYPE_MASK, vpp);
889 *dm_unlock = 0;
890 vn_lock(dvp, dvplocked | LK_RETRY);
891 return (error);
892 }
893
894 dd = dvp->v_data;
895 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
896 while (de == NULL) { /* While(...) so we can use break */
897
898 if (nameiop == DELETE)
899 return (ENOENT);
900
901 /*
902 * OK, we didn't have an entry for the name we were asked for
903 * so we try to see if anybody can create it on demand.
904 */
905 pname = devfs_fqpn(specname, dmp, dd, cnp);
906 if (pname == NULL)
907 break;
908
909 cdev = NULL;
910 DEVFS_DMP_HOLD(dmp);
911 sx_xunlock(&dmp->dm_lock);
912 sx_slock(&clone_drain_lock);
913 EVENTHANDLER_INVOKE(dev_clone,
914 td->td_ucred, pname, strlen(pname), &cdev);
915 sx_sunlock(&clone_drain_lock);
916
917 if (cdev == NULL)
918 sx_xlock(&dmp->dm_lock);
919 else if (devfs_populate_vp(dvp) != 0) {
920 *dm_unlock = 0;
921 sx_xlock(&dmp->dm_lock);
922 if (DEVFS_DMP_DROP(dmp)) {
923 sx_xunlock(&dmp->dm_lock);
924 devfs_unmount_final(dmp);
925 } else
926 sx_xunlock(&dmp->dm_lock);
927 dev_rel(cdev);
928 return (ENOENT);
929 }
930 if (DEVFS_DMP_DROP(dmp)) {
931 *dm_unlock = 0;
932 sx_xunlock(&dmp->dm_lock);
933 devfs_unmount_final(dmp);
934 if (cdev != NULL)
935 dev_rel(cdev);
936 return (ENOENT);
937 }
938
939 if (cdev == NULL)
940 break;
941
942 dev_lock();
943 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
944 if (dde != NULL && *dde != NULL)
945 de = *dde;
946 dev_unlock();
947 dev_rel(cdev);
948 break;
949 }
950
951 if (de == NULL || de->de_flags & DE_WHITEOUT) {
952 if ((nameiop == CREATE || nameiop == RENAME) &&
953 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
954 cnp->cn_flags |= SAVENAME;
955 return (EJUSTRETURN);
956 }
957 return (ENOENT);
958 }
959
960 if (devfs_prison_check(de, td))
961 return (ENOENT);
962
963 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
964 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
965 if (error)
966 return (error);
967 if (*vpp == dvp) {
968 VREF(dvp);
969 *vpp = dvp;
970 return (0);
971 }
972 }
973 error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK,
974 vpp);
975 *dm_unlock = 0;
976 return (error);
977 }
978
979 static int
980 devfs_lookup(struct vop_lookup_args *ap)
981 {
982 int j;
983 struct devfs_mount *dmp;
984 int dm_unlock;
985
986 if (devfs_populate_vp(ap->a_dvp) != 0)
987 return (ENOTDIR);
988
989 dmp = VFSTODEVFS(ap->a_dvp->v_mount);
990 dm_unlock = 1;
991 j = devfs_lookupx(ap, &dm_unlock);
992 if (dm_unlock == 1)
993 sx_xunlock(&dmp->dm_lock);
994 return (j);
995 }
996
997 static int
998 devfs_mknod(struct vop_mknod_args *ap)
999 {
1000 struct componentname *cnp;
1001 struct vnode *dvp, **vpp;
1002 struct devfs_dirent *dd, *de;
1003 struct devfs_mount *dmp;
1004 int error;
1005
1006 /*
1007 * The only type of node we should be creating here is a
1008 * character device, for anything else return EOPNOTSUPP.
1009 */
1010 if (ap->a_vap->va_type != VCHR)
1011 return (EOPNOTSUPP);
1012 dvp = ap->a_dvp;
1013 dmp = VFSTODEVFS(dvp->v_mount);
1014
1015 cnp = ap->a_cnp;
1016 vpp = ap->a_vpp;
1017 dd = dvp->v_data;
1018
1019 error = ENOENT;
1020 sx_xlock(&dmp->dm_lock);
1021 TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1022 if (cnp->cn_namelen != de->de_dirent->d_namlen)
1023 continue;
1024 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1025 de->de_dirent->d_namlen) != 0)
1026 continue;
1027 if (de->de_flags & DE_WHITEOUT)
1028 break;
1029 goto notfound;
1030 }
1031 if (de == NULL)
1032 goto notfound;
1033 de->de_flags &= ~DE_WHITEOUT;
1034 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1035 return (error);
1036 notfound:
1037 sx_xunlock(&dmp->dm_lock);
1038 return (error);
1039 }
1040
1041 /* ARGSUSED */
1042 static int
1043 devfs_open(struct vop_open_args *ap)
1044 {
1045 struct thread *td = ap->a_td;
1046 struct vnode *vp = ap->a_vp;
1047 struct cdev *dev = vp->v_rdev;
1048 struct file *fp = ap->a_fp;
1049 int error, ref, vlocked;
1050 struct cdevsw *dsw;
1051 struct file *fpop;
1052 struct mtx *mtxp;
1053
1054 if (vp->v_type == VBLK)
1055 return (ENXIO);
1056
1057 if (dev == NULL)
1058 return (ENXIO);
1059
1060 /* Make this field valid before any I/O in d_open. */
1061 if (dev->si_iosize_max == 0)
1062 dev->si_iosize_max = DFLTPHYS;
1063
1064 dsw = dev_refthread(dev, &ref);
1065 if (dsw == NULL)
1066 return (ENXIO);
1067 if (fp == NULL && dsw->d_fdopen != NULL) {
1068 dev_relthread(dev, ref);
1069 return (ENXIO);
1070 }
1071
1072 vlocked = VOP_ISLOCKED(vp);
1073 VOP_UNLOCK(vp, 0);
1074
1075 fpop = td->td_fpop;
1076 td->td_fpop = fp;
1077 if (fp != NULL) {
1078 fp->f_data = dev;
1079 fp->f_vnode = vp;
1080 }
1081 if (dsw->d_fdopen != NULL)
1082 error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1083 else
1084 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1085 /* cleanup any cdevpriv upon error */
1086 if (error != 0)
1087 devfs_clear_cdevpriv();
1088 td->td_fpop = fpop;
1089
1090 vn_lock(vp, vlocked | LK_RETRY);
1091 dev_relthread(dev, ref);
1092 if (error != 0) {
1093 if (error == ERESTART)
1094 error = EINTR;
1095 return (error);
1096 }
1097
1098 #if 0 /* /dev/console */
1099 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1100 #else
1101 if (fp == NULL)
1102 return (error);
1103 #endif
1104 if (fp->f_ops == &badfileops)
1105 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1106 mtxp = mtx_pool_find(mtxpool_sleep, fp);
1107
1108 /*
1109 * Hint to the dofilewrite() to not force the buffer draining
1110 * on the writer to the file. Most likely, the write would
1111 * not need normal buffers.
1112 */
1113 mtx_lock(mtxp);
1114 fp->f_vnread_flags |= FDEVFS_VNODE;
1115 mtx_unlock(mtxp);
1116 return (error);
1117 }
1118
1119 static int
1120 devfs_pathconf(struct vop_pathconf_args *ap)
1121 {
1122
1123 switch (ap->a_name) {
1124 case _PC_MAC_PRESENT:
1125 #ifdef MAC
1126 /*
1127 * If MAC is enabled, devfs automatically supports
1128 * trivial non-persistant label storage.
1129 */
1130 *ap->a_retval = 1;
1131 #else
1132 *ap->a_retval = 0;
1133 #endif
1134 return (0);
1135 default:
1136 return (vop_stdpathconf(ap));
1137 }
1138 /* NOTREACHED */
1139 }
1140
1141 /* ARGSUSED */
1142 static int
1143 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1144 {
1145 struct cdev *dev;
1146 struct cdevsw *dsw;
1147 int error, ref;
1148 struct file *fpop;
1149
1150 fpop = td->td_fpop;
1151 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1152 if (error)
1153 return (poll_no_poll(events));
1154 error = dsw->d_poll(dev, events, td);
1155 td->td_fpop = fpop;
1156 dev_relthread(dev, ref);
1157 return(error);
1158 }
1159
1160 /*
1161 * Print out the contents of a special device vnode.
1162 */
1163 static int
1164 devfs_print(struct vop_print_args *ap)
1165 {
1166
1167 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1168 return (0);
1169 }
1170
1171 static int
1172 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1173 int flags, struct thread *td)
1174 {
1175 struct cdev *dev;
1176 int ioflag, error, ref;
1177 ssize_t resid;
1178 struct cdevsw *dsw;
1179 struct file *fpop;
1180
1181 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1182 return (EINVAL);
1183 fpop = td->td_fpop;
1184 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1185 if (error)
1186 return (error);
1187 resid = uio->uio_resid;
1188 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1189 if (ioflag & O_DIRECT)
1190 ioflag |= IO_DIRECT;
1191
1192 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1193 error = dsw->d_read(dev, uio, ioflag);
1194 if (uio->uio_resid != resid || (error == 0 && resid != 0))
1195 vfs_timestamp(&dev->si_atime);
1196 td->td_fpop = fpop;
1197 dev_relthread(dev, ref);
1198
1199 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1200 return (error);
1201 }
1202
1203 static int
1204 devfs_readdir(struct vop_readdir_args *ap)
1205 {
1206 int error;
1207 struct uio *uio;
1208 struct dirent *dp;
1209 struct devfs_dirent *dd;
1210 struct devfs_dirent *de;
1211 struct devfs_mount *dmp;
1212 off_t off;
1213 int *tmp_ncookies = NULL;
1214
1215 if (ap->a_vp->v_type != VDIR)
1216 return (ENOTDIR);
1217
1218 uio = ap->a_uio;
1219 if (uio->uio_offset < 0)
1220 return (EINVAL);
1221
1222 /*
1223 * XXX: This is a temporary hack to get around this filesystem not
1224 * supporting cookies. We store the location of the ncookies pointer
1225 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1226 * and set the number of cookies to 0. We then set the pointer to
1227 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1228 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1229 * pointer to its original location before returning to the caller.
1230 */
1231 if (ap->a_ncookies != NULL) {
1232 tmp_ncookies = ap->a_ncookies;
1233 *ap->a_ncookies = 0;
1234 ap->a_ncookies = NULL;
1235 }
1236
1237 dmp = VFSTODEVFS(ap->a_vp->v_mount);
1238 if (devfs_populate_vp(ap->a_vp) != 0) {
1239 if (tmp_ncookies != NULL)
1240 ap->a_ncookies = tmp_ncookies;
1241 return (EIO);
1242 }
1243 error = 0;
1244 de = ap->a_vp->v_data;
1245 off = 0;
1246 TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1247 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1248 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1249 continue;
1250 if (devfs_prison_check(dd, uio->uio_td))
1251 continue;
1252 if (dd->de_dirent->d_type == DT_DIR)
1253 de = dd->de_dir;
1254 else
1255 de = dd;
1256 dp = dd->de_dirent;
1257 if (dp->d_reclen > uio->uio_resid)
1258 break;
1259 dp->d_fileno = de->de_inode;
1260 if (off >= uio->uio_offset) {
1261 error = vfs_read_dirent(ap, dp, off);
1262 if (error)
1263 break;
1264 }
1265 off += dp->d_reclen;
1266 }
1267 sx_xunlock(&dmp->dm_lock);
1268 uio->uio_offset = off;
1269
1270 /*
1271 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1272 * place.
1273 */
1274 if (tmp_ncookies != NULL)
1275 ap->a_ncookies = tmp_ncookies;
1276
1277 return (error);
1278 }
1279
1280 static int
1281 devfs_readlink(struct vop_readlink_args *ap)
1282 {
1283 struct devfs_dirent *de;
1284
1285 de = ap->a_vp->v_data;
1286 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1287 }
1288
1289 static int
1290 devfs_reclaim(struct vop_reclaim_args *ap)
1291 {
1292 struct vnode *vp = ap->a_vp;
1293 struct devfs_dirent *de;
1294 struct cdev *dev;
1295
1296 mtx_lock(&devfs_de_interlock);
1297 de = vp->v_data;
1298 if (de != NULL) {
1299 de->de_vnode = NULL;
1300 vp->v_data = NULL;
1301 }
1302 mtx_unlock(&devfs_de_interlock);
1303
1304 vnode_destroy_vobject(vp);
1305
1306 VI_LOCK(vp);
1307 dev_lock();
1308 dev = vp->v_rdev;
1309 vp->v_rdev = NULL;
1310
1311 if (dev == NULL) {
1312 dev_unlock();
1313 VI_UNLOCK(vp);
1314 return (0);
1315 }
1316
1317 dev->si_usecount -= vp->v_usecount;
1318 dev_unlock();
1319 VI_UNLOCK(vp);
1320 dev_rel(dev);
1321 return (0);
1322 }
1323
1324 static int
1325 devfs_remove(struct vop_remove_args *ap)
1326 {
1327 struct vnode *dvp = ap->a_dvp;
1328 struct vnode *vp = ap->a_vp;
1329 struct devfs_dirent *dd;
1330 struct devfs_dirent *de, *de_covered;
1331 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1332
1333 ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1334 ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1335
1336 sx_xlock(&dmp->dm_lock);
1337 dd = ap->a_dvp->v_data;
1338 de = vp->v_data;
1339 if (de->de_cdp == NULL) {
1340 TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1341 if (de->de_dirent->d_type == DT_LNK) {
1342 de_covered = devfs_find(dd, de->de_dirent->d_name,
1343 de->de_dirent->d_namlen, 0);
1344 if (de_covered != NULL)
1345 de_covered->de_flags &= ~DE_COVERED;
1346 }
1347 /* We need to unlock dvp because devfs_delete() may lock it. */
1348 VOP_UNLOCK(vp, 0);
1349 if (dvp != vp)
1350 VOP_UNLOCK(dvp, 0);
1351 devfs_delete(dmp, de, 0);
1352 sx_xunlock(&dmp->dm_lock);
1353 if (dvp != vp)
1354 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1355 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1356 } else {
1357 de->de_flags |= DE_WHITEOUT;
1358 sx_xunlock(&dmp->dm_lock);
1359 }
1360 return (0);
1361 }
1362
1363 /*
1364 * Revoke is called on a tty when a terminal session ends. The vnode
1365 * is orphaned by setting v_op to deadfs so we need to let go of it
1366 * as well so that we create a new one next time around.
1367 *
1368 */
1369 static int
1370 devfs_revoke(struct vop_revoke_args *ap)
1371 {
1372 struct vnode *vp = ap->a_vp, *vp2;
1373 struct cdev *dev;
1374 struct cdev_priv *cdp;
1375 struct devfs_dirent *de;
1376 int i;
1377
1378 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1379
1380 dev = vp->v_rdev;
1381 cdp = cdev2priv(dev);
1382
1383 dev_lock();
1384 cdp->cdp_inuse++;
1385 dev_unlock();
1386
1387 vhold(vp);
1388 vgone(vp);
1389 vdrop(vp);
1390
1391 VOP_UNLOCK(vp,0);
1392 loop:
1393 for (;;) {
1394 mtx_lock(&devfs_de_interlock);
1395 dev_lock();
1396 vp2 = NULL;
1397 for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1398 de = cdp->cdp_dirents[i];
1399 if (de == NULL)
1400 continue;
1401
1402 vp2 = de->de_vnode;
1403 if (vp2 != NULL) {
1404 dev_unlock();
1405 VI_LOCK(vp2);
1406 mtx_unlock(&devfs_de_interlock);
1407 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1408 curthread))
1409 goto loop;
1410 vhold(vp2);
1411 vgone(vp2);
1412 vdrop(vp2);
1413 vput(vp2);
1414 break;
1415 }
1416 }
1417 if (vp2 != NULL) {
1418 continue;
1419 }
1420 dev_unlock();
1421 mtx_unlock(&devfs_de_interlock);
1422 break;
1423 }
1424 dev_lock();
1425 cdp->cdp_inuse--;
1426 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1427 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1428 dev_unlock();
1429 dev_rel(&cdp->cdp_c);
1430 } else
1431 dev_unlock();
1432
1433 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1434 return (0);
1435 }
1436
1437 static int
1438 devfs_rioctl(struct vop_ioctl_args *ap)
1439 {
1440 struct vnode *vp;
1441 struct devfs_mount *dmp;
1442 int error;
1443
1444 vp = ap->a_vp;
1445 vn_lock(vp, LK_SHARED | LK_RETRY);
1446 if (vp->v_iflag & VI_DOOMED) {
1447 VOP_UNLOCK(vp, 0);
1448 return (EBADF);
1449 }
1450 dmp = VFSTODEVFS(vp->v_mount);
1451 sx_xlock(&dmp->dm_lock);
1452 VOP_UNLOCK(vp, 0);
1453 DEVFS_DMP_HOLD(dmp);
1454 devfs_populate(dmp);
1455 if (DEVFS_DMP_DROP(dmp)) {
1456 sx_xunlock(&dmp->dm_lock);
1457 devfs_unmount_final(dmp);
1458 return (ENOENT);
1459 }
1460 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1461 sx_xunlock(&dmp->dm_lock);
1462 return (error);
1463 }
1464
1465 static int
1466 devfs_rread(struct vop_read_args *ap)
1467 {
1468
1469 if (ap->a_vp->v_type != VDIR)
1470 return (EINVAL);
1471 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1472 }
1473
1474 static int
1475 devfs_setattr(struct vop_setattr_args *ap)
1476 {
1477 struct devfs_dirent *de;
1478 struct vattr *vap;
1479 struct vnode *vp;
1480 struct thread *td;
1481 int c, error;
1482 uid_t uid;
1483 gid_t gid;
1484
1485 vap = ap->a_vap;
1486 vp = ap->a_vp;
1487 td = curthread;
1488 if ((vap->va_type != VNON) ||
1489 (vap->va_nlink != VNOVAL) ||
1490 (vap->va_fsid != VNOVAL) ||
1491 (vap->va_fileid != VNOVAL) ||
1492 (vap->va_blocksize != VNOVAL) ||
1493 (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1494 (vap->va_rdev != VNOVAL) ||
1495 ((int)vap->va_bytes != VNOVAL) ||
1496 (vap->va_gen != VNOVAL)) {
1497 return (EINVAL);
1498 }
1499
1500 de = vp->v_data;
1501 if (vp->v_type == VDIR)
1502 de = de->de_dir;
1503
1504 error = c = 0;
1505 if (vap->va_uid == (uid_t)VNOVAL)
1506 uid = de->de_uid;
1507 else
1508 uid = vap->va_uid;
1509 if (vap->va_gid == (gid_t)VNOVAL)
1510 gid = de->de_gid;
1511 else
1512 gid = vap->va_gid;
1513 if (uid != de->de_uid || gid != de->de_gid) {
1514 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1515 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1516 error = priv_check(td, PRIV_VFS_CHOWN);
1517 if (error)
1518 return (error);
1519 }
1520 de->de_uid = uid;
1521 de->de_gid = gid;
1522 c = 1;
1523 }
1524
1525 if (vap->va_mode != (mode_t)VNOVAL) {
1526 if (ap->a_cred->cr_uid != de->de_uid) {
1527 error = priv_check(td, PRIV_VFS_ADMIN);
1528 if (error)
1529 return (error);
1530 }
1531 de->de_mode = vap->va_mode;
1532 c = 1;
1533 }
1534
1535 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1536 /* See the comment in ufs_vnops::ufs_setattr(). */
1537 if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) &&
1538 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1539 (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td))))
1540 return (error);
1541 if (vap->va_atime.tv_sec != VNOVAL) {
1542 if (vp->v_type == VCHR)
1543 vp->v_rdev->si_atime = vap->va_atime;
1544 else
1545 de->de_atime = vap->va_atime;
1546 }
1547 if (vap->va_mtime.tv_sec != VNOVAL) {
1548 if (vp->v_type == VCHR)
1549 vp->v_rdev->si_mtime = vap->va_mtime;
1550 else
1551 de->de_mtime = vap->va_mtime;
1552 }
1553 c = 1;
1554 }
1555
1556 if (c) {
1557 if (vp->v_type == VCHR)
1558 vfs_timestamp(&vp->v_rdev->si_ctime);
1559 else
1560 vfs_timestamp(&de->de_mtime);
1561 }
1562 return (0);
1563 }
1564
1565 #ifdef MAC
1566 static int
1567 devfs_setlabel(struct vop_setlabel_args *ap)
1568 {
1569 struct vnode *vp;
1570 struct devfs_dirent *de;
1571
1572 vp = ap->a_vp;
1573 de = vp->v_data;
1574
1575 mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1576 mac_devfs_update(vp->v_mount, de, vp);
1577
1578 return (0);
1579 }
1580 #endif
1581
1582 static int
1583 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1584 {
1585
1586 return (vnops.fo_stat(fp, sb, cred, td));
1587 }
1588
1589 static int
1590 devfs_symlink(struct vop_symlink_args *ap)
1591 {
1592 int i, error;
1593 struct devfs_dirent *dd;
1594 struct devfs_dirent *de, *de_covered, *de_dotdot;
1595 struct devfs_mount *dmp;
1596
1597 error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1598 if (error)
1599 return(error);
1600 dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1601 if (devfs_populate_vp(ap->a_dvp) != 0)
1602 return (ENOENT);
1603
1604 dd = ap->a_dvp->v_data;
1605 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1606 de->de_flags = DE_USER;
1607 de->de_uid = 0;
1608 de->de_gid = 0;
1609 de->de_mode = 0755;
1610 de->de_inode = alloc_unr(devfs_inos);
1611 de->de_dir = dd;
1612 de->de_dirent->d_type = DT_LNK;
1613 i = strlen(ap->a_target) + 1;
1614 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1615 bcopy(ap->a_target, de->de_symlink, i);
1616 #ifdef MAC
1617 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1618 #endif
1619 de_covered = devfs_find(dd, de->de_dirent->d_name,
1620 de->de_dirent->d_namlen, 0);
1621 if (de_covered != NULL) {
1622 if ((de_covered->de_flags & DE_USER) != 0) {
1623 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1624 sx_xunlock(&dmp->dm_lock);
1625 return (EEXIST);
1626 }
1627 KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1628 ("devfs_symlink: entry %p already covered", de_covered));
1629 de_covered->de_flags |= DE_COVERED;
1630 }
1631
1632 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */
1633 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */
1634 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1635 devfs_dir_ref_de(dmp, dd);
1636 devfs_rules_apply(dmp, de);
1637
1638 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1639 }
1640
1641 static int
1642 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1643 {
1644
1645 return (vnops.fo_truncate(fp, length, cred, td));
1646 }
1647
1648 static int
1649 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1650 int flags, struct thread *td)
1651 {
1652 struct cdev *dev;
1653 int error, ioflag, ref;
1654 ssize_t resid;
1655 struct cdevsw *dsw;
1656 struct file *fpop;
1657
1658 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1659 return (EINVAL);
1660 fpop = td->td_fpop;
1661 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1662 if (error)
1663 return (error);
1664 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1665 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1666 if (ioflag & O_DIRECT)
1667 ioflag |= IO_DIRECT;
1668 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1669
1670 resid = uio->uio_resid;
1671
1672 error = dsw->d_write(dev, uio, ioflag);
1673 if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1674 vfs_timestamp(&dev->si_ctime);
1675 dev->si_mtime = dev->si_ctime;
1676 }
1677 td->td_fpop = fpop;
1678 dev_relthread(dev, ref);
1679
1680 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1681 return (error);
1682 }
1683
1684 dev_t
1685 dev2udev(struct cdev *x)
1686 {
1687 if (x == NULL)
1688 return (NODEV);
1689 return (cdev2priv(x)->cdp_inode);
1690 }
1691
1692 static struct fileops devfs_ops_f = {
1693 .fo_read = devfs_read_f,
1694 .fo_write = devfs_write_f,
1695 .fo_truncate = devfs_truncate_f,
1696 .fo_ioctl = devfs_ioctl_f,
1697 .fo_poll = devfs_poll_f,
1698 .fo_kqfilter = devfs_kqfilter_f,
1699 .fo_stat = devfs_stat_f,
1700 .fo_close = devfs_close_f,
1701 .fo_chmod = vn_chmod,
1702 .fo_chown = vn_chown,
1703 .fo_sendfile = vn_sendfile,
1704 .fo_seek = vn_seek,
1705 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
1706 };
1707
1708 static struct vop_vector devfs_vnodeops = {
1709 .vop_default = &default_vnodeops,
1710
1711 .vop_access = devfs_access,
1712 .vop_getattr = devfs_getattr,
1713 .vop_ioctl = devfs_rioctl,
1714 .vop_lookup = devfs_lookup,
1715 .vop_mknod = devfs_mknod,
1716 .vop_pathconf = devfs_pathconf,
1717 .vop_read = devfs_rread,
1718 .vop_readdir = devfs_readdir,
1719 .vop_readlink = devfs_readlink,
1720 .vop_reclaim = devfs_reclaim,
1721 .vop_remove = devfs_remove,
1722 .vop_revoke = devfs_revoke,
1723 .vop_setattr = devfs_setattr,
1724 #ifdef MAC
1725 .vop_setlabel = devfs_setlabel,
1726 #endif
1727 .vop_symlink = devfs_symlink,
1728 .vop_vptocnp = devfs_vptocnp,
1729 };
1730
1731 static struct vop_vector devfs_specops = {
1732 .vop_default = &default_vnodeops,
1733
1734 .vop_access = devfs_access,
1735 .vop_bmap = VOP_PANIC,
1736 .vop_close = devfs_close,
1737 .vop_create = VOP_PANIC,
1738 .vop_fsync = devfs_fsync,
1739 .vop_getattr = devfs_getattr,
1740 .vop_link = VOP_PANIC,
1741 .vop_mkdir = VOP_PANIC,
1742 .vop_mknod = VOP_PANIC,
1743 .vop_open = devfs_open,
1744 .vop_pathconf = devfs_pathconf,
1745 .vop_print = devfs_print,
1746 .vop_read = VOP_PANIC,
1747 .vop_readdir = VOP_PANIC,
1748 .vop_readlink = VOP_PANIC,
1749 .vop_reallocblks = VOP_PANIC,
1750 .vop_reclaim = devfs_reclaim,
1751 .vop_remove = devfs_remove,
1752 .vop_rename = VOP_PANIC,
1753 .vop_revoke = devfs_revoke,
1754 .vop_rmdir = VOP_PANIC,
1755 .vop_setattr = devfs_setattr,
1756 #ifdef MAC
1757 .vop_setlabel = devfs_setlabel,
1758 #endif
1759 .vop_strategy = VOP_PANIC,
1760 .vop_symlink = VOP_PANIC,
1761 .vop_vptocnp = devfs_vptocnp,
1762 .vop_write = VOP_PANIC,
1763 };
1764
1765 /*
1766 * Our calling convention to the device drivers used to be that we passed
1767 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1768 * flags instead since that's what open(), close() and ioctl() takes and
1769 * we don't really want vnode.h in device drivers.
1770 * We solved the source compatibility by redefining some vnode flags to
1771 * be the same as the fcntl ones and by sending down the bitwise OR of
1772 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody
1773 * pulls the rug out under this.
1774 */
1775 CTASSERT(O_NONBLOCK == IO_NDELAY);
1776 CTASSERT(O_FSYNC == IO_SYNC);
Cache object: b52c10e3097733af6f3e9924c9b529ec
|