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: stable/10/sys/fs/devfs/devfs_vnops.c 277746 2015-01-26 11:37:16Z 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
65static struct vop_vector devfs_vnodeops;
66static struct vop_vector devfs_specops;
67static 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
74static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
75
76struct mtx devfs_de_interlock;
77MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
78struct sx clone_drain_lock;
79SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
80struct mtx cdevpriv_mtx;
81MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
82
83static int
84devfs_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
102int
103devfs_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
121int
122devfs_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
151void
152devfs_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
163void
164devfs_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
176void
177devfs_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 */
190static int
191devfs_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
232static int
233devfs_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 }
298finished:
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 */
307char *
308devfs_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
344static int
345devfs_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
369static void
370devfs_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 */
387int
388devfs_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 }
402loop:
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
497static int
498devfs_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 */
523static int
524devfs_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
599static int
600devfs_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
623static int
624devfs_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
649static int
650devfs_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 */
723static int
724devfs_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 != 0) {
738 error = vnops.fo_ioctl(fp, com, data, cred, td);
739 return (error);
740 }
741
742 if (com == FIODTYPE) {
743 *(int *)data = dsw->d_flags & D_TYPEMASK;
744 td->td_fpop = fpop;
745 dev_relthread(dev, ref);
746 return (0);
747 } else if (com == FIODGNAME) {
748 fgn = data;
749 p = devtoname(dev);
750 i = strlen(p) + 1;
751 if (i > fgn->len)
752 error = EINVAL;
753 else
754 error = copyout(p, fgn->buf, i);
755 td->td_fpop = fpop;
756 dev_relthread(dev, ref);
757 return (error);
758 }
759 error = dsw->d_ioctl(dev, com, data, fp->f_flag, td);
760 td->td_fpop = NULL;
761 dev_relthread(dev, ref);
762 if (error == ENOIOCTL)
763 error = ENOTTY;
764 if (error == 0 && com == TIOCSCTTY) {
765 vp = fp->f_vnode;
766
767 /* Do nothing if reassigning same control tty */
768 sx_slock(&proctree_lock);
769 if (td->td_proc->p_session->s_ttyvp == vp) {
770 sx_sunlock(&proctree_lock);
771 return (0);
772 }
773
774 vpold = td->td_proc->p_session->s_ttyvp;
775 VREF(vp);
776 SESS_LOCK(td->td_proc->p_session);
777 td->td_proc->p_session->s_ttyvp = vp;
778 td->td_proc->p_session->s_ttydp = cdev2priv(dev);
779 SESS_UNLOCK(td->td_proc->p_session);
780
781 sx_sunlock(&proctree_lock);
782
783 /* Get rid of reference to old control tty */
784 if (vpold)
785 vrele(vpold);
786 }
787 return (error);
788}
789
790/* ARGSUSED */
791static int
792devfs_kqfilter_f(struct file *fp, struct knote *kn)
793{
794 struct cdev *dev;
795 struct cdevsw *dsw;
796 int error, ref;
797 struct file *fpop;
798 struct thread *td;
799
800 td = curthread;
801 fpop = td->td_fpop;
802 error = devfs_fp_check(fp, &dev, &dsw, &ref);
803 if (error)
804 return (error);
805 error = dsw->d_kqfilter(dev, kn);
806 td->td_fpop = fpop;
807 dev_relthread(dev, ref);
808 return (error);
809}
810
811static inline int
812devfs_prison_check(struct devfs_dirent *de, struct thread *td)
813{
814 struct cdev_priv *cdp;
815 struct ucred *dcr;
816 int error;
817
818 cdp = de->de_cdp;
819 if (cdp == NULL)
820 return (0);
821 dcr = cdp->cdp_c.si_cred;
822 if (dcr == NULL)
823 return (0);
824
825 error = prison_check(td->td_ucred, dcr);
826 if (error == 0)
827 return (0);
828 /* We do, however, allow access to the controlling terminal */
829 if (!(td->td_proc->p_flag & P_CONTROLT))
830 return (error);
831 if (td->td_proc->p_session->s_ttydp == cdp)
832 return (0);
833 return (error);
834}
835
836static int
837devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
838{
839 struct componentname *cnp;
840 struct vnode *dvp, **vpp;
841 struct thread *td;
842 struct devfs_dirent *de, *dd;
843 struct devfs_dirent **dde;
844 struct devfs_mount *dmp;
845 struct cdev *cdev;
846 int error, flags, nameiop, dvplocked;
847 char specname[SPECNAMELEN + 1], *pname;
848
849 cnp = ap->a_cnp;
850 vpp = ap->a_vpp;
851 dvp = ap->a_dvp;
852 pname = cnp->cn_nameptr;
853 td = cnp->cn_thread;
854 flags = cnp->cn_flags;
855 nameiop = cnp->cn_nameiop;
856 dmp = VFSTODEVFS(dvp->v_mount);
857 dd = dvp->v_data;
858 *vpp = NULLVP;
859
860 if ((flags & ISLASTCN) && nameiop == RENAME)
861 return (EOPNOTSUPP);
862
863 if (dvp->v_type != VDIR)
864 return (ENOTDIR);
865
866 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
867 return (EIO);
868
869 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
870 if (error)
871 return (error);
872
873 if (cnp->cn_namelen == 1 && *pname == '.') {
874 if ((flags & ISLASTCN) && nameiop != LOOKUP)
875 return (EINVAL);
876 *vpp = dvp;
877 VREF(dvp);
878 return (0);
879 }
880
881 if (flags & ISDOTDOT) {
882 if ((flags & ISLASTCN) && nameiop != LOOKUP)
883 return (EINVAL);
884 de = devfs_parent_dirent(dd);
885 if (de == NULL)
886 return (ENOENT);
887 dvplocked = VOP_ISLOCKED(dvp);
888 VOP_UNLOCK(dvp, 0);
889 error = devfs_allocv(de, dvp->v_mount,
890 cnp->cn_lkflags & LK_TYPE_MASK, vpp);
891 *dm_unlock = 0;
892 vn_lock(dvp, dvplocked | LK_RETRY);
893 return (error);
894 }
895
896 dd = dvp->v_data;
897 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
898 while (de == NULL) { /* While(...) so we can use break */
899
900 if (nameiop == DELETE)
901 return (ENOENT);
902
903 /*
904 * OK, we didn't have an entry for the name we were asked for
905 * so we try to see if anybody can create it on demand.
906 */
907 pname = devfs_fqpn(specname, dmp, dd, cnp);
908 if (pname == NULL)
909 break;
910
911 cdev = NULL;
912 DEVFS_DMP_HOLD(dmp);
913 sx_xunlock(&dmp->dm_lock);
914 sx_slock(&clone_drain_lock);
915 EVENTHANDLER_INVOKE(dev_clone,
916 td->td_ucred, pname, strlen(pname), &cdev);
917 sx_sunlock(&clone_drain_lock);
918
919 if (cdev == NULL)
920 sx_xlock(&dmp->dm_lock);
921 else if (devfs_populate_vp(dvp) != 0) {
922 *dm_unlock = 0;
923 sx_xlock(&dmp->dm_lock);
924 if (DEVFS_DMP_DROP(dmp)) {
925 sx_xunlock(&dmp->dm_lock);
926 devfs_unmount_final(dmp);
927 } else
928 sx_xunlock(&dmp->dm_lock);
929 dev_rel(cdev);
930 return (ENOENT);
931 }
932 if (DEVFS_DMP_DROP(dmp)) {
933 *dm_unlock = 0;
934 sx_xunlock(&dmp->dm_lock);
935 devfs_unmount_final(dmp);
936 if (cdev != NULL)
937 dev_rel(cdev);
938 return (ENOENT);
939 }
940
941 if (cdev == NULL)
942 break;
943
944 dev_lock();
945 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
946 if (dde != NULL && *dde != NULL)
947 de = *dde;
948 dev_unlock();
949 dev_rel(cdev);
950 break;
951 }
952
953 if (de == NULL || de->de_flags & DE_WHITEOUT) {
954 if ((nameiop == CREATE || nameiop == RENAME) &&
955 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
956 cnp->cn_flags |= SAVENAME;
957 return (EJUSTRETURN);
958 }
959 return (ENOENT);
960 }
961
962 if (devfs_prison_check(de, td))
963 return (ENOENT);
964
965 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
966 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
967 if (error)
968 return (error);
969 if (*vpp == dvp) {
970 VREF(dvp);
971 *vpp = dvp;
972 return (0);
973 }
974 }
975 error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK,
976 vpp);
977 *dm_unlock = 0;
978 return (error);
979}
980
981static int
982devfs_lookup(struct vop_lookup_args *ap)
983{
984 int j;
985 struct devfs_mount *dmp;
986 int dm_unlock;
987
988 if (devfs_populate_vp(ap->a_dvp) != 0)
989 return (ENOTDIR);
990
991 dmp = VFSTODEVFS(ap->a_dvp->v_mount);
992 dm_unlock = 1;
993 j = devfs_lookupx(ap, &dm_unlock);
994 if (dm_unlock == 1)
995 sx_xunlock(&dmp->dm_lock);
996 return (j);
997}
998
999static int
1000devfs_mknod(struct vop_mknod_args *ap)
1001{
1002 struct componentname *cnp;
1003 struct vnode *dvp, **vpp;
1004 struct devfs_dirent *dd, *de;
1005 struct devfs_mount *dmp;
1006 int error;
1007
1008 /*
1009 * The only type of node we should be creating here is a
1010 * character device, for anything else return EOPNOTSUPP.
1011 */
1012 if (ap->a_vap->va_type != VCHR)
1013 return (EOPNOTSUPP);
1014 dvp = ap->a_dvp;
1015 dmp = VFSTODEVFS(dvp->v_mount);
1016
1017 cnp = ap->a_cnp;
1018 vpp = ap->a_vpp;
1019 dd = dvp->v_data;
1020
1021 error = ENOENT;
1022 sx_xlock(&dmp->dm_lock);
1023 TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1024 if (cnp->cn_namelen != de->de_dirent->d_namlen)
1025 continue;
1026 if (de->de_dirent->d_type == DT_CHR &&
1027 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
1028 continue;
1029 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1030 de->de_dirent->d_namlen) != 0)
1031 continue;
1032 if (de->de_flags & DE_WHITEOUT)
1033 break;
1034 goto notfound;
1035 }
1036 if (de == NULL)
1037 goto notfound;
1038 de->de_flags &= ~DE_WHITEOUT;
1039 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1040 return (error);
1041notfound:
1042 sx_xunlock(&dmp->dm_lock);
1043 return (error);
1044}
1045
1046/* ARGSUSED */
1047static int
1048devfs_open(struct vop_open_args *ap)
1049{
1050 struct thread *td = ap->a_td;
1051 struct vnode *vp = ap->a_vp;
1052 struct cdev *dev = vp->v_rdev;
1053 struct file *fp = ap->a_fp;
1054 int error, ref, vlocked;
1055 struct cdevsw *dsw;
1056 struct file *fpop;
1057 struct mtx *mtxp;
1058
1059 if (vp->v_type == VBLK)
1060 return (ENXIO);
1061
1062 if (dev == NULL)
1063 return (ENXIO);
1064
1065 /* Make this field valid before any I/O in d_open. */
1066 if (dev->si_iosize_max == 0)
1067 dev->si_iosize_max = DFLTPHYS;
1068
1069 dsw = dev_refthread(dev, &ref);
1070 if (dsw == NULL)
1071 return (ENXIO);
1072 if (fp == NULL && dsw->d_fdopen != NULL) {
1073 dev_relthread(dev, ref);
1074 return (ENXIO);
1075 }
1076
1077 vlocked = VOP_ISLOCKED(vp);
1078 VOP_UNLOCK(vp, 0);
1079
1080 fpop = td->td_fpop;
1081 td->td_fpop = fp;
1082 if (fp != NULL) {
1083 fp->f_data = dev;
1084 fp->f_vnode = vp;
1085 }
1086 if (dsw->d_fdopen != NULL)
1087 error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1088 else
1089 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1090 /* cleanup any cdevpriv upon error */
1091 if (error != 0)
1092 devfs_clear_cdevpriv();
1093 td->td_fpop = fpop;
1094
1095 vn_lock(vp, vlocked | LK_RETRY);
1096 dev_relthread(dev, ref);
1097 if (error != 0) {
1098 if (error == ERESTART)
1099 error = EINTR;
1100 return (error);
1101 }
1102
1103#if 0 /* /dev/console */
1104 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1105#else
1106 if (fp == NULL)
1107 return (error);
1108#endif
1109 if (fp->f_ops == &badfileops)
1110 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1111 mtxp = mtx_pool_find(mtxpool_sleep, fp);
1112
1113 /*
1114 * Hint to the dofilewrite() to not force the buffer draining
1115 * on the writer to the file. Most likely, the write would
1116 * not need normal buffers.
1117 */
1118 mtx_lock(mtxp);
1119 fp->f_vnread_flags |= FDEVFS_VNODE;
1120 mtx_unlock(mtxp);
1121 return (error);
1122}
1123
1124static int
1125devfs_pathconf(struct vop_pathconf_args *ap)
1126{
1127
1128 switch (ap->a_name) {
1129 case _PC_MAC_PRESENT:
1130#ifdef MAC
1131 /*
1132 * If MAC is enabled, devfs automatically supports
1133 * trivial non-persistant label storage.
1134 */
1135 *ap->a_retval = 1;
1136#else
1137 *ap->a_retval = 0;
1138#endif
1139 return (0);
1140 default:
1141 return (vop_stdpathconf(ap));
1142 }
1143 /* NOTREACHED */
1144}
1145
1146/* ARGSUSED */
1147static int
1148devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1149{
1150 struct cdev *dev;
1151 struct cdevsw *dsw;
1152 int error, ref;
1153 struct file *fpop;
1154
1155 fpop = td->td_fpop;
1156 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1157 if (error != 0) {
1158 error = vnops.fo_poll(fp, events, cred, td);
1159 return (error);
1160 }
1161 error = dsw->d_poll(dev, events, td);
1162 td->td_fpop = fpop;
1163 dev_relthread(dev, ref);
1164 return(error);
1165}
1166
1167/*
1168 * Print out the contents of a special device vnode.
1169 */
1170static int
1171devfs_print(struct vop_print_args *ap)
1172{
1173
1174 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1175 return (0);
1176}
1177
1178static int
1179devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1180 int flags, struct thread *td)
1181{
1182 struct cdev *dev;
1183 int ioflag, error, ref;
1184 ssize_t resid;
1185 struct cdevsw *dsw;
1186 struct file *fpop;
1187
1188 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1189 return (EINVAL);
1190 fpop = td->td_fpop;
1191 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1192 if (error != 0) {
1193 error = vnops.fo_read(fp, uio, cred, flags, td);
1194 return (error);
1195 }
1196 resid = uio->uio_resid;
1197 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1198 if (ioflag & O_DIRECT)
1199 ioflag |= IO_DIRECT;
1200
1201 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1202 error = dsw->d_read(dev, uio, ioflag);
1203 if (uio->uio_resid != resid || (error == 0 && resid != 0))
1204 vfs_timestamp(&dev->si_atime);
1205 td->td_fpop = fpop;
1206 dev_relthread(dev, ref);
1207
1208 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1209 return (error);
1210}
1211
1212static int
1213devfs_readdir(struct vop_readdir_args *ap)
1214{
1215 int error;
1216 struct uio *uio;
1217 struct dirent *dp;
1218 struct devfs_dirent *dd;
1219 struct devfs_dirent *de;
1220 struct devfs_mount *dmp;
1221 off_t off;
1222 int *tmp_ncookies = NULL;
1223
1224 if (ap->a_vp->v_type != VDIR)
1225 return (ENOTDIR);
1226
1227 uio = ap->a_uio;
1228 if (uio->uio_offset < 0)
1229 return (EINVAL);
1230
1231 /*
1232 * XXX: This is a temporary hack to get around this filesystem not
1233 * supporting cookies. We store the location of the ncookies pointer
1234 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1235 * and set the number of cookies to 0. We then set the pointer to
1236 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1237 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1238 * pointer to its original location before returning to the caller.
1239 */
1240 if (ap->a_ncookies != NULL) {
1241 tmp_ncookies = ap->a_ncookies;
1242 *ap->a_ncookies = 0;
1243 ap->a_ncookies = NULL;
1244 }
1245
1246 dmp = VFSTODEVFS(ap->a_vp->v_mount);
1247 if (devfs_populate_vp(ap->a_vp) != 0) {
1248 if (tmp_ncookies != NULL)
1249 ap->a_ncookies = tmp_ncookies;
1250 return (EIO);
1251 }
1252 error = 0;
1253 de = ap->a_vp->v_data;
1254 off = 0;
1255 TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1256 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1257 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1258 continue;
1259 if (devfs_prison_check(dd, uio->uio_td))
1260 continue;
1261 if (dd->de_dirent->d_type == DT_DIR)
1262 de = dd->de_dir;
1263 else
1264 de = dd;
1265 dp = dd->de_dirent;
1266 if (dp->d_reclen > uio->uio_resid)
1267 break;
1268 dp->d_fileno = de->de_inode;
1269 if (off >= uio->uio_offset) {
1270 error = vfs_read_dirent(ap, dp, off);
1271 if (error)
1272 break;
1273 }
1274 off += dp->d_reclen;
1275 }
1276 sx_xunlock(&dmp->dm_lock);
1277 uio->uio_offset = off;
1278
1279 /*
1280 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1281 * place.
1282 */
1283 if (tmp_ncookies != NULL)
1284 ap->a_ncookies = tmp_ncookies;
1285
1286 return (error);
1287}
1288
1289static int
1290devfs_readlink(struct vop_readlink_args *ap)
1291{
1292 struct devfs_dirent *de;
1293
1294 de = ap->a_vp->v_data;
1295 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1296}
1297
1298static int
1299devfs_reclaim(struct vop_reclaim_args *ap)
1300{
1301 struct vnode *vp = ap->a_vp;
1302 struct devfs_dirent *de;
1303 struct cdev *dev;
1304
1305 mtx_lock(&devfs_de_interlock);
1306 de = vp->v_data;
1307 if (de != NULL) {
1308 de->de_vnode = NULL;
1309 vp->v_data = NULL;
1310 }
1311 mtx_unlock(&devfs_de_interlock);
1312
1313 vnode_destroy_vobject(vp);
1314
1315 VI_LOCK(vp);
1316 dev_lock();
1317 dev = vp->v_rdev;
1318 vp->v_rdev = NULL;
1319
1320 if (dev == NULL) {
1321 dev_unlock();
1322 VI_UNLOCK(vp);
1323 return (0);
1324 }
1325
1326 dev->si_usecount -= vp->v_usecount;
1327 dev_unlock();
1328 VI_UNLOCK(vp);
1329 dev_rel(dev);
1330 return (0);
1331}
1332
1333static int
1334devfs_remove(struct vop_remove_args *ap)
1335{
1336 struct vnode *dvp = ap->a_dvp;
1337 struct vnode *vp = ap->a_vp;
1338 struct devfs_dirent *dd;
1339 struct devfs_dirent *de, *de_covered;
1340 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1341
1342 ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1343 ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1344
1345 sx_xlock(&dmp->dm_lock);
1346 dd = ap->a_dvp->v_data;
1347 de = vp->v_data;
1348 if (de->de_cdp == NULL) {
1349 TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1350 if (de->de_dirent->d_type == DT_LNK) {
1351 de_covered = devfs_find(dd, de->de_dirent->d_name,
1352 de->de_dirent->d_namlen, 0);
1353 if (de_covered != NULL)
1354 de_covered->de_flags &= ~DE_COVERED;
1355 }
1356 /* We need to unlock dvp because devfs_delete() may lock it. */
1357 VOP_UNLOCK(vp, 0);
1358 if (dvp != vp)
1359 VOP_UNLOCK(dvp, 0);
1360 devfs_delete(dmp, de, 0);
1361 sx_xunlock(&dmp->dm_lock);
1362 if (dvp != vp)
1363 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1364 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1365 } else {
1366 de->de_flags |= DE_WHITEOUT;
1367 sx_xunlock(&dmp->dm_lock);
1368 }
1369 return (0);
1370}
1371
1372/*
1373 * Revoke is called on a tty when a terminal session ends. The vnode
1374 * is orphaned by setting v_op to deadfs so we need to let go of it
1375 * as well so that we create a new one next time around.
1376 *
1377 */
1378static int
1379devfs_revoke(struct vop_revoke_args *ap)
1380{
1381 struct vnode *vp = ap->a_vp, *vp2;
1382 struct cdev *dev;
1383 struct cdev_priv *cdp;
1384 struct devfs_dirent *de;
1385 int i;
1386
1387 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1388
1389 dev = vp->v_rdev;
1390 cdp = cdev2priv(dev);
1391
1392 dev_lock();
1393 cdp->cdp_inuse++;
1394 dev_unlock();
1395
1396 vhold(vp);
1397 vgone(vp);
1398 vdrop(vp);
1399
1400 VOP_UNLOCK(vp,0);
1401 loop:
1402 for (;;) {
1403 mtx_lock(&devfs_de_interlock);
1404 dev_lock();
1405 vp2 = NULL;
1406 for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1407 de = cdp->cdp_dirents[i];
1408 if (de == NULL)
1409 continue;
1410
1411 vp2 = de->de_vnode;
1412 if (vp2 != NULL) {
1413 dev_unlock();
1414 VI_LOCK(vp2);
1415 mtx_unlock(&devfs_de_interlock);
1416 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1417 curthread))
1418 goto loop;
1419 vhold(vp2);
1420 vgone(vp2);
1421 vdrop(vp2);
1422 vput(vp2);
1423 break;
1424 }
1425 }
1426 if (vp2 != NULL) {
1427 continue;
1428 }
1429 dev_unlock();
1430 mtx_unlock(&devfs_de_interlock);
1431 break;
1432 }
1433 dev_lock();
1434 cdp->cdp_inuse--;
1435 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1436 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1437 dev_unlock();
1438 dev_rel(&cdp->cdp_c);
1439 } else
1440 dev_unlock();
1441
1442 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1443 return (0);
1444}
1445
1446static int
1447devfs_rioctl(struct vop_ioctl_args *ap)
1448{
1449 struct vnode *vp;
1450 struct devfs_mount *dmp;
1451 int error;
1452
1453 vp = ap->a_vp;
1454 vn_lock(vp, LK_SHARED | LK_RETRY);
1455 if (vp->v_iflag & VI_DOOMED) {
1456 VOP_UNLOCK(vp, 0);
1457 return (EBADF);
1458 }
1459 dmp = VFSTODEVFS(vp->v_mount);
1460 sx_xlock(&dmp->dm_lock);
1461 VOP_UNLOCK(vp, 0);
1462 DEVFS_DMP_HOLD(dmp);
1463 devfs_populate(dmp);
1464 if (DEVFS_DMP_DROP(dmp)) {
1465 sx_xunlock(&dmp->dm_lock);
1466 devfs_unmount_final(dmp);
1467 return (ENOENT);
1468 }
1469 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1470 sx_xunlock(&dmp->dm_lock);
1471 return (error);
1472}
1473
1474static int
1475devfs_rread(struct vop_read_args *ap)
1476{
1477
1478 if (ap->a_vp->v_type != VDIR)
1479 return (EINVAL);
1480 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1481}
1482
1483static int
1484devfs_setattr(struct vop_setattr_args *ap)
1485{
1486 struct devfs_dirent *de;
1487 struct vattr *vap;
1488 struct vnode *vp;
1489 struct thread *td;
1490 int c, error;
1491 uid_t uid;
1492 gid_t gid;
1493
1494 vap = ap->a_vap;
1495 vp = ap->a_vp;
1496 td = curthread;
1497 if ((vap->va_type != VNON) ||
1498 (vap->va_nlink != VNOVAL) ||
1499 (vap->va_fsid != VNOVAL) ||
1500 (vap->va_fileid != VNOVAL) ||
1501 (vap->va_blocksize != VNOVAL) ||
1502 (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1503 (vap->va_rdev != VNOVAL) ||
1504 ((int)vap->va_bytes != VNOVAL) ||
1505 (vap->va_gen != VNOVAL)) {
1506 return (EINVAL);
1507 }
1508
1509 de = vp->v_data;
1510 if (vp->v_type == VDIR)
1511 de = de->de_dir;
1512
1513 error = c = 0;
1514 if (vap->va_uid == (uid_t)VNOVAL)
1515 uid = de->de_uid;
1516 else
1517 uid = vap->va_uid;
1518 if (vap->va_gid == (gid_t)VNOVAL)
1519 gid = de->de_gid;
1520 else
1521 gid = vap->va_gid;
1522 if (uid != de->de_uid || gid != de->de_gid) {
1523 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1524 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1525 error = priv_check(td, PRIV_VFS_CHOWN);
1526 if (error)
1527 return (error);
1528 }
1529 de->de_uid = uid;
1530 de->de_gid = gid;
1531 c = 1;
1532 }
1533
1534 if (vap->va_mode != (mode_t)VNOVAL) {
1535 if (ap->a_cred->cr_uid != de->de_uid) {
1536 error = priv_check(td, PRIV_VFS_ADMIN);
1537 if (error)
1538 return (error);
1539 }
1540 de->de_mode = vap->va_mode;
1541 c = 1;
1542 }
1543
1544 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1545 error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1546 if (error != 0)
1547 return (error);
1548 if (vap->va_atime.tv_sec != VNOVAL) {
1549 if (vp->v_type == VCHR)
1550 vp->v_rdev->si_atime = vap->va_atime;
1551 else
1552 de->de_atime = vap->va_atime;
1553 }
1554 if (vap->va_mtime.tv_sec != VNOVAL) {
1555 if (vp->v_type == VCHR)
1556 vp->v_rdev->si_mtime = vap->va_mtime;
1557 else
1558 de->de_mtime = vap->va_mtime;
1559 }
1560 c = 1;
1561 }
1562
1563 if (c) {
1564 if (vp->v_type == VCHR)
1565 vfs_timestamp(&vp->v_rdev->si_ctime);
1566 else
1567 vfs_timestamp(&de->de_mtime);
1568 }
1569 return (0);
1570}
1571
1572#ifdef MAC
1573static int
1574devfs_setlabel(struct vop_setlabel_args *ap)
1575{
1576 struct vnode *vp;
1577 struct devfs_dirent *de;
1578
1579 vp = ap->a_vp;
1580 de = vp->v_data;
1581
1582 mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1583 mac_devfs_update(vp->v_mount, de, vp);
1584
1585 return (0);
1586}
1587#endif
1588
1589static int
1590devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1591{
1592
1593 return (vnops.fo_stat(fp, sb, cred, td));
1594}
1595
1596static int
1597devfs_symlink(struct vop_symlink_args *ap)
1598{
1599 int i, error;
1600 struct devfs_dirent *dd;
1601 struct devfs_dirent *de, *de_covered, *de_dotdot;
1602 struct devfs_mount *dmp;
1603
1604 error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1605 if (error)
1606 return(error);
1607 dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1608 if (devfs_populate_vp(ap->a_dvp) != 0)
1609 return (ENOENT);
1610
1611 dd = ap->a_dvp->v_data;
1612 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1613 de->de_flags = DE_USER;
1614 de->de_uid = 0;
1615 de->de_gid = 0;
1616 de->de_mode = 0755;
1617 de->de_inode = alloc_unr(devfs_inos);
1618 de->de_dir = dd;
1619 de->de_dirent->d_type = DT_LNK;
1620 i = strlen(ap->a_target) + 1;
1621 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1622 bcopy(ap->a_target, de->de_symlink, i);
1623#ifdef MAC
1624 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1625#endif
1626 de_covered = devfs_find(dd, de->de_dirent->d_name,
1627 de->de_dirent->d_namlen, 0);
1628 if (de_covered != NULL) {
1629 if ((de_covered->de_flags & DE_USER) != 0) {
1630 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1631 sx_xunlock(&dmp->dm_lock);
1632 return (EEXIST);
1633 }
1634 KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1635 ("devfs_symlink: entry %p already covered", de_covered));
1636 de_covered->de_flags |= DE_COVERED;
1637 }
1638
1639 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */
1640 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */
1641 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1642 devfs_dir_ref_de(dmp, dd);
1643 devfs_rules_apply(dmp, de);
1644
1645 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1646}
1647
1648static int
1649devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1650{
1651
1652 return (vnops.fo_truncate(fp, length, cred, td));
1653}
1654
1655static int
1656devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1657 int flags, struct thread *td)
1658{
1659 struct cdev *dev;
1660 int error, ioflag, ref;
1661 ssize_t resid;
1662 struct cdevsw *dsw;
1663 struct file *fpop;
1664
1665 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1666 return (EINVAL);
1667 fpop = td->td_fpop;
1668 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1669 if (error != 0) {
1670 error = vnops.fo_write(fp, uio, cred, flags, td);
1671 return (error);
1672 }
1673 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1674 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1675 if (ioflag & O_DIRECT)
1676 ioflag |= IO_DIRECT;
1677 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1678
1679 resid = uio->uio_resid;
1680
1681 error = dsw->d_write(dev, uio, ioflag);
1682 if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1683 vfs_timestamp(&dev->si_ctime);
1684 dev->si_mtime = dev->si_ctime;
1685 }
1686 td->td_fpop = fpop;
1687 dev_relthread(dev, ref);
1688
1689 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1690 return (error);
1691}
1692
1693dev_t
1694dev2udev(struct cdev *x)
1695{
1696 if (x == NULL)
1697 return (NODEV);
1698 return (cdev2priv(x)->cdp_inode);
1699}
1700
1701static struct fileops devfs_ops_f = {
1702 .fo_read = devfs_read_f,
1703 .fo_write = devfs_write_f,
1704 .fo_truncate = devfs_truncate_f,
1705 .fo_ioctl = devfs_ioctl_f,
1706 .fo_poll = devfs_poll_f,
1707 .fo_kqfilter = devfs_kqfilter_f,
1708 .fo_stat = devfs_stat_f,
1709 .fo_close = devfs_close_f,
1710 .fo_chmod = vn_chmod,
1711 .fo_chown = vn_chown,
1712 .fo_sendfile = vn_sendfile,
1713 .fo_seek = vn_seek,
1714 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
1715};
1716
1717static struct vop_vector devfs_vnodeops = {
1718 .vop_default = &default_vnodeops,
1719
1720 .vop_access = devfs_access,
1721 .vop_getattr = devfs_getattr,
1722 .vop_ioctl = devfs_rioctl,
1723 .vop_lookup = devfs_lookup,
1724 .vop_mknod = devfs_mknod,
1725 .vop_pathconf = devfs_pathconf,
1726 .vop_read = devfs_rread,
1727 .vop_readdir = devfs_readdir,
1728 .vop_readlink = devfs_readlink,
1729 .vop_reclaim = devfs_reclaim,
1730 .vop_remove = devfs_remove,
1731 .vop_revoke = devfs_revoke,
1732 .vop_setattr = devfs_setattr,
1733#ifdef MAC
1734 .vop_setlabel = devfs_setlabel,
1735#endif
1736 .vop_symlink = devfs_symlink,
1737 .vop_vptocnp = devfs_vptocnp,
1738};
1739
1740static struct vop_vector devfs_specops = {
1741 .vop_default = &default_vnodeops,
1742
1743 .vop_access = devfs_access,
1744 .vop_bmap = VOP_PANIC,
1745 .vop_close = devfs_close,
1746 .vop_create = VOP_PANIC,
1747 .vop_fsync = devfs_fsync,
1748 .vop_getattr = devfs_getattr,
1749 .vop_link = VOP_PANIC,
1750 .vop_mkdir = VOP_PANIC,
1751 .vop_mknod = VOP_PANIC,
1752 .vop_open = devfs_open,
1753 .vop_pathconf = devfs_pathconf,
1754 .vop_poll = dead_poll,
1755 .vop_print = devfs_print,
1756 .vop_read = dead_read,
1757 .vop_readdir = VOP_PANIC,
1758 .vop_readlink = VOP_PANIC,
1759 .vop_reallocblks = VOP_PANIC,
1760 .vop_reclaim = devfs_reclaim,
1761 .vop_remove = devfs_remove,
1762 .vop_rename = VOP_PANIC,
1763 .vop_revoke = devfs_revoke,
1764 .vop_rmdir = VOP_PANIC,
1765 .vop_setattr = devfs_setattr,
1766#ifdef MAC
1767 .vop_setlabel = devfs_setlabel,
1768#endif
1769 .vop_strategy = VOP_PANIC,
1770 .vop_symlink = VOP_PANIC,
1771 .vop_vptocnp = devfs_vptocnp,
1772 .vop_write = dead_write,
1773};
1774
1775/*
1776 * Our calling convention to the device drivers used to be that we passed
1777 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1778 * flags instead since that's what open(), close() and ioctl() takes and
1779 * we don't really want vnode.h in device drivers.
1780 * We solved the source compatibility by redefining some vnode flags to
1781 * be the same as the fcntl ones and by sending down the bitwise OR of
1782 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody
1783 * pulls the rug out under this.
1784 */
1785CTASSERT(O_NONBLOCK == IO_NDELAY);
1786CTASSERT(O_FSYNC == IO_SYNC);
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: stable/10/sys/fs/devfs/devfs_vnops.c 277746 2015-01-26 11:37:16Z 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
65static struct vop_vector devfs_vnodeops;
66static struct vop_vector devfs_specops;
67static 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
74static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
75
76struct mtx devfs_de_interlock;
77MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
78struct sx clone_drain_lock;
79SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
80struct mtx cdevpriv_mtx;
81MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
82
83static int
84devfs_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
102int
103devfs_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
121int
122devfs_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
151void
152devfs_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
163void
164devfs_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
176void
177devfs_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 */
190static int
191devfs_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
232static int
233devfs_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 }
298finished:
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 */
307char *
308devfs_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
344static int
345devfs_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
369static void
370devfs_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 */
387int
388devfs_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 }
402loop:
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
497static int
498devfs_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 */
523static int
524devfs_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
599static int
600devfs_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
623static int
624devfs_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
649static int
650devfs_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 */
723static int
724devfs_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 != 0) {
738 error = vnops.fo_ioctl(fp, com, data, cred, td);
739 return (error);
740 }
741
742 if (com == FIODTYPE) {
743 *(int *)data = dsw->d_flags & D_TYPEMASK;
744 td->td_fpop = fpop;
745 dev_relthread(dev, ref);
746 return (0);
747 } else if (com == FIODGNAME) {
748 fgn = data;
749 p = devtoname(dev);
750 i = strlen(p) + 1;
751 if (i > fgn->len)
752 error = EINVAL;
753 else
754 error = copyout(p, fgn->buf, i);
755 td->td_fpop = fpop;
756 dev_relthread(dev, ref);
757 return (error);
758 }
759 error = dsw->d_ioctl(dev, com, data, fp->f_flag, td);
760 td->td_fpop = NULL;
761 dev_relthread(dev, ref);
762 if (error == ENOIOCTL)
763 error = ENOTTY;
764 if (error == 0 && com == TIOCSCTTY) {
765 vp = fp->f_vnode;
766
767 /* Do nothing if reassigning same control tty */
768 sx_slock(&proctree_lock);
769 if (td->td_proc->p_session->s_ttyvp == vp) {
770 sx_sunlock(&proctree_lock);
771 return (0);
772 }
773
774 vpold = td->td_proc->p_session->s_ttyvp;
775 VREF(vp);
776 SESS_LOCK(td->td_proc->p_session);
777 td->td_proc->p_session->s_ttyvp = vp;
778 td->td_proc->p_session->s_ttydp = cdev2priv(dev);
779 SESS_UNLOCK(td->td_proc->p_session);
780
781 sx_sunlock(&proctree_lock);
782
783 /* Get rid of reference to old control tty */
784 if (vpold)
785 vrele(vpold);
786 }
787 return (error);
788}
789
790/* ARGSUSED */
791static int
792devfs_kqfilter_f(struct file *fp, struct knote *kn)
793{
794 struct cdev *dev;
795 struct cdevsw *dsw;
796 int error, ref;
797 struct file *fpop;
798 struct thread *td;
799
800 td = curthread;
801 fpop = td->td_fpop;
802 error = devfs_fp_check(fp, &dev, &dsw, &ref);
803 if (error)
804 return (error);
805 error = dsw->d_kqfilter(dev, kn);
806 td->td_fpop = fpop;
807 dev_relthread(dev, ref);
808 return (error);
809}
810
811static inline int
812devfs_prison_check(struct devfs_dirent *de, struct thread *td)
813{
814 struct cdev_priv *cdp;
815 struct ucred *dcr;
816 int error;
817
818 cdp = de->de_cdp;
819 if (cdp == NULL)
820 return (0);
821 dcr = cdp->cdp_c.si_cred;
822 if (dcr == NULL)
823 return (0);
824
825 error = prison_check(td->td_ucred, dcr);
826 if (error == 0)
827 return (0);
828 /* We do, however, allow access to the controlling terminal */
829 if (!(td->td_proc->p_flag & P_CONTROLT))
830 return (error);
831 if (td->td_proc->p_session->s_ttydp == cdp)
832 return (0);
833 return (error);
834}
835
836static int
837devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
838{
839 struct componentname *cnp;
840 struct vnode *dvp, **vpp;
841 struct thread *td;
842 struct devfs_dirent *de, *dd;
843 struct devfs_dirent **dde;
844 struct devfs_mount *dmp;
845 struct cdev *cdev;
846 int error, flags, nameiop, dvplocked;
847 char specname[SPECNAMELEN + 1], *pname;
848
849 cnp = ap->a_cnp;
850 vpp = ap->a_vpp;
851 dvp = ap->a_dvp;
852 pname = cnp->cn_nameptr;
853 td = cnp->cn_thread;
854 flags = cnp->cn_flags;
855 nameiop = cnp->cn_nameiop;
856 dmp = VFSTODEVFS(dvp->v_mount);
857 dd = dvp->v_data;
858 *vpp = NULLVP;
859
860 if ((flags & ISLASTCN) && nameiop == RENAME)
861 return (EOPNOTSUPP);
862
863 if (dvp->v_type != VDIR)
864 return (ENOTDIR);
865
866 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
867 return (EIO);
868
869 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
870 if (error)
871 return (error);
872
873 if (cnp->cn_namelen == 1 && *pname == '.') {
874 if ((flags & ISLASTCN) && nameiop != LOOKUP)
875 return (EINVAL);
876 *vpp = dvp;
877 VREF(dvp);
878 return (0);
879 }
880
881 if (flags & ISDOTDOT) {
882 if ((flags & ISLASTCN) && nameiop != LOOKUP)
883 return (EINVAL);
884 de = devfs_parent_dirent(dd);
885 if (de == NULL)
886 return (ENOENT);
887 dvplocked = VOP_ISLOCKED(dvp);
888 VOP_UNLOCK(dvp, 0);
889 error = devfs_allocv(de, dvp->v_mount,
890 cnp->cn_lkflags & LK_TYPE_MASK, vpp);
891 *dm_unlock = 0;
892 vn_lock(dvp, dvplocked | LK_RETRY);
893 return (error);
894 }
895
896 dd = dvp->v_data;
897 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
898 while (de == NULL) { /* While(...) so we can use break */
899
900 if (nameiop == DELETE)
901 return (ENOENT);
902
903 /*
904 * OK, we didn't have an entry for the name we were asked for
905 * so we try to see if anybody can create it on demand.
906 */
907 pname = devfs_fqpn(specname, dmp, dd, cnp);
908 if (pname == NULL)
909 break;
910
911 cdev = NULL;
912 DEVFS_DMP_HOLD(dmp);
913 sx_xunlock(&dmp->dm_lock);
914 sx_slock(&clone_drain_lock);
915 EVENTHANDLER_INVOKE(dev_clone,
916 td->td_ucred, pname, strlen(pname), &cdev);
917 sx_sunlock(&clone_drain_lock);
918
919 if (cdev == NULL)
920 sx_xlock(&dmp->dm_lock);
921 else if (devfs_populate_vp(dvp) != 0) {
922 *dm_unlock = 0;
923 sx_xlock(&dmp->dm_lock);
924 if (DEVFS_DMP_DROP(dmp)) {
925 sx_xunlock(&dmp->dm_lock);
926 devfs_unmount_final(dmp);
927 } else
928 sx_xunlock(&dmp->dm_lock);
929 dev_rel(cdev);
930 return (ENOENT);
931 }
932 if (DEVFS_DMP_DROP(dmp)) {
933 *dm_unlock = 0;
934 sx_xunlock(&dmp->dm_lock);
935 devfs_unmount_final(dmp);
936 if (cdev != NULL)
937 dev_rel(cdev);
938 return (ENOENT);
939 }
940
941 if (cdev == NULL)
942 break;
943
944 dev_lock();
945 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
946 if (dde != NULL && *dde != NULL)
947 de = *dde;
948 dev_unlock();
949 dev_rel(cdev);
950 break;
951 }
952
953 if (de == NULL || de->de_flags & DE_WHITEOUT) {
954 if ((nameiop == CREATE || nameiop == RENAME) &&
955 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
956 cnp->cn_flags |= SAVENAME;
957 return (EJUSTRETURN);
958 }
959 return (ENOENT);
960 }
961
962 if (devfs_prison_check(de, td))
963 return (ENOENT);
964
965 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
966 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
967 if (error)
968 return (error);
969 if (*vpp == dvp) {
970 VREF(dvp);
971 *vpp = dvp;
972 return (0);
973 }
974 }
975 error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK,
976 vpp);
977 *dm_unlock = 0;
978 return (error);
979}
980
981static int
982devfs_lookup(struct vop_lookup_args *ap)
983{
984 int j;
985 struct devfs_mount *dmp;
986 int dm_unlock;
987
988 if (devfs_populate_vp(ap->a_dvp) != 0)
989 return (ENOTDIR);
990
991 dmp = VFSTODEVFS(ap->a_dvp->v_mount);
992 dm_unlock = 1;
993 j = devfs_lookupx(ap, &dm_unlock);
994 if (dm_unlock == 1)
995 sx_xunlock(&dmp->dm_lock);
996 return (j);
997}
998
999static int
1000devfs_mknod(struct vop_mknod_args *ap)
1001{
1002 struct componentname *cnp;
1003 struct vnode *dvp, **vpp;
1004 struct devfs_dirent *dd, *de;
1005 struct devfs_mount *dmp;
1006 int error;
1007
1008 /*
1009 * The only type of node we should be creating here is a
1010 * character device, for anything else return EOPNOTSUPP.
1011 */
1012 if (ap->a_vap->va_type != VCHR)
1013 return (EOPNOTSUPP);
1014 dvp = ap->a_dvp;
1015 dmp = VFSTODEVFS(dvp->v_mount);
1016
1017 cnp = ap->a_cnp;
1018 vpp = ap->a_vpp;
1019 dd = dvp->v_data;
1020
1021 error = ENOENT;
1022 sx_xlock(&dmp->dm_lock);
1023 TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1024 if (cnp->cn_namelen != de->de_dirent->d_namlen)
1025 continue;
1026 if (de->de_dirent->d_type == DT_CHR &&
1027 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
1028 continue;
1029 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1030 de->de_dirent->d_namlen) != 0)
1031 continue;
1032 if (de->de_flags & DE_WHITEOUT)
1033 break;
1034 goto notfound;
1035 }
1036 if (de == NULL)
1037 goto notfound;
1038 de->de_flags &= ~DE_WHITEOUT;
1039 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1040 return (error);
1041notfound:
1042 sx_xunlock(&dmp->dm_lock);
1043 return (error);
1044}
1045
1046/* ARGSUSED */
1047static int
1048devfs_open(struct vop_open_args *ap)
1049{
1050 struct thread *td = ap->a_td;
1051 struct vnode *vp = ap->a_vp;
1052 struct cdev *dev = vp->v_rdev;
1053 struct file *fp = ap->a_fp;
1054 int error, ref, vlocked;
1055 struct cdevsw *dsw;
1056 struct file *fpop;
1057 struct mtx *mtxp;
1058
1059 if (vp->v_type == VBLK)
1060 return (ENXIO);
1061
1062 if (dev == NULL)
1063 return (ENXIO);
1064
1065 /* Make this field valid before any I/O in d_open. */
1066 if (dev->si_iosize_max == 0)
1067 dev->si_iosize_max = DFLTPHYS;
1068
1069 dsw = dev_refthread(dev, &ref);
1070 if (dsw == NULL)
1071 return (ENXIO);
1072 if (fp == NULL && dsw->d_fdopen != NULL) {
1073 dev_relthread(dev, ref);
1074 return (ENXIO);
1075 }
1076
1077 vlocked = VOP_ISLOCKED(vp);
1078 VOP_UNLOCK(vp, 0);
1079
1080 fpop = td->td_fpop;
1081 td->td_fpop = fp;
1082 if (fp != NULL) {
1083 fp->f_data = dev;
1084 fp->f_vnode = vp;
1085 }
1086 if (dsw->d_fdopen != NULL)
1087 error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1088 else
1089 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1090 /* cleanup any cdevpriv upon error */
1091 if (error != 0)
1092 devfs_clear_cdevpriv();
1093 td->td_fpop = fpop;
1094
1095 vn_lock(vp, vlocked | LK_RETRY);
1096 dev_relthread(dev, ref);
1097 if (error != 0) {
1098 if (error == ERESTART)
1099 error = EINTR;
1100 return (error);
1101 }
1102
1103#if 0 /* /dev/console */
1104 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1105#else
1106 if (fp == NULL)
1107 return (error);
1108#endif
1109 if (fp->f_ops == &badfileops)
1110 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1111 mtxp = mtx_pool_find(mtxpool_sleep, fp);
1112
1113 /*
1114 * Hint to the dofilewrite() to not force the buffer draining
1115 * on the writer to the file. Most likely, the write would
1116 * not need normal buffers.
1117 */
1118 mtx_lock(mtxp);
1119 fp->f_vnread_flags |= FDEVFS_VNODE;
1120 mtx_unlock(mtxp);
1121 return (error);
1122}
1123
1124static int
1125devfs_pathconf(struct vop_pathconf_args *ap)
1126{
1127
1128 switch (ap->a_name) {
1129 case _PC_MAC_PRESENT:
1130#ifdef MAC
1131 /*
1132 * If MAC is enabled, devfs automatically supports
1133 * trivial non-persistant label storage.
1134 */
1135 *ap->a_retval = 1;
1136#else
1137 *ap->a_retval = 0;
1138#endif
1139 return (0);
1140 default:
1141 return (vop_stdpathconf(ap));
1142 }
1143 /* NOTREACHED */
1144}
1145
1146/* ARGSUSED */
1147static int
1148devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1149{
1150 struct cdev *dev;
1151 struct cdevsw *dsw;
1152 int error, ref;
1153 struct file *fpop;
1154
1155 fpop = td->td_fpop;
1156 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1157 if (error != 0) {
1158 error = vnops.fo_poll(fp, events, cred, td);
1159 return (error);
1160 }
1161 error = dsw->d_poll(dev, events, td);
1162 td->td_fpop = fpop;
1163 dev_relthread(dev, ref);
1164 return(error);
1165}
1166
1167/*
1168 * Print out the contents of a special device vnode.
1169 */
1170static int
1171devfs_print(struct vop_print_args *ap)
1172{
1173
1174 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1175 return (0);
1176}
1177
1178static int
1179devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1180 int flags, struct thread *td)
1181{
1182 struct cdev *dev;
1183 int ioflag, error, ref;
1184 ssize_t resid;
1185 struct cdevsw *dsw;
1186 struct file *fpop;
1187
1188 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1189 return (EINVAL);
1190 fpop = td->td_fpop;
1191 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1192 if (error != 0) {
1193 error = vnops.fo_read(fp, uio, cred, flags, td);
1194 return (error);
1195 }
1196 resid = uio->uio_resid;
1197 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1198 if (ioflag & O_DIRECT)
1199 ioflag |= IO_DIRECT;
1200
1201 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1202 error = dsw->d_read(dev, uio, ioflag);
1203 if (uio->uio_resid != resid || (error == 0 && resid != 0))
1204 vfs_timestamp(&dev->si_atime);
1205 td->td_fpop = fpop;
1206 dev_relthread(dev, ref);
1207
1208 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1209 return (error);
1210}
1211
1212static int
1213devfs_readdir(struct vop_readdir_args *ap)
1214{
1215 int error;
1216 struct uio *uio;
1217 struct dirent *dp;
1218 struct devfs_dirent *dd;
1219 struct devfs_dirent *de;
1220 struct devfs_mount *dmp;
1221 off_t off;
1222 int *tmp_ncookies = NULL;
1223
1224 if (ap->a_vp->v_type != VDIR)
1225 return (ENOTDIR);
1226
1227 uio = ap->a_uio;
1228 if (uio->uio_offset < 0)
1229 return (EINVAL);
1230
1231 /*
1232 * XXX: This is a temporary hack to get around this filesystem not
1233 * supporting cookies. We store the location of the ncookies pointer
1234 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1235 * and set the number of cookies to 0. We then set the pointer to
1236 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1237 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1238 * pointer to its original location before returning to the caller.
1239 */
1240 if (ap->a_ncookies != NULL) {
1241 tmp_ncookies = ap->a_ncookies;
1242 *ap->a_ncookies = 0;
1243 ap->a_ncookies = NULL;
1244 }
1245
1246 dmp = VFSTODEVFS(ap->a_vp->v_mount);
1247 if (devfs_populate_vp(ap->a_vp) != 0) {
1248 if (tmp_ncookies != NULL)
1249 ap->a_ncookies = tmp_ncookies;
1250 return (EIO);
1251 }
1252 error = 0;
1253 de = ap->a_vp->v_data;
1254 off = 0;
1255 TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1256 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1257 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1258 continue;
1259 if (devfs_prison_check(dd, uio->uio_td))
1260 continue;
1261 if (dd->de_dirent->d_type == DT_DIR)
1262 de = dd->de_dir;
1263 else
1264 de = dd;
1265 dp = dd->de_dirent;
1266 if (dp->d_reclen > uio->uio_resid)
1267 break;
1268 dp->d_fileno = de->de_inode;
1269 if (off >= uio->uio_offset) {
1270 error = vfs_read_dirent(ap, dp, off);
1271 if (error)
1272 break;
1273 }
1274 off += dp->d_reclen;
1275 }
1276 sx_xunlock(&dmp->dm_lock);
1277 uio->uio_offset = off;
1278
1279 /*
1280 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1281 * place.
1282 */
1283 if (tmp_ncookies != NULL)
1284 ap->a_ncookies = tmp_ncookies;
1285
1286 return (error);
1287}
1288
1289static int
1290devfs_readlink(struct vop_readlink_args *ap)
1291{
1292 struct devfs_dirent *de;
1293
1294 de = ap->a_vp->v_data;
1295 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1296}
1297
1298static int
1299devfs_reclaim(struct vop_reclaim_args *ap)
1300{
1301 struct vnode *vp = ap->a_vp;
1302 struct devfs_dirent *de;
1303 struct cdev *dev;
1304
1305 mtx_lock(&devfs_de_interlock);
1306 de = vp->v_data;
1307 if (de != NULL) {
1308 de->de_vnode = NULL;
1309 vp->v_data = NULL;
1310 }
1311 mtx_unlock(&devfs_de_interlock);
1312
1313 vnode_destroy_vobject(vp);
1314
1315 VI_LOCK(vp);
1316 dev_lock();
1317 dev = vp->v_rdev;
1318 vp->v_rdev = NULL;
1319
1320 if (dev == NULL) {
1321 dev_unlock();
1322 VI_UNLOCK(vp);
1323 return (0);
1324 }
1325
1326 dev->si_usecount -= vp->v_usecount;
1327 dev_unlock();
1328 VI_UNLOCK(vp);
1329 dev_rel(dev);
1330 return (0);
1331}
1332
1333static int
1334devfs_remove(struct vop_remove_args *ap)
1335{
1336 struct vnode *dvp = ap->a_dvp;
1337 struct vnode *vp = ap->a_vp;
1338 struct devfs_dirent *dd;
1339 struct devfs_dirent *de, *de_covered;
1340 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1341
1342 ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1343 ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1344
1345 sx_xlock(&dmp->dm_lock);
1346 dd = ap->a_dvp->v_data;
1347 de = vp->v_data;
1348 if (de->de_cdp == NULL) {
1349 TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1350 if (de->de_dirent->d_type == DT_LNK) {
1351 de_covered = devfs_find(dd, de->de_dirent->d_name,
1352 de->de_dirent->d_namlen, 0);
1353 if (de_covered != NULL)
1354 de_covered->de_flags &= ~DE_COVERED;
1355 }
1356 /* We need to unlock dvp because devfs_delete() may lock it. */
1357 VOP_UNLOCK(vp, 0);
1358 if (dvp != vp)
1359 VOP_UNLOCK(dvp, 0);
1360 devfs_delete(dmp, de, 0);
1361 sx_xunlock(&dmp->dm_lock);
1362 if (dvp != vp)
1363 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1364 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1365 } else {
1366 de->de_flags |= DE_WHITEOUT;
1367 sx_xunlock(&dmp->dm_lock);
1368 }
1369 return (0);
1370}
1371
1372/*
1373 * Revoke is called on a tty when a terminal session ends. The vnode
1374 * is orphaned by setting v_op to deadfs so we need to let go of it
1375 * as well so that we create a new one next time around.
1376 *
1377 */
1378static int
1379devfs_revoke(struct vop_revoke_args *ap)
1380{
1381 struct vnode *vp = ap->a_vp, *vp2;
1382 struct cdev *dev;
1383 struct cdev_priv *cdp;
1384 struct devfs_dirent *de;
1385 int i;
1386
1387 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1388
1389 dev = vp->v_rdev;
1390 cdp = cdev2priv(dev);
1391
1392 dev_lock();
1393 cdp->cdp_inuse++;
1394 dev_unlock();
1395
1396 vhold(vp);
1397 vgone(vp);
1398 vdrop(vp);
1399
1400 VOP_UNLOCK(vp,0);
1401 loop:
1402 for (;;) {
1403 mtx_lock(&devfs_de_interlock);
1404 dev_lock();
1405 vp2 = NULL;
1406 for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1407 de = cdp->cdp_dirents[i];
1408 if (de == NULL)
1409 continue;
1410
1411 vp2 = de->de_vnode;
1412 if (vp2 != NULL) {
1413 dev_unlock();
1414 VI_LOCK(vp2);
1415 mtx_unlock(&devfs_de_interlock);
1416 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1417 curthread))
1418 goto loop;
1419 vhold(vp2);
1420 vgone(vp2);
1421 vdrop(vp2);
1422 vput(vp2);
1423 break;
1424 }
1425 }
1426 if (vp2 != NULL) {
1427 continue;
1428 }
1429 dev_unlock();
1430 mtx_unlock(&devfs_de_interlock);
1431 break;
1432 }
1433 dev_lock();
1434 cdp->cdp_inuse--;
1435 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1436 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1437 dev_unlock();
1438 dev_rel(&cdp->cdp_c);
1439 } else
1440 dev_unlock();
1441
1442 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1443 return (0);
1444}
1445
1446static int
1447devfs_rioctl(struct vop_ioctl_args *ap)
1448{
1449 struct vnode *vp;
1450 struct devfs_mount *dmp;
1451 int error;
1452
1453 vp = ap->a_vp;
1454 vn_lock(vp, LK_SHARED | LK_RETRY);
1455 if (vp->v_iflag & VI_DOOMED) {
1456 VOP_UNLOCK(vp, 0);
1457 return (EBADF);
1458 }
1459 dmp = VFSTODEVFS(vp->v_mount);
1460 sx_xlock(&dmp->dm_lock);
1461 VOP_UNLOCK(vp, 0);
1462 DEVFS_DMP_HOLD(dmp);
1463 devfs_populate(dmp);
1464 if (DEVFS_DMP_DROP(dmp)) {
1465 sx_xunlock(&dmp->dm_lock);
1466 devfs_unmount_final(dmp);
1467 return (ENOENT);
1468 }
1469 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1470 sx_xunlock(&dmp->dm_lock);
1471 return (error);
1472}
1473
1474static int
1475devfs_rread(struct vop_read_args *ap)
1476{
1477
1478 if (ap->a_vp->v_type != VDIR)
1479 return (EINVAL);
1480 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1481}
1482
1483static int
1484devfs_setattr(struct vop_setattr_args *ap)
1485{
1486 struct devfs_dirent *de;
1487 struct vattr *vap;
1488 struct vnode *vp;
1489 struct thread *td;
1490 int c, error;
1491 uid_t uid;
1492 gid_t gid;
1493
1494 vap = ap->a_vap;
1495 vp = ap->a_vp;
1496 td = curthread;
1497 if ((vap->va_type != VNON) ||
1498 (vap->va_nlink != VNOVAL) ||
1499 (vap->va_fsid != VNOVAL) ||
1500 (vap->va_fileid != VNOVAL) ||
1501 (vap->va_blocksize != VNOVAL) ||
1502 (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1503 (vap->va_rdev != VNOVAL) ||
1504 ((int)vap->va_bytes != VNOVAL) ||
1505 (vap->va_gen != VNOVAL)) {
1506 return (EINVAL);
1507 }
1508
1509 de = vp->v_data;
1510 if (vp->v_type == VDIR)
1511 de = de->de_dir;
1512
1513 error = c = 0;
1514 if (vap->va_uid == (uid_t)VNOVAL)
1515 uid = de->de_uid;
1516 else
1517 uid = vap->va_uid;
1518 if (vap->va_gid == (gid_t)VNOVAL)
1519 gid = de->de_gid;
1520 else
1521 gid = vap->va_gid;
1522 if (uid != de->de_uid || gid != de->de_gid) {
1523 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1524 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1525 error = priv_check(td, PRIV_VFS_CHOWN);
1526 if (error)
1527 return (error);
1528 }
1529 de->de_uid = uid;
1530 de->de_gid = gid;
1531 c = 1;
1532 }
1533
1534 if (vap->va_mode != (mode_t)VNOVAL) {
1535 if (ap->a_cred->cr_uid != de->de_uid) {
1536 error = priv_check(td, PRIV_VFS_ADMIN);
1537 if (error)
1538 return (error);
1539 }
1540 de->de_mode = vap->va_mode;
1541 c = 1;
1542 }
1543
1544 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1545 error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1546 if (error != 0)
1547 return (error);
1548 if (vap->va_atime.tv_sec != VNOVAL) {
1549 if (vp->v_type == VCHR)
1550 vp->v_rdev->si_atime = vap->va_atime;
1551 else
1552 de->de_atime = vap->va_atime;
1553 }
1554 if (vap->va_mtime.tv_sec != VNOVAL) {
1555 if (vp->v_type == VCHR)
1556 vp->v_rdev->si_mtime = vap->va_mtime;
1557 else
1558 de->de_mtime = vap->va_mtime;
1559 }
1560 c = 1;
1561 }
1562
1563 if (c) {
1564 if (vp->v_type == VCHR)
1565 vfs_timestamp(&vp->v_rdev->si_ctime);
1566 else
1567 vfs_timestamp(&de->de_mtime);
1568 }
1569 return (0);
1570}
1571
1572#ifdef MAC
1573static int
1574devfs_setlabel(struct vop_setlabel_args *ap)
1575{
1576 struct vnode *vp;
1577 struct devfs_dirent *de;
1578
1579 vp = ap->a_vp;
1580 de = vp->v_data;
1581
1582 mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1583 mac_devfs_update(vp->v_mount, de, vp);
1584
1585 return (0);
1586}
1587#endif
1588
1589static int
1590devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1591{
1592
1593 return (vnops.fo_stat(fp, sb, cred, td));
1594}
1595
1596static int
1597devfs_symlink(struct vop_symlink_args *ap)
1598{
1599 int i, error;
1600 struct devfs_dirent *dd;
1601 struct devfs_dirent *de, *de_covered, *de_dotdot;
1602 struct devfs_mount *dmp;
1603
1604 error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1605 if (error)
1606 return(error);
1607 dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1608 if (devfs_populate_vp(ap->a_dvp) != 0)
1609 return (ENOENT);
1610
1611 dd = ap->a_dvp->v_data;
1612 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1613 de->de_flags = DE_USER;
1614 de->de_uid = 0;
1615 de->de_gid = 0;
1616 de->de_mode = 0755;
1617 de->de_inode = alloc_unr(devfs_inos);
1618 de->de_dir = dd;
1619 de->de_dirent->d_type = DT_LNK;
1620 i = strlen(ap->a_target) + 1;
1621 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1622 bcopy(ap->a_target, de->de_symlink, i);
1623#ifdef MAC
1624 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1625#endif
1626 de_covered = devfs_find(dd, de->de_dirent->d_name,
1627 de->de_dirent->d_namlen, 0);
1628 if (de_covered != NULL) {
1629 if ((de_covered->de_flags & DE_USER) != 0) {
1630 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1631 sx_xunlock(&dmp->dm_lock);
1632 return (EEXIST);
1633 }
1634 KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1635 ("devfs_symlink: entry %p already covered", de_covered));
1636 de_covered->de_flags |= DE_COVERED;
1637 }
1638
1639 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */
1640 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */
1641 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1642 devfs_dir_ref_de(dmp, dd);
1643 devfs_rules_apply(dmp, de);
1644
1645 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1646}
1647
1648static int
1649devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1650{
1651
1652 return (vnops.fo_truncate(fp, length, cred, td));
1653}
1654
1655static int
1656devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1657 int flags, struct thread *td)
1658{
1659 struct cdev *dev;
1660 int error, ioflag, ref;
1661 ssize_t resid;
1662 struct cdevsw *dsw;
1663 struct file *fpop;
1664
1665 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1666 return (EINVAL);
1667 fpop = td->td_fpop;
1668 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1669 if (error != 0) {
1670 error = vnops.fo_write(fp, uio, cred, flags, td);
1671 return (error);
1672 }
1673 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1674 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1675 if (ioflag & O_DIRECT)
1676 ioflag |= IO_DIRECT;
1677 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1678
1679 resid = uio->uio_resid;
1680
1681 error = dsw->d_write(dev, uio, ioflag);
1682 if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1683 vfs_timestamp(&dev->si_ctime);
1684 dev->si_mtime = dev->si_ctime;
1685 }
1686 td->td_fpop = fpop;
1687 dev_relthread(dev, ref);
1688
1689 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1690 return (error);
1691}
1692
1693dev_t
1694dev2udev(struct cdev *x)
1695{
1696 if (x == NULL)
1697 return (NODEV);
1698 return (cdev2priv(x)->cdp_inode);
1699}
1700
1701static struct fileops devfs_ops_f = {
1702 .fo_read = devfs_read_f,
1703 .fo_write = devfs_write_f,
1704 .fo_truncate = devfs_truncate_f,
1705 .fo_ioctl = devfs_ioctl_f,
1706 .fo_poll = devfs_poll_f,
1707 .fo_kqfilter = devfs_kqfilter_f,
1708 .fo_stat = devfs_stat_f,
1709 .fo_close = devfs_close_f,
1710 .fo_chmod = vn_chmod,
1711 .fo_chown = vn_chown,
1712 .fo_sendfile = vn_sendfile,
1713 .fo_seek = vn_seek,
1714 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
1715};
1716
1717static struct vop_vector devfs_vnodeops = {
1718 .vop_default = &default_vnodeops,
1719
1720 .vop_access = devfs_access,
1721 .vop_getattr = devfs_getattr,
1722 .vop_ioctl = devfs_rioctl,
1723 .vop_lookup = devfs_lookup,
1724 .vop_mknod = devfs_mknod,
1725 .vop_pathconf = devfs_pathconf,
1726 .vop_read = devfs_rread,
1727 .vop_readdir = devfs_readdir,
1728 .vop_readlink = devfs_readlink,
1729 .vop_reclaim = devfs_reclaim,
1730 .vop_remove = devfs_remove,
1731 .vop_revoke = devfs_revoke,
1732 .vop_setattr = devfs_setattr,
1733#ifdef MAC
1734 .vop_setlabel = devfs_setlabel,
1735#endif
1736 .vop_symlink = devfs_symlink,
1737 .vop_vptocnp = devfs_vptocnp,
1738};
1739
1740static struct vop_vector devfs_specops = {
1741 .vop_default = &default_vnodeops,
1742
1743 .vop_access = devfs_access,
1744 .vop_bmap = VOP_PANIC,
1745 .vop_close = devfs_close,
1746 .vop_create = VOP_PANIC,
1747 .vop_fsync = devfs_fsync,
1748 .vop_getattr = devfs_getattr,
1749 .vop_link = VOP_PANIC,
1750 .vop_mkdir = VOP_PANIC,
1751 .vop_mknod = VOP_PANIC,
1752 .vop_open = devfs_open,
1753 .vop_pathconf = devfs_pathconf,
1754 .vop_poll = dead_poll,
1755 .vop_print = devfs_print,
1756 .vop_read = dead_read,
1757 .vop_readdir = VOP_PANIC,
1758 .vop_readlink = VOP_PANIC,
1759 .vop_reallocblks = VOP_PANIC,
1760 .vop_reclaim = devfs_reclaim,
1761 .vop_remove = devfs_remove,
1762 .vop_rename = VOP_PANIC,
1763 .vop_revoke = devfs_revoke,
1764 .vop_rmdir = VOP_PANIC,
1765 .vop_setattr = devfs_setattr,
1766#ifdef MAC
1767 .vop_setlabel = devfs_setlabel,
1768#endif
1769 .vop_strategy = VOP_PANIC,
1770 .vop_symlink = VOP_PANIC,
1771 .vop_vptocnp = devfs_vptocnp,
1772 .vop_write = dead_write,
1773};
1774
1775/*
1776 * Our calling convention to the device drivers used to be that we passed
1777 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1778 * flags instead since that's what open(), close() and ioctl() takes and
1779 * we don't really want vnode.h in device drivers.
1780 * We solved the source compatibility by redefining some vnode flags to
1781 * be the same as the fcntl ones and by sending down the bitwise OR of
1782 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody
1783 * pulls the rug out under this.
1784 */
1785CTASSERT(O_NONBLOCK == IO_NDELAY);
1786CTASSERT(O_FSYNC == IO_SYNC);