2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2000-2004
5 * Poul-Henning Kamp. All rights reserved.
6 * Copyright (c) 1989, 1992-1993, 1995
7 * The Regents of the University of California. All rights reserved.
9 * This code is derived from software donated to Berkeley by
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)kernfs_vnops.c 8.15 (Berkeley) 5/21/95
34 * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vnops.c 1.43
44 #include <sys/param.h>
45 #include <sys/systm.h>
47 #include <sys/dirent.h>
48 #include <sys/fcntl.h>
50 #include <sys/filedesc.h>
51 #include <sys/filio.h>
53 #include <sys/kernel.h>
54 #include <sys/limits.h>
56 #include <sys/malloc.h>
58 #include <sys/mount.h>
59 #include <sys/namei.h>
64 #include <sys/sysctl.h>
66 #include <sys/ttycom.h>
67 #include <sys/unistd.h>
68 #include <sys/vnode.h>
70 static struct vop_vector devfs_vnodeops;
71 static struct vop_vector devfs_specops;
72 static struct fileops devfs_ops_f;
74 #include <fs/devfs/devfs.h>
75 #include <fs/devfs/devfs_int.h>
77 #include <security/mac/mac_framework.h>
80 #include <vm/vm_extern.h>
81 #include <vm/vm_object.h>
83 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
85 struct mtx devfs_de_interlock;
86 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
87 struct sx clone_drain_lock;
88 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
89 struct mtx cdevpriv_mtx;
90 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
92 SYSCTL_DECL(_vfs_devfs);
94 static int devfs_dotimes;
95 SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW,
96 &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision");
99 * Update devfs node timestamp. Note that updates are unlocked and
100 * stat(2) could see partially updated times.
103 devfs_timestamp(struct timespec *tsp)
111 if (tsp->tv_sec != ts) {
119 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
123 *dswp = devvn_refthread(fp->f_vnode, devp, ref);
124 if (*devp != fp->f_data) {
126 dev_relthread(*devp, *ref);
129 KASSERT((*devp)->si_refcount > 0,
130 ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
133 curthread->td_fpop = fp;
138 devfs_get_cdevpriv(void **datap)
141 struct cdev_privdata *p;
144 fp = curthread->td_fpop;
150 *datap = p->cdpd_data;
157 devfs_set_cdevpriv(void *priv, d_priv_dtor_t *priv_dtr)
160 struct cdev_priv *cdp;
161 struct cdev_privdata *p;
164 fp = curthread->td_fpop;
167 cdp = cdev2priv((struct cdev *)fp->f_data);
168 p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
170 p->cdpd_dtr = priv_dtr;
172 mtx_lock(&cdevpriv_mtx);
173 if (fp->f_cdevpriv == NULL) {
174 LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
176 mtx_unlock(&cdevpriv_mtx);
179 mtx_unlock(&cdevpriv_mtx);
180 free(p, M_CDEVPDATA);
187 devfs_destroy_cdevpriv(struct cdev_privdata *p)
190 mtx_assert(&cdevpriv_mtx, MA_OWNED);
191 KASSERT(p->cdpd_fp->f_cdevpriv == p,
192 ("devfs_destoy_cdevpriv %p != %p", p->cdpd_fp->f_cdevpriv, p));
193 p->cdpd_fp->f_cdevpriv = NULL;
194 LIST_REMOVE(p, cdpd_list);
195 mtx_unlock(&cdevpriv_mtx);
196 (p->cdpd_dtr)(p->cdpd_data);
197 free(p, M_CDEVPDATA);
201 devfs_fpdrop(struct file *fp)
203 struct cdev_privdata *p;
205 mtx_lock(&cdevpriv_mtx);
206 if ((p = fp->f_cdevpriv) == NULL) {
207 mtx_unlock(&cdevpriv_mtx);
210 devfs_destroy_cdevpriv(p);
214 devfs_clear_cdevpriv(void)
218 fp = curthread->td_fpop;
225 * On success devfs_populate_vp() returns with dmp->dm_lock held.
228 devfs_populate_vp(struct vnode *vp)
230 struct devfs_dirent *de;
231 struct devfs_mount *dmp;
234 ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
236 dmp = VFSTODEVFS(vp->v_mount);
237 locked = VOP_ISLOCKED(vp);
239 sx_xlock(&dmp->dm_lock);
242 /* Can't call devfs_populate() with the vnode lock held. */
246 sx_xunlock(&dmp->dm_lock);
247 vn_lock(vp, locked | LK_RETRY);
248 sx_xlock(&dmp->dm_lock);
249 if (DEVFS_DMP_DROP(dmp)) {
250 sx_xunlock(&dmp->dm_lock);
251 devfs_unmount_final(dmp);
254 if ((vp->v_iflag & VI_DOOMED) != 0) {
255 sx_xunlock(&dmp->dm_lock);
260 ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
261 if ((de->de_flags & DE_DOOMED) != 0) {
262 sx_xunlock(&dmp->dm_lock);
270 devfs_vptocnp(struct vop_vptocnp_args *ap)
272 struct vnode *vp = ap->a_vp;
273 struct vnode **dvp = ap->a_vpp;
274 struct devfs_mount *dmp;
275 char *buf = ap->a_buf;
276 int *buflen = ap->a_buflen;
277 struct devfs_dirent *dd, *de;
280 dmp = VFSTODEVFS(vp->v_mount);
282 error = devfs_populate_vp(vp);
286 if (vp->v_type != VCHR && vp->v_type != VDIR) {
292 if (vp->v_type == VDIR && dd == dmp->dm_rootdir) {
299 i -= dd->de_dirent->d_namlen;
304 bcopy(dd->de_dirent->d_name, buf + i, dd->de_dirent->d_namlen);
306 de = devfs_parent_dirent(dd);
311 mtx_lock(&devfs_de_interlock);
315 mtx_unlock(&devfs_de_interlock);
321 mtx_unlock(&devfs_de_interlock);
325 sx_xunlock(&dmp->dm_lock);
330 * Construct the fully qualified path name relative to the mountpoint.
331 * If a NULL cnp is provided, no '/' is appended to the resulting path.
334 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
335 struct componentname *cnp)
338 struct devfs_dirent *de;
340 sx_assert(&dmp->dm_lock, SA_LOCKED);
345 i -= cnp->cn_namelen;
349 bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
351 while (de != dmp->dm_rootdir) {
352 if (cnp != NULL || i < SPECNAMELEN) {
358 i -= de->de_dirent->d_namlen;
361 bcopy(de->de_dirent->d_name, buf + i,
362 de->de_dirent->d_namlen);
363 de = devfs_parent_dirent(de);
371 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
372 struct devfs_dirent *de)
377 if (de->de_flags & DE_DOOMED)
379 if (DEVFS_DE_DROP(de)) {
380 KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
381 devfs_dirent_free(de);
383 if (DEVFS_DMP_DROP(dmp)) {
384 KASSERT(not_found == 1,
385 ("DEVFS mount struct freed before dirent"));
387 sx_xunlock(&dmp->dm_lock);
388 devfs_unmount_final(dmp);
390 if (not_found == 1 || (drop_dm_lock && not_found != 2))
391 sx_unlock(&dmp->dm_lock);
396 devfs_insmntque_dtr(struct vnode *vp, void *arg)
398 struct devfs_dirent *de;
400 de = (struct devfs_dirent *)arg;
401 mtx_lock(&devfs_de_interlock);
404 mtx_unlock(&devfs_de_interlock);
410 * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
414 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
420 struct devfs_mount *dmp;
423 dmp = VFSTODEVFS(mp);
424 if (de->de_flags & DE_DOOMED) {
425 sx_xunlock(&dmp->dm_lock);
431 mtx_lock(&devfs_de_interlock);
435 mtx_unlock(&devfs_de_interlock);
436 sx_xunlock(&dmp->dm_lock);
437 vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread);
438 sx_xlock(&dmp->dm_lock);
439 if (devfs_allocv_drop_refs(0, dmp, de)) {
443 else if ((vp->v_iflag & VI_DOOMED) != 0) {
444 mtx_lock(&devfs_de_interlock);
445 if (de->de_vnode == vp) {
449 mtx_unlock(&devfs_de_interlock);
453 sx_xunlock(&dmp->dm_lock);
457 mtx_unlock(&devfs_de_interlock);
458 if (de->de_dirent->d_type == DT_CHR) {
459 if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
460 devfs_allocv_drop_refs(1, dmp, de);
463 dev = &de->de_cdp->cdp_c;
467 error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
469 devfs_allocv_drop_refs(1, dmp, de);
470 printf("devfs_allocv: failed to allocate new vnode\n");
474 if (de->de_dirent->d_type == DT_CHR) {
479 /* XXX: v_rdev should be protect by vnode lock */
481 KASSERT(vp->v_usecount == 1,
482 ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount));
483 dev->si_usecount += vp->v_usecount;
484 /* Special casing of ttys for deadfs. Probably redundant. */
486 if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
487 vp->v_vflag |= VV_ISTTY;
490 if ((dev->si_flags & SI_ETERNAL) != 0)
491 vp->v_vflag |= VV_ETERNALDEV;
492 vp->v_op = &devfs_specops;
493 } else if (de->de_dirent->d_type == DT_DIR) {
495 } else if (de->de_dirent->d_type == DT_LNK) {
500 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
502 mtx_lock(&devfs_de_interlock);
505 mtx_unlock(&devfs_de_interlock);
506 error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
508 (void) devfs_allocv_drop_refs(1, dmp, de);
511 if (devfs_allocv_drop_refs(0, dmp, de)) {
516 mac_devfs_vnode_associate(mp, de, vp);
518 sx_xunlock(&dmp->dm_lock);
524 devfs_access(struct vop_access_args *ap)
526 struct vnode *vp = ap->a_vp;
527 struct devfs_dirent *de;
532 if (vp->v_type == VDIR)
535 error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
536 ap->a_accmode, ap->a_cred, NULL);
541 p = ap->a_td->td_proc;
542 /* We do, however, allow access to the controlling terminal */
544 if (!(p->p_flag & P_CONTROLT)) {
548 if (p->p_session->s_ttydp == de->de_cdp)
554 _Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0,
555 "devfs-only flag reuse failed");
558 devfs_close(struct vop_close_args *ap)
560 struct vnode *vp = ap->a_vp, *oldvp;
561 struct thread *td = ap->a_td;
563 struct cdev *dev = vp->v_rdev;
565 int dflags, error, ref, vp_locked;
568 * XXX: Don't call d_close() if we were called because of
569 * XXX: insmntque1() failure.
571 if (vp->v_data == NULL)
575 * Hack: a tty device that is a controlling terminal
576 * has a reference from the session structure.
577 * We cannot easily tell that a character device is
578 * a controlling terminal, unless it is the closing
579 * process' controlling terminal. In that case,
580 * if the reference count is 2 (this last descriptor
581 * plus the session), release the reference from the session.
586 if (vp == p->p_session->s_ttyvp) {
589 sx_xlock(&proctree_lock);
590 if (vp == p->p_session->s_ttyvp) {
591 SESS_LOCK(p->p_session);
593 if (count_dev(dev) == 2 &&
594 (vp->v_iflag & VI_DOOMED) == 0) {
595 p->p_session->s_ttyvp = NULL;
596 p->p_session->s_ttydp = NULL;
600 SESS_UNLOCK(p->p_session);
602 sx_xunlock(&proctree_lock);
609 * We do not want to really close the device if it
610 * is still in use unless we are trying to close it
611 * forcibly. Since every use (buffer, vnode, swap, cmap)
612 * holds a reference to the vnode, and because we mark
613 * any other vnodes that alias this device, when the
614 * sum of the reference counts on all the aliased
615 * vnodes descends to one, we are on last close.
617 dsw = dev_refthread(dev, &ref);
622 if (vp->v_iflag & VI_DOOMED) {
624 dflags |= FREVOKE | FNONBLOCK;
625 } else if (dsw->d_flags & D_TRACKCLOSE) {
626 /* Keep device updated on status. */
627 } else if (count_dev(dev) > 1) {
629 dev_relthread(dev, ref);
632 if (count_dev(dev) == 1)
633 dflags |= FLASTCLOSE;
636 vp_locked = VOP_ISLOCKED(vp);
638 KASSERT(dev->si_refcount > 0,
639 ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
640 error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td);
641 dev_relthread(dev, ref);
642 vn_lock(vp, vp_locked | LK_RETRY);
648 devfs_close_f(struct file *fp, struct thread *td)
654 * NB: td may be NULL if this descriptor is closed due to
655 * garbage collection from a closed UNIX domain socket.
657 fpop = curthread->td_fpop;
658 curthread->td_fpop = fp;
659 error = vnops.fo_close(fp, td);
660 curthread->td_fpop = fpop;
663 * The f_cdevpriv cannot be assigned non-NULL value while we
664 * are destroying the file.
666 if (fp->f_cdevpriv != NULL)
672 devfs_getattr(struct vop_getattr_args *ap)
674 struct vnode *vp = ap->a_vp;
675 struct vattr *vap = ap->a_vap;
676 struct devfs_dirent *de;
677 struct devfs_mount *dmp;
679 struct timeval boottime;
682 error = devfs_populate_vp(vp);
686 dmp = VFSTODEVFS(vp->v_mount);
687 sx_xunlock(&dmp->dm_lock);
690 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
691 if (vp->v_type == VDIR) {
694 ("Null dir dirent in devfs_getattr vp=%p", vp));
696 vap->va_uid = de->de_uid;
697 vap->va_gid = de->de_gid;
698 vap->va_mode = de->de_mode;
699 if (vp->v_type == VLNK)
700 vap->va_size = strlen(de->de_symlink);
701 else if (vp->v_type == VDIR)
702 vap->va_size = vap->va_bytes = DEV_BSIZE;
705 if (vp->v_type != VDIR)
707 vap->va_blocksize = DEV_BSIZE;
708 vap->va_type = vp->v_type;
710 getboottime(&boottime);
713 if ((aa).tv_sec <= 3600) { \
714 (aa).tv_sec = boottime.tv_sec; \
715 (aa).tv_nsec = boottime.tv_usec * 1000; \
719 if (vp->v_type != VCHR) {
721 vap->va_atime = de->de_atime;
723 vap->va_mtime = de->de_mtime;
725 vap->va_ctime = de->de_ctime;
729 vap->va_atime = dev->si_atime;
731 vap->va_mtime = dev->si_mtime;
733 vap->va_ctime = dev->si_ctime;
735 vap->va_rdev = cdev2priv(dev)->cdp_inode;
740 vap->va_nlink = de->de_links;
741 vap->va_fileid = de->de_inode;
748 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
755 error = vnops.fo_ioctl(fp, com, data, cred, td);
761 devfs_ioctl(struct vop_ioctl_args *ap)
763 struct fiodgname_arg *fgn;
764 struct vnode *vpold, *vp;
776 dsw = devvn_refthread(vp, &dev, &ref);
779 KASSERT(dev->si_refcount > 0,
780 ("devfs: un-referenced struct cdev *(%s)", devtoname(dev)));
782 if (com == FIODTYPE) {
783 *(int *)ap->a_data = dsw->d_flags & D_TYPEMASK;
786 } else if (com == FIODGNAME) {
793 error = copyout(p, fgn->buf, i);
797 error = dsw->d_ioctl(dev, com, ap->a_data, ap->a_fflag, td);
800 dev_relthread(dev, ref);
801 if (error == ENOIOCTL)
804 if (error == 0 && com == TIOCSCTTY) {
806 * Do nothing if reassigning same control tty, or if the
807 * control tty has already disappeared. If it disappeared,
808 * it's because we were racing with TIOCNOTTY. TIOCNOTTY
809 * already took care of releasing the old vnode and we have
810 * nothing left to do.
812 sx_slock(&proctree_lock);
813 if (td->td_proc->p_session->s_ttyvp == vp ||
814 td->td_proc->p_session->s_ttyp == NULL) {
815 sx_sunlock(&proctree_lock);
819 vpold = td->td_proc->p_session->s_ttyvp;
821 SESS_LOCK(td->td_proc->p_session);
822 td->td_proc->p_session->s_ttyvp = vp;
823 td->td_proc->p_session->s_ttydp = cdev2priv(dev);
824 SESS_UNLOCK(td->td_proc->p_session);
826 sx_sunlock(&proctree_lock);
828 /* Get rid of reference to old control tty */
837 devfs_kqfilter_f(struct file *fp, struct knote *kn)
847 error = devfs_fp_check(fp, &dev, &dsw, &ref);
850 error = dsw->d_kqfilter(dev, kn);
852 dev_relthread(dev, ref);
857 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
859 struct cdev_priv *cdp;
867 dcr = cdp->cdp_c.si_cred;
871 error = prison_check(td->td_ucred, dcr);
874 /* We do, however, allow access to the controlling terminal */
877 if (!(p->p_flag & P_CONTROLT)) {
881 if (p->p_session->s_ttydp == cdp)
888 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
890 struct componentname *cnp;
891 struct vnode *dvp, **vpp;
893 struct devfs_dirent *de, *dd;
894 struct devfs_dirent **dde;
895 struct devfs_mount *dmp;
898 int error, flags, nameiop, dvplocked;
899 char specname[SPECNAMELEN + 1], *pname;
904 pname = cnp->cn_nameptr;
906 flags = cnp->cn_flags;
907 nameiop = cnp->cn_nameiop;
909 dmp = VFSTODEVFS(mp);
913 if ((flags & ISLASTCN) && nameiop == RENAME)
916 if (dvp->v_type != VDIR)
919 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
922 error = vn_dir_check_exec(dvp, cnp);
926 if (cnp->cn_namelen == 1 && *pname == '.') {
927 if ((flags & ISLASTCN) && nameiop != LOOKUP)
934 if (flags & ISDOTDOT) {
935 if ((flags & ISLASTCN) && nameiop != LOOKUP)
937 de = devfs_parent_dirent(dd);
940 dvplocked = VOP_ISLOCKED(dvp);
942 error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK,
945 vn_lock(dvp, dvplocked | LK_RETRY);
950 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
951 while (de == NULL) { /* While(...) so we can use break */
953 if (nameiop == DELETE)
957 * OK, we didn't have an entry for the name we were asked for
958 * so we try to see if anybody can create it on demand.
960 pname = devfs_fqpn(specname, dmp, dd, cnp);
966 sx_xunlock(&dmp->dm_lock);
967 sx_slock(&clone_drain_lock);
968 EVENTHANDLER_INVOKE(dev_clone,
969 td->td_ucred, pname, strlen(pname), &cdev);
970 sx_sunlock(&clone_drain_lock);
973 sx_xlock(&dmp->dm_lock);
974 else if (devfs_populate_vp(dvp) != 0) {
976 sx_xlock(&dmp->dm_lock);
977 if (DEVFS_DMP_DROP(dmp)) {
978 sx_xunlock(&dmp->dm_lock);
979 devfs_unmount_final(dmp);
981 sx_xunlock(&dmp->dm_lock);
985 if (DEVFS_DMP_DROP(dmp)) {
987 sx_xunlock(&dmp->dm_lock);
988 devfs_unmount_final(dmp);
998 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
999 if (dde != NULL && *dde != NULL)
1006 if (de == NULL || de->de_flags & DE_WHITEOUT) {
1007 if ((nameiop == CREATE || nameiop == RENAME) &&
1008 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
1009 cnp->cn_flags |= SAVENAME;
1010 return (EJUSTRETURN);
1015 if (devfs_prison_check(de, td))
1018 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
1019 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
1028 error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp);
1034 devfs_lookup(struct vop_lookup_args *ap)
1037 struct devfs_mount *dmp;
1040 if (devfs_populate_vp(ap->a_dvp) != 0)
1043 dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1045 j = devfs_lookupx(ap, &dm_unlock);
1047 sx_xunlock(&dmp->dm_lock);
1052 devfs_mknod(struct vop_mknod_args *ap)
1054 struct componentname *cnp;
1055 struct vnode *dvp, **vpp;
1056 struct devfs_dirent *dd, *de;
1057 struct devfs_mount *dmp;
1061 * The only type of node we should be creating here is a
1062 * character device, for anything else return EOPNOTSUPP.
1064 if (ap->a_vap->va_type != VCHR)
1065 return (EOPNOTSUPP);
1067 dmp = VFSTODEVFS(dvp->v_mount);
1074 sx_xlock(&dmp->dm_lock);
1075 TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1076 if (cnp->cn_namelen != de->de_dirent->d_namlen)
1078 if (de->de_dirent->d_type == DT_CHR &&
1079 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
1081 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1082 de->de_dirent->d_namlen) != 0)
1084 if (de->de_flags & DE_WHITEOUT)
1090 de->de_flags &= ~DE_WHITEOUT;
1091 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1094 sx_xunlock(&dmp->dm_lock);
1100 devfs_open(struct vop_open_args *ap)
1102 struct thread *td = ap->a_td;
1103 struct vnode *vp = ap->a_vp;
1104 struct cdev *dev = vp->v_rdev;
1105 struct file *fp = ap->a_fp;
1106 int error, ref, vlocked;
1111 if (vp->v_type == VBLK)
1117 /* Make this field valid before any I/O in d_open. */
1118 if (dev->si_iosize_max == 0)
1119 dev->si_iosize_max = DFLTPHYS;
1121 dsw = dev_refthread(dev, &ref);
1124 if (fp == NULL && dsw->d_fdopen != NULL) {
1125 dev_relthread(dev, ref);
1129 vlocked = VOP_ISLOCKED(vp);
1138 if (dsw->d_fdopen != NULL)
1139 error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1141 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1142 /* Clean up any cdevpriv upon error. */
1144 devfs_clear_cdevpriv();
1147 vn_lock(vp, vlocked | LK_RETRY);
1148 dev_relthread(dev, ref);
1150 if (error == ERESTART)
1155 #if 0 /* /dev/console */
1156 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1161 if (fp->f_ops == &badfileops)
1162 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1163 mtxp = mtx_pool_find(mtxpool_sleep, fp);
1166 * Hint to the dofilewrite() to not force the buffer draining
1167 * on the writer to the file. Most likely, the write would
1168 * not need normal buffers.
1171 fp->f_vnread_flags |= FDEVFS_VNODE;
1177 devfs_pathconf(struct vop_pathconf_args *ap)
1180 switch (ap->a_name) {
1181 case _PC_FILESIZEBITS:
1185 *ap->a_retval = NAME_MAX;
1188 *ap->a_retval = INT_MAX;
1190 case _PC_SYMLINK_MAX:
1191 *ap->a_retval = MAXPATHLEN;
1194 if (ap->a_vp->v_vflag & VV_ISTTY) {
1195 *ap->a_retval = MAX_CANON;
1200 if (ap->a_vp->v_vflag & VV_ISTTY) {
1201 *ap->a_retval = MAX_INPUT;
1206 if (ap->a_vp->v_vflag & VV_ISTTY) {
1207 *ap->a_retval = _POSIX_VDISABLE;
1211 case _PC_MAC_PRESENT:
1214 * If MAC is enabled, devfs automatically supports
1215 * trivial non-persistant label storage.
1222 case _PC_CHOWN_RESTRICTED:
1226 return (vop_stdpathconf(ap));
1233 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1241 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1243 error = vnops.fo_poll(fp, events, cred, td);
1246 error = dsw->d_poll(dev, events, td);
1248 dev_relthread(dev, ref);
1253 * Print out the contents of a special device vnode.
1256 devfs_print(struct vop_print_args *ap)
1259 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1264 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1265 int flags, struct thread *td)
1268 int ioflag, error, ref;
1273 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1276 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1278 error = vnops.fo_read(fp, uio, cred, flags, td);
1281 resid = uio->uio_resid;
1282 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1283 if (ioflag & O_DIRECT)
1284 ioflag |= IO_DIRECT;
1286 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1287 error = dsw->d_read(dev, uio, ioflag);
1288 if (uio->uio_resid != resid || (error == 0 && resid != 0))
1289 devfs_timestamp(&dev->si_atime);
1291 dev_relthread(dev, ref);
1293 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1298 devfs_readdir(struct vop_readdir_args *ap)
1303 struct devfs_dirent *dd;
1304 struct devfs_dirent *de;
1305 struct devfs_mount *dmp;
1307 int *tmp_ncookies = NULL;
1309 if (ap->a_vp->v_type != VDIR)
1313 if (uio->uio_offset < 0)
1317 * XXX: This is a temporary hack to get around this filesystem not
1318 * supporting cookies. We store the location of the ncookies pointer
1319 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1320 * and set the number of cookies to 0. We then set the pointer to
1321 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1322 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1323 * pointer to its original location before returning to the caller.
1325 if (ap->a_ncookies != NULL) {
1326 tmp_ncookies = ap->a_ncookies;
1327 *ap->a_ncookies = 0;
1328 ap->a_ncookies = NULL;
1331 dmp = VFSTODEVFS(ap->a_vp->v_mount);
1332 if (devfs_populate_vp(ap->a_vp) != 0) {
1333 if (tmp_ncookies != NULL)
1334 ap->a_ncookies = tmp_ncookies;
1338 de = ap->a_vp->v_data;
1340 TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1341 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1342 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1344 if (devfs_prison_check(dd, uio->uio_td))
1346 if (dd->de_dirent->d_type == DT_DIR)
1351 MPASS(dp->d_reclen == GENERIC_DIRSIZ(dp));
1352 if (dp->d_reclen > uio->uio_resid)
1354 dp->d_fileno = de->de_inode;
1355 /* NOTE: d_off is the offset for the *next* entry. */
1356 dp->d_off = off + dp->d_reclen;
1357 if (off >= uio->uio_offset) {
1358 error = vfs_read_dirent(ap, dp, off);
1362 off += dp->d_reclen;
1364 sx_xunlock(&dmp->dm_lock);
1365 uio->uio_offset = off;
1368 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1371 if (tmp_ncookies != NULL)
1372 ap->a_ncookies = tmp_ncookies;
1378 devfs_readlink(struct vop_readlink_args *ap)
1380 struct devfs_dirent *de;
1382 de = ap->a_vp->v_data;
1383 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1387 devfs_reclaim(struct vop_reclaim_args *ap)
1390 struct devfs_dirent *de;
1393 mtx_lock(&devfs_de_interlock);
1396 de->de_vnode = NULL;
1399 mtx_unlock(&devfs_de_interlock);
1400 vnode_destroy_vobject(vp);
1405 devfs_reclaim_vchr(struct vop_reclaim_args *ap)
1411 MPASS(vp->v_type == VCHR);
1420 dev->si_usecount -= vp->v_usecount;
1429 devfs_remove(struct vop_remove_args *ap)
1431 struct vnode *dvp = ap->a_dvp;
1432 struct vnode *vp = ap->a_vp;
1433 struct devfs_dirent *dd;
1434 struct devfs_dirent *de, *de_covered;
1435 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1437 ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1438 ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1440 sx_xlock(&dmp->dm_lock);
1441 dd = ap->a_dvp->v_data;
1443 if (de->de_cdp == NULL) {
1444 TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1445 if (de->de_dirent->d_type == DT_LNK) {
1446 de_covered = devfs_find(dd, de->de_dirent->d_name,
1447 de->de_dirent->d_namlen, 0);
1448 if (de_covered != NULL)
1449 de_covered->de_flags &= ~DE_COVERED;
1451 /* We need to unlock dvp because devfs_delete() may lock it. */
1455 devfs_delete(dmp, de, 0);
1456 sx_xunlock(&dmp->dm_lock);
1458 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1459 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1461 de->de_flags |= DE_WHITEOUT;
1462 sx_xunlock(&dmp->dm_lock);
1468 * Revoke is called on a tty when a terminal session ends. The vnode
1469 * is orphaned by setting v_op to deadfs so we need to let go of it
1470 * as well so that we create a new one next time around.
1474 devfs_revoke(struct vop_revoke_args *ap)
1476 struct vnode *vp = ap->a_vp, *vp2;
1478 struct cdev_priv *cdp;
1479 struct devfs_dirent *de;
1482 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1485 cdp = cdev2priv(dev);
1498 mtx_lock(&devfs_de_interlock);
1501 for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1502 de = cdp->cdp_dirents[i];
1510 mtx_unlock(&devfs_de_interlock);
1511 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1525 mtx_unlock(&devfs_de_interlock);
1530 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1531 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1533 dev_rel(&cdp->cdp_c);
1537 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1542 devfs_rioctl(struct vop_ioctl_args *ap)
1545 struct devfs_mount *dmp;
1549 vn_lock(vp, LK_SHARED | LK_RETRY);
1550 if (vp->v_iflag & VI_DOOMED) {
1554 dmp = VFSTODEVFS(vp->v_mount);
1555 sx_xlock(&dmp->dm_lock);
1557 DEVFS_DMP_HOLD(dmp);
1558 devfs_populate(dmp);
1559 if (DEVFS_DMP_DROP(dmp)) {
1560 sx_xunlock(&dmp->dm_lock);
1561 devfs_unmount_final(dmp);
1564 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1565 sx_xunlock(&dmp->dm_lock);
1570 devfs_rread(struct vop_read_args *ap)
1573 if (ap->a_vp->v_type != VDIR)
1575 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1579 devfs_setattr(struct vop_setattr_args *ap)
1581 struct devfs_dirent *de;
1592 if ((vap->va_type != VNON) ||
1593 (vap->va_nlink != VNOVAL) ||
1594 (vap->va_fsid != VNOVAL) ||
1595 (vap->va_fileid != VNOVAL) ||
1596 (vap->va_blocksize != VNOVAL) ||
1597 (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1598 (vap->va_rdev != VNOVAL) ||
1599 ((int)vap->va_bytes != VNOVAL) ||
1600 (vap->va_gen != VNOVAL)) {
1604 error = devfs_populate_vp(vp);
1609 if (vp->v_type == VDIR)
1613 if (vap->va_uid == (uid_t)VNOVAL)
1617 if (vap->va_gid == (gid_t)VNOVAL)
1621 if (uid != de->de_uid || gid != de->de_gid) {
1622 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1623 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1624 error = priv_check(td, PRIV_VFS_CHOWN);
1633 if (vap->va_mode != (mode_t)VNOVAL) {
1634 if (ap->a_cred->cr_uid != de->de_uid) {
1635 error = priv_check(td, PRIV_VFS_ADMIN);
1639 de->de_mode = vap->va_mode;
1643 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1644 error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1647 if (vap->va_atime.tv_sec != VNOVAL) {
1648 if (vp->v_type == VCHR)
1649 vp->v_rdev->si_atime = vap->va_atime;
1651 de->de_atime = vap->va_atime;
1653 if (vap->va_mtime.tv_sec != VNOVAL) {
1654 if (vp->v_type == VCHR)
1655 vp->v_rdev->si_mtime = vap->va_mtime;
1657 de->de_mtime = vap->va_mtime;
1663 if (vp->v_type == VCHR)
1664 vfs_timestamp(&vp->v_rdev->si_ctime);
1666 vfs_timestamp(&de->de_mtime);
1670 sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock);
1676 devfs_setlabel(struct vop_setlabel_args *ap)
1679 struct devfs_dirent *de;
1684 mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1685 mac_devfs_update(vp->v_mount, de, vp);
1692 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1695 return (vnops.fo_stat(fp, sb, cred, td));
1699 devfs_symlink(struct vop_symlink_args *ap)
1702 struct devfs_dirent *dd;
1703 struct devfs_dirent *de, *de_covered, *de_dotdot;
1704 struct devfs_mount *dmp;
1706 error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1709 dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1710 if (devfs_populate_vp(ap->a_dvp) != 0)
1713 dd = ap->a_dvp->v_data;
1714 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1715 de->de_flags = DE_USER;
1719 de->de_inode = alloc_unr(devfs_inos);
1721 de->de_dirent->d_type = DT_LNK;
1722 i = strlen(ap->a_target) + 1;
1723 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1724 bcopy(ap->a_target, de->de_symlink, i);
1726 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1728 de_covered = devfs_find(dd, de->de_dirent->d_name,
1729 de->de_dirent->d_namlen, 0);
1730 if (de_covered != NULL) {
1731 if ((de_covered->de_flags & DE_USER) != 0) {
1732 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1733 sx_xunlock(&dmp->dm_lock);
1736 KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1737 ("devfs_symlink: entry %p already covered", de_covered));
1738 de_covered->de_flags |= DE_COVERED;
1741 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */
1742 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */
1743 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1744 devfs_dir_ref_de(dmp, dd);
1745 devfs_rules_apply(dmp, de);
1747 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1751 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1754 return (vnops.fo_truncate(fp, length, cred, td));
1758 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1759 int flags, struct thread *td)
1762 int error, ioflag, ref;
1767 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1770 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1772 error = vnops.fo_write(fp, uio, cred, flags, td);
1775 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1776 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1777 if (ioflag & O_DIRECT)
1778 ioflag |= IO_DIRECT;
1779 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1781 resid = uio->uio_resid;
1783 error = dsw->d_write(dev, uio, ioflag);
1784 if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1785 devfs_timestamp(&dev->si_ctime);
1786 dev->si_mtime = dev->si_ctime;
1789 dev_relthread(dev, ref);
1791 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1796 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1797 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1812 * Ensure that file and memory protections are
1816 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1817 maxprot = VM_PROT_NONE;
1818 if ((prot & VM_PROT_EXECUTE) != 0)
1821 maxprot = VM_PROT_EXECUTE;
1822 if ((fp->f_flag & FREAD) != 0)
1823 maxprot |= VM_PROT_READ;
1824 else if ((prot & VM_PROT_READ) != 0)
1828 * If we are sharing potential changes via MAP_SHARED and we
1829 * are trying to get write permission although we opened it
1830 * without asking for it, bail out.
1832 * Note that most character devices always share mappings.
1833 * The one exception is that D_MMAP_ANON devices
1834 * (i.e. /dev/zero) permit private writable mappings.
1836 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests
1837 * as well as updating maxprot to permit writing for
1838 * D_MMAP_ANON devices rather than doing that here.
1840 if ((flags & MAP_SHARED) != 0) {
1841 if ((fp->f_flag & FWRITE) != 0)
1842 maxprot |= VM_PROT_WRITE;
1843 else if ((prot & VM_PROT_WRITE) != 0)
1846 maxprot &= cap_maxprot;
1849 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1853 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff,
1856 dev_relthread(dev, ref);
1860 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1863 vm_object_deallocate(object);
1868 dev2udev(struct cdev *x)
1872 return (cdev2priv(x)->cdp_inode);
1875 static struct fileops devfs_ops_f = {
1876 .fo_read = devfs_read_f,
1877 .fo_write = devfs_write_f,
1878 .fo_truncate = devfs_truncate_f,
1879 .fo_ioctl = devfs_ioctl_f,
1880 .fo_poll = devfs_poll_f,
1881 .fo_kqfilter = devfs_kqfilter_f,
1882 .fo_stat = devfs_stat_f,
1883 .fo_close = devfs_close_f,
1884 .fo_chmod = vn_chmod,
1885 .fo_chown = vn_chown,
1886 .fo_sendfile = vn_sendfile,
1888 .fo_fill_kinfo = vn_fill_kinfo,
1889 .fo_mmap = devfs_mmap_f,
1890 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
1893 /* Vops for non-CHR vnodes in /dev. */
1894 static struct vop_vector devfs_vnodeops = {
1895 .vop_default = &default_vnodeops,
1897 .vop_access = devfs_access,
1898 .vop_getattr = devfs_getattr,
1899 .vop_ioctl = devfs_rioctl,
1900 .vop_lookup = devfs_lookup,
1901 .vop_mknod = devfs_mknod,
1902 .vop_pathconf = devfs_pathconf,
1903 .vop_read = devfs_rread,
1904 .vop_readdir = devfs_readdir,
1905 .vop_readlink = devfs_readlink,
1906 .vop_reclaim = devfs_reclaim,
1907 .vop_remove = devfs_remove,
1908 .vop_revoke = devfs_revoke,
1909 .vop_setattr = devfs_setattr,
1911 .vop_setlabel = devfs_setlabel,
1913 .vop_symlink = devfs_symlink,
1914 .vop_vptocnp = devfs_vptocnp,
1917 /* Vops for VCHR vnodes in /dev. */
1918 static struct vop_vector devfs_specops = {
1919 .vop_default = &default_vnodeops,
1921 .vop_access = devfs_access,
1922 .vop_bmap = VOP_PANIC,
1923 .vop_close = devfs_close,
1924 .vop_create = VOP_PANIC,
1925 .vop_fsync = vop_stdfsync,
1926 .vop_getattr = devfs_getattr,
1927 .vop_ioctl = devfs_ioctl,
1928 .vop_link = VOP_PANIC,
1929 .vop_mkdir = VOP_PANIC,
1930 .vop_mknod = VOP_PANIC,
1931 .vop_open = devfs_open,
1932 .vop_pathconf = devfs_pathconf,
1933 .vop_poll = dead_poll,
1934 .vop_print = devfs_print,
1935 .vop_read = dead_read,
1936 .vop_readdir = VOP_PANIC,
1937 .vop_readlink = VOP_PANIC,
1938 .vop_reallocblks = VOP_PANIC,
1939 .vop_reclaim = devfs_reclaim_vchr,
1940 .vop_remove = devfs_remove,
1941 .vop_rename = VOP_PANIC,
1942 .vop_revoke = devfs_revoke,
1943 .vop_rmdir = VOP_PANIC,
1944 .vop_setattr = devfs_setattr,
1946 .vop_setlabel = devfs_setlabel,
1948 .vop_strategy = VOP_PANIC,
1949 .vop_symlink = VOP_PANIC,
1950 .vop_vptocnp = devfs_vptocnp,
1951 .vop_write = dead_write,
1955 * Our calling convention to the device drivers used to be that we passed
1956 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1957 * flags instead since that's what open(), close() and ioctl() takes and
1958 * we don't really want vnode.h in device drivers.
1959 * We solved the source compatibility by redefining some vnode flags to
1960 * be the same as the fcntl ones and by sending down the bitwise OR of
1961 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody
1962 * pulls the rug out under this.
1964 CTASSERT(O_NONBLOCK == IO_NDELAY);
1965 CTASSERT(O_FSYNC == IO_SYNC);