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.
7 * This code is derived from software donated to Berkeley by
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
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.
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
31 * @(#)kernfs_vnops.c 8.15 (Berkeley) 5/21/95
32 * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vnops.c 1.43
42 #include <sys/param.h>
43 #include <sys/systm.h>
45 #include <sys/dirent.h>
46 #include <sys/fcntl.h>
48 #include <sys/filedesc.h>
49 #include <sys/filio.h>
51 #include <sys/kernel.h>
53 #include <sys/malloc.h>
55 #include <sys/mount.h>
56 #include <sys/namei.h>
61 #include <sys/sysctl.h>
63 #include <sys/ttycom.h>
64 #include <sys/unistd.h>
65 #include <sys/vnode.h>
67 static struct vop_vector devfs_vnodeops;
68 static struct vop_vector devfs_specops;
69 static struct fileops devfs_ops_f;
71 #include <fs/devfs/devfs.h>
72 #include <fs/devfs/devfs_int.h>
74 #include <security/mac/mac_framework.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_object.h>
80 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
82 struct mtx devfs_de_interlock;
83 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
84 struct sx clone_drain_lock;
85 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
86 struct mtx cdevpriv_mtx;
87 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
89 SYSCTL_DECL(_vfs_devfs);
91 static int devfs_dotimes;
92 SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW,
93 &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision");
96 * Update devfs node timestamp. Note that updates are unlocked and
97 * stat(2) could see partially updated times.
100 devfs_timestamp(struct timespec *tsp)
108 if (tsp->tv_sec != ts) {
116 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
120 *dswp = devvn_refthread(fp->f_vnode, devp, ref);
121 if (*devp != fp->f_data) {
123 dev_relthread(*devp, *ref);
126 KASSERT((*devp)->si_refcount > 0,
127 ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
130 curthread->td_fpop = fp;
135 devfs_get_cdevpriv(void **datap)
138 struct cdev_privdata *p;
141 fp = curthread->td_fpop;
147 *datap = p->cdpd_data;
154 devfs_set_cdevpriv(void *priv, d_priv_dtor_t *priv_dtr)
157 struct cdev_priv *cdp;
158 struct cdev_privdata *p;
161 fp = curthread->td_fpop;
164 cdp = cdev2priv((struct cdev *)fp->f_data);
165 p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
167 p->cdpd_dtr = priv_dtr;
169 mtx_lock(&cdevpriv_mtx);
170 if (fp->f_cdevpriv == NULL) {
171 LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
173 mtx_unlock(&cdevpriv_mtx);
176 mtx_unlock(&cdevpriv_mtx);
177 free(p, M_CDEVPDATA);
184 devfs_destroy_cdevpriv(struct cdev_privdata *p)
187 mtx_assert(&cdevpriv_mtx, MA_OWNED);
188 KASSERT(p->cdpd_fp->f_cdevpriv == p,
189 ("devfs_destoy_cdevpriv %p != %p", p->cdpd_fp->f_cdevpriv, p));
190 p->cdpd_fp->f_cdevpriv = NULL;
191 LIST_REMOVE(p, cdpd_list);
192 mtx_unlock(&cdevpriv_mtx);
193 (p->cdpd_dtr)(p->cdpd_data);
194 free(p, M_CDEVPDATA);
198 devfs_fpdrop(struct file *fp)
200 struct cdev_privdata *p;
202 mtx_lock(&cdevpriv_mtx);
203 if ((p = fp->f_cdevpriv) == NULL) {
204 mtx_unlock(&cdevpriv_mtx);
207 devfs_destroy_cdevpriv(p);
211 devfs_clear_cdevpriv(void)
215 fp = curthread->td_fpop;
222 * On success devfs_populate_vp() returns with dmp->dm_lock held.
225 devfs_populate_vp(struct vnode *vp)
227 struct devfs_dirent *de;
228 struct devfs_mount *dmp;
231 ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
233 dmp = VFSTODEVFS(vp->v_mount);
234 locked = VOP_ISLOCKED(vp);
236 sx_xlock(&dmp->dm_lock);
239 /* Can't call devfs_populate() with the vnode lock held. */
243 sx_xunlock(&dmp->dm_lock);
244 vn_lock(vp, locked | LK_RETRY);
245 sx_xlock(&dmp->dm_lock);
246 if (DEVFS_DMP_DROP(dmp)) {
247 sx_xunlock(&dmp->dm_lock);
248 devfs_unmount_final(dmp);
251 if ((vp->v_iflag & VI_DOOMED) != 0) {
252 sx_xunlock(&dmp->dm_lock);
257 ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
258 if ((de->de_flags & DE_DOOMED) != 0) {
259 sx_xunlock(&dmp->dm_lock);
267 devfs_vptocnp(struct vop_vptocnp_args *ap)
269 struct vnode *vp = ap->a_vp;
270 struct vnode **dvp = ap->a_vpp;
271 struct devfs_mount *dmp;
272 char *buf = ap->a_buf;
273 int *buflen = ap->a_buflen;
274 struct devfs_dirent *dd, *de;
277 dmp = VFSTODEVFS(vp->v_mount);
279 error = devfs_populate_vp(vp);
286 if (vp->v_type == VCHR) {
287 i -= strlen(dd->de_cdp->cdp_c.si_name);
292 bcopy(dd->de_cdp->cdp_c.si_name, buf + i,
293 strlen(dd->de_cdp->cdp_c.si_name));
295 } else if (vp->v_type == VDIR) {
296 if (dd == dmp->dm_rootdir) {
301 i -= dd->de_dirent->d_namlen;
306 bcopy(dd->de_dirent->d_name, buf + i,
307 dd->de_dirent->d_namlen);
314 de = devfs_parent_dirent(de);
319 mtx_lock(&devfs_de_interlock);
323 mtx_unlock(&devfs_de_interlock);
329 mtx_unlock(&devfs_de_interlock);
333 sx_xunlock(&dmp->dm_lock);
338 * Construct the fully qualified path name relative to the mountpoint.
339 * If a NULL cnp is provided, no '/' is appended to the resulting path.
342 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
343 struct componentname *cnp)
346 struct devfs_dirent *de;
348 sx_assert(&dmp->dm_lock, SA_LOCKED);
353 i -= cnp->cn_namelen;
357 bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
359 while (de != dmp->dm_rootdir) {
360 if (cnp != NULL || i < SPECNAMELEN) {
366 i -= de->de_dirent->d_namlen;
369 bcopy(de->de_dirent->d_name, buf + i,
370 de->de_dirent->d_namlen);
371 de = devfs_parent_dirent(de);
379 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
380 struct devfs_dirent *de)
385 if (de->de_flags & DE_DOOMED)
387 if (DEVFS_DE_DROP(de)) {
388 KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
389 devfs_dirent_free(de);
391 if (DEVFS_DMP_DROP(dmp)) {
392 KASSERT(not_found == 1,
393 ("DEVFS mount struct freed before dirent"));
395 sx_xunlock(&dmp->dm_lock);
396 devfs_unmount_final(dmp);
398 if (not_found == 1 || (drop_dm_lock && not_found != 2))
399 sx_unlock(&dmp->dm_lock);
404 devfs_insmntque_dtr(struct vnode *vp, void *arg)
406 struct devfs_dirent *de;
408 de = (struct devfs_dirent *)arg;
409 mtx_lock(&devfs_de_interlock);
412 mtx_unlock(&devfs_de_interlock);
418 * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
422 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
428 struct devfs_mount *dmp;
431 dmp = VFSTODEVFS(mp);
432 if (de->de_flags & DE_DOOMED) {
433 sx_xunlock(&dmp->dm_lock);
439 mtx_lock(&devfs_de_interlock);
443 mtx_unlock(&devfs_de_interlock);
444 sx_xunlock(&dmp->dm_lock);
445 vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread);
446 sx_xlock(&dmp->dm_lock);
447 if (devfs_allocv_drop_refs(0, dmp, de)) {
451 else if ((vp->v_iflag & VI_DOOMED) != 0) {
452 mtx_lock(&devfs_de_interlock);
453 if (de->de_vnode == vp) {
457 mtx_unlock(&devfs_de_interlock);
461 sx_xunlock(&dmp->dm_lock);
465 mtx_unlock(&devfs_de_interlock);
466 if (de->de_dirent->d_type == DT_CHR) {
467 if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
468 devfs_allocv_drop_refs(1, dmp, de);
471 dev = &de->de_cdp->cdp_c;
475 error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
477 devfs_allocv_drop_refs(1, dmp, de);
478 printf("devfs_allocv: failed to allocate new vnode\n");
482 if (de->de_dirent->d_type == DT_CHR) {
487 /* XXX: v_rdev should be protect by vnode lock */
489 KASSERT(vp->v_usecount == 1,
490 ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount));
491 dev->si_usecount += vp->v_usecount;
492 /* Special casing of ttys for deadfs. Probably redundant. */
494 if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
495 vp->v_vflag |= VV_ISTTY;
498 if ((dev->si_flags & SI_ETERNAL) != 0)
499 vp->v_vflag |= VV_ETERNALDEV;
500 vp->v_op = &devfs_specops;
501 } else if (de->de_dirent->d_type == DT_DIR) {
503 } else if (de->de_dirent->d_type == DT_LNK) {
508 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
510 mtx_lock(&devfs_de_interlock);
513 mtx_unlock(&devfs_de_interlock);
514 error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
516 (void) devfs_allocv_drop_refs(1, dmp, de);
519 if (devfs_allocv_drop_refs(0, dmp, de)) {
524 mac_devfs_vnode_associate(mp, de, vp);
526 sx_xunlock(&dmp->dm_lock);
532 devfs_access(struct vop_access_args *ap)
534 struct vnode *vp = ap->a_vp;
535 struct devfs_dirent *de;
540 if (vp->v_type == VDIR)
543 error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
544 ap->a_accmode, ap->a_cred, NULL);
549 p = ap->a_td->td_proc;
550 /* We do, however, allow access to the controlling terminal */
552 if (!(p->p_flag & P_CONTROLT)) {
556 if (p->p_session->s_ttydp == de->de_cdp)
562 _Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0,
563 "devfs-only flag reuse failed");
566 devfs_close(struct vop_close_args *ap)
568 struct vnode *vp = ap->a_vp, *oldvp;
569 struct thread *td = ap->a_td;
571 struct cdev *dev = vp->v_rdev;
573 int dflags, error, ref, vp_locked;
576 * XXX: Don't call d_close() if we were called because of
577 * XXX: insmntque1() failure.
579 if (vp->v_data == NULL)
583 * Hack: a tty device that is a controlling terminal
584 * has a reference from the session structure.
585 * We cannot easily tell that a character device is
586 * a controlling terminal, unless it is the closing
587 * process' controlling terminal. In that case,
588 * if the reference count is 2 (this last descriptor
589 * plus the session), release the reference from the session.
594 if (vp == p->p_session->s_ttyvp) {
597 sx_xlock(&proctree_lock);
598 if (vp == p->p_session->s_ttyvp) {
599 SESS_LOCK(p->p_session);
601 if (count_dev(dev) == 2 &&
602 (vp->v_iflag & VI_DOOMED) == 0) {
603 p->p_session->s_ttyvp = NULL;
604 p->p_session->s_ttydp = NULL;
608 SESS_UNLOCK(p->p_session);
610 sx_xunlock(&proctree_lock);
617 * We do not want to really close the device if it
618 * is still in use unless we are trying to close it
619 * forcibly. Since every use (buffer, vnode, swap, cmap)
620 * holds a reference to the vnode, and because we mark
621 * any other vnodes that alias this device, when the
622 * sum of the reference counts on all the aliased
623 * vnodes descends to one, we are on last close.
625 dsw = dev_refthread(dev, &ref);
630 if (vp->v_iflag & VI_DOOMED) {
632 dflags |= FREVOKE | FNONBLOCK;
633 } else if (dsw->d_flags & D_TRACKCLOSE) {
634 /* Keep device updated on status. */
635 } else if (count_dev(dev) > 1) {
637 dev_relthread(dev, ref);
640 if (count_dev(dev) == 1)
641 dflags |= FLASTCLOSE;
644 vp_locked = VOP_ISLOCKED(vp);
646 KASSERT(dev->si_refcount > 0,
647 ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
648 error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td);
649 dev_relthread(dev, ref);
650 vn_lock(vp, vp_locked | LK_RETRY);
656 devfs_close_f(struct file *fp, struct thread *td)
662 * NB: td may be NULL if this descriptor is closed due to
663 * garbage collection from a closed UNIX domain socket.
665 fpop = curthread->td_fpop;
666 curthread->td_fpop = fp;
667 error = vnops.fo_close(fp, td);
668 curthread->td_fpop = fpop;
671 * The f_cdevpriv cannot be assigned non-NULL value while we
672 * are destroying the file.
674 if (fp->f_cdevpriv != NULL)
680 devfs_getattr(struct vop_getattr_args *ap)
682 struct vnode *vp = ap->a_vp;
683 struct vattr *vap = ap->a_vap;
684 struct devfs_dirent *de;
685 struct devfs_mount *dmp;
687 struct timeval boottime;
690 error = devfs_populate_vp(vp);
694 dmp = VFSTODEVFS(vp->v_mount);
695 sx_xunlock(&dmp->dm_lock);
698 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
699 if (vp->v_type == VDIR) {
702 ("Null dir dirent in devfs_getattr vp=%p", vp));
704 vap->va_uid = de->de_uid;
705 vap->va_gid = de->de_gid;
706 vap->va_mode = de->de_mode;
707 if (vp->v_type == VLNK)
708 vap->va_size = strlen(de->de_symlink);
709 else if (vp->v_type == VDIR)
710 vap->va_size = vap->va_bytes = DEV_BSIZE;
713 if (vp->v_type != VDIR)
715 vap->va_blocksize = DEV_BSIZE;
716 vap->va_type = vp->v_type;
718 getboottime(&boottime);
721 if ((aa).tv_sec <= 3600) { \
722 (aa).tv_sec = boottime.tv_sec; \
723 (aa).tv_nsec = boottime.tv_usec * 1000; \
727 if (vp->v_type != VCHR) {
729 vap->va_atime = de->de_atime;
731 vap->va_mtime = de->de_mtime;
733 vap->va_ctime = de->de_ctime;
737 vap->va_atime = dev->si_atime;
739 vap->va_mtime = dev->si_mtime;
741 vap->va_ctime = dev->si_ctime;
743 vap->va_rdev = cdev2priv(dev)->cdp_inode;
748 vap->va_nlink = de->de_links;
749 vap->va_fileid = de->de_inode;
756 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
763 error = vnops.fo_ioctl(fp, com, data, cred, td);
769 devfs_ioctl(struct vop_ioctl_args *ap)
771 struct fiodgname_arg *fgn;
772 struct vnode *vpold, *vp;
784 dsw = devvn_refthread(vp, &dev, &ref);
787 KASSERT(dev->si_refcount > 0,
788 ("devfs: un-referenced struct cdev *(%s)", devtoname(dev)));
790 if (com == FIODTYPE) {
791 *(int *)ap->a_data = dsw->d_flags & D_TYPEMASK;
794 } else if (com == FIODGNAME) {
801 error = copyout(p, fgn->buf, i);
805 error = dsw->d_ioctl(dev, com, ap->a_data, ap->a_fflag, td);
808 dev_relthread(dev, ref);
809 if (error == ENOIOCTL)
812 if (error == 0 && com == TIOCSCTTY) {
813 /* Do nothing if reassigning same control tty */
814 sx_slock(&proctree_lock);
815 if (td->td_proc->p_session->s_ttyvp == vp) {
816 sx_sunlock(&proctree_lock);
820 vpold = td->td_proc->p_session->s_ttyvp;
822 SESS_LOCK(td->td_proc->p_session);
823 td->td_proc->p_session->s_ttyvp = vp;
824 td->td_proc->p_session->s_ttydp = cdev2priv(dev);
825 SESS_UNLOCK(td->td_proc->p_session);
827 sx_sunlock(&proctree_lock);
829 /* Get rid of reference to old control tty */
838 devfs_kqfilter_f(struct file *fp, struct knote *kn)
848 error = devfs_fp_check(fp, &dev, &dsw, &ref);
851 error = dsw->d_kqfilter(dev, kn);
853 dev_relthread(dev, ref);
858 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
860 struct cdev_priv *cdp;
868 dcr = cdp->cdp_c.si_cred;
872 error = prison_check(td->td_ucred, dcr);
875 /* We do, however, allow access to the controlling terminal */
878 if (!(p->p_flag & P_CONTROLT)) {
882 if (p->p_session->s_ttydp == cdp)
889 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
891 struct componentname *cnp;
892 struct vnode *dvp, **vpp;
894 struct devfs_dirent *de, *dd;
895 struct devfs_dirent **dde;
896 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;
908 dmp = VFSTODEVFS(dvp->v_mount);
912 if ((flags & ISLASTCN) && nameiop == RENAME)
915 if (dvp->v_type != VDIR)
918 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
921 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
925 if (cnp->cn_namelen == 1 && *pname == '.') {
926 if ((flags & ISLASTCN) && nameiop != LOOKUP)
933 if (flags & ISDOTDOT) {
934 if ((flags & ISLASTCN) && nameiop != LOOKUP)
936 de = devfs_parent_dirent(dd);
939 dvplocked = VOP_ISLOCKED(dvp);
941 error = devfs_allocv(de, dvp->v_mount,
942 cnp->cn_lkflags & LK_TYPE_MASK, vpp);
944 vn_lock(dvp, dvplocked | LK_RETRY);
949 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
950 while (de == NULL) { /* While(...) so we can use break */
952 if (nameiop == DELETE)
956 * OK, we didn't have an entry for the name we were asked for
957 * so we try to see if anybody can create it on demand.
959 pname = devfs_fqpn(specname, dmp, dd, cnp);
965 sx_xunlock(&dmp->dm_lock);
966 sx_slock(&clone_drain_lock);
967 EVENTHANDLER_INVOKE(dev_clone,
968 td->td_ucred, pname, strlen(pname), &cdev);
969 sx_sunlock(&clone_drain_lock);
972 sx_xlock(&dmp->dm_lock);
973 else if (devfs_populate_vp(dvp) != 0) {
975 sx_xlock(&dmp->dm_lock);
976 if (DEVFS_DMP_DROP(dmp)) {
977 sx_xunlock(&dmp->dm_lock);
978 devfs_unmount_final(dmp);
980 sx_xunlock(&dmp->dm_lock);
984 if (DEVFS_DMP_DROP(dmp)) {
986 sx_xunlock(&dmp->dm_lock);
987 devfs_unmount_final(dmp);
997 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
998 if (dde != NULL && *dde != NULL)
1005 if (de == NULL || de->de_flags & DE_WHITEOUT) {
1006 if ((nameiop == CREATE || nameiop == RENAME) &&
1007 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
1008 cnp->cn_flags |= SAVENAME;
1009 return (EJUSTRETURN);
1014 if (devfs_prison_check(de, td))
1017 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
1018 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
1027 error = devfs_allocv(de, dvp->v_mount, cnp->cn_lkflags & LK_TYPE_MASK,
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) {
1182 if (ap->a_vp->v_vflag & VV_ISTTY) {
1183 *ap->a_retval = MAX_CANON;
1188 if (ap->a_vp->v_vflag & VV_ISTTY) {
1189 *ap->a_retval = MAX_INPUT;
1194 if (ap->a_vp->v_vflag & VV_ISTTY) {
1195 *ap->a_retval = _POSIX_VDISABLE;
1199 case _PC_MAC_PRESENT:
1202 * If MAC is enabled, devfs automatically supports
1203 * trivial non-persistant label storage.
1211 return (vop_stdpathconf(ap));
1218 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1226 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1228 error = vnops.fo_poll(fp, events, cred, td);
1231 error = dsw->d_poll(dev, events, td);
1233 dev_relthread(dev, ref);
1238 * Print out the contents of a special device vnode.
1241 devfs_print(struct vop_print_args *ap)
1244 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1249 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1250 int flags, struct thread *td)
1253 int ioflag, error, ref;
1258 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1261 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1263 error = vnops.fo_read(fp, uio, cred, flags, td);
1266 resid = uio->uio_resid;
1267 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1268 if (ioflag & O_DIRECT)
1269 ioflag |= IO_DIRECT;
1271 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1272 error = dsw->d_read(dev, uio, ioflag);
1273 if (uio->uio_resid != resid || (error == 0 && resid != 0))
1274 devfs_timestamp(&dev->si_atime);
1276 dev_relthread(dev, ref);
1278 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1283 devfs_readdir(struct vop_readdir_args *ap)
1288 struct devfs_dirent *dd;
1289 struct devfs_dirent *de;
1290 struct devfs_mount *dmp;
1292 int *tmp_ncookies = NULL;
1294 if (ap->a_vp->v_type != VDIR)
1298 if (uio->uio_offset < 0)
1302 * XXX: This is a temporary hack to get around this filesystem not
1303 * supporting cookies. We store the location of the ncookies pointer
1304 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1305 * and set the number of cookies to 0. We then set the pointer to
1306 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1307 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1308 * pointer to its original location before returning to the caller.
1310 if (ap->a_ncookies != NULL) {
1311 tmp_ncookies = ap->a_ncookies;
1312 *ap->a_ncookies = 0;
1313 ap->a_ncookies = NULL;
1316 dmp = VFSTODEVFS(ap->a_vp->v_mount);
1317 if (devfs_populate_vp(ap->a_vp) != 0) {
1318 if (tmp_ncookies != NULL)
1319 ap->a_ncookies = tmp_ncookies;
1323 de = ap->a_vp->v_data;
1325 TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1326 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1327 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1329 if (devfs_prison_check(dd, uio->uio_td))
1331 if (dd->de_dirent->d_type == DT_DIR)
1336 MPASS(dp->d_reclen == GENERIC_DIRSIZ(dp));
1337 if (dp->d_reclen > uio->uio_resid)
1339 dp->d_fileno = de->de_inode;
1340 if (off >= uio->uio_offset) {
1341 error = vfs_read_dirent(ap, dp, off);
1345 off += dp->d_reclen;
1347 sx_xunlock(&dmp->dm_lock);
1348 uio->uio_offset = off;
1351 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1354 if (tmp_ncookies != NULL)
1355 ap->a_ncookies = tmp_ncookies;
1361 devfs_readlink(struct vop_readlink_args *ap)
1363 struct devfs_dirent *de;
1365 de = ap->a_vp->v_data;
1366 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1370 devfs_reclaim(struct vop_reclaim_args *ap)
1373 struct devfs_dirent *de;
1376 mtx_lock(&devfs_de_interlock);
1379 de->de_vnode = NULL;
1382 mtx_unlock(&devfs_de_interlock);
1383 vnode_destroy_vobject(vp);
1388 devfs_reclaim_vchr(struct vop_reclaim_args *ap)
1394 MPASS(vp->v_type == VCHR);
1403 dev->si_usecount -= vp->v_usecount;
1412 devfs_remove(struct vop_remove_args *ap)
1414 struct vnode *dvp = ap->a_dvp;
1415 struct vnode *vp = ap->a_vp;
1416 struct devfs_dirent *dd;
1417 struct devfs_dirent *de, *de_covered;
1418 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1420 ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1421 ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1423 sx_xlock(&dmp->dm_lock);
1424 dd = ap->a_dvp->v_data;
1426 if (de->de_cdp == NULL) {
1427 TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1428 if (de->de_dirent->d_type == DT_LNK) {
1429 de_covered = devfs_find(dd, de->de_dirent->d_name,
1430 de->de_dirent->d_namlen, 0);
1431 if (de_covered != NULL)
1432 de_covered->de_flags &= ~DE_COVERED;
1434 /* We need to unlock dvp because devfs_delete() may lock it. */
1438 devfs_delete(dmp, de, 0);
1439 sx_xunlock(&dmp->dm_lock);
1441 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1442 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1444 de->de_flags |= DE_WHITEOUT;
1445 sx_xunlock(&dmp->dm_lock);
1451 * Revoke is called on a tty when a terminal session ends. The vnode
1452 * is orphaned by setting v_op to deadfs so we need to let go of it
1453 * as well so that we create a new one next time around.
1457 devfs_revoke(struct vop_revoke_args *ap)
1459 struct vnode *vp = ap->a_vp, *vp2;
1461 struct cdev_priv *cdp;
1462 struct devfs_dirent *de;
1465 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1468 cdp = cdev2priv(dev);
1481 mtx_lock(&devfs_de_interlock);
1484 for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1485 de = cdp->cdp_dirents[i];
1493 mtx_unlock(&devfs_de_interlock);
1494 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1508 mtx_unlock(&devfs_de_interlock);
1513 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1514 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1516 dev_rel(&cdp->cdp_c);
1520 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1525 devfs_rioctl(struct vop_ioctl_args *ap)
1528 struct devfs_mount *dmp;
1532 vn_lock(vp, LK_SHARED | LK_RETRY);
1533 if (vp->v_iflag & VI_DOOMED) {
1537 dmp = VFSTODEVFS(vp->v_mount);
1538 sx_xlock(&dmp->dm_lock);
1540 DEVFS_DMP_HOLD(dmp);
1541 devfs_populate(dmp);
1542 if (DEVFS_DMP_DROP(dmp)) {
1543 sx_xunlock(&dmp->dm_lock);
1544 devfs_unmount_final(dmp);
1547 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1548 sx_xunlock(&dmp->dm_lock);
1553 devfs_rread(struct vop_read_args *ap)
1556 if (ap->a_vp->v_type != VDIR)
1558 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1562 devfs_setattr(struct vop_setattr_args *ap)
1564 struct devfs_dirent *de;
1575 if ((vap->va_type != VNON) ||
1576 (vap->va_nlink != VNOVAL) ||
1577 (vap->va_fsid != VNOVAL) ||
1578 (vap->va_fileid != VNOVAL) ||
1579 (vap->va_blocksize != VNOVAL) ||
1580 (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1581 (vap->va_rdev != VNOVAL) ||
1582 ((int)vap->va_bytes != VNOVAL) ||
1583 (vap->va_gen != VNOVAL)) {
1587 error = devfs_populate_vp(vp);
1592 if (vp->v_type == VDIR)
1596 if (vap->va_uid == (uid_t)VNOVAL)
1600 if (vap->va_gid == (gid_t)VNOVAL)
1604 if (uid != de->de_uid || gid != de->de_gid) {
1605 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1606 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1607 error = priv_check(td, PRIV_VFS_CHOWN);
1616 if (vap->va_mode != (mode_t)VNOVAL) {
1617 if (ap->a_cred->cr_uid != de->de_uid) {
1618 error = priv_check(td, PRIV_VFS_ADMIN);
1622 de->de_mode = vap->va_mode;
1626 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1627 error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1630 if (vap->va_atime.tv_sec != VNOVAL) {
1631 if (vp->v_type == VCHR)
1632 vp->v_rdev->si_atime = vap->va_atime;
1634 de->de_atime = vap->va_atime;
1636 if (vap->va_mtime.tv_sec != VNOVAL) {
1637 if (vp->v_type == VCHR)
1638 vp->v_rdev->si_mtime = vap->va_mtime;
1640 de->de_mtime = vap->va_mtime;
1646 if (vp->v_type == VCHR)
1647 vfs_timestamp(&vp->v_rdev->si_ctime);
1649 vfs_timestamp(&de->de_mtime);
1653 sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock);
1659 devfs_setlabel(struct vop_setlabel_args *ap)
1662 struct devfs_dirent *de;
1667 mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1668 mac_devfs_update(vp->v_mount, de, vp);
1675 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1678 return (vnops.fo_stat(fp, sb, cred, td));
1682 devfs_symlink(struct vop_symlink_args *ap)
1685 struct devfs_dirent *dd;
1686 struct devfs_dirent *de, *de_covered, *de_dotdot;
1687 struct devfs_mount *dmp;
1689 error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1692 dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1693 if (devfs_populate_vp(ap->a_dvp) != 0)
1696 dd = ap->a_dvp->v_data;
1697 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1698 de->de_flags = DE_USER;
1702 de->de_inode = alloc_unr(devfs_inos);
1704 de->de_dirent->d_type = DT_LNK;
1705 i = strlen(ap->a_target) + 1;
1706 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1707 bcopy(ap->a_target, de->de_symlink, i);
1709 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1711 de_covered = devfs_find(dd, de->de_dirent->d_name,
1712 de->de_dirent->d_namlen, 0);
1713 if (de_covered != NULL) {
1714 if ((de_covered->de_flags & DE_USER) != 0) {
1715 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1716 sx_xunlock(&dmp->dm_lock);
1719 KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1720 ("devfs_symlink: entry %p already covered", de_covered));
1721 de_covered->de_flags |= DE_COVERED;
1724 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */
1725 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */
1726 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1727 devfs_dir_ref_de(dmp, dd);
1728 devfs_rules_apply(dmp, de);
1730 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1734 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1737 return (vnops.fo_truncate(fp, length, cred, td));
1741 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1742 int flags, struct thread *td)
1745 int error, ioflag, ref;
1750 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1753 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1755 error = vnops.fo_write(fp, uio, cred, flags, td);
1758 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1759 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1760 if (ioflag & O_DIRECT)
1761 ioflag |= IO_DIRECT;
1762 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1764 resid = uio->uio_resid;
1766 error = dsw->d_write(dev, uio, ioflag);
1767 if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1768 devfs_timestamp(&dev->si_ctime);
1769 dev->si_mtime = dev->si_ctime;
1772 dev_relthread(dev, ref);
1774 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1779 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1780 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1795 * Ensure that file and memory protections are
1799 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1800 maxprot = VM_PROT_NONE;
1801 if ((prot & VM_PROT_EXECUTE) != 0)
1804 maxprot = VM_PROT_EXECUTE;
1805 if ((fp->f_flag & FREAD) != 0)
1806 maxprot |= VM_PROT_READ;
1807 else if ((prot & VM_PROT_READ) != 0)
1811 * If we are sharing potential changes via MAP_SHARED and we
1812 * are trying to get write permission although we opened it
1813 * without asking for it, bail out.
1815 * Note that most character devices always share mappings.
1816 * The one exception is that D_MMAP_ANON devices
1817 * (i.e. /dev/zero) permit private writable mappings.
1819 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests
1820 * as well as updating maxprot to permit writing for
1821 * D_MMAP_ANON devices rather than doing that here.
1823 if ((flags & MAP_SHARED) != 0) {
1824 if ((fp->f_flag & FWRITE) != 0)
1825 maxprot |= VM_PROT_WRITE;
1826 else if ((prot & VM_PROT_WRITE) != 0)
1829 maxprot &= cap_maxprot;
1832 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1836 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff,
1839 dev_relthread(dev, ref);
1843 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1846 vm_object_deallocate(object);
1851 dev2udev(struct cdev *x)
1855 return (cdev2priv(x)->cdp_inode);
1858 static struct fileops devfs_ops_f = {
1859 .fo_read = devfs_read_f,
1860 .fo_write = devfs_write_f,
1861 .fo_truncate = devfs_truncate_f,
1862 .fo_ioctl = devfs_ioctl_f,
1863 .fo_poll = devfs_poll_f,
1864 .fo_kqfilter = devfs_kqfilter_f,
1865 .fo_stat = devfs_stat_f,
1866 .fo_close = devfs_close_f,
1867 .fo_chmod = vn_chmod,
1868 .fo_chown = vn_chown,
1869 .fo_sendfile = vn_sendfile,
1871 .fo_fill_kinfo = vn_fill_kinfo,
1872 .fo_mmap = devfs_mmap_f,
1873 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
1876 /* Vops for non-CHR vnodes in /dev. */
1877 static struct vop_vector devfs_vnodeops = {
1878 .vop_default = &default_vnodeops,
1880 .vop_access = devfs_access,
1881 .vop_getattr = devfs_getattr,
1882 .vop_ioctl = devfs_rioctl,
1883 .vop_lookup = devfs_lookup,
1884 .vop_mknod = devfs_mknod,
1885 .vop_pathconf = devfs_pathconf,
1886 .vop_read = devfs_rread,
1887 .vop_readdir = devfs_readdir,
1888 .vop_readlink = devfs_readlink,
1889 .vop_reclaim = devfs_reclaim,
1890 .vop_remove = devfs_remove,
1891 .vop_revoke = devfs_revoke,
1892 .vop_setattr = devfs_setattr,
1894 .vop_setlabel = devfs_setlabel,
1896 .vop_symlink = devfs_symlink,
1897 .vop_vptocnp = devfs_vptocnp,
1900 /* Vops for VCHR vnodes in /dev. */
1901 static struct vop_vector devfs_specops = {
1902 .vop_default = &default_vnodeops,
1904 .vop_access = devfs_access,
1905 .vop_bmap = VOP_PANIC,
1906 .vop_close = devfs_close,
1907 .vop_create = VOP_PANIC,
1908 .vop_fsync = vop_stdfsync,
1909 .vop_getattr = devfs_getattr,
1910 .vop_ioctl = devfs_ioctl,
1911 .vop_link = VOP_PANIC,
1912 .vop_mkdir = VOP_PANIC,
1913 .vop_mknod = VOP_PANIC,
1914 .vop_open = devfs_open,
1915 .vop_pathconf = devfs_pathconf,
1916 .vop_poll = dead_poll,
1917 .vop_print = devfs_print,
1918 .vop_read = dead_read,
1919 .vop_readdir = VOP_PANIC,
1920 .vop_readlink = VOP_PANIC,
1921 .vop_reallocblks = VOP_PANIC,
1922 .vop_reclaim = devfs_reclaim_vchr,
1923 .vop_remove = devfs_remove,
1924 .vop_rename = VOP_PANIC,
1925 .vop_revoke = devfs_revoke,
1926 .vop_rmdir = VOP_PANIC,
1927 .vop_setattr = devfs_setattr,
1929 .vop_setlabel = devfs_setlabel,
1931 .vop_strategy = VOP_PANIC,
1932 .vop_symlink = VOP_PANIC,
1933 .vop_vptocnp = devfs_vptocnp,
1934 .vop_write = dead_write,
1938 * Our calling convention to the device drivers used to be that we passed
1939 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1940 * flags instead since that's what open(), close() and ioctl() takes and
1941 * we don't really want vnode.h in device drivers.
1942 * We solved the source compatibility by redefining some vnode flags to
1943 * be the same as the fcntl ones and by sending down the bitwise OR of
1944 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody
1945 * pulls the rug out under this.
1947 CTASSERT(O_NONBLOCK == IO_NDELAY);
1948 CTASSERT(O_FSYNC == IO_SYNC);