BSD 4_4 release
[unix-history] / usr / src / sys / ufs / ufs / ufs_vnops.c
index 0830fd1..71a395e 100644 (file)
-/*     ufs_vnops.c     4.51    83/02/10        */
-
-#include "../h/param.h"
-#include "../h/systm.h"
-#include "../h/dir.h"
-#include "../h/user.h"
-#include "../h/kernel.h"
-#include "../h/file.h"
-#include "../h/stat.h"
-#include "../h/inode.h"
-#include "../h/fs.h"
-#include "../h/buf.h"
-#include "../h/proc.h"
-#include "../h/quota.h"
-#include "../h/descrip.h"
-#include "../h/uio.h"
-#include "../h/socket.h"
-#include "../h/socketvar.h"
-#include "../h/nami.h"
+/*
+ * Copyright (c) 1982, 1986, 1989, 1993
+ *     The Regents of the University of California.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *     This product includes software developed by the University of
+ *     California, Berkeley and its contributors.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ *     @(#)ufs_vnops.c 8.1 (Berkeley) 6/11/93
+ */
+
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/namei.h>
+#include <sys/resourcevar.h>
+#include <sys/kernel.h>
+#include <sys/file.h>
+#include <sys/stat.h>
+#include <sys/buf.h>
+#include <sys/proc.h>
+#include <sys/conf.h>
+#include <sys/mount.h>
+#include <sys/vnode.h>
+#include <sys/malloc.h>
+#include <sys/dirent.h>
+
+#include <vm/vm.h>
+
+#include <miscfs/specfs/specdev.h>
+
+#include <ufs/ufs/lockf.h>
+#include <ufs/ufs/quota.h>
+#include <ufs/ufs/inode.h>
+#include <ufs/ufs/dir.h>
+#include <ufs/ufs/ufsmount.h>
+#include <ufs/ufs/ufs_extern.h>
+
+static int ufs_chmod __P((struct vnode *, int, struct ucred *, struct proc *));
+static int ufs_chown
+       __P((struct vnode *, uid_t, gid_t, struct ucred *, struct proc *));
+
+union _qcvt {
+       quad_t qcvt;
+       long val[2];
+};
+#define SETHIGH(q, h) { \
+       union _qcvt tmp; \
+       tmp.qcvt = (q); \
+       tmp.val[_QUAD_HIGHWORD] = (h); \
+       (q) = tmp.qcvt; \
+}
+#define SETLOW(q, l) { \
+       union _qcvt tmp; \
+       tmp.qcvt = (q); \
+       tmp.val[_QUAD_LOWWORD] = (l); \
+       (q) = tmp.qcvt; \
+}
 
 /*
 
 /*
- * Change current working directory (``.'').
+ * Create a regular file
  */
  */
-chdir()
+int
+ufs_create(ap)
+       struct vop_create_args /* {
+               struct vnode *a_dvp;
+               struct vnode **a_vpp;
+               struct componentname *a_cnp;
+               struct vattr *a_vap;
+       } */ *ap;
 {
 {
+       int error;
 
 
-       chdirec(&u.u_cdir);
+       if (error =
+           ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
+           ap->a_dvp, ap->a_vpp, ap->a_cnp))
+               return (error);
+       return (0);
 }
 
 /*
 }
 
 /*
- * Change notion of root (``/'') directory.
+ * Mknod vnode call
  */
  */
-chroot()
+/* ARGSUSED */
+int
+ufs_mknod(ap)
+       struct vop_mknod_args /* {
+               struct vnode *a_dvp;
+               struct vnode **a_vpp;
+               struct componentname *a_cnp;
+               struct vattr *a_vap;
+       } */ *ap;
 {
 {
+       register struct vattr *vap = ap->a_vap;
+       register struct vnode **vpp = ap->a_vpp;
+       register struct inode *ip;
+       int error;
 
 
-       if (suser())
-               chdirec(&u.u_rdir);
+       if (error =
+           ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
+           ap->a_dvp, vpp, ap->a_cnp))
+               return (error);
+       ip = VTOI(*vpp);
+       ip->i_flag |= IACC|IUPD|ICHG;
+       if (vap->va_rdev != VNOVAL) {
+               /*
+                * Want to be able to use this to make badblock
+                * inodes, so don't truncate the dev number.
+                */
+               ip->i_rdev = vap->va_rdev;
+       }
+       /*
+        * Remove inode so that it will be reloaded by iget and
+        * checked to see if it is an alias of an existing entry
+        * in the inode cache.
+        */
+       vput(*vpp);
+       (*vpp)->v_type = VNON;
+       vgone(*vpp);
+       *vpp = 0;
+       return (0);
 }
 
 /*
 }
 
 /*
- * Common routine for chroot and chdir.
+ * Open called.
+ *
+ * Nothing to do.
  */
  */
-chdirec(ipp)
-       register struct inode **ipp;
+/* ARGSUSED */
+int
+ufs_open(ap)
+       struct vop_open_args /* {
+               struct vnode *a_vp;
+               int  a_mode;
+               struct ucred *a_cred;
+               struct proc *a_p;
+       } */ *ap;
 {
 {
-       register struct inode *ip;
-       struct a {
-               char    *fname;
-       };
-
-       ip = namei(uchar, LOOKUP, 1);
-       if (ip == NULL)
-               return;
-       if ((ip->i_mode&IFMT) != IFDIR) {
-               u.u_error = ENOTDIR;
-               goto bad;
+
+       /*
+        * Files marked append-only must be opened for appending.
+        */
+       if ((VTOI(ap->a_vp)->i_flags & APPEND) &&
+           (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
+               return (EPERM);
+       return (0);
+}
+
+/*
+ * Close called
+ *
+ * Update the times on the inode.
+ */
+/* ARGSUSED */
+int
+ufs_close(ap)
+       struct vop_close_args /* {
+               struct vnode *a_vp;
+               int  a_fflag;
+               struct ucred *a_cred;
+               struct proc *a_p;
+       } */ *ap;
+{
+       register struct vnode *vp = ap->a_vp;
+       register struct inode *ip = VTOI(vp);
+
+       if (vp->v_usecount > 1 && !(ip->i_flag & ILOCKED))
+               ITIMES(ip, &time, &time);
+       return (0);
+}
+
+/*
+ * Check mode permission on inode pointer. Mode is READ, WRITE or EXEC.
+ * The mode is shifted to select the owner/group/other fields. The
+ * super user is granted all permissions.
+ */
+int
+ufs_access(ap)
+       struct vop_access_args /* {
+               struct vnode *a_vp;
+               int  a_mode;
+               struct ucred *a_cred;
+               struct proc *a_p;
+       } */ *ap;
+{
+       register struct vnode *vp = ap->a_vp;
+       register struct inode *ip = VTOI(vp);
+       register struct ucred *cred = ap->a_cred;
+       mode_t mode = ap->a_mode;
+       register gid_t *gp;
+       int i, error;
+
+#ifdef DIAGNOSTIC
+       if (!VOP_ISLOCKED(vp)) {
+               vprint("ufs_access: not locked", vp);
+               panic("ufs_access: not locked");
        }
        }
-       if (access(ip, IEXEC))
-               goto bad;
-       iunlock(ip);
-       if (*ipp)
-               irele(*ipp);
-       *ipp = ip;
-       return;
+#endif
+#ifdef QUOTA
+       if (mode & VWRITE) {
+               switch (vp->v_type) {
+               case VREG: case VDIR: case VLNK:
+                       if (error = getinoquota(ip))
+                               return (error);
+               }
+       }
+#endif /* QUOTA */
+       if ((mode & VWRITE) && (ip->i_flags & IMMUTABLE))
+               return (EPERM);
+       /*
+        * If you're the super-user, you always get access.
+        */
+       if (cred->cr_uid == 0)
+               return (0);
+       /*
+        * Access check is based on only one of owner, group, public.
+        * If not owner, then check group. If not a member of the
+        * group, then check public access.
+        */
+       if (cred->cr_uid != ip->i_uid) {
+               mode >>= 3;
+               gp = cred->cr_groups;
+               for (i = 0; i < cred->cr_ngroups; i++, gp++)
+                       if (ip->i_gid == *gp)
+                               goto found;
+               mode >>= 3;
+found:
+               ;
+       }
+       return ((ip->i_mode & mode) == mode ? 0 : EACCES);
+}
 
 
-bad:
-       iput(ip);
+/* ARGSUSED */
+int
+ufs_getattr(ap)
+       struct vop_getattr_args /* {
+               struct vnode *a_vp;
+               struct vattr *a_vap;
+               struct ucred *a_cred;
+               struct proc *a_p;
+       } */ *ap;
+{
+       register struct vnode *vp = ap->a_vp;
+       register struct inode *ip = VTOI(vp);
+       register struct vattr *vap = ap->a_vap;
+
+       ITIMES(ip, &time, &time);
+       /*
+        * Copy from inode table
+        */
+       vap->va_fsid = ip->i_dev;
+       vap->va_fileid = ip->i_number;
+       vap->va_mode = ip->i_mode & ~IFMT;
+       vap->va_nlink = ip->i_nlink;
+       vap->va_uid = ip->i_uid;
+       vap->va_gid = ip->i_gid;
+       vap->va_rdev = (dev_t)ip->i_rdev;
+       vap->va_size = ip->i_din.di_size;
+       vap->va_atime = ip->i_atime;
+       vap->va_mtime = ip->i_mtime;
+       vap->va_ctime = ip->i_ctime;
+       vap->va_flags = ip->i_flags;
+       vap->va_gen = ip->i_gen;
+       /* this doesn't belong here */
+       if (vp->v_type == VBLK)
+               vap->va_blocksize = BLKDEV_IOSIZE;
+       else if (vp->v_type == VCHR)
+               vap->va_blocksize = MAXBSIZE;
+       else
+               vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
+       vap->va_bytes = dbtob(ip->i_blocks);
+       vap->va_type = vp->v_type;
+       vap->va_filerev = ip->i_modrev;
+       return (0);
 }
 
 /*
 }
 
 /*
- * Open system call.
+ * Set attribute vnode op. called from several syscalls
  */
  */
-open()
+int
+ufs_setattr(ap)
+       struct vop_setattr_args /* {
+               struct vnode *a_vp;
+               struct vattr *a_vap;
+               struct ucred *a_cred;
+               struct proc *a_p;
+       } */ *ap;
 {
 {
-       register struct inode *ip;
-       register struct a {
-               char    *fname;
-               int     flags;
-               int     mode;
-       } *uap;
-       int checkpermissions = 1, flags;
-
-       uap = (struct a *)u.u_ap;
-       flags = uap->flags + 1;
-       if ((flags&FTRUNCATE) && (flags&FWRITE) == 0) {
-               u.u_error = EINVAL;
-               return;
+       register struct vattr *vap = ap->a_vap;
+       register struct vnode *vp = ap->a_vp;
+       register struct inode *ip = VTOI(vp);
+       register struct ucred *cred = ap->a_cred;
+       register struct proc *p = ap->a_p;
+       struct timeval atimeval, mtimeval;
+       int error;
+
+       /*
+        * Check for unsettable attributes.
+        */
+       if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
+           (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
+           (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
+           ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
+               return (EINVAL);
        }
        }
-       if (flags&FCREATE) {
-               ip = namei(uchar, CREATE, 1);
-               if (ip == NULL) {
-                       if (u.u_error)
-                               return;
-                       ip = maknode(uap->mode&07777&(~ISVTX));
-                       checkpermissions = 0;
-                       flags &= ~FTRUNCATE;
+       if (vap->va_flags != VNOVAL) {
+               if (cred->cr_uid != ip->i_uid &&
+                   (error = suser(cred, &p->p_acflag)))
+                       return (error);
+               if (cred->cr_uid == 0) {
+                       if ((ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) &&
+                           securelevel > 0)
+                               return (EPERM);
+                       ip->i_flags = vap->va_flags;
+               } else {
+                       if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND))
+                               return (EPERM);
+                       ip->i_flags &= SF_SETTABLE;
+                       ip->i_flags |= (vap->va_flags & UF_SETTABLE);
                }
                }
-       } else
-               ip = namei(uchar, LOOKUP, 1);
-       if (ip == NULL)
-               return;
-       open1(ip, flags, checkpermissions);
+               ip->i_flag |= ICHG;
+               if (vap->va_flags & (IMMUTABLE | APPEND))
+                       return (0);
+       }
+       if (ip->i_flags & (IMMUTABLE | APPEND))
+               return (EPERM);
+       /*
+        * Go through the fields and update iff not VNOVAL.
+        */
+       if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL)
+               if (error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred, p))
+                       return (error);
+       if (vap->va_size != VNOVAL) {
+               if (vp->v_type == VDIR)
+                       return (EISDIR);
+               if (error = VOP_TRUNCATE(vp, vap->va_size, 0, cred, p))
+                       return (error);
+       }
+       ip = VTOI(vp);
+       if (vap->va_atime.ts_sec != VNOVAL || vap->va_mtime.ts_sec != VNOVAL) {
+               if (cred->cr_uid != ip->i_uid &&
+                   (error = suser(cred, &p->p_acflag)) &&
+                   ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || 
+                   (error = VOP_ACCESS(vp, VWRITE, cred, p))))
+                       return (error);
+               if (vap->va_atime.ts_sec != VNOVAL)
+                       ip->i_flag |= IACC;
+               if (vap->va_mtime.ts_sec != VNOVAL)
+                       ip->i_flag |= IUPD | ICHG;
+               atimeval.tv_sec = vap->va_atime.ts_sec;
+               atimeval.tv_usec = vap->va_atime.ts_nsec / 1000;
+               mtimeval.tv_sec = vap->va_mtime.ts_sec;
+               mtimeval.tv_usec = vap->va_mtime.ts_nsec / 1000;
+               if (error = VOP_UPDATE(vp, &atimeval, &mtimeval, 1))
+                       return (error);
+       }
+       error = 0;
+       if (vap->va_mode != (mode_t)VNOVAL)
+               error = ufs_chmod(vp, (int)vap->va_mode, cred, p);
+       return (error);
 }
 
 }
 
-#ifndef NOCOMPAT
 /*
 /*
- * Creat system call.
+ * Change the mode on a file.
+ * Inode must be locked before calling.
  */
  */
-ocreat()
+static int
+ufs_chmod(vp, mode, cred, p)
+       register struct vnode *vp;
+       register int mode;
+       register struct ucred *cred;
+       struct proc *p;
 {
 {
-       register struct inode *ip;
-       register struct a {
-               char    *fname;
-               int     fmode;
-       } *uap;
-
-       uap = (struct a *)u.u_ap;
-       ip = namei(uchar, CREATE, 1);
-       if (ip == NULL) {
-               if (u.u_error)
-                       return;
-               ip = maknode(uap->fmode&07777&(~ISVTX));
-               if (ip == NULL)
-                       return;
-               open1(ip, FWRITE, 0);
-       } else
-               open1(ip, FWRITE|FTRUNCATE, 1);
+       register struct inode *ip = VTOI(vp);
+       int error;
+
+       if (cred->cr_uid != ip->i_uid &&
+           (error = suser(cred, &p->p_acflag)))
+               return (error);
+       if (cred->cr_uid) {
+               if (vp->v_type != VDIR && (mode & ISVTX))
+                       return (EFTYPE);
+               if (!groupmember(ip->i_gid, cred) && (mode & ISGID))
+                       return (EPERM);
+       }
+       ip->i_mode &= ~07777;
+       ip->i_mode |= mode & 07777;
+       ip->i_flag |= ICHG;
+       if ((vp->v_flag & VTEXT) && (ip->i_mode & ISVTX) == 0)
+               (void) vnode_pager_uncache(vp);
+       return (0);
 }
 }
-#endif
 
 /*
 
 /*
- * Common code for open and creat.
- * Check permissions (if we haven't done so already),
- * allocate an open file structure, and call
- * the device open routine, if any.
+ * Perform chown operation on inode ip;
+ * inode must be locked prior to call.
  */
  */
-open1(ip, mode, checkpermissions)
-       register struct inode *ip;
-       register mode;
+static int
+ufs_chown(vp, uid, gid, cred, p)
+       register struct vnode *vp;
+       uid_t uid;
+       gid_t gid;
+       struct ucred *cred;
+       struct proc *p;
 {
 {
-       register struct file *fp;
-       int i, flags;
+       register struct inode *ip = VTOI(vp);
+       uid_t ouid;
+       gid_t ogid;
+       int error = 0;
+#ifdef QUOTA
+       register int i;
+       long change;
+#endif
 
 
-       if (checkpermissions) {
-               if (mode&FREAD)
-                       if (access(ip, IREAD))
-                               goto bad;
-               if (mode&FWRITE) {
-                       if (access(ip, IWRITE))
-                               goto bad;
-                       if ((ip->i_mode&IFMT) == IFDIR) {
-                               u.u_error = EISDIR;
-                               goto bad;
-                       }
+       if (uid == (uid_t)VNOVAL)
+               uid = ip->i_uid;
+       if (gid == (gid_t)VNOVAL)
+               gid = ip->i_gid;
+       /*
+        * If we don't own the file, are trying to change the owner
+        * of the file, or are not a member of the target group,
+        * the caller must be superuser or the call fails.
+        */
+       if ((cred->cr_uid != ip->i_uid || uid != ip->i_uid ||
+           !groupmember((gid_t)gid, cred)) &&
+           (error = suser(cred, &p->p_acflag)))
+               return (error);
+       ouid = ip->i_uid;
+       ogid = ip->i_gid;
+#ifdef QUOTA
+       if (error = getinoquota(ip))
+               return (error);
+       if (ouid == uid) {
+               dqrele(vp, ip->i_dquot[USRQUOTA]);
+               ip->i_dquot[USRQUOTA] = NODQUOT;
+       }
+       if (ogid == gid) {
+               dqrele(vp, ip->i_dquot[GRPQUOTA]);
+               ip->i_dquot[GRPQUOTA] = NODQUOT;
+       }
+       change = ip->i_blocks;
+       (void) chkdq(ip, -change, cred, CHOWN);
+       (void) chkiq(ip, -1, cred, CHOWN);
+       for (i = 0; i < MAXQUOTAS; i++) {
+               dqrele(vp, ip->i_dquot[i]);
+               ip->i_dquot[i] = NODQUOT;
+       }
+#endif
+       ip->i_uid = uid;
+       ip->i_gid = gid;
+#ifdef QUOTA
+       if ((error = getinoquota(ip)) == 0) {
+               if (ouid == uid) {
+                       dqrele(vp, ip->i_dquot[USRQUOTA]);
+                       ip->i_dquot[USRQUOTA] = NODQUOT;
+               }
+               if (ogid == gid) {
+                       dqrele(vp, ip->i_dquot[GRPQUOTA]);
+                       ip->i_dquot[GRPQUOTA] = NODQUOT;
+               }
+               if ((error = chkdq(ip, change, cred, CHOWN)) == 0) {
+                       if ((error = chkiq(ip, 1, cred, CHOWN)) == 0)
+                               goto good;
+                       else
+                               (void) chkdq(ip, -change, cred, CHOWN|FORCE);
+               }
+               for (i = 0; i < MAXQUOTAS; i++) {
+                       dqrele(vp, ip->i_dquot[i]);
+                       ip->i_dquot[i] = NODQUOT;
                }
        }
                }
        }
+       ip->i_uid = ouid;
+       ip->i_gid = ogid;
+       if (getinoquota(ip) == 0) {
+               if (ouid == uid) {
+                       dqrele(vp, ip->i_dquot[USRQUOTA]);
+                       ip->i_dquot[USRQUOTA] = NODQUOT;
+               }
+               if (ogid == gid) {
+                       dqrele(vp, ip->i_dquot[GRPQUOTA]);
+                       ip->i_dquot[GRPQUOTA] = NODQUOT;
+               }
+               (void) chkdq(ip, change, cred, FORCE|CHOWN);
+               (void) chkiq(ip, 1, cred, FORCE|CHOWN);
+               (void) getinoquota(ip);
+       }
+       return (error);
+good:
+       if (getinoquota(ip))
+               panic("chown: lost quota");
+#endif /* QUOTA */
+       if (ouid != uid || ogid != gid)
+               ip->i_flag |= ICHG;
+       if (ouid != uid && cred->cr_uid != 0)
+               ip->i_mode &= ~ISUID;
+       if (ogid != gid && cred->cr_uid != 0)
+               ip->i_mode &= ~ISGID;
+       return (0);
+}
+
+/* ARGSUSED */
+int
+ufs_ioctl(ap)
+       struct vop_ioctl_args /* {
+               struct vnode *a_vp;
+               int  a_command;
+               caddr_t  a_data;
+               int  a_fflag;
+               struct ucred *a_cred;
+               struct proc *a_p;
+       } */ *ap;
+{
+
+       return (ENOTTY);
+}
+
+/* ARGSUSED */
+int
+ufs_select(ap)
+       struct vop_select_args /* {
+               struct vnode *a_vp;
+               int  a_which;
+               int  a_fflags;
+               struct ucred *a_cred;
+               struct proc *a_p;
+       } */ *ap;
+{
 
        /*
 
        /*
-        * Check locking on inode.  Release "inode lock"
-        * while doing so in case we block inside flocki.
+        * We should really check to see if I/O is possible.
         */
         */
-       flags = 0;
-       if (mode&(FSHLOCK|FEXLOCK)) {
-               iunlock(ip);
-               flags = flocki(ip, 0, mode);
-               ilock(ip);
-               if (u.u_error)
-                       goto bad;
-       }
-       if (mode&FTRUNCATE)
-               itrunc(ip, (u_long)0);
-       iunlock(ip);
-       if ((fp = falloc()) == NULL)
+       return (1);
+}
+
+/*
+ * Mmap a file
+ *
+ * NB Currently unsupported.
+ */
+/* ARGSUSED */
+int
+ufs_mmap(ap)
+       struct vop_mmap_args /* {
+               struct vnode *a_vp;
+               int  a_fflags;
+               struct ucred *a_cred;
+               struct proc *a_p;
+       } */ *ap;
+{
+
+       return (EINVAL);
+}
+
+/*
+ * Seek on a file
+ *
+ * Nothing to do, so just return.
+ */
+/* ARGSUSED */
+int
+ufs_seek(ap)
+       struct vop_seek_args /* {
+               struct vnode *a_vp;
+               off_t  a_oldoff;
+               off_t  a_newoff;
+               struct ucred *a_cred;
+       } */ *ap;
+{
+
+       return (0);
+}
+
+/*
+ * ufs remove
+ * Hard to avoid races here, especially
+ * in unlinking directories.
+ */
+int
+ufs_remove(ap)
+       struct vop_remove_args /* {
+               struct vnode *a_dvp;
+               struct vnode *a_vp;
+               struct componentname *a_cnp;
+       } */ *ap;
+{
+       register struct inode *ip;
+       register struct vnode *vp = ap->a_vp;
+       register struct vnode *dvp = ap->a_dvp;
+       int error;
+
+       ip = VTOI(vp);
+       if ((ip->i_flags & (IMMUTABLE | APPEND)) ||
+           (VTOI(dvp)->i_flags & APPEND)) {
+               error = EPERM;
                goto out;
                goto out;
-       fp->f_flag = mode & FMODES;
-       fp->f_type = DTYPE_FILE;
-       i = u.u_r.r_val1;
-       fp->f_inode = ip;
-       u.u_error = openi(ip, mode);
-       if (u.u_error == 0) {
-               u.u_pofile[i] = flags;
-               return;
        }
        }
-       u.u_ofile[i] = NULL;
-       fp->f_count--;
+       if ((error = ufs_dirremove(dvp, ap->a_cnp)) == 0) {
+               ip->i_nlink--;
+               ip->i_flag |= ICHG;
+       }
 out:
 out:
-       irele(ip);
-       return;
-bad:
-       iput(ip);
+       if (dvp == vp)
+               vrele(vp);
+       else
+               vput(vp);
+       vput(dvp);
+       return (error);
 }
 
 /*
 }
 
 /*
- * Mknod system call
+ * link vnode call
  */
  */
-mknod()
+int
+ufs_link(ap)
+       struct vop_link_args /* {
+               struct vnode *a_vp;
+               struct vnode *a_tdvp;
+               struct componentname *a_cnp;
+       } */ *ap;
 {
 {
+       register struct vnode *vp = ap->a_vp;
+       register struct vnode *tdvp = ap->a_tdvp;
+       register struct componentname *cnp = ap->a_cnp;
        register struct inode *ip;
        register struct inode *ip;
-       register struct a {
-               char    *fname;
-               int     fmode;
-               int     dev;
-       } *uap;
-
-       uap = (struct a *)u.u_ap;
-       if (suser()) {
-               ip = namei(uchar, CREATE, 0);
-               if (ip != NULL) {
-                       u.u_error = EEXIST;
-                       goto out;
+       struct timeval tv;
+       int error;
+
+#ifdef DIAGNOSTIC
+       if ((cnp->cn_flags & HASBUF) == 0)
+               panic("ufs_link: no name");
+#endif
+       if (vp->v_mount != tdvp->v_mount) {
+               VOP_ABORTOP(vp, cnp);
+               error = EXDEV;
+               goto out2;
+       }
+       if (vp != tdvp && (error = VOP_LOCK(tdvp))) {
+               VOP_ABORTOP(vp, cnp);
+               goto out2;
+       }
+       ip = VTOI(tdvp);
+       if ((nlink_t)ip->i_nlink >= LINK_MAX) {
+               VOP_ABORTOP(vp, cnp);
+               error = EMLINK;
+               goto out1;
+       }
+       if (ip->i_flags & (IMMUTABLE | APPEND)) {
+               VOP_ABORTOP(vp, cnp);
+               error = EPERM;
+               goto out1;
+       }
+       ip->i_nlink++;
+       ip->i_flag |= ICHG;
+       tv = time;
+       error = VOP_UPDATE(tdvp, &tv, &tv, 1);
+       if (!error)
+               error = ufs_direnter(ip, vp, cnp);
+       if (error) {
+               ip->i_nlink--;
+               ip->i_flag |= ICHG;
+       }
+       FREE(cnp->cn_pnbuf, M_NAMEI);
+out1:
+       if (vp != tdvp)
+               VOP_UNLOCK(tdvp);
+out2:
+       vput(vp);
+       return (error);
+}
+
+
+
+/*
+ * relookup - lookup a path name component
+ *    Used by lookup to re-aquire things.
+ */
+int
+relookup(dvp, vpp, cnp)
+       struct vnode *dvp, **vpp;
+       struct componentname *cnp;
+{
+       register struct vnode *dp = 0;  /* the directory we are searching */
+       struct vnode *tdp;              /* saved dp */
+       struct mount *mp;               /* mount table entry */
+       int docache;                    /* == 0 do not cache last component */
+       int wantparent;                 /* 1 => wantparent or lockparent flag */
+       int rdonly;                     /* lookup read-only flag bit */
+       char *cp;                       /* DEBUG: check name ptr/len */
+       int newhash;                    /* DEBUG: check name hash */
+       int error = 0;
+
+       /*
+        * Setup: break out flag bits into variables.
+        */
+       wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
+       docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
+       if (cnp->cn_nameiop == DELETE ||
+           (wantparent && cnp->cn_nameiop != CREATE))
+               docache = 0;
+       rdonly = cnp->cn_flags & RDONLY;
+       cnp->cn_flags &= ~ISSYMLINK;
+       dp = dvp;
+       VOP_LOCK(dp);
+
+/* dirloop: */
+       /*
+        * Search a new directory.
+        *
+        * The cn_hash value is for use by vfs_cache.
+        * The last component of the filename is left accessible via
+        * cnp->cn_nameptr for callers that need the name. Callers needing
+        * the name set the SAVENAME flag. When done, they assume
+        * responsibility for freeing the pathname buffer.
+        */
+#ifdef NAMEI_DIAGNOSTIC
+       for (newhash = 0, cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
+               newhash += (unsigned char)*cp;
+       if (newhash != cnp->cn_hash)
+               panic("relookup: bad hash");
+       if (cnp->cn_namelen != cp - cnp->cn_nameptr)
+               panic ("relookup: bad len");
+       if (*cp != 0)
+               panic("relookup: not last component");
+       printf("{%s}: ", cnp->cn_nameptr);
+#endif
+
+       /*
+        * Check for degenerate name (e.g. / or "")
+        * which is a way of talking about a directory,
+        * e.g. like "/." or ".".
+        */
+       if (cnp->cn_nameptr[0] == '\0') {
+               if (cnp->cn_nameiop != LOOKUP || wantparent) {
+                       error = EISDIR;
+                       goto bad;
+               }
+               if (dp->v_type != VDIR) {
+                       error = ENOTDIR;
+                       goto bad;
                }
                }
+               if (!(cnp->cn_flags & LOCKLEAF))
+                       VOP_UNLOCK(dp);
+               *vpp = dp;
+               if (cnp->cn_flags & SAVESTART)
+                       panic("lookup: SAVESTART");
+               return (0);
        }
        }
-       if (u.u_error)
-               return;
-       ip = maknode(uap->fmode);
-       if (ip == NULL)
-               return;
-       if (uap->dev) {
+
+       if (cnp->cn_flags & ISDOTDOT)
+               panic ("relookup: lookup on dot-dot");
+
+       /*
+        * We now have a segment name to search for, and a directory to search.
+        */
+       if (error = VOP_LOOKUP(dp, vpp, cnp)) {
+#ifdef DIAGNOSTIC
+               if (*vpp != NULL)
+                       panic("leaf should be empty");
+#endif
+               if (error != EJUSTRETURN)
+                       goto bad;
                /*
                /*
-                * Want to be able to use this to make badblock
-                * inodes, so don't truncate the dev number.
+                * If creating and at end of pathname, then can consider
+                * allowing file to be created.
+                */
+               if (rdonly || (dvp->v_mount->mnt_flag & MNT_RDONLY)) {
+                       error = EROFS;
+                       goto bad;
+               }
+               /* ASSERT(dvp == ndp->ni_startdir) */
+               if (cnp->cn_flags & SAVESTART)
+                       VREF(dvp);
+               /*
+                * We return with ni_vp NULL to indicate that the entry
+                * doesn't currently exist, leaving a pointer to the
+                * (possibly locked) directory inode in ndp->ni_dvp.
                 */
                 */
-               ip->i_rdev = uap->dev;
-               ip->i_flag |= IACC|IUPD|ICHG;
+               return (0);
        }
        }
+       dp = *vpp;
 
 
-out:
-       iput(ip);
+#ifdef DIAGNOSTIC
+       /*
+        * Check for symbolic link
+        */
+       if (dp->v_type == VLNK && (cnp->cn_flags & FOLLOW))
+               panic ("relookup: symlink found.\n");
+#endif
+
+nextname:
+       /*
+        * Check for read-only file systems.
+        */
+       if (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME) {
+               /*
+                * Disallow directory write attempts on read-only
+                * file systems.
+                */
+               if (rdonly || (dp->v_mount->mnt_flag & MNT_RDONLY) ||
+                   (wantparent &&
+                    (dvp->v_mount->mnt_flag & MNT_RDONLY))) {
+                       error = EROFS;
+                       goto bad2;
+               }
+       }
+       /* ASSERT(dvp == ndp->ni_startdir) */
+       if (cnp->cn_flags & SAVESTART)
+               VREF(dvp);
+       
+       if (!wantparent)
+               vrele(dvp);
+       if ((cnp->cn_flags & LOCKLEAF) == 0)
+               VOP_UNLOCK(dp);
+       return (0);
+
+bad2:
+       if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
+               VOP_UNLOCK(dvp);
+       vrele(dvp);
+bad:
+       vput(dp);
+       *vpp = NULL;
+       return (error);
 }
 
 }
 
+
 /*
 /*
- * link system call
+ * Rename system call.
+ *     rename("foo", "bar");
+ * is essentially
+ *     unlink("bar");
+ *     link("foo", "bar");
+ *     unlink("foo");
+ * but ``atomically''.  Can't do full commit without saving state in the
+ * inode on disk which isn't feasible at this time.  Best we can do is
+ * always guarantee the target exists.
+ *
+ * Basic algorithm is:
+ *
+ * 1) Bump link count on source while we're linking it to the
+ *    target.  This also ensure the inode won't be deleted out
+ *    from underneath us while we work (it may be truncated by
+ *    a concurrent `trunc' or `open' for creation).
+ * 2) Link source to destination.  If destination already exists,
+ *    delete it first.
+ * 3) Unlink source reference to inode if still around. If a
+ *    directory was moved and the parent of the destination
+ *    is different from the source, patch the ".." entry in the
+ *    directory.
  */
  */
-link()
+int
+ufs_rename(ap)
+       struct vop_rename_args  /* {
+               struct vnode *a_fdvp;
+               struct vnode *a_fvp;
+               struct componentname *a_fcnp;
+               struct vnode *a_tdvp;
+               struct vnode *a_tvp;
+               struct componentname *a_tcnp;
+       } */ *ap;
 {
 {
-       register struct inode *ip, *xp;
-       register struct a {
-               char    *target;
-               char    *linkname;
-       } *uap;
-
-       uap = (struct a *)u.u_ap;
-       ip = namei(uchar, LOOKUP, 1); /* well, this routine is doomed anyhow */
-       if (ip == NULL)
-               return;
-       if ((ip->i_mode&IFMT) == IFDIR && !suser()) {
-               iput(ip);
-               return;
+       struct vnode *tvp = ap->a_tvp;
+       register struct vnode *tdvp = ap->a_tdvp;
+       struct vnode *fvp = ap->a_fvp;
+       register struct vnode *fdvp = ap->a_fdvp;
+       register struct componentname *tcnp = ap->a_tcnp;
+       register struct componentname *fcnp = ap->a_fcnp;
+       register struct inode *ip, *xp, *dp;
+       struct dirtemplate dirbuf;
+       struct timeval tv;
+       int doingdirectory = 0, oldparent = 0, newparent = 0;
+       int error = 0;
+       int fdvpneedsrele = 1, tdvpneedsrele = 1;
+       u_char namlen;
+
+#ifdef DIAGNOSTIC
+       if ((tcnp->cn_flags & HASBUF) == 0 ||
+           (fcnp->cn_flags & HASBUF) == 0)
+               panic("ufs_rename: no name");
+#endif
+       /*
+        * Check for cross-device rename.
+        */
+       if ((fvp->v_mount != tdvp->v_mount) ||
+           (tvp && (fvp->v_mount != tvp->v_mount))) {
+               error = EXDEV;
+abortit:
+               VOP_ABORTOP(tdvp, tcnp); /* XXX, why not in NFS? */
+               if (tdvp == tvp)
+                       vrele(tdvp);
+               else
+                       vput(tdvp);
+               if (tvp)
+                       vput(tvp);
+               VOP_ABORTOP(fdvp, fcnp); /* XXX, why not in NFS? */
+               vrele(fdvp);
+               vrele(fvp);
+               return (error);
+       }
+
+       /*
+        * Check if just deleting a link name.
+        */
+       if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) ||
+           (VTOI(tdvp)->i_flags & APPEND))) {
+               error = EPERM;
+               goto abortit;
+       }
+       if (fvp == tvp) {
+               if (fvp->v_type == VDIR) {
+                       error = EINVAL;
+                       goto abortit;
+               }
+               VOP_ABORTOP(fdvp, fcnp);
+               vrele(fdvp);
+               vrele(fvp);
+               vput(tdvp);
+               vput(tvp);
+               tcnp->cn_flags &= ~MODMASK;
+               tcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
+               if ((tcnp->cn_flags & SAVESTART) == 0)
+                       panic("ufs_rename: lost from startdir");
+               tcnp->cn_nameiop = DELETE;
+               (void) relookup(tdvp, &tvp, tcnp);
+               return (VOP_REMOVE(tdvp, tvp, tcnp));
+       }
+       if (error = VOP_LOCK(fvp))
+               goto abortit;
+       dp = VTOI(fdvp);
+       ip = VTOI(fvp);
+       if ((ip->i_flags & (IMMUTABLE | APPEND)) || (dp->i_flags & APPEND)) {
+               VOP_UNLOCK(fvp);
+               error = EPERM;
+               goto abortit;
+       }
+       if ((ip->i_mode & IFMT) == IFDIR) {
+               /*
+                * Avoid ".", "..", and aliases of "." for obvious reasons.
+                */
+               if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
+                   dp == ip || (fcnp->cn_flags&ISDOTDOT) ||
+                   (ip->i_flag & IRENAME)) {
+                       VOP_UNLOCK(fvp);
+                       error = EINVAL;
+                       goto abortit;
+               }
+               ip->i_flag |= IRENAME;
+               oldparent = dp->i_number;
+               doingdirectory++;
        }
        }
+       vrele(fdvp);
+
+       /*
+        * When the target exists, both the directory
+        * and target vnodes are returned locked.
+        */
+       dp = VTOI(tdvp);
+       xp = NULL;
+       if (tvp)
+               xp = VTOI(tvp);
+
+       /*
+        * 1) Bump link count while we're moving stuff
+        *    around.  If we crash somewhere before
+        *    completing our work, the link count
+        *    may be wrong, but correctable.
+        */
        ip->i_nlink++;
        ip->i_flag |= ICHG;
        ip->i_nlink++;
        ip->i_flag |= ICHG;
-       iupdat(ip, &time, &time, 1);
-       iunlock(ip);
-       u.u_dirp = (caddr_t)uap->linkname;
-       xp = namei(uchar, CREATE, 0);
-       if (xp != NULL) {
-               u.u_error = EEXIST;
-               iput(xp);
-               goto out;
+       tv = time;
+       if (error = VOP_UPDATE(fvp, &tv, &tv, 1)) {
+               VOP_UNLOCK(fvp);
+               goto bad;
        }
        }
-       if (u.u_error)
-               goto out;
-       if (u.u_pdir->i_dev != ip->i_dev) {
-               iput(u.u_pdir);
-               u.u_error = EXDEV;
-               goto out;
+
+       /*
+        * If ".." must be changed (ie the directory gets a new
+        * parent) then the source directory must not be in the
+        * directory heirarchy above the target, as this would
+        * orphan everything below the source directory. Also
+        * the user must have write permission in the source so
+        * as to be able to change "..". We must repeat the call 
+        * to namei, as the parent directory is unlocked by the
+        * call to checkpath().
+        */
+       error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_proc);
+       VOP_UNLOCK(fvp);
+       if (oldparent != dp->i_number)
+               newparent = dp->i_number;
+       if (doingdirectory && newparent) {
+               if (error)      /* write access check above */
+                       goto bad;
+               if (xp != NULL)
+                       vput(tvp);
+               if (error = ufs_checkpath(ip, dp, tcnp->cn_cred))
+                       goto out;
+               if ((tcnp->cn_flags & SAVESTART) == 0)
+                       panic("ufs_rename: lost to startdir");
+               if (error = relookup(tdvp, &tvp, tcnp))
+                       goto out;
+               dp = VTOI(tdvp);
+               xp = NULL;
+               if (tvp)
+                       xp = VTOI(tvp);
        }
        }
-       u.u_error = direnter(ip);
+       /*
+        * 2) If target doesn't exist, link the target
+        *    to the source and unlink the source. 
+        *    Otherwise, rewrite the target directory
+        *    entry to reference the source inode and
+        *    expunge the original entry's existence.
+        */
+       if (xp == NULL) {
+               if (dp->i_dev != ip->i_dev)
+                       panic("rename: EXDEV");
+               /*
+                * Account for ".." in new directory.
+                * When source and destination have the same
+                * parent we don't fool with the link count.
+                */
+               if (doingdirectory && newparent) {
+                       if ((nlink_t)dp->i_nlink >= LINK_MAX) {
+                               error = EMLINK;
+                               goto bad;
+                       }
+                       dp->i_nlink++;
+                       dp->i_flag |= ICHG;
+                       if (error = VOP_UPDATE(tdvp, &tv, &tv, 1))
+                               goto bad;
+               }
+               if (error = ufs_direnter(ip, tdvp, tcnp)) {
+                       if (doingdirectory && newparent) {
+                               dp->i_nlink--;
+                               dp->i_flag |= ICHG;
+                               (void)VOP_UPDATE(tdvp, &tv, &tv, 1);
+                       }
+                       goto bad;
+               }
+               vput(tdvp);
+       } else {
+               if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev)
+                       panic("rename: EXDEV");
+               /*
+                * Short circuit rename(foo, foo).
+                */
+               if (xp->i_number == ip->i_number)
+                       panic("rename: same file");
+               /*
+                * If the parent directory is "sticky", then the user must
+                * own the parent directory, or the destination of the rename,
+                * otherwise the destination may not be changed (except by
+                * root). This implements append-only directories.
+                */
+               if ((dp->i_mode & ISVTX) && tcnp->cn_cred->cr_uid != 0 &&
+                   tcnp->cn_cred->cr_uid != dp->i_uid &&
+                   xp->i_uid != tcnp->cn_cred->cr_uid) {
+                       error = EPERM;
+                       goto bad;
+               }
+               /*
+                * Target must be empty if a directory and have no links
+                * to it. Also, ensure source and target are compatible
+                * (both directories, or both not directories).
+                */
+               if ((xp->i_mode&IFMT) == IFDIR) {
+                       if (!ufs_dirempty(xp, dp->i_number, tcnp->cn_cred) || 
+                           xp->i_nlink > 2) {
+                               error = ENOTEMPTY;
+                               goto bad;
+                       }
+                       if (!doingdirectory) {
+                               error = ENOTDIR;
+                               goto bad;
+                       }
+                       cache_purge(tdvp);
+               } else if (doingdirectory) {
+                       error = EISDIR;
+                       goto bad;
+               }
+               if (error = ufs_dirrewrite(dp, ip, tcnp))
+                       goto bad;
+               /*
+                * If the target directory is in the same
+                * directory as the source directory,
+                * decrement the link count on the parent
+                * of the target directory.
+                */
+                if (doingdirectory && !newparent) {
+                       dp->i_nlink--;
+                       dp->i_flag |= ICHG;
+               }
+               vput(tdvp);
+               /*
+                * Adjust the link count of the target to
+                * reflect the dirrewrite above.  If this is
+                * a directory it is empty and there are
+                * no links to it, so we can squash the inode and
+                * any space associated with it.  We disallowed
+                * renaming over top of a directory with links to
+                * it above, as the remaining link would point to
+                * a directory without "." or ".." entries.
+                */
+               xp->i_nlink--;
+               if (doingdirectory) {
+                       if (--xp->i_nlink != 0)
+                               panic("rename: linked directory");
+                       error = VOP_TRUNCATE(tvp, (off_t)0, IO_SYNC,
+                           tcnp->cn_cred, tcnp->cn_proc);
+               }
+               xp->i_flag |= ICHG;
+               vput(tvp);
+               xp = NULL;
+       }
+
+       /*
+        * 3) Unlink the source.
+        */
+       fcnp->cn_flags &= ~MODMASK;
+       fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
+       if ((fcnp->cn_flags & SAVESTART) == 0)
+               panic("ufs_rename: lost from startdir");
+       (void) relookup(fdvp, &fvp, fcnp);
+       if (fvp != NULL) {
+               xp = VTOI(fvp);
+               dp = VTOI(fdvp);
+       } else {
+               /*
+                * From name has disappeared.
+                */
+               if (doingdirectory)
+                       panic("rename: lost dir entry");
+               vrele(ap->a_fvp);
+               return (0);
+       }
+       /*
+        * Ensure that the directory entry still exists and has not
+        * changed while the new name has been entered. If the source is
+        * a file then the entry may have been unlinked or renamed. In
+        * either case there is no further work to be done. If the source
+        * is a directory then it cannot have been rmdir'ed; its link
+        * count of three would cause a rmdir to fail with ENOTEMPTY.
+        * The IRENAME flag ensures that it cannot be moved by another
+        * rename.
+        */
+       if (xp != ip) {
+               if (doingdirectory)
+                       panic("rename: lost dir entry");
+       } else {
+               /*
+                * If the source is a directory with a
+                * new parent, the link count of the old
+                * parent directory must be decremented
+                * and ".." set to point to the new parent.
+                */
+               if (doingdirectory && newparent) {
+                       dp->i_nlink--;
+                       dp->i_flag |= ICHG;
+                       error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf,
+                               sizeof (struct dirtemplate), (off_t)0,
+                               UIO_SYSSPACE, IO_NODELOCKED, 
+                               tcnp->cn_cred, (int *)0, (struct proc *)0);
+                       if (error == 0) {
+#                              if (BYTE_ORDER == LITTLE_ENDIAN)
+                                       if (fvp->v_mount->mnt_maxsymlinklen <= 0)
+                                               namlen = dirbuf.dotdot_type;
+                                       else
+                                               namlen = dirbuf.dotdot_namlen;
+#                              else
+                                       namlen = dirbuf.dotdot_namlen;
+#                              endif
+                               if (namlen != 2 ||
+                                   dirbuf.dotdot_name[0] != '.' ||
+                                   dirbuf.dotdot_name[1] != '.') {
+                                       ufs_dirbad(xp, (doff_t)12,
+                                           "rename: mangled dir");
+                               } else {
+                                       dirbuf.dotdot_ino = newparent;
+                                       (void) vn_rdwr(UIO_WRITE, fvp,
+                                           (caddr_t)&dirbuf,
+                                           sizeof (struct dirtemplate),
+                                           (off_t)0, UIO_SYSSPACE,
+                                           IO_NODELOCKED|IO_SYNC,
+                                           tcnp->cn_cred, (int *)0,
+                                           (struct proc *)0);
+                                       cache_purge(fdvp);
+                               }
+                       }
+               }
+               error = ufs_dirremove(fdvp, fcnp);
+               if (!error) {
+                       xp->i_nlink--;
+                       xp->i_flag |= ICHG;
+               }
+               xp->i_flag &= ~IRENAME;
+       }
+       if (dp)
+               vput(fdvp);
+       if (xp)
+               vput(fvp);
+       vrele(ap->a_fvp);
+       return (error);
+
+bad:
+       if (xp)
+               vput(ITOV(xp));
+       vput(ITOV(dp));
 out:
 out:
-       if (u.u_error) {
+       if (VOP_LOCK(fvp) == 0) {
                ip->i_nlink--;
                ip->i_flag |= ICHG;
                ip->i_nlink--;
                ip->i_flag |= ICHG;
-       }
-       irele(ip);
+               vput(fvp);
+       } else
+               vrele(fvp);
+       return (error);
 }
 
 /*
 }
 
 /*
- * symlink -- make a symbolic link
+ * A virgin directory (no blushing please).
+ */
+static struct dirtemplate mastertemplate = {
+       0, 12, DT_DIR, 1, ".",
+       0, DIRBLKSIZ - 12, DT_DIR, 2, ".."
+};
+static struct odirtemplate omastertemplate = {
+       0, 12, 1, ".",
+       0, DIRBLKSIZ - 12, 2, ".."
+};
+
+/*
+ * Mkdir system call
  */
  */
-symlink()
+int
+ufs_mkdir(ap)
+       struct vop_mkdir_args /* {
+               struct vnode *a_dvp;
+               struct vnode **a_vpp;
+               struct componentname *a_cnp;
+               struct vattr *a_vap;
+       } */ *ap;
 {
 {
-       register struct a {
-               char    *target;
-               char    *linkname;
-       } *uap;
-       register struct inode *ip;
-       register char *tp;
-       register c, nc;
-
-       uap = (struct a *)u.u_ap;
-       tp = uap->target;
-       nc = 0;
-       while (c = fubyte(tp)) {
-               if (c < 0) {
-                       u.u_error = EFAULT;
-                       return;
-               }
-               tp++;
-               nc++;
+       register struct vnode *dvp = ap->a_dvp;
+       register struct vattr *vap = ap->a_vap;
+       register struct componentname *cnp = ap->a_cnp;
+       register struct inode *ip, *dp;
+       struct vnode *tvp;
+       struct dirtemplate dirtemplate, *dtp;
+       struct timeval tv;
+       int error, dmode;
+
+#ifdef DIAGNOSTIC
+       if ((cnp->cn_flags & HASBUF) == 0)
+               panic("ufs_mkdir: no name");
+#endif
+       dp = VTOI(dvp);
+       if ((nlink_t)dp->i_nlink >= LINK_MAX) {
+               error = EMLINK;
+               goto out;
        }
        }
-       u.u_dirp = uap->linkname;
-       ip = namei(uchar, CREATE, 0);
-       if (ip) {
-               iput(ip);
-               u.u_error = EEXIST;
-               return;
+       dmode = vap->va_mode & 0777;
+       dmode |= IFDIR;
+       /*
+        * Must simulate part of ufs_makeinode here to acquire the inode,
+        * but not have it entered in the parent directory. The entry is
+        * made later after writing "." and ".." entries.
+        */
+       if (error = VOP_VALLOC(dvp, dmode, cnp->cn_cred, &tvp))
+               goto out;
+       ip = VTOI(tvp);
+       ip->i_uid = cnp->cn_cred->cr_uid;
+       ip->i_gid = dp->i_gid;
+#ifdef QUOTA
+       if ((error = getinoquota(ip)) ||
+           (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
+               free(cnp->cn_pnbuf, M_NAMEI);
+               VOP_VFREE(tvp, ip->i_number, dmode);
+               vput(tvp);
+               vput(dvp);
+               return (error);
        }
        }
-       if (u.u_error)
-               return;
-       ip = maknode(IFLNK | 0777);
-       if (ip == NULL)
-               return;
-       u.u_error = rdwri(UIO_WRITE, ip, uap->target, nc, 0, 0, (int *)0);
-       /* handle u.u_error != 0 */
-       iput(ip);
+#endif
+       ip->i_flag |= IACC|IUPD|ICHG;
+       ip->i_mode = dmode;
+       tvp->v_type = VDIR;     /* Rest init'd in iget() */
+       ip->i_nlink = 2;
+       tv = time;
+       error = VOP_UPDATE(tvp, &tv, &tv, 1);
+
+       /*
+        * Bump link count in parent directory
+        * to reflect work done below.  Should
+        * be done before reference is created
+        * so reparation is possible if we crash.
+        */
+       dp->i_nlink++;
+       dp->i_flag |= ICHG;
+       if (error = VOP_UPDATE(dvp, &tv, &tv, 1))
+               goto bad;
+
+       /* Initialize directory with "." and ".." from static template. */
+       if (dvp->v_mount->mnt_maxsymlinklen > 0)
+               dtp = &mastertemplate;
+       else
+               dtp = (struct dirtemplate *)&omastertemplate;
+       dirtemplate = *dtp;
+       dirtemplate.dot_ino = ip->i_number;
+       dirtemplate.dotdot_ino = dp->i_number;
+       error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate,
+           sizeof (dirtemplate), (off_t)0, UIO_SYSSPACE,
+           IO_NODELOCKED|IO_SYNC, cnp->cn_cred, (int *)0, (struct proc *)0);
+       if (error) {
+               dp->i_nlink--;
+               dp->i_flag |= ICHG;
+               goto bad;
+       }
+       if (DIRBLKSIZ > VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
+               panic("ufs_mkdir: blksize"); /* XXX should grow with balloc() */
+       else {
+               ip->i_size = DIRBLKSIZ;
+               ip->i_flag |= ICHG;
+       }
+
+       /* Directory set up, now install it's entry in the parent directory. */
+       if (error = ufs_direnter(ip, dvp, cnp)) {
+               dp->i_nlink--;
+               dp->i_flag |= ICHG;
+       }
+bad:
+       /*
+        * No need to do an explicit VOP_TRUNCATE here, vrele will do this
+        * for us because we set the link count to 0.
+        */
+       if (error) {
+               ip->i_nlink = 0;
+               ip->i_flag |= ICHG;
+               vput(tvp);
+       } else
+               *ap->a_vpp = tvp;
+out:
+       FREE(cnp->cn_pnbuf, M_NAMEI);
+       vput(dvp);
+       return (error);
 }
 
 /*
 }
 
 /*
- * Unlink system call.
- * Hard to avoid races here, especially
- * in unlinking directories.
+ * Rmdir system call.
  */
  */
-unlink()
+int
+ufs_rmdir(ap)
+       struct vop_rmdir_args /* {
+               struct vnode *a_dvp;
+               struct vnode *a_vp;
+               struct componentname *a_cnp;
+       } */ *ap;
 {
 {
-       struct a {
-               char    *fname;
-       };
+       register struct vnode *vp = ap->a_vp;
+       register struct vnode *dvp = ap->a_dvp;
+       register struct componentname *cnp = ap->a_cnp;
        register struct inode *ip, *dp;
        register struct inode *ip, *dp;
+       int error;
 
 
-       ip = namei(uchar, DELETE | LOCKPARENT, 0);
-       if (ip == NULL)
-               return;
-       dp = u.u_pdir;
-       if ((ip->i_mode&IFMT) == IFDIR && !suser())
-               goto out;
+       ip = VTOI(vp);
+       dp = VTOI(dvp);
+       /*
+        * No rmdir "." please.
+        */
+       if (dp == ip) {
+               vrele(dvp);
+               vput(vp);
+               return (EINVAL);
+       }
        /*
        /*
-        * Don't unlink a mounted file.
+        * Verify the directory is empty (and valid).
+        * (Rmdir ".." won't be valid since
+        *  ".." will contain a reference to
+        *  the current directory and thus be
+        *  non-empty.)
         */
         */
-       if (ip->i_dev != dp->i_dev) {
-               u.u_error = EBUSY;
+       error = 0;
+       if (ip->i_nlink != 2 ||
+           !ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
+               error = ENOTEMPTY;
                goto out;
        }
                goto out;
        }
-       if (ip->i_flag&ITEXT)
-               xrele(ip);      /* try once to free text */
-       if (dirremove()) {
-               ip->i_nlink--;
-               ip->i_flag |= ICHG;
+       if ((dp->i_flags & APPEND) || (ip->i_flags & (IMMUTABLE | APPEND))) {
+               error = EPERM;
+               goto out;
        }
        }
+       /*
+        * Delete reference to directory before purging
+        * inode.  If we crash in between, the directory
+        * will be reattached to lost+found,
+        */
+       if (error = ufs_dirremove(dvp, cnp))
+               goto out;
+       dp->i_nlink--;
+       dp->i_flag |= ICHG;
+       cache_purge(dvp);
+       vput(dvp);
+       dvp = NULL;
+       /*
+        * Truncate inode.  The only stuff left
+        * in the directory is "." and "..".  The
+        * "." reference is inconsequential since
+        * we're quashing it.  The ".." reference
+        * has already been adjusted above.  We've
+        * removed the "." reference and the reference
+        * in the parent directory, but there may be
+        * other hard links so decrement by 2 and
+        * worry about them later.
+        */
+       ip->i_nlink -= 2;
+       error = VOP_TRUNCATE(vp, (off_t)0, IO_SYNC, cnp->cn_cred,
+           cnp->cn_proc);
+       cache_purge(ITOV(ip));
 out:
 out:
-       if (dp == ip)
-               irele(ip);
-       else
-               iput(ip);
-       iput(dp);
-}
-
-/*
- * Seek system call
- */
-lseek()
-{
-       register struct file *fp;
-       register struct a {
-               int     fd;
-               off_t   off;
-               int     sbase;
-       } *uap;
-
-       uap = (struct a *)u.u_ap;
-       fp = getf(uap->fd);
-       if (fp == NULL)
-               return;
-       if (fp->f_type == DTYPE_SOCKET) {
-               u.u_error = ESPIPE;
-               return;
-       }
-       if (uap->sbase == FSEEK_RELATIVE)
-               uap->off += fp->f_offset;
-       else if (uap->sbase == FSEEK_EOF)
-               uap->off += fp->f_inode->i_size;
-       fp->f_offset = uap->off;
-       u.u_r.r_off = uap->off;
+       if (dvp)
+               vput(dvp);
+       vput(vp);
+       return (error);
 }
 
 /*
 }
 
 /*
- * Access system call
+ * symlink -- make a symbolic link
  */
  */
-saccess()
+int
+ufs_symlink(ap)
+       struct vop_symlink_args /* {
+               struct vnode *a_dvp;
+               struct vnode **a_vpp;
+               struct componentname *a_cnp;
+               struct vattr *a_vap;
+               char *a_target;
+       } */ *ap;
 {
 {
-       register svuid, svgid;
+       register struct vnode *vp, **vpp = ap->a_vpp;
        register struct inode *ip;
        register struct inode *ip;
-       register struct a {
-               char    *fname;
-               int     fmode;
-       } *uap;
-
-       uap = (struct a *)u.u_ap;
-       svuid = u.u_uid;
-       svgid = u.u_gid;
-       u.u_uid = u.u_ruid;
-       u.u_gid = u.u_rgid;
-       ip = namei(uchar, LOOKUP, 1);
-       if (ip != NULL) {
-               if ((uap->fmode&FACCESS_READ) && access(ip, IREAD))
-                       goto done;
-               if ((uap->fmode&FACCESS_WRITE) && access(ip, IWRITE))
-                       goto done;
-               if ((uap->fmode&FACCESS_EXECUTE) && access(ip, IEXEC))
-                       goto done;
-done:
-               iput(ip);
-       }
-       u.u_uid = svuid;
-       u.u_gid = svgid;
+       int len, error;
+
+       if (error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
+           vpp, ap->a_cnp))
+               return (error);
+       vp = *vpp;
+       len = strlen(ap->a_target);
+       if (len < vp->v_mount->mnt_maxsymlinklen) {
+               ip = VTOI(vp);
+               bcopy(ap->a_target, (char *)ip->i_shortlink, len);
+               ip->i_size = len;
+               ip->i_flag |= IUPD|ICHG;
+       } else
+               error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0,
+                   UIO_SYSSPACE, IO_NODELOCKED, ap->a_cnp->cn_cred, (int *)0,
+                   (struct proc *)0);
+       vput(vp);
+       return (error);
 }
 
 /*
 }
 
 /*
- * the fstat system call.
+ * Vnode op for reading directories.
+ * 
+ * The routine below assumes that the on-disk format of a directory
+ * is the same as that defined by <sys/dirent.h>. If the on-disk
+ * format changes, then it will be necessary to do a conversion
+ * from the on-disk format that read returns to the format defined
+ * by <sys/dirent.h>.
  */
  */
-fstat()
+int
+ufs_readdir(ap)
+       struct vop_readdir_args /* {
+               struct vnode *a_vp;
+               struct uio *a_uio;
+               struct ucred *a_cred;
+       } */ *ap;
 {
 {
-       register struct file *fp;
-       register struct a {
-               int     fd;
-               struct stat *sb;
-       } *uap;
-
-       uap = (struct a *)u.u_ap;
-       fp = getf(uap->fd);
-       if (fp == NULL)
-               return;
-       if (fp->f_type == DTYPE_SOCKET)
-               u.u_error = sostat(fp->f_socket, uap->sb);
-       else
-               stat1(fp->f_inode, uap->sb);
+       register struct uio *uio = ap->a_uio;
+       int count, lost, error;
+
+       count = uio->uio_resid;
+       count &= ~(DIRBLKSIZ - 1);
+       lost = uio->uio_resid - count;
+       if (count < DIRBLKSIZ || (uio->uio_offset & (DIRBLKSIZ -1)))
+               return (EINVAL);
+       uio->uio_resid = count;
+       uio->uio_iov->iov_len = count;
+#      if (BYTE_ORDER == LITTLE_ENDIAN)
+               if (ap->a_vp->v_mount->mnt_maxsymlinklen > 0) {
+                       error = VOP_READ(ap->a_vp, uio, 0, ap->a_cred);
+               } else {
+                       struct dirent *dp, *edp;
+                       struct uio auio;
+                       struct iovec aiov;
+                       caddr_t dirbuf;
+                       int readcnt;
+                       u_char tmp;
+
+                       auio = *uio;
+                       auio.uio_iov = &aiov;
+                       auio.uio_iovcnt = 1;
+                       auio.uio_segflg = UIO_SYSSPACE;
+                       aiov.iov_len = count;
+                       MALLOC(dirbuf, caddr_t, count, M_TEMP, M_WAITOK);
+                       aiov.iov_base = dirbuf;
+                       error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
+                       if (error == 0) {
+                               readcnt = count - auio.uio_resid;
+                               edp = (struct dirent *)&dirbuf[readcnt];
+                               for (dp = (struct dirent *)dirbuf; dp < edp; ) {
+                                       tmp = dp->d_namlen;
+                                       dp->d_namlen = dp->d_type;
+                                       dp->d_type = tmp;
+                                       if (dp->d_reclen > 0) {
+                                               dp = (struct dirent *)
+                                                   ((char *)dp + dp->d_reclen);
+                                       } else {
+                                               error = EIO;
+                                               break;
+                                       }
+                               }
+                               if (dp >= edp)
+                                       error = uiomove(dirbuf, readcnt, uio);
+                       }
+                       FREE(dirbuf, M_TEMP);
+               }
+#      else
+               error = VOP_READ(ap->a_vp, uio, 0, ap->a_cred);
+#      endif
+       uio->uio_resid += lost;
+       return (error);
 }
 
 /*
 }
 
 /*
- * Stat system call.  This version follows links.
+ * Return target name of a symbolic link
  */
  */
-stat()
+int
+ufs_readlink(ap)
+       struct vop_readlink_args /* {
+               struct vnode *a_vp;
+               struct uio *a_uio;
+               struct ucred *a_cred;
+       } */ *ap;
 {
 {
-       register struct inode *ip;
-       register struct a {
-               char    *fname;
-               struct stat *sb;
-       } *uap;
-
-       uap = (struct a *)u.u_ap;
-       ip = namei(uchar, LOOKUP, 1);
-       if (ip == NULL)
-               return;
-       stat1(ip, uap->sb);
-       iput(ip);
+       register struct vnode *vp = ap->a_vp;
+       register struct inode *ip = VTOI(vp);
+       int isize;
+
+       isize = ip->i_size;
+       if (isize < vp->v_mount->mnt_maxsymlinklen) {
+               uiomove((char *)ip->i_shortlink, isize, ap->a_uio);
+               return (0);
+       }
+       return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
 }
 
 /*
 }
 
 /*
- * Lstat system call.  This version does not follow links.
+ * Ufs abort op, called after namei() when a CREATE/DELETE isn't actually
+ * done. If a buffer has been saved in anticipation of a CREATE, delete it.
  */
  */
-lstat()
+/* ARGSUSED */
+int
+ufs_abortop(ap)
+       struct vop_abortop_args /* {
+               struct vnode *a_dvp;
+               struct componentname *a_cnp;
+       } */ *ap;
 {
 {
-       register struct inode *ip;
-       register struct a {
-               char    *fname;
-               struct stat *sb;
-       } *uap;
-
-       uap = (struct a *)u.u_ap;
-       ip = namei(uchar, LOOKUP, 0);
-       if (ip == NULL)
-               return;
-       stat1(ip, uap->sb);
-       iput(ip);
+       if ((ap->a_cnp->cn_flags & (HASBUF | SAVESTART)) == HASBUF)
+               FREE(ap->a_cnp->cn_pnbuf, M_NAMEI);
+       return (0);
 }
 
 /*
 }
 
 /*
- * The basic routine for fstat and stat:
- * get the inode and pass appropriate parts back.
+ * Lock an inode. If its already locked, set the WANT bit and sleep.
  */
  */
-stat1(ip, ub)
-       register struct inode *ip;
-       struct stat *ub;
+int
+ufs_lock(ap)
+       struct vop_lock_args /* {
+               struct vnode *a_vp;
+       } */ *ap;
 {
 {
-       struct stat ds;
+       register struct vnode *vp = ap->a_vp;
+       register struct inode *ip;
+       struct proc *p = curproc;       /* XXX */
 
 
-       IUPDAT(ip, &time, &time, 0);
-       /*
-        * Copy from inode table
-        */
-       ds.st_dev = ip->i_dev;
-       ds.st_ino = ip->i_number;
-       ds.st_mode = ip->i_mode;
-       ds.st_nlink = ip->i_nlink;
-       ds.st_uid = ip->i_uid;
-       ds.st_gid = ip->i_gid;
-       ds.st_rdev = (dev_t)ip->i_rdev;
-       ds.st_size = ip->i_size;
-       ds.st_atime = ip->i_atime;
-       ds.st_spare1 = 0;
-       ds.st_mtime = ip->i_mtime;
-       ds.st_spare2 = 0;
-       ds.st_ctime = ip->i_ctime;
-       ds.st_spare3 = 0;
-       /* this doesn't belong here */
-       if ((ip->i_mode&IFMT) == IFBLK)
-               ds.st_blksize = BLKDEV_IOSIZE;
-       else if ((ip->i_mode&IFMT) == IFCHR)
-               ds.st_blksize = MAXBSIZE;
+start:
+       while (vp->v_flag & VXLOCK) {
+               vp->v_flag |= VXWANT;
+               sleep((caddr_t)vp, PINOD);
+       }
+       if (vp->v_tag == VT_NON)
+               return (ENOENT);
+       ip = VTOI(vp);
+       if (ip->i_flag & ILOCKED) {
+               ip->i_flag |= IWANT;
+#ifdef DIAGNOSTIC
+               if (p) {
+                       if (p->p_pid == ip->i_lockholder)
+                               panic("locking against myself");
+                       ip->i_lockwaiter = p->p_pid;
+               } else
+                       ip->i_lockwaiter = -1;
+#endif
+               (void) sleep((caddr_t)ip, PINOD);
+               goto start;
+       }
+#ifdef DIAGNOSTIC
+       ip->i_lockwaiter = 0;
+       if (ip->i_lockholder != 0)
+               panic("lockholder (%d) != 0", ip->i_lockholder);
+       if (p && p->p_pid == 0)
+               printf("locking by process 0\n");
+       if (p)
+               ip->i_lockholder = p->p_pid;
        else
        else
-               ds.st_blksize = ip->i_fs->fs_bsize;
-       ds.st_spare4[0] = ds.st_spare4[1] = ds.st_spare4[2] = 0;
-       u.u_error = copyout((caddr_t)&ds, (caddr_t)ub, sizeof(ds));
+               ip->i_lockholder = -1;
+#endif
+       ip->i_flag |= ILOCKED;
+       return (0);
 }
 
 /*
 }
 
 /*
- * Return target name of a symbolic link
+ * Unlock an inode.  If WANT bit is on, wakeup.
  */
  */
-readlink()
+int lockcount = 90;
+int
+ufs_unlock(ap)
+       struct vop_unlock_args /* {
+               struct vnode *a_vp;
+       } */ *ap;
 {
 {
-       register struct inode *ip;
-       register struct a {
-               char    *name;
-               char    *buf;
-               int     count;
-       } *uap = (struct a *)u.u_ap;
-       int resid;
-
-       ip = namei(uchar, LOOKUP, 0);
-       if (ip == NULL)
-               return;
-       if ((ip->i_mode&IFMT) != IFLNK) {
-               u.u_error = ENXIO;
-               goto out;
+       register struct inode *ip = VTOI(ap->a_vp);
+       struct proc *p = curproc;       /* XXX */
+
+#ifdef DIAGNOSTIC
+       if ((ip->i_flag & ILOCKED) == 0) {
+               vprint("ufs_unlock: unlocked inode", ap->a_vp);
+               panic("ufs_unlock NOT LOCKED");
        }
        }
-       u.u_error = rdwri(UIO_READ, ip, uap->buf, uap->count, 0, 0, &resid);
-out:
-       iput(ip);
-       u.u_r.r_val1 = uap->count - resid;
+       if (p && p->p_pid != ip->i_lockholder && p->p_pid > -1 &&
+           ip->i_lockholder > -1 && lockcount++ < 100)
+               panic("unlocker (%d) != lock holder (%d)",
+                   p->p_pid, ip->i_lockholder);
+       ip->i_lockholder = 0;
+#endif
+       ip->i_flag &= ~ILOCKED;
+       if (ip->i_flag & IWANT) {
+               ip->i_flag &= ~IWANT;
+               wakeup((caddr_t)ip);
+       }
+       return (0);
 }
 
 /*
 }
 
 /*
- * Change mode of a file given path name.
+ * Check for a locked inode.
  */
  */
-chmod()
+int
+ufs_islocked(ap)
+       struct vop_islocked_args /* {
+               struct vnode *a_vp;
+       } */ *ap;
 {
 {
-       struct inode *ip;
-       struct a {
-               char    *fname;
-               int     fmode;
-       } *uap;
-
-       uap = (struct a *)u.u_ap;
-       if ((ip = owner(1)) == NULL)
-               return;
-       chmod1(ip, uap->fmode);
-       iput(ip);
+
+       if (VTOI(ap->a_vp)->i_flag & ILOCKED)
+               return (1);
+       return (0);
 }
 
 /*
 }
 
 /*
- * Change mode of a file given a file descriptor.
+ * Calculate the logical to physical mapping if not done already,
+ * then call the device strategy routine.
  */
  */
-fchmod()
+int
+ufs_strategy(ap)
+       struct vop_strategy_args /* {
+               struct buf *a_bp;
+       } */ *ap;
 {
 {
-       struct a {
-               int     fd;
-               int     fmode;
-       } *uap;
+       register struct buf *bp = ap->a_bp;
+       register struct vnode *vp = bp->b_vp;
        register struct inode *ip;
        register struct inode *ip;
-       register struct file *fp;
-
-       uap = (struct a *)u.u_ap;
-       fp = getf(uap->fd);
-       if (fp == NULL)
-               return;
-       if (fp->f_type == DTYPE_SOCKET) {
-               u.u_error = EINVAL;
-               return;
+       int error;
+
+       ip = VTOI(vp);
+       if (vp->v_type == VBLK || vp->v_type == VCHR)
+               panic("ufs_strategy: spec");
+       if (bp->b_blkno == bp->b_lblkno) {
+               if (error =
+                   VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, NULL)) {
+                       bp->b_error = error;
+                       bp->b_flags |= B_ERROR;
+                       biodone(bp);
+                       return (error);
+               }
+               if ((long)bp->b_blkno == -1)
+                       clrbuf(bp);
        }
        }
-       ip = fp->f_inode;
-       if (u.u_uid != ip->i_uid && !suser())
-               return;
-       ilock(ip);
-       chmod1(ip, uap->fmode);
-       iunlock(ip);
-}
-
-/*
- * Change the mode on a file.
- * Inode must be locked before calling.
- */
-chmod1(ip, mode)
-       register struct inode *ip;
-       register int mode;
-{
-       register int *gp;
-
-       ip->i_mode &= ~07777;
-       if (u.u_uid) {
-               mode &= ~ISVTX;
-               for (gp = u.u_groups; gp < &u.u_groups[NGROUPS]; gp++)
-                       if (*gp == ip->i_gid)
-                               goto ok;
-               mode &= ~ISGID;
-ok:
-               ;
-#ifdef MUSH
-               if (u.u_quota->q_syflags & QF_UMASK && u.u_uid != 0 &&
-                   (ip->i_mode & IFMT) != IFCHR)
-                       mode &= ~u.u_cmask;
-#endif
+       if ((long)bp->b_blkno == -1) {
+               biodone(bp);
+               return (0);
        }
        }
-       ip->i_mode |= mode&07777;
-       ip->i_flag |= ICHG;
-       if (ip->i_flag&ITEXT && (ip->i_mode&ISVTX)==0)
-               xrele(ip);
+       vp = ip->i_devvp;
+       bp->b_dev = vp->v_rdev;
+       VOCALL (vp->v_op, VOFFSET(vop_strategy), ap);
+       return (0);
 }
 
 /*
 }
 
 /*
- * Set ownership given a path name.
+ * Print out the contents of an inode.
  */
  */
-chown()
+int
+ufs_print(ap)
+       struct vop_print_args /* {
+               struct vnode *a_vp;
+       } */ *ap;
 {
 {
-       struct inode *ip;
-       struct a {
-               char    *fname;
-               int     uid;
-               int     gid;
-       } *uap;
-
-       uap = (struct a *)u.u_ap;
-       if (!suser() || (ip = owner(0)) == NULL)
-               return;
-       chown1(ip, uap->uid, uap->gid);
-       iput(ip);
+       register struct vnode *vp = ap->a_vp;
+       register struct inode *ip = VTOI(vp);
+
+       printf("tag VT_UFS, ino %d, on dev %d, %d", ip->i_number,
+               major(ip->i_dev), minor(ip->i_dev));
+#ifdef FIFO
+       if (vp->v_type == VFIFO)
+               fifo_printinfo(vp);
+#endif /* FIFO */
+       printf("%s\n", (ip->i_flag & ILOCKED) ? " (LOCKED)" : "");
+       if (ip->i_lockholder == 0)
+               return (0);
+       printf("\towner pid %d", ip->i_lockholder);
+       if (ip->i_lockwaiter)
+               printf(" waiting pid %d", ip->i_lockwaiter);
+       printf("\n");
+       return (0);
 }
 
 /*
 }
 
 /*
- * Set ownership given a file descriptor.
+ * Read wrapper for special devices.
  */
  */
-fchown()
+int
+ufsspec_read(ap)
+       struct vop_read_args /* {
+               struct vnode *a_vp;
+               struct uio *a_uio;
+               int  a_ioflag;
+               struct ucred *a_cred;
+       } */ *ap;
 {
 {
-       struct a {
-               int     fd;
-               int     uid;
-               int     gid;
-       } *uap;
-       register struct inode *ip;
-       register struct file *fp;
-
-       uap = (struct a *)u.u_ap;
-       fp = getf(uap->fd);
-       if (fp == NULL)
-               return;
-       if (fp->f_type == DTYPE_SOCKET) {
-               u.u_error = EINVAL;
-               return;
-       }
-       ip = fp->f_inode;
-       if (!suser())
-               return;
-       ilock(ip);
-       chown1(ip, uap->uid, uap->gid);
-       iunlock(ip);
+
+       /*
+        * Set access flag.
+        */
+       VTOI(ap->a_vp)->i_flag |= IACC;
+       return (VOCALL (spec_vnodeop_p, VOFFSET(vop_read), ap));
 }
 
 /*
 }
 
 /*
- * Perform chown operation on inode ip;
- * inode must be locked prior to call.
+ * Write wrapper for special devices.
  */
  */
-chown1(ip, uid, gid)
-       register struct inode *ip;
-       int uid, gid;
+int
+ufsspec_write(ap)
+       struct vop_write_args /* {
+               struct vnode *a_vp;
+               struct uio *a_uio;
+               int  a_ioflag;
+               struct ucred *a_cred;
+       } */ *ap;
 {
 {
-#ifdef QUOTA
-       register long change;
-
-       /*
-        * This doesn't allow for holes in files (which hopefully don't
-        * happen often in files that we chown), and is not accurate anyway
-        * (eg: it totally ignores 3 level indir blk files - but hopefully
-        * noone who can make a file that big will have a quota)
-        */
-       if (ip->i_uid == uid)
-               change = 0;
-       else {
-               register struct fs *fs = ip->i_fs;
 
 
-               if (ip->i_size > (change = NDADDR * fs->fs_bsize)) {
-                       register off_t size;
-
-                       size = blkroundup(fs, ip->i_size) - change;
-                       change += size;
-                       change += fs->fs_bsize;
-                       /* this assumes NIADDR <= 2 */
-                       if (size > NINDIR(fs) * fs->fs_bsize)
-                               change += fs->fs_bsize;
-               } else
-                       change = fragroundup(fs, ip->i_size);
-               change /= DEV_BSIZE;
-       }
-       (void)chkdq(ip, -change, 1);
-       (void)chkiq(ip->i_dev, ip, ip->i_uid, 1);
-       dqrele(ip->i_dquot);
-#endif
        /*
        /*
-        * keep uid/gid's in sane range -- no err,
-        * so chown(file, uid, -1) will do something useful
+        * Set update and change flags.
         */
         */
-       if (uid >= 0 && uid <= 32767)   /* should have a constant */
-               ip->i_uid = uid;
-       if (gid >= 0 && gid <= 32767)   /* same here */
-               ip->i_gid = gid;
-       ip->i_flag |= ICHG;
-       if (u.u_ruid != 0)
-               ip->i_mode &= ~(ISUID|ISGID);
-#ifdef QUOTA
-       ip->i_dquot = inoquota(ip);
-       (void)chkdq(ip, change, 1);
-       (void)chkiq(ip->i_dev, (struct inode *)NULL, uid, 1);
-#endif
+       VTOI(ap->a_vp)->i_flag |= IUPD|ICHG;
+       return (VOCALL (spec_vnodeop_p, VOFFSET(vop_write), ap));
 }
 
 /*
 }
 
 /*
- * Set IUPD and IACC times on file.
- * Can't set ICHG.
+ * Close wrapper for special devices.
+ *
+ * Update the times on the inode then do device close.
  */
  */
-outime()
+int
+ufsspec_close(ap)
+       struct vop_close_args /* {
+               struct vnode *a_vp;
+               int  a_fflag;
+               struct ucred *a_cred;
+               struct proc *a_p;
+       } */ *ap;
 {
 {
-       register struct a {
-               char    *fname;
-               time_t  *tptr;
-       } *uap;
-       register struct inode *ip;
-       time_t tv[2];
-       struct timeval tv0, tv1;
-
-       uap = (struct a *)u.u_ap;
-       if ((ip = owner(1)) == NULL)
-               return;
-       u.u_error = copyin((caddr_t)uap->tptr, (caddr_t)tv, sizeof(tv));
-       if (u.u_error == 0) {
-               ip->i_flag |= IACC|IUPD|ICHG;
-               tv0.tv_sec = tv[0]; tv0.tv_usec = 0;
-               tv1.tv_sec = tv[1]; tv1.tv_usec = 0;
-               iupdat(ip, &tv0, &tv1, 0);
-       }
-       iput(ip);
+       register struct inode *ip = VTOI(ap->a_vp);
+
+       if (ap->a_vp->v_usecount > 1 && !(ip->i_flag & ILOCKED))
+               ITIMES(ip, &time, &time);
+       return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap));
 }
 
 }
 
+#ifdef FIFO
 /*
 /*
- * Flush any pending I/O.
+ * Read wrapper for fifo's
  */
  */
-sync()
+int
+ufsfifo_read(ap)
+       struct vop_read_args /* {
+               struct vnode *a_vp;
+               struct uio *a_uio;
+               int  a_ioflag;
+               struct ucred *a_cred;
+       } */ *ap;
 {
 {
+       extern int (**fifo_vnodeop_p)();
 
 
-       update();
+       /*
+        * Set access flag.
+        */
+       VTOI(ap->a_vp)->i_flag |= IACC;
+       return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_read), ap));
 }
 
 /*
 }
 
 /*
- * Apply an advisory lock on a file descriptor.
+ * Write wrapper for fifo's.
  */
  */
-flock()
+int
+ufsfifo_write(ap)
+       struct vop_write_args /* {
+               struct vnode *a_vp;
+               struct uio *a_uio;
+               int  a_ioflag;
+               struct ucred *a_cred;
+       } */ *ap;
 {
 {
-       struct a {
-               int     fd;
-               int     how;
-       } *uap;
-       register struct file *fp;
-       register int cmd, flags;
-
-       uap = (struct a *)u.u_ap;
-       fp = getf(uap->fd);
-       if (fp == NULL)
-               return;
-       if (fp->f_type == DTYPE_SOCKET) {               /* XXX */
-               u.u_error = EINVAL;
-               return;
-       }
-       cmd = uap->how;
-       flags = u.u_pofile[uap->fd] & (UF_SHLOCK|UF_EXLOCK);
-       if (cmd&FUNLOCK) {
-               if (flags == 0) {
-                       u.u_error = EINVAL;
-                       return;
-               }
-               funlocki(fp->f_inode, flags);
-               u.u_pofile[uap->fd] &= ~(UF_SHLOCK|UF_EXLOCK);
-               return;
-       }
+       extern int (**fifo_vnodeop_p)();
+
        /*
        /*
-        * No reason to write lock a file we've already
-        * write locked, similarly with a read lock.
+        * Set update and change flags.
         */
         */
-       if ((flags&UF_EXLOCK) && (cmd&FEXLOCK) ||
-           (flags&UF_SHLOCK) && (cmd&FSHLOCK))
-               return;
-       u.u_pofile[uap->fd] = flocki(fp->f_inode, u.u_pofile[uap->fd], cmd);
+       VTOI(ap->a_vp)->i_flag |= IUPD|ICHG;
+       return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_write), ap));
 }
 
 /*
 }
 
 /*
- * Truncate a file given its path name.
+ * Close wrapper for fifo's.
+ *
+ * Update the times on the inode then do device close.
  */
  */
-truncate()
+ufsfifo_close(ap)
+       struct vop_close_args /* {
+               struct vnode *a_vp;
+               int  a_fflag;
+               struct ucred *a_cred;
+               struct proc *a_p;
+       } */ *ap;
 {
 {
-       struct a {
-               char    *fname;
-               u_long  length;
-       } *uap = (struct a *)u.u_ap;
-       struct inode *ip;
+       extern int (**fifo_vnodeop_p)();
+       register struct inode *ip = VTOI(ap->a_vp);
 
 
-       ip = namei(uchar, LOOKUP, 1);
-       if (ip == NULL)
-               return;
-       if (access(ip, IWRITE))
-               goto bad;
-       if ((ip->i_mode&IFMT) == IFDIR) {
-               u.u_error = EISDIR;
-               goto bad;
-       }
-       itrunc(ip, uap->length);
-bad:
-       iput(ip);
+       if (ap->a_vp->v_usecount > 1 && !(ip->i_flag & ILOCKED))
+               ITIMES(ip, &time, &time);
+       return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap));
 }
 }
+#endif /* FIFO */
 
 /*
 
 /*
- * Truncate a file given a file descriptor.
+ * Return POSIX pathconf information applicable to ufs filesystems.
  */
  */
-ftruncate()
+ufs_pathconf(ap)
+       struct vop_pathconf_args /* {
+               struct vnode *a_vp;
+               int a_name;
+               int *a_retval;
+       } */ *ap;
 {
 {
-       struct a {
-               int     fd;
-               u_long  length;
-       } *uap = (struct a *)u.u_ap;
-       struct inode *ip;
-       struct file *fp;
-
-       fp = getf(uap->fd);
-       if (fp == NULL)
-               return;
-       if (fp->f_type == DTYPE_SOCKET) {
-               u.u_error = EINVAL;
-               return;
-       }
-       if ((fp->f_flag&FWRITE) == 0) {
-               u.u_error = EINVAL;
-               return;
-       }
-       ip = fp->f_inode;
-       ilock(ip);
-       itrunc(ip, uap->length);
-       iunlock(ip);
-}
 
 
-/*
- * Synch an open file.
- */
-fsync()
-{
-       struct a {
-               int     fd;
-       } *uap = (struct a *)u.u_ap;
-       struct inode *ip;
-       struct file *fp;
-
-       fp = getf(uap->fd);
-       if (fp == NULL)
-               return;
-       if (fp->f_type == DTYPE_SOCKET) {
-               u.u_error = EINVAL;
-               return;
+       switch (ap->a_name) {
+       case _PC_LINK_MAX:
+               *ap->a_retval = LINK_MAX;
+               return (0);
+       case _PC_NAME_MAX:
+               *ap->a_retval = NAME_MAX;
+               return (0);
+       case _PC_PATH_MAX:
+               *ap->a_retval = PATH_MAX;
+               return (0);
+       case _PC_PIPE_BUF:
+               *ap->a_retval = PIPE_BUF;
+               return (0);
+       case _PC_CHOWN_RESTRICTED:
+               *ap->a_retval = 1;
+               return (0);
+       case _PC_NO_TRUNC:
+               *ap->a_retval = 1;
+               return (0);
+       default:
+               return (EINVAL);
        }
        }
-       ip = fp->f_inode;
-       ilock(ip);
-       syncip(ip);
-       iunlock(ip);
+       /* NOTREACHED */
 }
 
 /*
 }
 
 /*
- * Rename system call.
- *     rename("foo", "bar");
- * is essentially
- *     unlink("bar");
- *     link("foo", "bar");
- *     unlink("foo");
- * but ``atomically''.  Can't do full commit without saving state in the
- * inode on disk which isn't feasible at this time.  Best we can do is
- * always guarantee the target exists.
- *
- * Basic algorithm is:
- *
- * 1) Bump link count on source while we're linking it to the
- *    target.  This also insure the inode won't be deleted out
- *    from underneath us while we work.
- * 2) Link source to destination.  If destination already exists,
- *    delete it first.
- * 3) Unlink source reference to inode if still around.
- * 4) If a directory was moved and the parent of the destination
- *    is different from the source, patch the ".." entry in the
- *    directory.
- *
- * Source and destination must either both be directories, or both
- * not be directories.  If target is a directory, it must be empty.
+ * Advisory record locking support
  */
  */
-rename()
+int
+ufs_advlock(ap)
+       struct vop_advlock_args /* {
+               struct vnode *a_vp;
+               caddr_t  a_id;
+               int  a_op;
+               struct flock *a_fl;
+               int  a_flags;
+       } */ *ap;
 {
 {
-       struct a {
-               char    *from;
-               char    *to;
-       } *uap;
-       register struct inode *ip, *xp, *dp;
-       int oldparent, parentdifferent, doingdirectory;
-       int error = 0;
-
-       uap = (struct a *)u.u_ap;
-       ip = namei(uchar, LOOKUP | LOCKPARENT, 0);
-       if (ip == NULL)
-               return;
-       dp = u.u_pdir;
-       oldparent = 0, doingdirectory = 0;
-       if ((ip->i_mode&IFMT) == IFDIR) {
-               register struct direct *d;
-
-               d = &u.u_dent;
-               /*
-                * Avoid "." and ".." for obvious reasons.
-                */
-               if (d->d_name[0] == '.') {
-                       if (d->d_namlen == 1 ||
-                           (d->d_namlen == 2 && d->d_name[1] == '.')) {
-                               iput(ip);
-                               u.u_error = EINVAL;
-                               return;
-                       }
-               }
-               oldparent = dp->i_number;
-               doingdirectory++;
-       }
-       irele(dp);
-
-       /*
-        * 1) Bump link count while we're moving stuff
-        *    around.  If we crash somewhere before
-        *    completing our work, the link count
-        *    may be wrong, but correctable.
-        */
-       ip->i_nlink++;
-       ip->i_flag |= ICHG;
-       iupdat(ip, &time, &time, 1);
-       iunlock(ip);
+       register struct inode *ip = VTOI(ap->a_vp);
+       register struct flock *fl = ap->a_fl;
+       register struct lockf *lock;
+       off_t start, end;
+       int error;
 
        /*
 
        /*
-        * When the target exists, both the directory
-        * and target inodes are returned locked.
+        * Avoid the common case of unlocking when inode has no locks.
         */
         */
-       u.u_dirp = (caddr_t)uap->to;
-       xp = namei(uchar, CREATE | LOCKPARENT, 0);
-       if (u.u_error) {
-               error = u.u_error;
-               goto out;
+       if (ip->i_lockf == (struct lockf *)0) {
+               if (ap->a_op != F_SETLK) {
+                       fl->l_type = F_UNLCK;
+                       return (0);
+               }
        }
        }
-       dp = u.u_pdir;
        /*
        /*
-        * 2) If target doesn't exist, link the target
-        *    to the source and unlink the source. 
-        *    Otherwise, rewrite the target directory
-        *    entry to reference the source inode and
-        *    expunge the original entry's existence.
+        * Convert the flock structure into a start and end.
         */
         */
-       parentdifferent = oldparent != dp->i_number;
-       if (xp == NULL) {
-               if (dp->i_dev != ip->i_dev) {
-                       error = EXDEV;
-                       goto bad;
-               }
-               /*
-                * Disallow rename(foo, foo/bar).
-                */
-               if (dp->i_number == ip->i_number) {
-                       error = EEXIST;
-                       goto bad;
-               }
-               /*
-                * Account for ".." in directory.
-                * When source and destination have the
-                * same parent we don't fool with the
-                * link count -- this isn't required
-                * because we do a similar check below.
-                */
-               if (doingdirectory && parentdifferent) {
-                       dp->i_nlink++;
-                       dp->i_flag |= ICHG;
-                       iupdat(dp, &time, &time, 1);
-               }
-               error = direnter(ip);
-               if (error)
-                       goto out;
-       } else {
-               if (xp->i_dev != dp->i_dev || xp->i_dev != ip->i_dev) {
-                       error = EXDEV;
-                       goto bad;
-               }
-               /*
-                * Short circuit rename(foo, foo).
-                */
-               if (xp->i_number == ip->i_number)
-                       goto bad;
-               /*
-                * Target must be empty if a directory
-                * and have no links to it.
-                * Also, insure source and target are
-                * compatible (both directories, or both
-                * not directories).
-                */
-               if ((xp->i_mode&IFMT) == IFDIR) {
-                       if (!dirempty(xp) || xp->i_nlink > 2) {
-                               error = ENOTEMPTY;
-                               goto bad;
-                       }
-                       if (!doingdirectory) {
-                               error = ENOTDIR;
-                               goto bad;
-                       }
-               } else if (doingdirectory) {
-                       error = EISDIR;
-                       goto bad;
-               }
-               dirrewrite(dp, ip);
-               if (u.u_error) {
-                       error = u.u_error;
-                       goto bad1;
-               }
+       switch (fl->l_whence) {
+
+       case SEEK_SET:
+       case SEEK_CUR:
                /*
                /*
-                * Adjust the link count of the target to
-                * reflect the dirrewrite above.  If this is
-                * a directory it is empty and there are
-                * no links to it, so we can squash the inode and
-                * any space associated with it.  We disallowed
-                * renaming over top of a directory with links to
-                * it above, as we've no way to determine if
-                * we've got a link or the directory itself, and
-                * if we get a link, then ".." will be screwed up.
+                * Caller is responsible for adding any necessary offset
+                * when SEEK_CUR is used.
                 */
                 */
-               xp->i_nlink--;
-               if (doingdirectory) {
-                       if (--xp->i_nlink != 0)
-                               panic("rename: linked directory");
-                       itrunc(xp, (u_long)0);
-               }
-               xp->i_flag |= ICHG;
-               iput(xp);
-               xp = NULL;
-       }
+               start = fl->l_start;
+               break;
 
 
+       case SEEK_END:
+               start = ip->i_size + fl->l_start;
+               break;
+
+       default:
+               return (EINVAL);
+       }
+       if (start < 0)
+               return (EINVAL);
+       if (fl->l_len == 0)
+               end = -1;
+       else
+               end = start + fl->l_len - 1;
        /*
        /*
-        * 3) Unlink the source.
+        * Create the lockf structure
         */
         */
-       u.u_dirp = uap->from;
-       dp = namei(uchar, DELETE, 0);
+       MALLOC(lock, struct lockf *, sizeof *lock, M_LOCKF, M_WAITOK);
+       lock->lf_start = start;
+       lock->lf_end = end;
+       lock->lf_id = ap->a_id;
+       lock->lf_inode = ip;
+       lock->lf_type = fl->l_type;
+       lock->lf_next = (struct lockf *)0;
+       lock->lf_block = (struct lockf *)0;
+       lock->lf_flags = ap->a_flags;
        /*
        /*
-        * Insure directory entry still exists and
-        * has not changed since the start of all
-        * this.  If either has occured, forget about
-        * about deleting the original entry and just
-        * adjust the link count in the inode.
+        * Do the requested operation.
         */
         */
-       if (dp == NULL || u.u_dent.d_ino != ip->i_number) {
-               ip->i_nlink--;
-               ip->i_flag |= ICHG;
-       } else {
-               /*
-                * If source is a directory, must adjust
-                * link count of parent directory also.
-                * If target didn't exist and source and
-                * target have the same parent, then we
-                * needn't touch the link count, it all
-                * balances out in the end.  Otherwise, we
-                * must do so to reflect deletion of ".."
-                * done above.
-                */
-               if (doingdirectory && (xp != NULL || parentdifferent)) {
-                       dp->i_nlink--;
-                       dp->i_flag |= ICHG;
-               }
-               if (dirremove()) {
-                       ip->i_nlink--;
-                       ip->i_flag |= ICHG;
-               }
-               if (error == 0)         /* conservative */
-                       error = u.u_error;
+       switch(ap->a_op) {
+       case F_SETLK:
+               return (lf_setlock(lock));
+
+       case F_UNLCK:
+               error = lf_clearlock(lock);
+               FREE(lock, M_LOCKF);
+               return (error);
+
+       case F_GETLK:
+               error = lf_getlock(lock, fl);
+               FREE(lock, M_LOCKF);
+               return (error);
+       
+       default:
+               free(lock, M_LOCKF);
+               return (EINVAL);
        }
        }
-       irele(ip);
-       if (dp)
-               iput(dp);
+       /* NOTREACHED */
+}
 
 
-       /*
-        * 4) Renaming a directory with the parent
-        *    different requires ".." to be rewritten.
-        *    The window is still there for ".." to
-        *    be inconsistent, but this is unavoidable,
-        *    and a lot shorter than when it was done
-        *    in a user process.
-        */
-       if (doingdirectory && parentdifferent && error == 0) {
-               struct dirtemplate dirbuf;
-
-               u.u_dirp = uap->to;
-               ip = namei(uchar, LOOKUP | LOCKPARENT, 0);
-               if (ip == NULL) {
-                       printf("rename: .. went away\n");
-                       return;
-               }
-               dp = u.u_pdir;
-               if ((ip->i_mode&IFMT) != IFDIR) {
-                       printf("rename: .. not a directory\n");
-                       goto stuck;
-               }
-               error = rdwri(UIO_READ, ip, (caddr_t)&dirbuf,
-                       sizeof (struct dirtemplate), (off_t)0, 1, (int *)0);
-               if (error == 0) {
-                       dirbuf.dotdot_ino = dp->i_number;
-                       (void) rdwri(UIO_WRITE, ip, (caddr_t)&dirbuf,
-                         sizeof (struct dirtemplate), (off_t)0, 1, (int *)0);
+/*
+ * Initialize the vnode associated with a new inode, handle aliased
+ * vnodes.
+ */
+int
+ufs_vinit(mntp, specops, fifoops, vpp)
+       struct mount *mntp;
+       int (**specops)();
+       int (**fifoops)();
+       struct vnode **vpp;
+{
+       struct inode *ip;
+       struct vnode *vp, *nvp;
+
+       vp = *vpp;
+       ip = VTOI(vp);
+       switch(vp->v_type = IFTOVT(ip->i_mode)) {
+       case VCHR:
+       case VBLK:
+               vp->v_op = specops;
+               if (nvp = checkalias(vp, ip->i_rdev, mntp)) {
+                       /*
+                        * Discard unneeded vnode, but save its inode.
+                        */
+                       ufs_ihashrem(ip);
+                       VOP_UNLOCK(vp);
+                       nvp->v_data = vp->v_data;
+                       vp->v_data = NULL;
+                       vp->v_op = spec_vnodeop_p;
+                       vrele(vp);
+                       vgone(vp);
+                       /*
+                        * Reinitialize aliased inode.
+                        */
+                       vp = nvp;
+                       ip->i_vnode = vp;
+                       ufs_ihashins(ip);
                }
                }
-stuck:
-               irele(dp);
-               iput(ip);
+               break;
+       case VFIFO:
+#ifdef FIFO
+               vp->v_op = fifoops;
+               break;
+#else
+               return (EOPNOTSUPP);
+#endif
        }
        }
-       goto done;
-
-bad:
-       iput(dp);
-bad1:
-       if (xp)
-               iput(xp);
-out:
-       ip->i_nlink--;
-       ip->i_flag |= ICHG;
-       irele(ip);
-done:
-       if (error)
-               u.u_error = error;
+       if (ip->i_number == ROOTINO)
+                vp->v_flag |= VROOT;
+       /*
+        * Initialize modrev times
+        */
+       SETHIGH(ip->i_modrev, mono_time.tv_sec);
+       SETLOW(ip->i_modrev, mono_time.tv_usec * 4294);
+       *vpp = vp;
+       return (0);
 }
 
 /*
 }
 
 /*
- * Make a new file.
+ * Allocate a new inode.
  */
  */
-struct inode *
-maknode(mode)
+int
+ufs_makeinode(mode, dvp, vpp, cnp)
        int mode;
        int mode;
+       struct vnode *dvp;
+       struct vnode **vpp;
+       struct componentname *cnp;
 {
 {
-       register struct inode *ip;
-       ino_t ipref;
+       register struct inode *ip, *pdir;
+       struct timeval tv;
+       struct vnode *tvp;
+       int error;
+
+       pdir = VTOI(dvp);
+#ifdef DIAGNOSTIC
+       if ((cnp->cn_flags & HASBUF) == 0)
+               panic("ufs_makeinode: no name");
+#endif
+       *vpp = NULL;
+       if ((mode & IFMT) == 0)
+               mode |= IFREG;
 
 
-       if ((mode & IFMT) == IFDIR)
-               ipref = dirpref(u.u_pdir->i_fs);
-       else
-               ipref = u.u_pdir->i_number;
-       ip = ialloc(u.u_pdir, ipref, mode);
-       if (ip == NULL) {
-               iput(u.u_pdir);
-               return (NULL);
+       if (error = VOP_VALLOC(dvp, mode, cnp->cn_cred, &tvp)) {
+               free(cnp->cn_pnbuf, M_NAMEI);
+               vput(dvp);
+               return (error);
        }
        }
+       ip = VTOI(tvp);
+       ip->i_gid = pdir->i_gid;
+       if ((mode & IFMT) == IFLNK)
+               ip->i_uid = pdir->i_uid;
+       else
+               ip->i_uid = cnp->cn_cred->cr_uid;
 #ifdef QUOTA
 #ifdef QUOTA
-       if (ip->i_dquot != NODQUOT)
-               panic("maknode: dquot");
+       if ((error = getinoquota(ip)) ||
+           (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
+               free(cnp->cn_pnbuf, M_NAMEI);
+               VOP_VFREE(tvp, ip->i_number, mode);
+               vput(tvp);
+               vput(dvp);
+               return (error);
+       }
 #endif
        ip->i_flag |= IACC|IUPD|ICHG;
 #endif
        ip->i_flag |= IACC|IUPD|ICHG;
-       if ((mode & IFMT) == 0)
-               mode |= IFREG;
-       ip->i_mode = mode & ~u.u_cmask;
+       ip->i_mode = mode;
+       tvp->v_type = IFTOVT(mode);     /* Rest init'd in iget() */
        ip->i_nlink = 1;
        ip->i_nlink = 1;
-       ip->i_uid = u.u_uid;
-       ip->i_gid = u.u_pdir->i_gid;
-#ifdef QUOTA
-       ip->i_dquot = inoquota(ip);
-#endif
+       if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
+           suser(cnp->cn_cred, NULL))
+               ip->i_mode &= ~ISGID;
 
        /*
         * Make sure inode goes to disk before directory entry.
         */
 
        /*
         * Make sure inode goes to disk before directory entry.
         */
-       iupdat(ip, &time, &time, 1);
-       u.u_error = direnter(ip);
-       if (u.u_error) {
-               /*
-                * Write error occurred trying to update directory
-                * so must deallocate the inode.
-                */
-               ip->i_nlink = 0;
-               ip->i_flag |= ICHG;
-               iput(ip);
-               return (NULL);
-       }
-       return (ip);
+       tv = time;
+       if (error = VOP_UPDATE(tvp, &tv, &tv, 1))
+               goto bad;
+       if (error = ufs_direnter(ip, dvp, cnp))
+               goto bad;
+       if ((cnp->cn_flags & SAVESTART) == 0)
+               FREE(cnp->cn_pnbuf, M_NAMEI);
+       vput(dvp);
+       *vpp = tvp;
+       return (0);
+
+bad:
+       /*
+        * Write error occurred trying to update the inode
+        * or the directory so must deallocate the inode.
+        */
+       free(cnp->cn_pnbuf, M_NAMEI);
+       vput(dvp);
+       ip->i_nlink = 0;
+       ip->i_flag |= ICHG;
+       vput(tvp);
+       return (error);
 }
 }