name resolution checking (need kern/kern_malloc.c 7.25.1.1,
[unix-history] / usr / src / sys / kern / vfs_lookup.c
index 02c6c98..7fc4eb7 100644 (file)
-/*     vfs_lookup.c    4.28    82/10/19        */
+/*
+ * Copyright (c) 1982, 1986, 1989 Regents of the University of California.
+ * All rights reserved.
+ *
+ * %sccs.include.redist.c%
+ *
+ *     @(#)vfs_lookup.c        7.31.1.1 (Berkeley) %G%
+ */
 
 
-#include "../h/param.h"
-#include "../h/systm.h"
-#include "../h/inode.h"
-#include "../h/fs.h"
-#include "../h/mount.h"
-#include "../h/dir.h"
-#include "../h/user.h"
-#include "../h/buf.h"
-#include "../h/conf.h"
-#include "../h/uio.h"
+#include "param.h"
+#include "syslimits.h"
+#include "time.h"
+#include "namei.h"
+#include "vnode.h"
+#include "mount.h"
+#include "errno.h"
+#include "malloc.h"
+#include "filedesc.h"
+#include "proc.h"
+
+#ifdef KTRACE
+#include "ktrace.h"
+#endif
 
 
-struct buf *blkatoff();
-int    dirchk = 1;
 /*
 /*
- * Convert a pathname into a pointer to a locked inode,
- * with side effects usable in creating and removing files.
- * This is a very central and rather complicated routine.
+ * Convert a pathname into a pointer to a locked inode.
+ *
+ * The FOLLOW flag is set when symbolic links are to be followed
+ * when they occur at the end of the name translation process.
+ * Symbolic links are always followed for all other pathname
+ * components other than the last.
  *
  *
- * The func argument gives the routine which returns successive
- * characters of the name to be translated.  The flag
- * argument is (0, 1, 2) depending on whether the name is to be
- * (looked up, created, deleted).  The follow argument is 1 when
- * symbolic links are to be followed when they occur at the end of
- * the name translation process.
+ * The segflg defines whether the name is to be copied from user
+ * space or kernel space.
  *
  *
- * Overall outline:
+ * Overall outline of namei:
  *
  *     copy in name
  *     get starting directory
  *
  *     copy in name
  *     get starting directory
- * dirloop:
- *     check accessibility of directory
- * dirloop2:
- *     copy next component of name to u.u_dent
- *     handle degenerate case where name is null string
- *     search for name in directory, to found or notfound
- * notfound:
- *     if creating, return locked inode, leaving information on avail. slots
- *     else return error
- * found:
- *     if at end of path and deleting, return information to allow delete
- *     if .. and on mounted filesys, look in mount table for parent
- *     if symbolic link, massage name in buffer and continue at dirloop
- *     if more components of name, do next level at dirloop
- *     return the answer as locked inode
+ *     while (!done && !error) {
+ *             call lookup to search path.
+ *             if symbolic link, massage name in buffer and continue
+ *     }
  */
  */
-struct inode *
-namei(func, flag, follow)
-       int (*func)(), flag, follow;
+namei(ndp, p)
+       register struct nameidata *ndp;
+       struct proc *p;
 {
 {
+       register struct filedesc *fdp;  /* pointer to file descriptor state */
        register char *cp;              /* pointer into pathname argument */
        register char *cp;              /* pointer into pathname argument */
-/* these variables refer to things which must be freed or unlocked */
-       register struct inode *dp = 0;  /* the directory we are searching */
-       register struct fs *fs;         /* file system that directory is in */
-       register struct buf *bp = 0;    /* a buffer of directory entries */
-       register struct direct *ep;     /* the current directory entry */
-       int entryoffsetinblock;         /* offset of ep in bp's buffer */
-       register struct buf *nbp;       /* buffer storing path name argument */
-/* these variables hold information about the search for a slot */
-       enum {NONE, COMPACT, FOUND} slotstatus;
-       int slotoffset = -1;            /* offset of area with free space */
-       int slotsize;                   /* size of area at slotoffset */
-       int slotfreespace;              /* amount of space free in slot */
-       int slotneeded;                 /* size of the entry we're seeking */
-/* */
-       int dirsize;
-       int prevoff;                    /* u.u_offset of previous entry */
-       int nlink = 0;                  /* number of symbolic links taken */
-       struct inode *pdp;              /* saved dp during symlink work */
-       int i;
+       register struct vnode *dp;      /* the directory we are searching */
+       struct iovec aiov;              /* uio for reading symbolic links */
+       struct uio auio;
+       int error, linklen;
+
+       ndp->ni_cred = p->p_ucred;
+       fdp = p->p_fd;
 
        /*
         * Get a buffer for the name to be translated, and copy the
         * name into the buffer.
         */
 
        /*
         * Get a buffer for the name to be translated, and copy the
         * name into the buffer.
         */
-       nbp = geteblk(MAXPATHLEN);
-       for (cp = nbp->b_un.b_addr; *cp = (*func)(); ) {
-               if ((*cp&0377) == ('/'|0200) || (*cp&0200) && flag != 2) {
-                       u.u_error = EPERM;
-                       goto bad;
-               }
-               cp++;
-               if (cp >= nbp->b_un.b_addr + MAXPATHLEN) {
-                       u.u_error = ENOENT;
-                       goto bad;
-               }
+       if ((ndp->ni_nameiop & HASBUF) == 0)
+               MALLOC(ndp->ni_pnbuf, caddr_t, MAXPATHLEN, M_NAMEI, M_WAITOK);
+       if (ndp->ni_segflg == UIO_SYSSPACE)
+               error = copystr(ndp->ni_dirp, ndp->ni_pnbuf,
+                           MAXPATHLEN, &ndp->ni_pathlen);
+       else
+               error = copyinstr(ndp->ni_dirp, ndp->ni_pnbuf,
+                           MAXPATHLEN, &ndp->ni_pathlen);
+       if (error) {
+               free(ndp->ni_pnbuf, M_NAMEI);
+               ndp->ni_vp = NULL;
+               return (error);
        }
        }
-       if (u.u_error)
-               goto bad;
+       ndp->ni_loopcnt = 0;
+#ifdef KTRACE
+       if (KTRPOINT(p, KTR_NAMEI))
+               ktrnamei(p->p_tracep, ndp->ni_pnbuf);
+#endif
 
        /*
 
        /*
-        * Get starting directory.
+        * Get starting point for the translation.
         */
         */
-       cp = nbp->b_un.b_addr;
-       if (*cp == '/') {
-               while (*cp == '/')
-                       cp++;
-               if ((dp = u.u_rdir) == NULL)
-                       dp = rootdir;
-       } else
-               dp = u.u_cdir;
-       fs = dp->i_fs;
-       ilock(dp);
-       dp->i_count++;
-       u.u_pdir = (struct inode *)0xc0000000;          /* illegal */
+       if ((ndp->ni_rootdir = fdp->fd_rdir) == NULL)
+               ndp->ni_rootdir = rootdir;
+       dp = fdp->fd_cdir;
+       VREF(dp);
+       for (;;) {
+               /*
+                * Check if root directory should replace current directory.
+                * Done at start of translation and after symbolic link.
+                */
+               ndp->ni_ptr = ndp->ni_pnbuf;
+               if (*ndp->ni_ptr == '/') {
+                       vrele(dp);
+                       while (*ndp->ni_ptr == '/') {
+                               ndp->ni_ptr++;
+                               ndp->ni_pathlen--;
+                       }
+                       dp = ndp->ni_rootdir;
+                       VREF(dp);
+               }
+               ndp->ni_startdir = dp;
+               if (error = lookup(ndp, p)) {
+                       FREE(ndp->ni_pnbuf, M_NAMEI);
+                       return (error);
+               }
+               /*
+                * Check for symbolic link
+                */
+               if (ndp->ni_more == 0) {
+                       if ((ndp->ni_nameiop & (SAVENAME | SAVESTART)) == 0)
+                               FREE(ndp->ni_pnbuf, M_NAMEI);
+                       else
+                               ndp->ni_nameiop |= HASBUF;
+                       return (0);
+               }
+               if ((ndp->ni_nameiop & LOCKPARENT) && ndp->ni_pathlen == 1)
+                       VOP_UNLOCK(ndp->ni_dvp);
+               if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
+                       error = ELOOP;
+                       break;
+               }
+               if (ndp->ni_pathlen > 1)
+                       MALLOC(cp, char *, MAXPATHLEN, M_NAMEI, M_WAITOK);
+               else
+                       cp = ndp->ni_pnbuf;
+               aiov.iov_base = cp;
+               aiov.iov_len = MAXPATHLEN;
+               auio.uio_iov = &aiov;
+               auio.uio_iovcnt = 1;
+               auio.uio_offset = 0;
+               auio.uio_rw = UIO_READ;
+               auio.uio_segflg = UIO_SYSSPACE;
+               auio.uio_procp = (struct proc *)0;
+               auio.uio_resid = MAXPATHLEN;
+               if (error = VOP_READLINK(ndp->ni_vp, &auio, p->p_ucred)) {
+                       if (ndp->ni_pathlen > 1)
+                               free(cp, M_NAMEI);
+                       break;
+               }
+               linklen = MAXPATHLEN - auio.uio_resid;
+               if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
+                       if (ndp->ni_pathlen > 1)
+                               free(cp, M_NAMEI);
+                       error = ENAMETOOLONG;
+                       break;
+               }
+               if (ndp->ni_pathlen > 1) {
+                       bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
+                       FREE(ndp->ni_pnbuf, M_NAMEI);
+                       ndp->ni_pnbuf = cp;
+               } else
+                       ndp->ni_pnbuf[linklen] = '\0';
+               ndp->ni_pathlen += linklen;
+               vput(ndp->ni_vp);
+               dp = ndp->ni_dvp;
+       }
+       FREE(ndp->ni_pnbuf, M_NAMEI);
+       vrele(ndp->ni_dvp);
+       vput(ndp->ni_vp);
+       ndp->ni_vp = NULL;
+       return (error);
+}
+
+/*
+ * Search a pathname.
+ * This is a very central and rather complicated routine.
+ *
+ * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
+ * The starting directory is taken from ni_startdir. The pathname is
+ * descended until done, or a symbolic link is encountered. The variable
+ * ni_more is clear if the path is completed; it is set to one if a
+ * symbolic link needing interpretation is encountered.
+ *
+ * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
+ * whether the name is to be looked up, created, renamed, or deleted.
+ * When CREATE, RENAME, or DELETE is specified, information usable in
+ * creating, renaming, or deleting a directory entry may be calculated.
+ * If flag has LOCKPARENT or'ed into it, the parent directory is returned
+ * locked. If flag has WANTPARENT or'ed into it, the parent directory is
+ * returned unlocked. Otherwise the parent directory is not returned. If
+ * the target of the pathname exists and LOCKLEAF is or'ed into the flag
+ * the target is returned locked, otherwise it is returned unlocked.
+ * When creating or renaming and LOCKPARENT is specified, the target may not
+ * be ".".  When deleting and LOCKPARENT is specified, the target may be ".".
+ * NOTE: (LOOKUP | LOCKPARENT) currently returns the parent vnode unlocked.
+ * 
+ * Overall outline of lookup:
+ *
+ * dirloop:
+ *     identify next component of name at ndp->ni_ptr
+ *     handle degenerate case where name is null string
+ *     if .. and crossing mount points and on mounted filesys, find parent
+ *     call VOP_LOOKUP routine for next component name
+ *         directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
+ *         component vnode returned in ni_vp (if it exists), locked.
+ *     if result vnode is mounted on and crossing mount points,
+ *         find mounted on vnode
+ *     if more components of name, do next level at dirloop
+ *     return the answer in ni_vp, locked if LOCKLEAF set
+ *         if LOCKPARENT set, return locked parent in ni_dvp
+ *         if WANTPARENT set, return unlocked parent in ni_dvp
+ */
+lookup(ndp, p)
+       register struct nameidata *ndp;
+       struct proc *p;
+{
+       register char *cp;              /* pointer into pathname argument */
+       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 flag;                       /* LOOKUP, CREATE, RENAME or DELETE */
+       int wantparent;                 /* 1 => wantparent or lockparent flag */
+       int rdonly;                     /* mounted read-only flag bit(s) */
+       int error = 0;
 
        /*
 
        /*
-        * We come to dirloop to search a new directory.
-        * The directory must be locked so that it can be
-        * iput, and fs must be already set to dp->i_fs.
-        */
-dirloop:
-       /*
-        * Check accessiblity of directory.
+        * Setup: break out flag bits into variables.
         */
         */
-       if ((dp->i_mode&IFMT) != IFDIR) {
-               u.u_error = ENOTDIR;
-               goto bad;
-       }
-       if (access(dp, IEXEC))
-               goto bad;
+       flag = ndp->ni_nameiop & OPMASK;
+       wantparent = ndp->ni_nameiop & (LOCKPARENT|WANTPARENT);
+       docache = (ndp->ni_nameiop & NOCACHE) ^ NOCACHE;
+       if (flag == DELETE || (wantparent && flag != CREATE))
+               docache = 0;
+       rdonly = MNT_RDONLY;
+       if (ndp->ni_nameiop & REMOTE)
+               rdonly |= MNT_EXRDONLY;
+       ndp->ni_dvp = NULL;
+       ndp->ni_more = 0;
+       dp = ndp->ni_startdir;
+       ndp->ni_startdir = NULLVP;
+       VOP_LOCK(dp);
 
 
-dirloop2:
+dirloop:
        /*
        /*
-        * Copy next component of name to u.u_dent.
+        * Search a new directory.
+        *
+        * The ni_hash value is for use by vfs_cache.
+        * The last component of the filename is left accessible via
+        * ndp->ptr for callers that need the name. Callers needing
+        * the name set the SAVENAME flag. When done, they assume
+        * responsibility for freeing the pathname buffer.
         */
         */
-       for (i = 0; *cp != 0 && *cp != '/'; cp++) {
-               if (i >= MAXNAMLEN) {
-                       u.u_error = ENOENT;
+       ndp->ni_hash = 0;
+       for (cp = ndp->ni_ptr; *cp != 0 && *cp != '/'; cp++) {
+               ndp->ni_hash += (unsigned char)*cp;
+               if ((*cp & 0200) == 0)
+                       continue;
+               if ((*cp & 0377) == ('/' | 0200) || flag != DELETE) {
+                       error = EINVAL;
                        goto bad;
                }
                        goto bad;
                }
-               u.u_dent.d_name[i++] = *cp;
        }
        }
-       u.u_dent.d_namlen = i;
-       u.u_dent.d_name[i] = 0;
+       ndp->ni_namelen = cp - ndp->ni_ptr;
+       if (ndp->ni_namelen >= NAME_MAX) {
+               error = ENAMETOOLONG;
+               goto bad;
+       }
+#ifdef NAMEI_DIAGNOSTIC
+       { char c = *cp;
+       *cp = '\0';
+       printf("{%s}: ", ndp->ni_ptr);
+       *cp = c; }
+#endif
+       ndp->ni_pathlen -= ndp->ni_namelen;
+       ndp->ni_next = cp;
+       ndp->ni_makeentry = 1;
+       if (*cp == '\0' && docache == 0)
+               ndp->ni_makeentry = 0;
+       ndp->ni_isdotdot = (ndp->ni_namelen == 2 &&
+               ndp->ni_ptr[1] == '.' && ndp->ni_ptr[0] == '.');
 
        /*
         * Check for degenerate name (e.g. / or "")
         * which is a way of talking about a directory,
         * e.g. like "/." or ".".
         */
 
        /*
         * Check for degenerate name (e.g. / or "")
         * which is a way of talking about a directory,
         * e.g. like "/." or ".".
         */
-       if (u.u_dent.d_name[0] == 0) {
-               if (flag) {
-                       u.u_error = ENOENT;
+       if (ndp->ni_ptr[0] == '\0') {
+               if (flag != LOOKUP || wantparent) {
+                       error = EISDIR;
                        goto bad;
                }
                        goto bad;
                }
-               brelse(nbp);
-               return (dp);
+               if (dp->v_type != VDIR) {
+                       error = ENOTDIR;
+                       goto bad;
+               }
+               if (!(ndp->ni_nameiop & LOCKLEAF))
+                       VOP_UNLOCK(dp);
+               ndp->ni_vp = dp;
+               if (ndp->ni_nameiop & SAVESTART)
+                       panic("lookup: SAVESTART");
+               return (0);
        }
 
        /*
        }
 
        /*
-        * Suppress search for slots unless creating
-        * file and at end of pathname, in which case
-        * we watch for a place to put the new file in
-        * case it doesn't already exist.
+        * Handle "..": two special cases.
+        * 1. If at root directory (e.g. after chroot)
+        *    then ignore it so can't get out.
+        * 2. If this vnode is the root of a mounted
+        *    filesystem, then replace it with the
+        *    vnode which was mounted on so we take the
+        *    .. in the other file system.
         */
         */
-       slotstatus = FOUND;
-       if (flag == 1 && *cp == 0) {
-               slotstatus = NONE;
-               slotfreespace = 0;
-               slotneeded = DIRSIZ(&u.u_dent);
-       }
-
-       dirsize = roundup(dp->i_size, DIRBLKSIZ);
-       u.u_offset = 0;
-       while (u.u_offset < dirsize) {
-               /*
-                * If offset is on a block boundary,
-                * read the next directory block.
-                * Release previous if it exists.
-                */
-               if (blkoff(fs, u.u_offset) == 0) {
-                       if (bp != NULL)
-                               brelse(bp);
-                       bp = blkatoff(dp, u.u_offset, (char **)0);
-                       if (bp == 0)
-                               goto bad;
-                       entryoffsetinblock = 0;
-               }
-
-               /*
-                * If still looking for a slot, and at a DIRBLKSIZE
-                * boundary, have to start looking for free space
-                * again.
-                */
-               if (slotstatus == NONE &&
-                   (entryoffsetinblock&(DIRBLKSIZ-1)) == 0) {
-                       slotoffset = -1;
-                       slotfreespace = 0;
-               }
-
-               /*
-                * Get pointer to next entry, and do consistency checking:
-                *      record length must be multiple of 4
-                *      record length must not be zero
-                *      entry must fit in rest of this DIRBLKSIZ block
-                *      record must be large enough to contain name
-                * When dirchk is set we also check:
-                *      name is not longer than MAXNAMLEN
-                *      name must be as long as advertised, and null terminated
-                * Checking last two conditions is done only when dirchk is
-                * set, to save time.
-                */
-               ep = (struct direct *)(bp->b_un.b_addr + entryoffsetinblock);
-               i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
-               if ((ep->d_reclen & 0x3) || ep->d_reclen == 0 ||
-                   ep->d_reclen > i || DIRSIZ(ep) > ep->d_reclen ||
-                   dirchk && (ep->d_namlen > MAXNAMLEN || dirbadname(ep))) {
-                       dirbad(dp, "mangled entry");
-                       u.u_offset += i;
-                       entryoffsetinblock += i;
-                       continue;
-               }
-
-               /*
-                * If an appropriate sized slot has not yet been found,
-                * check to see if one is available. Also accumulate space
-                * in the current block so that we can determine if
-                * compaction is viable.
-                */
-               if (slotstatus != FOUND) {
-                       int size = ep->d_reclen;
-
-                       if (ep->d_ino != 0)
-                               size -= DIRSIZ(ep);
-                       if (size > 0) {
-                               if (size >= slotneeded) {
-                                       slotstatus = FOUND;
-                                       slotoffset = u.u_offset;
-                                       slotsize = ep->d_reclen;
-                               } else if (slotstatus == NONE) {
-                                       slotfreespace += size;
-                                       if (slotoffset == -1)
-                                               slotoffset = u.u_offset;
-                                       if (slotfreespace >= slotneeded) {
-                                               slotstatus = COMPACT;
-                                               slotsize =
-                                                   u.u_offset+ep->d_reclen -
-                                                     slotoffset;
-                                       }
-                               }
+       if (ndp->ni_isdotdot) {
+               for (;;) {
+                       if (dp == ndp->ni_rootdir) {
+                               ndp->ni_dvp = dp;
+                               ndp->ni_vp = dp;
+                               VREF(dp);
+                               goto nextname;
                        }
                        }
+                       if ((dp->v_flag & VROOT) == 0 ||
+                           (ndp->ni_nameiop & NOCROSSMOUNT))
+                               break;
+                       tdp = dp;
+                       dp = dp->v_mount->mnt_vnodecovered;
+                       vput(tdp);
+                       VREF(dp);
+                       VOP_LOCK(dp);
                }
                }
-
-               /*
-                * Check for a name match.
-                */
-               if (ep->d_ino) {
-                       if (ep->d_namlen == u.u_dent.d_namlen &&
-                           !bcmp(u.u_dent.d_name, ep->d_name, ep->d_namlen))
-                               goto found;
-               }
-               prevoff = u.u_offset;
-               u.u_offset += ep->d_reclen;
-               entryoffsetinblock += ep->d_reclen;
-       }
-/* notfound: */
-       /*
-        * If creating, and at end of pathname and current
-        * directory has not been removed, then can consider allowing
-        * file to be created.
-        */
-       if (flag == 1 && *cp == 0 && dp->i_nlink != 0) {
-               /*
-                * Access for write is interpreted as allowing
-                * creation of files in the directory.
-                */
-               if (access(dp, IWRITE))
-                       goto bad;
-               /*
-                * Return an indication of where the new directory
-                * entry should be put.  If we didn't find a slot,
-                * then set u.u_count to 0 indicating that the
-                * new slot belongs at the end of the directory.
-                * If we found a slot, then the new entry can be
-                * put in the range [u.u_offset..u.u_offset+u.u_count)
-                */
-               if (slotstatus == NONE)
-                       u.u_count = 0;
-               else {
-                       u.u_offset = slotoffset;
-                       u.u_count = slotsize;
-               }
-               dp->i_flag |= IUPD|ICHG;
-               if (bp)
-                       brelse(bp);
-               brelse(nbp);
-               /*
-                * We return with the directory locked, so that
-                * the parameters we set up above will still be
-                * valid if we actually decide to do a direnter().
-                * We return NULL to indicate that the entry doesn't
-                * currently exist, leaving a pointer to the (locked)
-                * directory inode in u.u_pdir.
-                */
-               u.u_pdir = dp;
-               return (NULL);
        }
        }
-       u.u_error = ENOENT;
-       goto bad;
-found:
-       /*
-        * Check that directory length properly reflects presence
-        * of this entry.
-        */
-       if (entryoffsetinblock + DIRSIZ(ep) > dp->i_size) {
-               dirbad(dp, "i_size too small");
-               dp->i_size = entryoffsetinblock + DIRSIZ(ep);
-               dp->i_flag |= IUPD|ICHG;
-       }
-
-       /*
-        * Found component in pathname; save directory
-        * entry in u.u_dent, and release directory buffer.
-        */
-       bcopy((caddr_t)ep, (caddr_t)&u.u_dent, (u_int)DIRSIZ(ep));
-       brelse(bp);
-       bp = NULL;
 
        /*
 
        /*
-        * If deleting, and at end of pathname, return
-        * parameters which can be used to remove file.
-        * Note that in this case we return the directory
-        * inode, not the inode of the file being deleted.
+        * We now have a segment name to search for, and a directory to search.
         */
         */
-       if (flag == 2 && *cp == 0) {
+       if (error = VOP_LOOKUP(dp, ndp, p)) {
+#ifdef DIAGNOSTIC
+               if (ndp->ni_vp != NULL)
+                       panic("leaf should be empty");
+#endif
+#ifdef NAMEI_DIAGNOSTIC
+               printf("not found\n");
+#endif
+               if (flag == LOOKUP || flag == DELETE ||
+                   error != ENOENT || *cp != 0)
+                       goto bad;
                /*
                /*
-                * Write access to directory required to delete files.
+                * If creating and at end of pathname, then can consider
+                * allowing file to be created.
                 */
                 */
-               if (access(dp, IWRITE))
+               if (ndp->ni_dvp->v_mount->mnt_flag & rdonly) {
+                       error = EROFS;
                        goto bad;
                        goto bad;
+               }
                /*
                /*
-                * Return pointer to current entry in u.u_offset,
-                * and distance past previous entry (if there
-                * is a previous entry in this block) in u.u_count.
-                * Save directory inode pointer in u.u_pdir for dirremove().
+                * 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.
                 */
                 */
-               if ((u.u_offset&(DIRBLKSIZ-1)) == 0)
-                       u.u_count = 0;
-               else
-                       u.u_count = u.u_offset - prevoff;
-               brelse(nbp);
-               u.u_pdir = dp;          /* for dirremove() */
-               return (dp);
-       }
-
-       /*
-        * Special handling for ".." allowing chdir out of mounted
-        * file system: indirect .. in root inode to reevaluate
-        * in directory file system was mounted on.
-        */
-       if (u.u_dent.d_name[0] == '.' && u.u_dent.d_name[1] == '.' &&
-           u.u_dent.d_name[2] == '\0') {
-               if (dp == u.u_rdir)
-                       u.u_dent.d_ino = dp->i_number;
-               else if (u.u_dent.d_ino == ROOTINO &&
-                  dp->i_number == ROOTINO) {
-                       for (i = 1; i < NMOUNT; i++)
-                       if (mount[i].m_bufp != NULL &&
-                          mount[i].m_dev == dp->i_dev) {
-                               iput(dp);
-                               dp = mount[i].m_inodp;
-                               ilock(dp);
-                               dp->i_count++;
-                               fs = dp->i_fs;
-                               cp -= 2;     /* back over .. */
-                               goto dirloop2;
-                       }
+               if (ndp->ni_nameiop & SAVESTART) {
+                       ndp->ni_startdir = ndp->ni_dvp;
+                       VREF(ndp->ni_startdir);
+                       p->p_spare[1]++;
                }
                }
+               return (0);
        }
        }
+#ifdef NAMEI_DIAGNOSTIC
+       printf("found\n");
+#endif
 
 
+       dp = ndp->ni_vp;
        /*
        /*
-        * Check for symbolic link, which may require us
-        * to massage the name before we continue translation.
-        * To avoid deadlock have to unlock the current directory,
-        * but don't iput it because we may need it again (if
-        * the symbolic link is relative to .).  Instead save
-        * it (unlocked) as pdp.
+        * Check for symbolic link
         */
         */
-       pdp = dp;
-       iunlock(pdp);
-       dp = iget(dp->i_dev, fs, u.u_dent.d_ino);
-       if (dp == NULL)
-               goto bad2;
-       fs = dp->i_fs;
+       if ((dp->v_type == VLNK) &&
+           ((ndp->ni_nameiop & FOLLOW) || *ndp->ni_next == '/')) {
+               ndp->ni_more = 1;
+               return (0);
+       }
 
        /*
 
        /*
-        * Check for symbolic link
+        * Check to see if the vnode has been mounted on;
+        * if so find the root of the mounted file system.
         */
         */
-       if ((dp->i_mode & IFMT) == IFLNK && (follow || *cp == '/')) {
-               u_int pathlen = strlen(cp) + 1;
-
-               if (dp->i_size + pathlen >= MAXPATHLEN - 1 ||
-                   ++nlink > MAXSYMLINKS) {
-                       u.u_error = ELOOP;
-                       goto bad2;
+mntloop:
+       while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
+              (ndp->ni_nameiop & NOCROSSMOUNT) == 0) {
+               while(mp->mnt_flag & MNT_MLOCK) {
+                       mp->mnt_flag |= MNT_MWAIT;
+                       sleep((caddr_t)mp, PVFS);
+                       goto mntloop;
                }
                }
-               bcopy(cp, nbp->b_un.b_addr + dp->i_size, pathlen);
-               u.u_error =
-                   rdwri(UIO_READ, dp, nbp->b_un.b_addr, dp->i_size,
-                       0, 1, (int *)0);
-               if (u.u_error)
+               if (error = VFS_ROOT(dp->v_mountedhere, &tdp))
                        goto bad2;
                        goto bad2;
-               cp = nbp->b_un.b_addr;
-               iput(dp);
-               if (*cp == '/') {
-                       irele(pdp);
-                       while (*cp == '/')
-                               cp++;
-                       if ((dp = u.u_rdir) == NULL)
-                               dp = rootdir;
-                       ilock(dp);
-                       dp->i_count++;
-               } else {
-                       dp = pdp;
-                       ilock(dp);
-               }
-               fs = dp->i_fs;
-               goto dirloop;
+               vput(dp);
+               ndp->ni_vp = dp = tdp;
        }
        }
-       irele(pdp);
 
 
+nextname:
        /*
         * Not a symbolic link.  If more pathname,
         * continue at next component, else return.
         */
        /*
         * Not a symbolic link.  If more pathname,
         * continue at next component, else return.
         */
-       if (*cp == '/') {
-               while (*cp == '/')
-                       cp++;
-               goto dirloop;
-       }
-       brelse(nbp);
-       return (dp);
-bad2:
-       irele(pdp);
-bad:
-       if (bp)
-               brelse(bp);
-       if (dp)
-               iput(dp);
-       brelse(nbp);
-       return (NULL);
-}
-
-dirbad(ip, how)
-       struct inode *ip;
-       char *how;
-{
-
-       printf("%s: bad dir ino %d at offset %d: %s\n",
-           ip->i_fs->fs_fsmnt, ip->i_number, u.u_offset, how);
-}
-
-dirbadname(ep)
-       register struct direct *ep;
-{
-       register int i;
-
-       for (i = 0; i < ep->d_namlen; i++)
-               if (ep->d_name[i] == 0)
-                       return (1);
-       return (ep->d_name[i]);
-}
-
-/*
- * Write a directory entry after a call to namei, using the parameters
- * which it left in the u. area.  The argument ip is the inode which
- * the new directory entry will refer to.  The u. area field u.u_pdir is
- * a pointer to the directory to be written, which was left locked by
- * namei.  Remaining parameters (u.u_offset, u.u_count) indicate
- * how the space for the new entry is to be gotten.
- */
-direnter(ip)
-       struct inode *ip;
-{
-       register struct direct *ep, *nep;
-       struct buf *bp;
-       int loc, freespace;
-       u_int dsize;
-       int newentrysize;
-       char *dirbuf;
-
-       u.u_dent.d_ino = ip->i_number;
-       u.u_segflg = 1;
-       newentrysize = DIRSIZ(&u.u_dent);
-       if (u.u_count == 0) {
-               /*
-                * If u.u_count is 0, then namei could find no space in the
-                * directory.  In this case u.u_offset will be on a directory
-                * block boundary and we will write the new entry into a fresh
-                * block.
-                */
-               if (u.u_offset&(DIRBLKSIZ-1))
-                       panic("wdir: newblk");
-               u.u_dent.d_reclen = DIRBLKSIZ;
-               (void) rdwri(UIO_WRITE, u.u_pdir, (caddr_t)&u.u_dent,
-                   newentrysize, u.u_offset, 1, (int *)0);
-               iput(u.u_pdir);
-               return;
-       }
-
-       /*
-        * If u.u_count is non-zero, then namei found space for the
-        * new entry in the range u.u_offset to u.u_offset+u.u_count.
-        * in the directory.  To use this space, we may have to compact
-        * the entries located there, by copying them together towards
-        * the beginning of the block, leaving the free space in
-        * one usable chunk at the end.
-        */
-
-       /*
-        * Increase size of directory if entry eats into new space.
-        * This should never push the size past a new multiple of
-        * DIRBLKSIZE.
-        */
-       if (u.u_offset+u.u_count > u.u_pdir->i_size)
-               u.u_pdir->i_size = u.u_offset + u.u_count;
-
-       /*
-        * Get the block containing the space for the new directory
-        * entry.
-        */
-       bp = blkatoff(u.u_pdir, u.u_offset, (char **)&dirbuf);
-       if (bp == 0)
-               return;
-
-       /*
-        * Find space for the new entry.  In the simple case, the
-        * entry at offset base will have the space.  If it does
-        * not, then namei arranged that compacting the region
-        * u.u_offset to u.u_offset+u.u_count would yield the space.
-        */
-       ep = (struct direct *)dirbuf;
-       dsize = DIRSIZ(ep);
-       freespace = ep->d_reclen - dsize;
-       for (loc = ep->d_reclen; loc < u.u_count; ) {
-               nep = (struct direct *)(dirbuf + loc);
-               if (ep->d_ino) {
-                       /* trim the existing slot */
-                       ep->d_reclen = dsize;
-                       ep = (struct direct *)((char *)ep + dsize);
-               } else {
-                       /* overwrite; nothing there; header is ours */
-                       freespace += dsize;     
+       if (*ndp->ni_next == '/') {
+               ndp->ni_ptr = ndp->ni_next;
+               while (*ndp->ni_ptr == '/') {
+                       ndp->ni_ptr++;
+                       ndp->ni_pathlen--;
                }
                }
-               dsize = DIRSIZ(nep);
-               freespace += nep->d_reclen - dsize;
-               loc += nep->d_reclen;
-               bcopy((caddr_t)nep, (caddr_t)ep, dsize);
+               vrele(ndp->ni_dvp);
+               goto dirloop;
        }
        /*
        }
        /*
-        * Update the pointer fields in the previous entry (if any),
-        * copy in the new entry, and write out the block.
+        * Check for read-only file systems.
         */
         */
-       if (ep->d_ino == 0) {
-               if (freespace + dsize < newentrysize)
-                       panic("wdir: compact1");
-               u.u_dent.d_reclen = freespace + dsize;
-       } else {
-               if (freespace < newentrysize)
-                       panic("wdir: compact2");
-               u.u_dent.d_reclen = freespace;
-               ep->d_reclen = dsize;
-               ep = (struct direct *)((char *)ep + dsize);
-       }
-       bcopy((caddr_t)&u.u_dent, (caddr_t)ep, (u_int)newentrysize);
-       bwrite(bp);
-       u.u_pdir->i_flag |= IUPD|ICHG;
-       iput(u.u_pdir);
-}
-
-dirremove()
-{
-       register struct inode *dp = u.u_pdir;
-       register struct buf *bp;
-       struct direct *ep;
-
-       if (u.u_count == 0) {
-               /*
-                * First entry in block: set d_ino to zero.
-                */
-               u.u_dent.d_ino = 0;
-               (void) rdwri(UIO_WRITE, dp, (caddr_t)&u.u_dent,
-                   (int)DIRSIZ(&u.u_dent), u.u_offset, 1, (int *)0);
-       } else {
+       if (flag == DELETE || flag == RENAME) {
                /*
                /*
-                * Collapse new free space into previous entry.
+                * Disallow directory write attempts on read-only
+                * file systems.
                 */
                 */
-               bp = blkatoff(dp, (int)(u.u_offset - u.u_count), (char **)&ep);
-               if (bp == 0)
-                       return (0);
-               ep->d_reclen += u.u_dent.d_reclen;
-               bwrite(bp);
-               dp->i_flag |= IUPD|ICHG;
+               if ((dp->v_mount->mnt_flag & rdonly) ||
+                   (wantparent && (ndp->ni_dvp->v_mount->mnt_flag & rdonly))) {
+                       error = EROFS;
+                       goto bad2;
+               }
        }
        }
-       return (1);
-}
-
-/*
- * Return buffer with contents of block "offset"
- * from the beginning of directory "ip".  If "res"
- * is non-zero, fill it in with a pointer to the
- * remaining space in the directory.
- */
-struct buf *
-blkatoff(ip, offset, res)
-       struct inode *ip;
-       off_t offset;
-       char **res;
-{
-       register struct fs *fs = ip->i_fs;
-       daddr_t lbn = lblkno(fs, offset);
-       int base = blkoff(fs, offset);
-       int bsize = blksize(fs, ip, lbn);
-       daddr_t bn = fsbtodb(fs, bmap(ip, lbn, B_WRITE, base, bsize));
-       register struct buf *bp;
-
-       if (u.u_error)
-               return (0);
-       bp = bread(ip->i_dev, bn, bsize);
-       if (bp->b_flags & B_ERROR) {
-               brelse(bp);
-               return (0);
+       if (ndp->ni_nameiop & SAVESTART) {
+               ndp->ni_startdir = ndp->ni_dvp;
+               p->p_spare[1]++;
+               VREF(ndp->ni_startdir);
        }
        }
-       if (res)
-               *res = bp->b_un.b_addr + base;
-       return (bp);
+       if (!wantparent)
+               vrele(ndp->ni_dvp);
+       if ((ndp->ni_nameiop & LOCKLEAF) == 0)
+               VOP_UNLOCK(dp);
+       return (0);
+
+bad2:
+       if ((ndp->ni_nameiop & LOCKPARENT) && *ndp->ni_next == '\0')
+               VOP_UNLOCK(ndp->ni_dvp);
+       vrele(ndp->ni_dvp);
+bad:
+       vput(dp);
+       ndp->ni_vp = NULL;
+       return (error);
 }
 }