X-Git-Url: https://git.subgeniuskitty.com/unix-history/.git/blobdiff_plain/f2a5ad78e1ece492aea02f3096e9d49416fd29c4..cc1730771008ddc8f1bccc5ed3aa8a987856137e:/usr/src/sys/ufs/ffs/ufs_lookup.c

diff --git a/usr/src/sys/ufs/ffs/ufs_lookup.c b/usr/src/sys/ufs/ffs/ufs_lookup.c
index 015dc16e76..124a0ca411 100644
--- a/usr/src/sys/ufs/ffs/ufs_lookup.c
+++ b/usr/src/sys/ufs/ffs/ufs_lookup.c
@@ -1,170 +1,164 @@
-/*	ufs_lookup.c	4.35	83/02/10	*/
-
-#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 "../h/nami.h"
-
-struct	buf *blkatoff();
-int	dirchk = 0;
 /*
- * 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.
+ * Copyright (c) 1989 The Regents of the University of California.
+ * All rights reserved.
  *
- * The func argument gives the routine which returns successive
- * characters of the name to be translated. 
+ * %sccs.include.redist.c%
  *
- * The flag argument is (LOOKUP, CREATE, DELETE) depending on whether
- * the name is to be (looked up, created, deleted).  If flag has
- * LOCKPARENT or'ed into it and the target of the pathname exists,
- * namei returns both the target and its parent directory locked. 
+ *	@(#)ufs_lookup.c	7.36 (Berkeley) %G%
+ */
+
+#include <sys/param.h>
+#include <sys/namei.h>
+#include <sys/buf.h>
+#include <sys/file.h>
+#include <sys/mount.h>
+#include <sys/vnode.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>
+
+struct	nchstats nchstats;
+#ifdef DIAGNOSTIC
+int	dirchk = 1;
+#else
+int	dirchk = 0;
+#endif
+
+/*
+ * Convert a component of a pathname into a pointer to a locked inode.
+ * This is a very central and rather complicated routine.
  * If the file system is not maintained in a strict tree hierarchy,
- * this can result in a deadlock situation.  When creating and
- * LOCKPARENT is specified, the target may not be ".".  When deleting
- * and LOCKPARENT is specified, the target may be ".", but the caller
- * must check to insure it does an irele and iput instead of two iputs.
+ * this can result in a deadlock situation (see comments in code below).
  *
- * The follow argument is 1 when symbolic links are to be followed
- * when they occur at the end of the name translation process.
+ * 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 and the target of the pathname
+ * exists, lookup returns both the target and its parent directory locked.
+ * When creating or renaming and LOCKPARENT is specified, the target may
+ * not be ".".  When deleting and LOCKPARENT is specified, the target may
+ * be "."., but the caller must check to ensure it does an vrele and iput
+ * instead of two iputs.
  *
- * Overall outline:
+ * Overall outline of ufs_lookup:
  *
- *	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
+ *	look for name in cache, if found, then if at end of path
+ *	  and deleting or creating, drop it, else return name
  *	search for name in directory, to found or notfound
  * notfound:
- *	if creating, return locked directory, leaving info on avail. slots
+ *	if creating, return locked directory, leaving info on available slots
  *	else return error
  * found:
  *	if at end of path and deleting, return information to allow delete
- *	if at end of path and rewriting (create and LOCKPARENT), lock targe
+ *	if at end of path and rewriting (RENAME and LOCKPARENT), lock target
  *	  inode and return info to allow rewrite
- *	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
+ *	if not at end, add name to cache; if at end and neither creating
+ *	  nor deleting, add name to cache
  *
- * NOTE: (LOOKUP | LOCKPARENT) currently returns the parent inode,
- *	 but unlocked.
+ * NOTE: (LOOKUP | LOCKPARENT) currently returns the parent inode unlocked.
  */
-struct inode *
-namei(func, flag, follow)
-	int (*func)(), flag, follow;
+int
+ufs_lookup(vdp, ndp, p)
+	register struct vnode *vdp;
+	register struct nameidata *ndp;
+	struct proc *p;
 {
-	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 inode *dp;	/* the directory we are searching */
+	struct buf *bp;			/* 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 slotoffset;			/* 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 */
+	int numdirpasses;		/* strategy for directory search */
+	int endsearch;			/* offset to end directory search */
+	int prevoff;			/* prev entry ndp->ni_ufs.ufs_offset */
 	struct inode *pdp;		/* saved dp during symlink work */
-	int i;
-	int lockparent;
+	struct vnode *tdp;		/* returned by VOP_VGET */
+	off_t enduseful;		/* pointer past last used dir slot */
+	u_long bmask;			/* block offset mask */
+	int flag;			/* LOOKUP, CREATE, RENAME, or DELETE */
+	int lockparent;			/* 1 => lockparent flag is set */
+	int wantparent;			/* 1 => wantparent or lockparent flag */
+	int error;
 
-	lockparent = flag & LOCKPARENT;
-	flag &= ~LOCKPARENT;
-	/*
-	 * 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 (u.u_error)
-		goto bad;
-
-	/*
-	 * Get starting directory.
-	 */
-	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 */
+	bp = NULL;
+	slotoffset = -1;
+	ndp->ni_dvp = vdp;
+	ndp->ni_vp = NULL;
+	dp = VTOI(vdp);
+	lockparent = ndp->ni_nameiop & LOCKPARENT;
+	flag = ndp->ni_nameiop & OPMASK;
+	wantparent = ndp->ni_nameiop & (LOCKPARENT|WANTPARENT);
 
-	/*
-	 * 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.
 	 */
-	if ((dp->i_mode&IFMT) != IFDIR) {
-		u.u_error = ENOTDIR;
-		goto bad;
-	}
-	if (access(dp, IEXEC))
-		goto bad;
+	if ((dp->i_mode&IFMT) != IFDIR)
+		return (ENOTDIR);
+	if (error = ufs_access(vdp, VEXEC, ndp->ni_cred, p))
+		return (error);
 
-dirloop2:
 	/*
-	 * Copy next component of name to u.u_dent.
+	 * We now have a segment name to search for, and a directory to search.
+	 *
+	 * Before tediously performing a linear scan of the directory,
+	 * check the name cache to see if the directory/name pair
+	 * we are looking for is known already.
 	 */
-	for (i = 0; *cp != 0 && *cp != '/'; cp++) {
-		if (i >= MAXNAMLEN) {
-			u.u_error = ENOENT;
-			goto bad;
+	if (error = cache_lookup(ndp)) {
+		int vpid;	/* capability number of vnode */
+
+		if (error == ENOENT)
+			return (error);
+#ifdef PARANOID
+		if (vdp == ndp->ni_rdir && ndp->ni_isdotdot)
+			panic("ufs_lookup: .. through root");
+#endif
+		/*
+		 * Get the next vnode in the path.
+		 * See comment below starting `Step through' for
+		 * an explaination of the locking protocol.
+		 */
+		pdp = dp;
+		dp = VTOI(ndp->ni_vp);
+		vdp = ndp->ni_vp;
+		vpid = vdp->v_id;
+		if (pdp == dp) {
+			VREF(vdp);
+			error = 0;
+		} else if (ndp->ni_isdotdot) {
+			IUNLOCK(pdp);
+			error = vget(vdp);
+			if (!error && lockparent && *ndp->ni_next == '\0')
+				ILOCK(pdp);
+		} else {
+			error = vget(vdp);
+			if (!lockparent || error || *ndp->ni_next != '\0')
+				IUNLOCK(pdp);
 		}
-		u.u_dent.d_name[i++] = *cp;
-	}
-	u.u_dent.d_namlen = i;
-	u.u_dent.d_name[i] = 0;
-
-	/*
-	 * 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 || lockparent) {
-			u.u_error = ENOENT;
-			goto bad;
+		/*
+		 * Check that the capability number did not change
+		 * while we were waiting for the lock.
+		 */
+		if (!error) {
+			if (vpid == vdp->v_id)
+				return (0);
+			ufs_iput(dp);
+			if (lockparent && pdp != dp && *ndp->ni_next == '\0')
+				IUNLOCK(pdp);
 		}
-		brelse(nbp);
-		return (dp);
+		ILOCK(pdp);
+		dp = pdp;
+		vdp = ITOV(dp);
+		ndp->ni_vp = NULL;
 	}
 
 	/*
@@ -174,59 +168,78 @@ dirloop2:
 	 * case it doesn't already exist.
 	 */
 	slotstatus = FOUND;
-	if (flag == CREATE && *cp == 0) {
+	if ((flag == CREATE || flag == RENAME) && *ndp->ni_next == 0) {
 		slotstatus = NONE;
 		slotfreespace = 0;
-		slotneeded = DIRSIZ(&u.u_dent);
+		slotneeded = ((sizeof (struct direct) - (MAXNAMLEN + 1)) +
+			((ndp->ni_namelen + 1 + 3) &~ 3));
 	}
 
-	dirsize = roundup(dp->i_size, DIRBLKSIZ);
-	u.u_offset = 0;
-	while (u.u_offset < dirsize) {
+	/*
+	 * If there is cached information on a previous search of
+	 * this directory, pick up where we last left off.
+	 * We cache only lookups as these are the most common
+	 * and have the greatest payoff. Caching CREATE has little
+	 * benefit as it usually must search the entire directory
+	 * to determine that the entry does not exist. Caching the
+	 * location of the last DELETE or RENAME has not reduced
+	 * profiling time and hence has been removed in the interest
+	 * of simplicity.
+	 */
+	bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
+	if (flag != LOOKUP || dp->i_diroff == 0 || dp->i_diroff > dp->i_size) {
+		ndp->ni_ufs.ufs_offset = 0;
+		numdirpasses = 1;
+	} else {
+		ndp->ni_ufs.ufs_offset = dp->i_diroff;
+		if ((entryoffsetinblock = ndp->ni_ufs.ufs_offset & bmask) &&
+		    (error = VOP_BLKATOFF(vdp, ndp->ni_ufs.ufs_offset, NULL,
+		     &bp)))
+			return (error);
+		numdirpasses = 2;
+		nchstats.ncs_2passes++;
+	}
+	endsearch = roundup(dp->i_size, DIRBLKSIZ);
+	enduseful = 0;
+
+searchloop:
+	while (ndp->ni_ufs.ufs_offset < endsearch) {
 		/*
-		 * If offset is on a block boundary,
-		 * read the next directory block.
-		 * Release previous if it exists.
+		 * 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 ((ndp->ni_ufs.ufs_offset & bmask) == 0) {
 			if (bp != NULL)
 				brelse(bp);
-			bp = blkatoff(dp, u.u_offset, (char **)0);
-			if (bp == 0)
-				goto bad;
+			if (error = VOP_BLKATOFF(vdp, ndp->ni_ufs.ufs_offset,
+			    NULL, &bp))
+				return (error);
 			entryoffsetinblock = 0;
 		}
-
 		/*
 		 * If still looking for a slot, and at a DIRBLKSIZE
-		 * boundary, have to start looking for free space
-		 * again.
+		 * boundary, have to start looking for free space again.
 		 */
 		if (slotstatus == NONE &&
-		    (entryoffsetinblock&(DIRBLKSIZ-1)) == 0) {
+		    (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.
+		 * Get pointer to next entry.
+		 * Full validation checks are slow, so we only check
+		 * enough to insure forward progress through the
+		 * directory. Complete checks can be run by patching
+		 * "dirchk" to be true.
 		 */
 		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;
+		if (ep->d_reclen == 0 ||
+		    dirchk && ufs_dirbadentry(ep, entryoffsetinblock)) {
+			int i;
+
+			ufs_dirbad(dp, ndp->ni_ufs.ufs_offset, "mangled entry");
+			i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
+			ndp->ni_ufs.ufs_offset += i;
 			entryoffsetinblock += i;
 			continue;
 		}
@@ -245,17 +258,18 @@ dirloop2:
 			if (size > 0) {
 				if (size >= slotneeded) {
 					slotstatus = FOUND;
-					slotoffset = u.u_offset;
+					slotoffset = ndp->ni_ufs.ufs_offset;
 					slotsize = ep->d_reclen;
 				} else if (slotstatus == NONE) {
 					slotfreespace += size;
 					if (slotoffset == -1)
-						slotoffset = u.u_offset;
+						slotoffset =
+						      ndp->ni_ufs.ufs_offset;
 					if (slotfreespace >= slotneeded) {
 						slotstatus = COMPACT;
 						slotsize =
-						    u.u_offset+ep->d_reclen -
-						      slotoffset;
+						      ndp->ni_ufs.ufs_offset +
+						      ep->d_reclen - slotoffset;
 					}
 				}
 			}
@@ -265,339 +279,391 @@ dirloop2:
 		 * 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))
+			if (ep->d_namlen == ndp->ni_namelen &&
+			    !bcmp(ndp->ni_ptr, ep->d_name,
+				(unsigned)ep->d_namlen)) {
+				/*
+				 * Save directory entry's inode number and
+				 * reclen in ndp->ni_ufs area, and release
+				 * directory buffer.
+				 */
+				ndp->ni_ufs.ufs_ino = ep->d_ino;
+				ndp->ni_ufs.ufs_reclen = ep->d_reclen;
+				brelse(bp);
 				goto found;
+			}
 		}
-		prevoff = u.u_offset;
-		u.u_offset += ep->d_reclen;
+		prevoff = ndp->ni_ufs.ufs_offset;
+		ndp->ni_ufs.ufs_offset += ep->d_reclen;
 		entryoffsetinblock += ep->d_reclen;
+		if (ep->d_ino)
+			enduseful = ndp->ni_ufs.ufs_offset;
 	}
 /* notfound: */
+	/*
+	 * If we started in the middle of the directory and failed
+	 * to find our target, we must check the beginning as well.
+	 */
+	if (numdirpasses == 2) {
+		numdirpasses--;
+		ndp->ni_ufs.ufs_offset = 0;
+		endsearch = dp->i_diroff;
+		goto searchloop;
+	}
+	if (bp != NULL)
+		brelse(bp);
 	/*
 	 * If creating, and at end of pathname and current
 	 * directory has not been removed, then can consider
 	 * allowing file to be created.
 	 */
-	if (flag == CREATE && *cp == 0 && dp->i_nlink != 0) {
+	if ((flag == CREATE || flag == RENAME) &&
+	    *ndp->ni_next == 0 && dp->i_nlink != 0) {
 		/*
 		 * Access for write is interpreted as allowing
 		 * creation of files in the directory.
 		 */
-		if (access(dp, IWRITE))
-			goto bad;
+		if (error = ufs_access(vdp, VWRITE, ndp->ni_cred, p))
+			return (error);
 		/*
 		 * 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)
+		 * then set ndp->ni_ufs.ufs_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 from ndp->ni_ufs.ufs_offset
+		 * to ndp->ni_ufs.ufs_offset + ndp->ni_ufs.ufs_count.
 		 */
-		if (slotstatus == NONE)
-			u.u_count = 0;
-		else {
-			u.u_offset = slotoffset;
-			u.u_count = slotsize;
+		if (slotstatus == NONE) {
+			ndp->ni_ufs.ufs_offset = roundup(dp->i_size, DIRBLKSIZ);
+			ndp->ni_ufs.ufs_count = 0;
+			enduseful = ndp->ni_ufs.ufs_offset;
+		} else {
+			ndp->ni_ufs.ufs_offset = slotoffset;
+			ndp->ni_ufs.ufs_count = slotsize;
+			if (enduseful < slotoffset + slotsize)
+				enduseful = slotoffset + slotsize;
 		}
+		ndp->ni_ufs.ufs_endoff = roundup(enduseful, DIRBLKSIZ);
 		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.
+		 * We return ni_vp == NULL to indicate that the entry
+		 * does not currently exist; we leave a pointer to
+		 * the (locked) directory inode in ndp->ni_dvp.
+		 * The pathname buffer is saved so that the name
+		 * can be obtained later.
+		 *
+		 * NB - if the directory is unlocked, then this
+		 * information cannot be used.
 		 */
-		u.u_pdir = dp;
-		return (NULL);
+		ndp->ni_nameiop |= SAVENAME;
+		if (!lockparent)
+			IUNLOCK(dp);
 	}
-	u.u_error = ENOENT;
-	goto bad;
+	/*
+	 * Insert name into cache (as non-existent) if appropriate.
+	 */
+	if (ndp->ni_makeentry && flag != CREATE)
+		cache_enter(ndp);
+	return (ENOENT);
+
 found:
+	if (numdirpasses == 2)
+		nchstats.ncs_pass2++;
 	/*
 	 * Check that directory length properly reflects presence
 	 * of this entry.
 	 */
 	if (entryoffsetinblock + DIRSIZ(ep) > dp->i_size) {
-		dirbad(dp, "i_size too small");
+		ufs_dirbad(dp, ndp->ni_ufs.ufs_offset, "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.
+	 * Found component in pathname.
+	 * If the final component of path name, save information
+	 * in the cache as to where the entry was found.
 	 */
-	bcopy((caddr_t)ep, (caddr_t)&u.u_dent, (u_int)DIRSIZ(ep));
-	brelse(bp);
-	bp = NULL;
+	if (*ndp->ni_next == '\0' && flag == LOOKUP)
+		dp->i_diroff = ndp->ni_ufs.ufs_offset &~ (DIRBLKSIZ - 1);
 
 	/*
 	 * If deleting, and at end of pathname, return
 	 * parameters which can be used to remove file.
-	 * If the lockparent flag isn't set, we return only
-	 * the directory (in u.u_pdir), otherwise we go
+	 * If the wantparent flag isn't set, we return only
+	 * the directory (in ndp->ni_dvp), otherwise we go
 	 * on and lock the inode, being careful with ".".
 	 */
-	if (flag == DELETE && *cp == 0) {
+	if (flag == DELETE && *ndp->ni_next == 0) {
 		/*
 		 * Write access to directory required to delete files.
 		 */
-		if (access(dp, IWRITE))
-			goto bad;
-		u.u_pdir = dp;		/* for dirremove() */
+		if (error = ufs_access(vdp, VWRITE, ndp->ni_cred, p))
+			return (error);
 		/*
-		 * Return pointer to current entry in u.u_offset,
+		 * Return pointer to current entry in ndp->ni_ufs.ufs_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().
+		 * is a previous entry in this block) in ndp->ni_ufs.ufs_count.
+		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
 		 */
-		if ((u.u_offset&(DIRBLKSIZ-1)) == 0)
-			u.u_count = 0;
+		if ((ndp->ni_ufs.ufs_offset&(DIRBLKSIZ-1)) == 0)
+			ndp->ni_ufs.ufs_count = 0;
 		else
-			u.u_count = u.u_offset - prevoff;
-		if (lockparent) {
-			if (dp->i_number == u.u_dent.d_ino)
-				dp->i_count++;
-			else {
-				dp = iget(dp->i_dev, fs, u.u_dent.d_ino);
-				if (dp == NULL) {
-					iput(u.u_pdir);
-					goto bad;
-				}
-			}
+			ndp->ni_ufs.ufs_count =
+			    ndp->ni_ufs.ufs_offset - prevoff;
+		if (dp->i_number == ndp->ni_ufs.ufs_ino) {
+			VREF(vdp);
+			ndp->ni_vp = vdp;
+			return (0);
 		}
-		brelse(nbp);
-		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 (error = VOP_VGET(vdp, ndp->ni_ufs.ufs_ino, &tdp))
+			return (error);
+		/*
+		 * If directory is "sticky", then user must own
+		 * the directory, or the file in it, else she
+		 * may not delete it (unless she's root). This
+		 * implements append-only directories.
+		 */
+		if ((dp->i_mode & ISVTX) &&
+		    ndp->ni_cred->cr_uid != 0 &&
+		    ndp->ni_cred->cr_uid != dp->i_uid &&
+		    VTOI(tdp)->i_uid != ndp->ni_cred->cr_uid) {
+			vput(tdp);
+			return (EPERM);
 		}
+		ndp->ni_vp = tdp;
+		if (!lockparent)
+			IUNLOCK(dp);
+		return (0);
 	}
 
 	/*
-	 * If rewriting (rename), return the inode and the
+	 * If rewriting (RENAME), return the inode and the
 	 * information required to rewrite the present directory
 	 * Must get inode of directory entry to verify it's a
-	 * regular file, or empty directory.  
+	 * regular file, or empty directory.
 	 */
-	if ((flag == CREATE && lockparent) && *cp == 0) {
-		if (access(dp, IWRITE))
-			goto bad;
-		u.u_pdir = dp;		/* for dirrewrite() */
+	if (flag == RENAME && wantparent && *ndp->ni_next == 0) {
+		if (error = ufs_access(vdp, VWRITE, ndp->ni_cred, p))
+			return (error);
 		/*
-		 * Careful about locking second inode. 
-		 * This can only occur if the target is ".". 
+		 * Careful about locking second inode.
+		 * This can only occur if the target is ".".
 		 */
-		if (dp->i_number == u.u_dent.d_ino) {
-			u.u_error = EISDIR;		/* XXX */
-			goto bad;
-		}
-		dp = iget(dp->i_dev, fs, u.u_dent.d_ino);
-		if (dp == NULL) {
-			iput(u.u_pdir);
-			goto bad;
-		}
-		brelse(nbp);
-		return (dp);
+		if (dp->i_number == ndp->ni_ufs.ufs_ino)
+			return (EISDIR);
+		if (error = VOP_VGET(vdp, ndp->ni_ufs.ufs_ino, &tdp))
+			return (error);
+		ndp->ni_vp = tdp;
+		ndp->ni_nameiop |= SAVENAME;
+		if (!lockparent)
+			IUNLOCK(dp);
+		return (0);
 	}
 
 	/*
-	 * 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.
+	 * Step through the translation in the name.  We do not `iput' the
+	 * directory because we may need it again if a symbolic link
+	 * is relative to the current directory.  Instead we save it
+	 * unlocked as "pdp".  We must get the target inode before unlocking
+	 * the directory to insure that the inode will not be removed
+	 * before we get it.  We prevent deadlock by always fetching
+	 * inodes from the root, moving down the directory tree. Thus
+	 * when following backward pointers ".." we must unlock the
+	 * parent directory before getting the requested directory.
+	 * There is a potential race condition here if both the current
+	 * and parent directories are removed before the `iget' for the
+	 * inode associated with ".." returns.  We hope that this occurs
+	 * infrequently since we cannot avoid this race condition without
+	 * implementing a sophisticated deadlock detection algorithm.
+	 * Note also that this simple deadlock detection scheme will not
+	 * work if the file system has any hard links other than ".."
+	 * that point backwards in the directory structure.
 	 */
 	pdp = dp;
-	iunlock(pdp);
-	dp = iget(dp->i_dev, fs, u.u_dent.d_ino);
-	if (dp == NULL)
-		goto bad2;
-	fs = dp->i_fs;
-
-	/*
-	 * Check for symbolic link
-	 */
-	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;
+	if (ndp->ni_isdotdot) {
+		IUNLOCK(pdp);	/* race to get the inode */
+		if (error = VOP_VGET(vdp, ndp->ni_ufs.ufs_ino, &tdp)) {
+			ILOCK(pdp);
+			return (error);
 		}
-		ovbcopy(cp, nbp->b_un.b_addr + dp->i_size, pathlen);
-		u.u_error =
-		    rdwri(UIO_READ, dp, nbp->b_un.b_addr, (int)dp->i_size,
-			0, 1, (int *)0);
-		if (u.u_error)
-			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;
+		if (lockparent && *ndp->ni_next == '\0')
+			ILOCK(pdp);
+		ndp->ni_vp = tdp;
+	} else if (dp->i_number == ndp->ni_ufs.ufs_ino) {
+		VREF(vdp);	/* we want ourself, ie "." */
+		ndp->ni_vp = vdp;
+	} else {
+		if (error = VOP_VGET(vdp, ndp->ni_ufs.ufs_ino, &tdp))
+			return (error);
+		if (!lockparent || *ndp->ni_next != '\0')
+			IUNLOCK(pdp);
+		ndp->ni_vp = tdp;
 	}
 
 	/*
-	 * Not a symbolic link.  If more pathname,
-	 * continue at next component, else return.
+	 * Insert name into cache if appropriate.
 	 */
-	if (*cp == '/') {
-		while (*cp == '/')
-			cp++;
-		irele(pdp);
-		goto dirloop;
-	}
-	brelse(nbp);
-	if (lockparent)
-		u.u_pdir = pdp;
-	else
-		irele(pdp);
-	return (dp);
-bad2:
-	irele(pdp);
-bad:
-	if (bp)
-		brelse(bp);
-	if (dp)
-		iput(dp);
-	brelse(nbp);
-	return (NULL);
+	if (ndp->ni_makeentry)
+		cache_enter(ndp);
+	return (0);
 }
 
-dirbad(ip, how)
+void
+ufs_dirbad(ip, offset, how)
 	struct inode *ip;
+	off_t offset;
 	char *how;
 {
+	struct mount *mp;
 
-	printf("%s: bad dir ino %d at offset %d: %s\n",
-	    ip->i_fs->fs_fsmnt, ip->i_number, u.u_offset, how);
+	mp = ITOV(ip)->v_mount;
+	(void)printf("%s: bad dir ino %d at offset %d: %s\n",
+	    mp->mnt_stat.f_mntonname, ip->i_number, offset, how);
+	if ((mp->mnt_stat.f_flags & MNT_RDONLY) == 0)
+		panic("bad dir");
 }
 
-dirbadname(ep)
+/*
+ * Do consistency checking on a directory entry:
+ *	record length must be multiple of 4
+ *	entry must fit in rest of its DIRBLKSIZ block
+ *	record must be large enough to contain entry
+ *	name is not longer than MAXNAMLEN
+ *	name must be as long as advertised, and null terminated
+ */
+int
+ufs_dirbadentry(ep, entryoffsetinblock)
 	register struct direct *ep;
+	int entryoffsetinblock;
 {
 	register int i;
 
+	if ((ep->d_reclen & 0x3) != 0 ||
+	    ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
+	    ep->d_reclen < DIRSIZ(ep) || ep->d_namlen > MAXNAMLEN) {
+		/*return (1); */
+		printf("First bad\n");
+		goto bad;
+	}
 	for (i = 0; i < ep->d_namlen; i++)
-		if (ep->d_name[i] == 0)
-			return (1);
+		if (ep->d_name[i] == '\0') {
+			/*return (1); */
+			printf("Second bad\n");
+			goto bad;
+	}
+	if (ep->d_name[i])
+		goto bad;
 	return (ep->d_name[i]);
+bad:
+printf("ufs_dirbadentry: jumping out: reclen: %d namlen %d ino %d name %s\n",
+	ep->d_reclen, ep->d_namlen, ep->d_ino, ep->d_name );
+	return(1);
 }
 
 /*
  * 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.
+ * that it left in nameidata.  The argument ip is the inode which the new
+ * directory entry will refer to.  The nameidata field ndp->ni_dvp is a
+ * pointer to the directory to be written, which was left locked by namei.
+ * Remaining parameters (ndp->ni_ufs.ufs_offset, ndp->ni_ufs.ufs_count)
+ * indicate how the space for the new entry is to be obtained.
  */
-direnter(ip)
+int
+ufs_direnter(ip, ndp)
 	struct inode *ip;
+	register struct nameidata *ndp;
 {
 	register struct direct *ep, *nep;
+	register struct inode *dp;
+	register struct vnode *dvp;
 	struct buf *bp;
-	int loc, freespace, error = 0;
+	struct direct newdir;
+	struct iovec aiov;
+	struct uio auio;
 	u_int dsize;
-	int newentrysize;
+	int error, loc, newentrysize, spacefree;
 	char *dirbuf;
 
-	u.u_dent.d_ino = ip->i_number;
-	u.u_segflg = 1;
-	newentrysize = DIRSIZ(&u.u_dent);
-	if (u.u_count == 0) {
+#ifdef DIAGNOSTIC
+	if ((ndp->ni_nameiop & SAVENAME) == 0)
+		panic("direnter: missing name");
+#endif
+	dvp = ndp->ni_dvp;
+	dp = VTOI(dvp);
+	newdir.d_ino = ip->i_number;
+	newdir.d_namlen = ndp->ni_namelen;
+	bcopy(ndp->ni_ptr, newdir.d_name, (unsigned)ndp->ni_namelen + 1);
+	newentrysize = DIRSIZ(&newdir);
+	if (ndp->ni_ufs.ufs_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 ndp->ni_ufs.ufs_count is 0, then namei could find no
+		 * space in the directory. Here, ndp->ni_ufs.ufs_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))
+		if (ndp->ni_ufs.ufs_offset & (DIRBLKSIZ - 1))
 			panic("wdir: newblk");
-		u.u_dent.d_reclen = DIRBLKSIZ;
-		error = rdwri(UIO_WRITE, u.u_pdir, (caddr_t)&u.u_dent,
-		    newentrysize, u.u_offset, 1, (int *)0);
-		iput(u.u_pdir);
+		auio.uio_offset = ndp->ni_ufs.ufs_offset;
+		newdir.d_reclen = DIRBLKSIZ;
+		auio.uio_resid = newentrysize;
+		aiov.iov_len = newentrysize;
+		aiov.iov_base = (caddr_t)&newdir;
+		auio.uio_iov = &aiov;
+		auio.uio_iovcnt = 1;
+		auio.uio_rw = UIO_WRITE;
+		auio.uio_segflg = UIO_SYSSPACE;
+		auio.uio_procp = (struct proc *)0;
+		error = VOP_WRITE(dvp, &auio, IO_SYNC, ndp->ni_cred);
+		if (DIRBLKSIZ >
+		    VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_bsize)
+			/* XXX should grow with balloc() */
+			panic("ufs_direnter: frag size");
+		else if (!error) {
+			dp->i_size = roundup(dp->i_size, DIRBLKSIZ);
+			dp->i_flag |= ICHG;
+		}
 		return (error);
 	}
 
 	/*
-	 * 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.
+	 * If ndp->ni_ufs.ufs_count is non-zero, then namei found space
+	 * for the new entry in the range ndp->ni_ufs.ufs_offset to
+	 * ndp->ni_ufs.ufs_offset + ndp->ni_ufs.ufs_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.
+	 *
+	 * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
 	 */
-	if (u.u_offset + u.u_count > u.u_pdir->i_size)
-		u.u_pdir->i_size = u.u_offset + u.u_count;
-
+	if (ndp->ni_ufs.ufs_offset + ndp->ni_ufs.ufs_count > dp->i_size)
+		dp->i_size = ndp->ni_ufs.ufs_offset + ndp->ni_ufs.ufs_count;
 	/*
-	 * Get the block containing the space for the new directory
-	 * entry.  Should return error by result instead of u.u_error.
+	 * Get the block containing the space for the new directory entry.
 	 */
-	bp = blkatoff(u.u_pdir, u.u_offset, (char **)&dirbuf);
-	if (bp == 0) {
-		iput(u.u_pdir);
-		return (u.u_error);
-	}
-
+	if (error = VOP_BLKATOFF(dvp, ndp->ni_ufs.ufs_offset, &dirbuf, &bp))
+		return (error);
 	/*
-	 * 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.
+	 * 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 ndp->ni_ufs.ufs_offset to
+	 * ndp->ni_ufs.ufs_offset + ndp->ni_ufs.ufs_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; ) {
+	spacefree = ep->d_reclen - dsize;
+	for (loc = ep->d_reclen; loc < ndp->ni_ufs.ufs_count; ) {
 		nep = (struct direct *)(dirbuf + loc);
 		if (ep->d_ino) {
 			/* trim the existing slot */
@@ -605,10 +671,10 @@ direnter(ip)
 			ep = (struct direct *)((char *)ep + dsize);
 		} else {
 			/* overwrite; nothing there; header is ours */
-			freespace += dsize;	
+			spacefree += dsize;
 		}
 		dsize = DIRSIZ(nep);
-		freespace += nep->d_reclen - dsize;
+		spacefree += nep->d_reclen - dsize;
 		loc += nep->d_reclen;
 		bcopy((caddr_t)nep, (caddr_t)ep, dsize);
 	}
@@ -617,60 +683,70 @@ direnter(ip)
 	 * copy in the new entry, and write out the block.
 	 */
 	if (ep->d_ino == 0) {
-		if (freespace + dsize < newentrysize)
+		if (spacefree + dsize < newentrysize)
 			panic("wdir: compact1");
-		u.u_dent.d_reclen = freespace + dsize;
+		newdir.d_reclen = spacefree + dsize;
 	} else {
-		if (freespace < newentrysize)
+		if (spacefree < newentrysize)
 			panic("wdir: compact2");
-		u.u_dent.d_reclen = freespace;
+		newdir.d_reclen = spacefree;
 		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);
+	bcopy((caddr_t)&newdir, (caddr_t)ep, (u_int)newentrysize);
+	error = VOP_BWRITE(bp);
+	dp->i_flag |= IUPD|ICHG;
+	if (!error && ndp->ni_ufs.ufs_endoff &&
+	    ndp->ni_ufs.ufs_endoff < dp->i_size)
+		error = VOP_TRUNCATE(dvp, (u_long)ndp->ni_ufs.ufs_endoff,
+		    IO_SYNC);
 	return (error);
 }
 
 /*
- * Remove a directory entry after a call to namei, using the
- * parameters which it left in the u. area.  The u. entry
- * u_offset contains the offset into the directory of the
- * entry to be eliminated.  The u_count field contains the
+ * Remove a directory entry after a call to namei, using
+ * the parameters which it left in nameidata. The entry
+ * ni_ufs.ufs_offset contains the offset into the directory of the
+ * entry to be eliminated.  The ni_ufs.ufs_count field contains the
  * size of the previous record in the directory.  If this
  * is 0, the first entry is being deleted, so we need only
  * zero the inode number to mark the entry as free.  If the
- * entry isn't the first in the directory, we must reclaim
+ * entry is not the first in the directory, we must reclaim
  * the space of the now empty record by adding the record size
  * to the size of the previous entry.
  */
-dirremove()
+int
+ufs_dirremove(ndp)
+	register struct nameidata *ndp;
 {
-	register struct inode *dp = u.u_pdir;
-	register struct buf *bp;
+	register struct inode *dp;
 	struct direct *ep;
+	struct buf *bp;
+	int error;
 
-	if (u.u_count == 0) {
+	dp = VTOI(ndp->ni_dvp);
+	if (ndp->ni_ufs.ufs_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 {
-		/*
-		 * Collapse new free space into previous entry.
-		 */
-		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);
+		if (error = VOP_BLKATOFF(ndp->ni_dvp, ndp->ni_ufs.ufs_offset,
+		    (char **)&ep, &bp))
+			return (error);
+		ep->d_ino = 0;
+		error = VOP_BWRITE(bp);
 		dp->i_flag |= IUPD|ICHG;
+		return (error);
 	}
-	return (1);
+	/*
+	 * Collapse new free space into previous entry.
+	 */
+	if (error = VOP_BLKATOFF(ndp->ni_dvp,
+	    ndp->ni_ufs.ufs_offset - ndp->ni_ufs.ufs_count, (char **)&ep, &bp))
+		return (error);
+	ep->d_reclen += ndp->ni_ufs.ufs_reclen;
+	error = VOP_BWRITE(bp);
+	dp->i_flag |= IUPD|ICHG;
+	return (error);
 }
 
 /*
@@ -678,74 +754,137 @@ dirremove()
  * supplied.  The parameters describing the directory entry are
  * set up by a call to namei.
  */
-dirrewrite(dp, ip)
+int
+ufs_dirrewrite(dp, ip, ndp)
 	struct inode *dp, *ip;
+	struct nameidata *ndp;
 {
+	struct buf *bp;
+	struct direct *ep;
+	int error;
 
-	u.u_dent.d_ino = ip->i_number;
-	u.u_error = rdwri(UIO_WRITE, dp, (caddr_t)&u.u_dent,
-		(int)DIRSIZ(&u.u_dent), u.u_offset, 1, (int *)0);
-	iput(dp);
-}
-
-/*
- * 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 (res)
-		*res = bp->b_un.b_addr + base;
-	return (bp);
+	if (error = VOP_BLKATOFF(ITOV(dp), ndp->ni_ufs.ufs_offset,
+	    (char **)&ep, &bp))
+		return (error);
+	ep->d_ino = ip->i_number;
+	error = VOP_BWRITE(bp);
+	dp->i_flag |= IUPD|ICHG;
+	return (error);
 }
 
 /*
  * Check if a directory is empty or not.
  * Inode supplied must be locked.
+ *
+ * Using a struct dirtemplate here is not precisely
+ * what we want, but better than using a struct direct.
+ *
+ * NB: does not handle corrupted directories.
  */
-dirempty(ip)
+int
+ufs_dirempty(ip, parentino, cred)
 	register struct inode *ip;
+	ino_t parentino;
+	struct ucred *cred;
 {
 	register off_t off;
-	struct direct dbuf;
-	register struct direct *dp = &dbuf;
+	struct dirtemplate dbuf;
+	register struct direct *dp = (struct direct *)&dbuf;
 	int error, count;
+#define	MINDIRSIZ (sizeof (struct dirtemplate) / 2)
 
 	for (off = 0; off < ip->i_size; off += dp->d_reclen) {
-		error = rdwri(UIO_READ, ip, (caddr_t)dp,
-			sizeof (struct direct), off, 1, &count);
-		count = sizeof (struct direct) - count;
-#define	MINDIRSIZ (sizeof (struct direct) - (MAXNAMLEN + 1))
-		if (error || count < MINDIRSIZ || count < DIRSIZ(dp))
+		error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ, off,
+		   UIO_SYSSPACE, IO_NODELOCKED, cred, &count, (struct proc *)0);
+		/*
+		 * Since we read MINDIRSIZ, residual must
+		 * be 0 unless we're at end of file.
+		 */
+		if (error || count != 0)
+			return (0);
+		/* avoid infinite loops */
+		if (dp->d_reclen == 0)
 			return (0);
+		/* skip empty entries */
 		if (dp->d_ino == 0)
 			continue;
+		/* accept only "." and ".." */
+		if (dp->d_namlen > 2)
+			return (0);
 		if (dp->d_name[0] != '.')
 			return (0);
-		if (dp->d_namlen == 1 ||
-		    (dp->d_namlen == 2 && dp->d_name[1] == '.'))
+		/*
+		 * At this point d_namlen must be 1 or 2.
+		 * 1 implies ".", 2 implies ".." if second
+		 * char is also "."
+		 */
+		if (dp->d_namlen == 1)
+			continue;
+		if (dp->d_name[1] == '.' && dp->d_ino == parentino)
 			continue;
 		return (0);
 	}
 	return (1);
 }
+
+/*
+ * Check if source directory is in the path of the target directory.
+ * Target is supplied locked, source is unlocked.
+ * The target is always iput before returning.
+ */
+int
+ufs_checkpath(source, target, cred)
+	struct inode *source, *target;
+	struct ucred *cred;
+{
+	struct dirtemplate dirbuf;
+	register struct inode *ip;
+	struct vnode *vp;
+	int error, rootino;
+
+	ip = target;
+	if (ip->i_number == source->i_number) {
+		error = EEXIST;
+		goto out;
+	}
+	rootino = ROOTINO;
+	error = 0;
+	if (ip->i_number == rootino)
+		goto out;
+
+	for (;;) {
+		if ((ip->i_mode&IFMT) != IFDIR) {
+			error = ENOTDIR;
+			break;
+		}
+		vp = ITOV(ip);
+		error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
+			sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
+			IO_NODELOCKED, cred, (int *)0, (struct proc *)0);
+		if (error != 0)
+			break;
+		if (dirbuf.dotdot_namlen != 2 ||
+		    dirbuf.dotdot_name[0] != '.' ||
+		    dirbuf.dotdot_name[1] != '.') {
+			error = ENOTDIR;
+			break;
+		}
+		if (dirbuf.dotdot_ino == source->i_number) {
+			error = EINVAL;
+			break;
+		}
+		if (dirbuf.dotdot_ino == rootino)
+			break;
+		ufs_iput(ip);
+		if (error = VOP_VGET(vp, dirbuf.dotdot_ino, &vp))
+			break;
+		ip = VTOI(vp);
+	}
+
+out:
+	if (error == ENOTDIR)
+		printf("checkpath: .. not a directory\n");
+	if (ip != NULL)
+		ufs_iput(ip);
+	return (error);
+}