Import 4.4BSD-Lite2 onto the vendor branch, note that in the kernel, all

files are off the vendor branch, so this should not change anything.

A "U" marker generally means that the file was not changed in between
the 4.4Lite and Lite-2 releases, and does not need a merge.  "C" generally
means that there was a change.
[two new auxillary files in miscfs/union]

8 files changed, 2152 insertions, 590 deletions

diff --git a/sys/miscfs/union/README b/sys/miscfs/union/README
index 14a4769..265caa1 100644
--- a/sys/miscfs/union/README
+++ b/sys/miscfs/union/README
@@ -1,7 +1,8 @@
 If you plan on using union mounts, then you should consider replacing
-"libc/gen/opendir.c" in the C library with the file "libc.opendir.c"
+three files in "libc/gen" in the C library with the files in "libc"  
 in this directory.  The replacement version of opendir() automatically
-removes duplicate names when a union stack is encountered.  You will
-then need to rebuild the C library and all commands.
+removes duplicate names when a union stack is encountered. The other
+two files do special handling of whiteouts. You will then need to 
+rebuild the C library and all commands.
 
-@(#)README	8.1 (Berkeley) 2/15/94
+@(#)README	8.2 (Berkeley) 11/4/94
diff --git a/sys/miscfs/union/libc.fts.c b/sys/miscfs/union/libc.fts.c
new file mode 100644
index 0000000..234d076
--- /dev/null
+++ b/sys/miscfs/union/libc.fts.c
@@ -0,0 +1,995 @@
+/*-
+ * Copyright (c) 1990, 1993, 1994
+ *	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.
+ */
+
+#if defined(LIBC_SCCS) && !defined(lint)
+static char sccsid[] = "@(#)fts.c	8.6 (Berkeley) 8/14/94";
+#endif /* LIBC_SCCS and not lint */
+
+#include <sys/param.h>
+#include <sys/stat.h>
+
+#include <dirent.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <fts.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+static FTSENT	*fts_alloc __P((FTS *, char *, int));
+static FTSENT	*fts_build __P((FTS *, int));
+static void	 fts_lfree __P((FTSENT *));
+static void	 fts_load __P((FTS *, FTSENT *));
+static size_t	 fts_maxarglen __P((char * const *));
+static void	 fts_padjust __P((FTS *, void *));
+static int	 fts_palloc __P((FTS *, size_t));
+static FTSENT	*fts_sort __P((FTS *, FTSENT *, int));
+static u_short	 fts_stat __P((FTS *, FTSENT *, int));
+
+#define	ISDOT(a)	(a[0] == '.' && (!a[1] || a[1] == '.' && !a[2]))
+
+#define	ISSET(opt)	(sp->fts_options & opt)
+#define	SET(opt)	(sp->fts_options |= opt)
+
+#define	CHDIR(sp, path)	(!ISSET(FTS_NOCHDIR) && chdir(path))
+#define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
+
+/* fts_build flags */
+#define	BCHILD		1		/* fts_children */
+#define	BNAMES		2		/* fts_children, names only */
+#define	BREAD		3		/* fts_read */
+
+FTS *
+fts_open(argv, options, compar)
+	char * const *argv;
+	register int options;
+	int (*compar)();
+{
+	register FTS *sp;
+	register FTSENT *p, *root;
+	register int nitems;
+	FTSENT *parent, *tmp;
+	int len;
+
+	/* Options check. */
+	if (options & ~FTS_OPTIONMASK) {
+		errno = EINVAL;
+		return (NULL);
+	}
+
+	/* Allocate/initialize the stream */
+	if ((sp = malloc((u_int)sizeof(FTS))) == NULL)
+		return (NULL);
+	memset(sp, 0, sizeof(FTS));
+	sp->fts_compar = compar;
+	sp->fts_options = options;
+
+	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
+	if (ISSET(FTS_LOGICAL))
+		SET(FTS_NOCHDIR);
+
+	/*
+	 * Start out with 1K of path space, and enough, in any case,
+	 * to hold the user's paths.
+	 */
+	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
+		goto mem1;
+
+	/* Allocate/initialize root's parent. */
+	if ((parent = fts_alloc(sp, "", 0)) == NULL)
+		goto mem2;
+	parent->fts_level = FTS_ROOTPARENTLEVEL;
+
+	/* Allocate/initialize root(s). */
+	for (root = NULL, nitems = 0; *argv; ++argv, ++nitems) {
+		/* Don't allow zero-length paths. */
+		if ((len = strlen(*argv)) == 0) {
+			errno = ENOENT;
+			goto mem3;
+		}
+
+		p = fts_alloc(sp, *argv, len);
+		p->fts_level = FTS_ROOTLEVEL;
+		p->fts_parent = parent;
+		p->fts_accpath = p->fts_name;
+		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW));
+
+		/* Command-line "." and ".." are real directories. */
+		if (p->fts_info == FTS_DOT)
+			p->fts_info = FTS_D;
+
+		/*
+		 * If comparison routine supplied, traverse in sorted
+		 * order; otherwise traverse in the order specified.
+		 */
+		if (compar) {
+			p->fts_link = root;
+			root = p;
+		} else {
+			p->fts_link = NULL;
+			if (root == NULL)
+				tmp = root = p;
+			else {
+				tmp->fts_link = p;
+				tmp = p;
+			}
+		}
+	}
+	if (compar && nitems > 1)
+		root = fts_sort(sp, root, nitems);
+
+	/*
+	 * Allocate a dummy pointer and make fts_read think that we've just
+	 * finished the node before the root(s); set p->fts_info to FTS_INIT
+	 * so that everything about the "current" node is ignored.
+	 */
+	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
+		goto mem3;
+	sp->fts_cur->fts_link = root;
+	sp->fts_cur->fts_info = FTS_INIT;
+
+	/*
+	 * If using chdir(2), grab a file descriptor pointing to dot to insure
+	 * that we can get back here; this could be avoided for some paths,
+	 * but almost certainly not worth the effort.  Slashes, symbolic links,
+	 * and ".." are all fairly nasty problems.  Note, if we can't get the
+	 * descriptor we run anyway, just more slowly.
+	 */
+	if (!ISSET(FTS_NOCHDIR) && (sp->fts_rfd = open(".", O_RDONLY, 0)) < 0)
+		SET(FTS_NOCHDIR);
+
+	return (sp);
+
+mem3:	fts_lfree(root);
+	free(parent);
+mem2:	free(sp->fts_path);
+mem1:	free(sp);
+	return (NULL);
+}
+
+static void
+fts_load(sp, p)
+	FTS *sp;
+	register FTSENT *p;
+{
+	register int len;
+	register char *cp;
+
+	/*
+	 * Load the stream structure for the next traversal.  Since we don't
+	 * actually enter the directory until after the preorder visit, set
+	 * the fts_accpath field specially so the chdir gets done to the right
+	 * place and the user can access the first node.  From fts_open it's
+	 * known that the path will fit.
+	 */
+	len = p->fts_pathlen = p->fts_namelen;
+	memmove(sp->fts_path, p->fts_name, len + 1);
+	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
+		len = strlen(++cp);
+		memmove(p->fts_name, cp, len + 1);
+		p->fts_namelen = len;
+	}
+	p->fts_accpath = p->fts_path = sp->fts_path;
+	sp->fts_dev = p->fts_dev;
+}
+
+int
+fts_close(sp)
+	FTS *sp;
+{
+	register FTSENT *freep, *p;
+	int saved_errno;
+
+	/*
+	 * This still works if we haven't read anything -- the dummy structure
+	 * points to the root list, so we step through to the end of the root
+	 * list which has a valid parent pointer.
+	 */
+	if (sp->fts_cur) {
+		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
+			freep = p;
+			p = p->fts_link ? p->fts_link : p->fts_parent;
+			free(freep);
+		}
+		free(p);
+	}
+
+	/* Free up child linked list, sort array, path buffer. */
+	if (sp->fts_child)
+		fts_lfree(sp->fts_child);
+	if (sp->fts_array)
+		free(sp->fts_array);
+	free(sp->fts_path);
+
+	/* Return to original directory, save errno if necessary. */
+	if (!ISSET(FTS_NOCHDIR)) {
+		saved_errno = fchdir(sp->fts_rfd) ? errno : 0;
+		(void)close(sp->fts_rfd);
+	}
+
+	/* Free up the stream pointer. */
+	free(sp);
+
+	/* Set errno and return. */
+	if (!ISSET(FTS_NOCHDIR) && saved_errno) {
+		errno = saved_errno;
+		return (-1);
+	}
+	return (0);
+}
+
+/*
+ * Special case a root of "/" so that slashes aren't appended which would
+ * cause paths to be written as "//foo".
+ */
+#define	NAPPEND(p)							\
+	(p->fts_level == FTS_ROOTLEVEL && p->fts_pathlen == 1 &&	\
+	    p->fts_path[0] == '/' ? 0 : p->fts_pathlen)
+
+FTSENT *
+fts_read(sp)
+	register FTS *sp;
+{
+	register FTSENT *p, *tmp;
+	register int instr;
+	register char *t;
+	int saved_errno;
+
+	/* If finished or unrecoverable error, return NULL. */
+	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
+		return (NULL);
+
+	/* Set current node pointer. */
+	p = sp->fts_cur;
+
+	/* Save and zero out user instructions. */
+	instr = p->fts_instr;
+	p->fts_instr = FTS_NOINSTR;
+
+	/* Any type of file may be re-visited; re-stat and re-turn. */
+	if (instr == FTS_AGAIN) {
+		p->fts_info = fts_stat(sp, p, 0);
+		return (p);
+	}
+
+	/*
+	 * Following a symlink -- SLNONE test allows application to see
+	 * SLNONE and recover.  If indirecting through a symlink, have
+	 * keep a pointer to current location.  If unable to get that
+	 * pointer, follow fails.
+	 */
+	if (instr == FTS_FOLLOW &&
+	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
+		p->fts_info = fts_stat(sp, p, 1);
+		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR))
+			if ((p->fts_symfd = open(".", O_RDONLY, 0)) < 0) {
+				p->fts_errno = errno;
+				p->fts_info = FTS_ERR;
+			} else
+				p->fts_flags |= FTS_SYMFOLLOW;
+		return (p);
+	}
+
+	/* Directory in pre-order. */
+	if (p->fts_info == FTS_D) {
+		/* If skipped or crossed mount point, do post-order visit. */
+		if (instr == FTS_SKIP ||
+		    ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev) {
+			if (p->fts_flags & FTS_SYMFOLLOW)
+				(void)close(p->fts_symfd);
+			if (sp->fts_child) {
+				fts_lfree(sp->fts_child);
+				sp->fts_child = NULL;
+			}
+			p->fts_info = FTS_DP;
+			return (p);
+		} 
+
+		/* Rebuild if only read the names and now traversing. */
+		if (sp->fts_child && sp->fts_options & FTS_NAMEONLY) {
+			sp->fts_options &= ~FTS_NAMEONLY;
+			fts_lfree(sp->fts_child);
+			sp->fts_child = NULL;
+		}
+
+		/*
+		 * Cd to the subdirectory.
+		 *
+		 * If have already read and now fail to chdir, whack the list
+		 * to make the names come out right, and set the parent errno
+		 * so the application will eventually get an error condition.
+		 * Set the FTS_DONTCHDIR flag so that when we logically change
+		 * directories back to the parent we don't do a chdir.
+		 *
+		 * If haven't read do so.  If the read fails, fts_build sets
+		 * FTS_STOP or the fts_info field of the node.
+		 */
+		if (sp->fts_child) {
+			if (CHDIR(sp, p->fts_accpath)) {
+				p->fts_errno = errno;
+				p->fts_flags |= FTS_DONTCHDIR;
+				for (p = sp->fts_child; p; p = p->fts_link)
+					p->fts_accpath =
+					    p->fts_parent->fts_accpath;
+			}
+		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
+			if (ISSET(FTS_STOP))
+				return (NULL);
+			return (p);
+		}
+		p = sp->fts_child;
+		sp->fts_child = NULL;
+		goto name;
+	}
+
+	/* Move to the next node on this level. */
+next:	tmp = p;
+	if (p = p->fts_link) {
+		free(tmp);
+
+		/*
+		 * If reached the top, return to the original directory, and
+		 * load the paths for the next root.
+		 */
+		if (p->fts_level == FTS_ROOTLEVEL) {
+			if (!ISSET(FTS_NOCHDIR) && FCHDIR(sp, sp->fts_rfd)) {
+				SET(FTS_STOP);
+				return (NULL);
+			}
+			fts_load(sp, p);
+			return (sp->fts_cur = p);
+		}
+
+		/*
+		 * User may have called fts_set on the node.  If skipped,
+		 * ignore.  If followed, get a file descriptor so we can
+		 * get back if necessary.
+		 */
+		if (p->fts_instr == FTS_SKIP)
+			goto next;
+		if (p->fts_instr == FTS_FOLLOW) {
+			p->fts_info = fts_stat(sp, p, 1);
+			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR))
+				if ((p->fts_symfd =
+				    open(".", O_RDONLY, 0)) < 0) {
+					p->fts_errno = errno;
+					p->fts_info = FTS_ERR;
+				} else
+					p->fts_flags |= FTS_SYMFOLLOW;
+			p->fts_instr = FTS_NOINSTR;
+		}
+
+name:		t = sp->fts_path + NAPPEND(p->fts_parent);
+		*t++ = '/';
+		memmove(t, p->fts_name, p->fts_namelen + 1);
+		return (sp->fts_cur = p);
+	}
+
+	/* Move up to the parent node. */
+	p = tmp->fts_parent;
+	free(tmp);
+
+	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
+		/*
+		 * Done; free everything up and set errno to 0 so the user
+		 * can distinguish between error and EOF.
+		 */
+		free(p);
+		errno = 0;
+		return (sp->fts_cur = NULL);
+	}
+
+	/* Nul terminate the pathname. */
+	sp->fts_path[p->fts_pathlen] = '\0';
+
+	/*
+	 * Return to the parent directory.  If at a root node or came through
+	 * a symlink, go back through the file descriptor.  Otherwise, cd up
+	 * one directory.
+	 */
+	if (p->fts_level == FTS_ROOTLEVEL) {
+		if (!ISSET(FTS_NOCHDIR) && FCHDIR(sp, sp->fts_rfd)) {
+			SET(FTS_STOP);
+			return (NULL);
+		}
+	} else if (p->fts_flags & FTS_SYMFOLLOW) {
+		if (FCHDIR(sp, p->fts_symfd)) {
+			saved_errno = errno;
+			(void)close(p->fts_symfd);
+			errno = saved_errno;
+			SET(FTS_STOP);
+			return (NULL);
+		}
+		(void)close(p->fts_symfd);
+	} else if (!(p->fts_flags & FTS_DONTCHDIR)) {
+		if (CHDIR(sp, "..")) {
+			SET(FTS_STOP);
+			return (NULL);
+		}
+	}
+	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
+	return (sp->fts_cur = p);
+}
+
+/*
+ * Fts_set takes the stream as an argument although it's not used in this
+ * implementation; it would be necessary if anyone wanted to add global
+ * semantics to fts using fts_set.  An error return is allowed for similar
+ * reasons.
+ */
+/* ARGSUSED */
+int
+fts_set(sp, p, instr)
+	FTS *sp;
+	FTSENT *p;
+	int instr;
+{
+	if (instr && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
+	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
+		errno = EINVAL;
+		return (1);
+	}
+	p->fts_instr = instr;
+	return (0);
+}
+
+FTSENT *
+fts_children(sp, instr)
+	register FTS *sp;
+	int instr;
+{
+	register FTSENT *p;
+	int fd;
+
+	if (instr && instr != FTS_NAMEONLY) {
+		errno = EINVAL;
+		return (NULL);
+	}
+
+	/* Set current node pointer. */
+	p = sp->fts_cur;
+
+	/*
+	 * Errno set to 0 so user can distinguish empty directory from
+	 * an error.
+	 */
+	errno = 0;
+
+	/* Fatal errors stop here. */
+	if (ISSET(FTS_STOP))
+		return (NULL);
+
+	/* Return logical hierarchy of user's arguments. */
+	if (p->fts_info == FTS_INIT)
+		return (p->fts_link);
+
+	/*
+	 * If not a directory being visited in pre-order, stop here.  Could
+	 * allow FTS_DNR, assuming the user has fixed the problem, but the
+	 * same effect is available with FTS_AGAIN.
+	 */
+	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
+		return (NULL);
+
+	/* Free up any previous child list. */
+	if (sp->fts_child)
+		fts_lfree(sp->fts_child);
+
+	if (instr == FTS_NAMEONLY) {
+		sp->fts_options |= FTS_NAMEONLY;
+		instr = BNAMES;
+	} else 
+		instr = BCHILD;
+
+	/*
+	 * If using chdir on a relative path and called BEFORE fts_read does
+	 * its chdir to the root of a traversal, we can lose -- we need to
+	 * chdir into the subdirectory, and we don't know where the current
+	 * directory is, so we can't get back so that the upcoming chdir by
+	 * fts_read will work.
+	 */
+	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
+	    ISSET(FTS_NOCHDIR))
+		return (sp->fts_child = fts_build(sp, instr));
+
+	if ((fd = open(".", O_RDONLY, 0)) < 0)
+		return (NULL);
+	sp->fts_child = fts_build(sp, instr);
+	if (fchdir(fd))
+		return (NULL);
+	(void)close(fd);
+	return (sp->fts_child);
+}
+
+/*
+ * This is the tricky part -- do not casually change *anything* in here.  The
+ * idea is to build the linked list of entries that are used by fts_children
+ * and fts_read.  There are lots of special cases.
+ *
+ * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
+ * set and it's a physical walk (so that symbolic links can't be directories),
+ * we can do things quickly.  First, if it's a 4.4BSD file system, the type
+ * of the file is in the directory entry.  Otherwise, we assume that the number
+ * of subdirectories in a node is equal to the number of links to the parent.
+ * The former skips all stat calls.  The latter skips stat calls in any leaf
+ * directories and for any files after the subdirectories in the directory have
+ * been found, cutting the stat calls by about 2/3.
+ */
+static FTSENT *
+fts_build(sp, type)
+	register FTS *sp;
+	int type;
+{
+	register struct dirent *dp;
+	register FTSENT *p, *head;
+	register int nitems;
+	FTSENT *cur, *tail;
+	DIR *dirp;
+	void *adjaddr;
+	int cderrno, descend, len, level, maxlen, nlinks, oflag, saved_errno;
+	char *cp;
+
+	/* Set current node pointer. */
+	cur = sp->fts_cur;
+
+	/*
+	 * Open the directory for reading.  If this fails, we're done.
+	 * If being called from fts_read, set the fts_info field.
+	 */
+#ifdef FTS_WHITEOUT
+	if (ISSET(FTS_WHITEOUT))
+		oflag = DTF_NODUP|DTF_REWIND;
+	else
+		oflag = DTF_HIDEW|DTF_NODUP|DTF_REWIND;
+#else
+#define __opendir2(path, flag) opendir(path)
+#endif
+	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
+		if (type == BREAD) {
+			cur->fts_info = FTS_DNR;
+			cur->fts_errno = errno;
+		}
+		return (NULL);
+	}
+
+	/*
+	 * Nlinks is the number of possible entries of type directory in the
+	 * directory if we're cheating on stat calls, 0 if we're not doing
+	 * any stat calls at all, -1 if we're doing stats on everything.
+	 */
+	if (type == BNAMES)
+		nlinks = 0;
+	else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL))
+		nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
+	else
+		nlinks = -1;
+
+#ifdef notdef
+	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
+	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
+	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
+#endif
+	/*
+	 * If we're going to need to stat anything or we want to descend
+	 * and stay in the directory, chdir.  If this fails we keep going,
+	 * but set a flag so we don't chdir after the post-order visit.
+	 * We won't be able to stat anything, but we can still return the
+	 * names themselves.  Note, that since fts_read won't be able to
+	 * chdir into the directory, it will have to return different path
+	 * names than before, i.e. "a/b" instead of "b".  Since the node
+	 * has already been visited in pre-order, have to wait until the
+	 * post-order visit to return the error.  There is a special case
+	 * here, if there was nothing to stat then it's not an error to
+	 * not be able to stat.  This is all fairly nasty.  If a program
+	 * needed sorted entries or stat information, they had better be
+	 * checking FTS_NS on the returned nodes.
+	 */
+	cderrno = 0;
+	if (nlinks || type == BREAD)
+		if (FCHDIR(sp, dirfd(dirp))) {
+			if (nlinks && type == BREAD)
+				cur->fts_errno = errno;
+			cur->fts_flags |= FTS_DONTCHDIR;
+			descend = 0;
+			cderrno = errno;
+		} else
+			descend = 1;
+	else
+		descend = 0;
+
+	/*
+	 * Figure out the max file name length that can be stored in the
+	 * current path -- the inner loop allocates more path as necessary.
+	 * We really wouldn't have to do the maxlen calculations here, we
+	 * could do them in fts_read before returning the path, but it's a
+	 * lot easier here since the length is part of the dirent structure.
+	 *
+	 * If not changing directories set a pointer so that can just append
+	 * each new name into the path.
+	 */
+	maxlen = sp->fts_pathlen - cur->fts_pathlen - 1;
+	len = NAPPEND(cur);
+	if (ISSET(FTS_NOCHDIR)) {
+		cp = sp->fts_path + len;
+		*cp++ = '/';
+	}
+
+	level = cur->fts_level + 1;
+
+	/* Read the directory, attaching each entry to the `link' pointer. */
+	adjaddr = NULL;
+	for (head = tail = NULL, nitems = 0; dp = readdir(dirp);) {
+		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
+			continue;
+
+		if ((p = fts_alloc(sp, dp->d_name, (int)dp->d_namlen)) == NULL)
+			goto mem1;
+		if (dp->d_namlen > maxlen) {
+			if (fts_palloc(sp, (size_t)dp->d_namlen)) {
+				/*
+				 * No more memory for path or structures.  Save
+				 * errno, free up the current structure and the
+				 * structures already allocated.
+				 */
+mem1:				saved_errno = errno;
+				if (p)
+					free(p);
+				fts_lfree(head);
+				(void)closedir(dirp);
+				errno = saved_errno;
+				cur->fts_info = FTS_ERR;
+				SET(FTS_STOP);
+				return (NULL);
+			}
+			adjaddr = sp->fts_path;
+			maxlen = sp->fts_pathlen - sp->fts_cur->fts_pathlen - 1;
+		}
+
+		p->fts_pathlen = len + dp->d_namlen + 1;
+		p->fts_parent = sp->fts_cur;
+		p->fts_level = level;
+
+#ifdef FTS_WHITEOUT
+		if (dp->d_type == DT_WHT)
+			p->fts_flags |= FTS_ISW;
+#endif
+
+		if (cderrno) {
+			if (nlinks) {
+				p->fts_info = FTS_NS;
+				p->fts_errno = cderrno;
+			} else
+				p->fts_info = FTS_NSOK;
+			p->fts_accpath = cur->fts_accpath;
+		} else if (nlinks == 0
+#ifdef DT_DIR
+		    || nlinks > 0 && 
+		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN
+#endif
+		    ) {
+			p->fts_accpath =
+			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
+			p->fts_info = FTS_NSOK;
+		} else {
+			/* Build a file name for fts_stat to stat. */
+			if (ISSET(FTS_NOCHDIR)) {
+				p->fts_accpath = p->fts_path;
+				memmove(cp, p->fts_name, p->fts_namelen + 1);
+			} else
+				p->fts_accpath = p->fts_name;
+			/* Stat it. */
+			p->fts_info = fts_stat(sp, p, 0);
+
+			/* Decrement link count if applicable. */
+			if (nlinks > 0 && (p->fts_info == FTS_D ||
+			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
+				--nlinks;
+		}
+
+		/* We walk in directory order so "ls -f" doesn't get upset. */
+		p->fts_link = NULL;
+		if (head == NULL)
+			head = tail = p;
+		else {
+			tail->fts_link = p;
+			tail = p;
+		}
+		++nitems;
+	}
+	(void)closedir(dirp);
+
+	/*
+	 * If had to realloc the path, adjust the addresses for the rest
+	 * of the tree.
+	 */
+	if (adjaddr)
+		fts_padjust(sp, adjaddr);
+
+	/*
+	 * If not changing directories, reset the path back to original
+	 * state.
+	 */
+	if (ISSET(FTS_NOCHDIR)) {
+		if (cp - 1 > sp->fts_path)
+			--cp;
+		*cp = '\0';
+	}
+
+	/*
+	 * If descended after called from fts_children or after called from
+	 * fts_read and nothing found, get back.  At the root level we use
+	 * the saved fd; if one of fts_open()'s arguments is a relative path
+	 * to an empty directory, we wind up here with no other way back.  If
+	 * can't get back, we're done.
+	 */
+	if (descend && (type == BCHILD || !nitems) &&
+	    (cur->fts_level == FTS_ROOTLEVEL ?
+	    FCHDIR(sp, sp->fts_rfd) : CHDIR(sp, ".."))) {
+		cur->fts_info = FTS_ERR;
+		SET(FTS_STOP);
+		return (NULL);
+	}
+
+	/* If didn't find anything, return NULL. */
+	if (!nitems) {
+		if (type == BREAD)
+			cur->fts_info = FTS_DP;
+		return (NULL);
+	}
+
+	/* Sort the entries. */
+	if (sp->fts_compar && nitems > 1)
+		head = fts_sort(sp, head, nitems);
+	return (head);
+}
+
+static u_short
+fts_stat(sp, p, follow)
+	FTS *sp;
+	register FTSENT *p;
+	int follow;
+{
+	register FTSENT *t;
+	register dev_t dev;
+	register ino_t ino;
+	struct stat *sbp, sb;
+	int saved_errno;
+
+	/* If user needs stat info, stat buffer already allocated. */
+	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
+
+#ifdef FTS_WHITEOUT
+	/* check for whiteout */
+	if (p->fts_flags & FTS_ISW) {
+		if (sbp != &sb) {
+			memset(sbp, '\0', sizeof (*sbp));
+			sbp->st_mode = S_IFWHT;
+		}
+		return (FTS_W);
+	}
+#endif
+	
+	/*
+	 * If doing a logical walk, or application requested FTS_FOLLOW, do
+	 * a stat(2).  If that fails, check for a non-existent symlink.  If
+	 * fail, set the errno from the stat call.
+	 */
+	if (ISSET(FTS_LOGICAL) || follow) {
+		if (stat(p->fts_accpath, sbp)) {
+			saved_errno = errno;
+			if (!lstat(p->fts_accpath, sbp)) {
+				errno = 0;
+				return (FTS_SLNONE);
+			} 
+			p->fts_errno = saved_errno;
+			goto err;
+		}
+	} else if (lstat(p->fts_accpath, sbp)) {
+		p->fts_errno = errno;
+err:		memset(sbp, 0, sizeof(struct stat));
+		return (FTS_NS);
+	}
+
+	if (S_ISDIR(sbp->st_mode)) {
+		/*
+		 * Set the device/inode.  Used to find cycles and check for
+		 * crossing mount points.  Also remember the link count, used
+		 * in fts_build to limit the number of stat calls.  It is
+		 * understood that these fields are only referenced if fts_info
+		 * is set to FTS_D.
+		 */
+		dev = p->fts_dev = sbp->st_dev;
+		ino = p->fts_ino = sbp->st_ino;
+		p->fts_nlink = sbp->st_nlink;
+
+		if (ISDOT(p->fts_name))
+			return (FTS_DOT);
+
+		/*
+		 * Cycle detection is done by brute force when the directory
+		 * is first encountered.  If the tree gets deep enough or the
+		 * number of symbolic links to directories is high enough,
+		 * something faster might be worthwhile.
+		 */
+		for (t = p->fts_parent;
+		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
+			if (ino == t->fts_ino && dev == t->fts_dev) {
+				p->fts_cycle = t;
+				return (FTS_DC);
+			}
+		return (FTS_D);
+	}
+	if (S_ISLNK(sbp->st_mode))
+		return (FTS_SL);
+	if (S_ISREG(sbp->st_mode))
+		return (FTS_F);
+	return (FTS_DEFAULT);
+}
+
+static FTSENT *
+fts_sort(sp, head, nitems)
+	FTS *sp;
+	FTSENT *head;
+	register int nitems;
+{
+	register FTSENT **ap, *p;
+
+	/*
+	 * Construct an array of pointers to the structures and call qsort(3).
+	 * Reassemble the array in the order returned by qsort.  If unable to
+	 * sort for memory reasons, return the directory entries in their
+	 * current order.  Allocate enough space for the current needs plus
+	 * 40 so don't realloc one entry at a time.
+	 */
+	if (nitems > sp->fts_nitems) {
+		sp->fts_nitems = nitems + 40;
+		if ((sp->fts_array = realloc(sp->fts_array,
+		    (size_t)(sp->fts_nitems * sizeof(FTSENT *)))) == NULL) {
+			sp->fts_nitems = 0;
+			return (head);
+		}
+	}
+	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
+		*ap++ = p;
+	qsort((void *)sp->fts_array, nitems, sizeof(FTSENT *), sp->fts_compar);
+	for (head = *(ap = sp->fts_array); --nitems; ++ap)
+		ap[0]->fts_link = ap[1];
+	ap[0]->fts_link = NULL;
+	return (head);
+}
+
+static FTSENT *
+fts_alloc(sp, name, namelen)
+	FTS *sp;
+	char *name;
+	register int namelen;
+{
+	register FTSENT *p;
+	size_t len;
+
+	/*
+	 * The file name is a variable length array and no stat structure is
+	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
+	 * structure, the file name and the stat structure in one chunk, but
+	 * be careful that the stat structure is reasonably aligned.  Since the
+	 * fts_name field is declared to be of size 1, the fts_name pointer is
+	 * namelen + 2 before the first possible address of the stat structure.
+	 */
+	len = sizeof(FTSENT) + namelen;
+	if (!ISSET(FTS_NOSTAT))
+		len += sizeof(struct stat) + ALIGNBYTES;
+	if ((p = malloc(len)) == NULL)
+		return (NULL);
+
+	/* Copy the name plus the trailing NULL. */
+	memmove(p->fts_name, name, namelen + 1);
+
+	if (!ISSET(FTS_NOSTAT))
+		p->fts_statp = (struct stat *)ALIGN(p->fts_name + namelen + 2);
+	p->fts_namelen = namelen;
+	p->fts_path = sp->fts_path;
+	p->fts_errno = 0;
+	p->fts_flags = 0;
+	p->fts_instr = FTS_NOINSTR;
+	p->fts_number = 0;
+	p->fts_pointer = NULL;
+	return (p);
+}
+
+static void
+fts_lfree(head)
+	register FTSENT *head;
+{
+	register FTSENT *p;
+
+	/* Free a linked list of structures. */
+	while (p = head) {
+		head = head->fts_link;
+		free(p);
+	}
+}
+
+/*
+ * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
+ * Most systems will allow creation of paths much longer than MAXPATHLEN, even
+ * though the kernel won't resolve them.  Add the size (not just what's needed)
+ * plus 256 bytes so don't realloc the path 2 bytes at a time. 
+ */
+static int
+fts_palloc(sp, more)
+	FTS *sp;
+	size_t more;
+{
+	sp->fts_pathlen += more + 256;
+	sp->fts_path = realloc(sp->fts_path, (size_t)sp->fts_pathlen);
+	return (sp->fts_path == NULL);
+}
+
+/*
+ * When the path is realloc'd, have to fix all of the pointers in structures
+ * already returned.
+ */
+static void
+fts_padjust(sp, addr)
+	FTS *sp;
+	void *addr;
+{
+	FTSENT *p;
+
+#define	ADJUST(p) {							\
+	(p)->fts_accpath =						\
+	    (char *)addr + ((p)->fts_accpath - (p)->fts_path);		\
+	(p)->fts_path = addr;						\
+}
+	/* Adjust the current set of children. */
+	for (p = sp->fts_child; p; p = p->fts_link)
+		ADJUST(p);
+
+	/* Adjust the rest of the tree. */
+	for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
+		ADJUST(p);
+		p = p->fts_link ? p->fts_link : p->fts_parent;
+	}
+}
+
+static size_t
+fts_maxarglen(argv)
+	char * const *argv;
+{
+	size_t len, max;
+
+	for (max = 0; *argv; ++argv)
+		if ((len = strlen(*argv)) > max)
+			max = len;
+	return (max);
+}
diff --git a/sys/miscfs/union/libc.opendir.c b/sys/miscfs/union/libc.opendir.c
index 99ed58b..c1c6f5b 100644
--- a/sys/miscfs/union/libc.opendir.c
+++ b/sys/miscfs/union/libc.opendir.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1983, 1993, 1994
+ * Copyright (c) 1983, 1993
  *	The Regents of the University of California.  All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -32,8 +32,7 @@
  */
 
 #if defined(LIBC_SCCS) && !defined(lint)
-static char orig_sccsid[] = "@(#)opendir.c	8.2 (Berkeley) 2/12/94";
-static char sccsid[] = "@(#)libc.opendir.c	8.1 (Berkeley) 2/15/94";
+static char sccsid[] = "@(#)opendir.c	8.6 (Berkeley) 8/14/94";
 #endif /* LIBC_SCCS and not lint */
 
 #include <sys/param.h>
@@ -45,18 +44,27 @@ static char sccsid[] = "@(#)libc.opendir.c	8.1 (Berkeley) 2/15/94";
 #include <unistd.h>
 
 /*
- * open a directory.
+ * Open a directory.
  */
 DIR *
 opendir(name)
 	const char *name;
 {
+
+	return (__opendir2(name, DTF_HIDEW|DTF_NODUP));
+}
+
+DIR *
+__opendir2(name, flags)
+	const char *name;
+	int flags;
+{
 	DIR *dirp;
 	int fd;
 	int incr;
-	struct statfs sfb;
+	int unionstack;
 
-	if ((fd = open(name, 0)) == -1)
+	if ((fd = open(name, O_RDONLY)) == -1)
 		return (NULL);
 	if (fcntl(fd, F_SETFD, FD_CLOEXEC) == -1 ||
 	    (dirp = (DIR *)malloc(sizeof(DIR))) == NULL) {
@@ -68,24 +76,30 @@ opendir(name)
 	 * If CLBYTES is an exact multiple of DIRBLKSIZ, use a CLBYTES
 	 * buffer that it cluster boundary aligned.
 	 * Hopefully this can be a big win someday by allowing page
-	 * trades trade to user space to be done by getdirentries()
+	 * trades to user space to be done by getdirentries()
 	 */
 	if ((CLBYTES % DIRBLKSIZ) == 0)
 		incr = CLBYTES;
 	else
 		incr = DIRBLKSIZ;
 
-#ifdef MOUNT_UNION
 	/*
 	 * Determine whether this directory is the top of a union stack.
 	 */
-	if (fstatfs(fd, &sfb) < 0) {
-		free(dirp);
-		close(fd);
-		return (NULL);
+	if (flags & DTF_NODUP) {
+		struct statfs sfb;
+
+		if (fstatfs(fd, &sfb) < 0) {
+			free(dirp);
+			close(fd);
+			return (NULL);
+		}
+		unionstack = (sfb.f_type == MOUNT_UNION);
+	} else {
+		unionstack = 0;
 	}
 
-	if (sfb.f_type == MOUNT_UNION) {
+	if (unionstack) {
 		int len = 0;
 		int space = 0;
 		char *buf = 0;
@@ -100,11 +114,6 @@ opendir(name)
 		 * number to zero.
 		 */
 
-		/*
-		 * Fixup dd_loc to be non-zero to fake out readdir
-		 */
-		dirp->dd_loc = sizeof(void *);
-
 		do {
 			/*
 			 * Always make at least DIRBLKSIZ bytes
@@ -119,7 +128,7 @@ opendir(name)
 					close(fd);
 					return (NULL);
 				}
-				ddptr = buf + (len - space) + dirp->dd_loc;
+				ddptr = buf + (len - space);
 			}
 
 			n = getdirentries(fd, ddptr, space, &dirp->dd_seek);
@@ -129,6 +138,24 @@ opendir(name)
 			}
 		} while (n > 0);
 
+		flags |= __DTF_READALL;
+
+		/*
+		 * Re-open the directory.
+		 * This has the effect of rewinding back to the
+		 * top of the union stack and is needed by
+		 * programs which plan to fchdir to a descriptor
+		 * which has also been read -- see fts.c.
+		 */
+		if (flags & DTF_REWIND) {
+			(void) close(fd);
+			if ((fd = open(name, O_RDONLY)) == -1) {
+				free(buf);
+				free(dirp);
+				return (NULL);
+			}
+		}
+
 		/*
 		 * There is now a buffer full of (possibly) duplicate
 		 * names.
@@ -144,7 +171,7 @@ opendir(name)
 		 */
 		for (dpv = 0;;) {
 			n = 0;
-			ddptr = buf + dirp->dd_loc;
+			ddptr = buf;
 			while (ddptr < buf + len) {
 				struct dirent *dp;
 
@@ -166,10 +193,9 @@ opendir(name)
 				struct dirent *xp;
 
 				/*
-				 * If and when whiteouts happen,
-				 * this sort would need to be stable.
+				 * This sort must be stable.
 				 */
-				heapsort(dpv, n, sizeof(*dpv), alphasort);
+				mergesort(dpv, n, sizeof(*dpv), alphasort);
 
 				dpv[n] = NULL;
 				xp = NULL;
@@ -183,9 +209,13 @@ opendir(name)
 					struct dirent *dp = dpv[n];
 
 					if ((xp == NULL) ||
-					    strcmp(dp->d_name, xp->d_name))
+					    strcmp(dp->d_name, xp->d_name)) {
 						xp = dp;
-					else
+					} else {
+						dp->d_fileno = 0;
+					}
+					if (dp->d_type == DT_WHT &&
+					    (flags & DTF_HIDEW))
 						dp->d_fileno = 0;
 				}
 
@@ -200,9 +230,7 @@ opendir(name)
 
 		dirp->dd_len = len;
 		dirp->dd_size = ddptr - dirp->dd_buf;
-	} else
-#endif /* MOUNT_UNION */
-	{
+	} else {
 		dirp->dd_len = incr;
 		dirp->dd_buf = malloc(dirp->dd_len);
 		if (dirp->dd_buf == NULL) {
@@ -211,10 +239,12 @@ opendir(name)
 			return (NULL);
 		}
 		dirp->dd_seek = 0;
-		dirp->dd_loc = 0;
+		flags &= ~DTF_REWIND;
 	}
 
+	dirp->dd_loc = 0;
 	dirp->dd_fd = fd;
+	dirp->dd_flags = flags;
 
 	/*
 	 * Set up seek point for rewinddir.
diff --git a/sys/miscfs/union/libc.readdir.c b/sys/miscfs/union/libc.readdir.c
new file mode 100644
index 0000000..2e97661
--- /dev/null
+++ b/sys/miscfs/union/libc.readdir.c
@@ -0,0 +1,75 @@
+/*
+ * Copyright (c) 1983, 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.
+ */
+
+#if defined(LIBC_SCCS) && !defined(lint)
+static char sccsid[] = "@(#)readdir.c	8.3 (Berkeley) 9/29/94";
+#endif /* LIBC_SCCS and not lint */
+
+#include <sys/param.h>
+#include <dirent.h>
+
+/*
+ * get next entry in a directory.
+ */
+struct dirent *
+readdir(dirp)
+	register DIR *dirp;
+{
+	register struct dirent *dp;
+
+	for (;;) {
+		if (dirp->dd_loc >= dirp->dd_size) {
+			if (dirp->dd_flags & __DTF_READALL)
+				return (NULL);
+			dirp->dd_loc = 0;
+		}
+		if (dirp->dd_loc == 0 && !(dirp->dd_flags & __DTF_READALL)) {
+			dirp->dd_size = getdirentries(dirp->dd_fd,
+			    dirp->dd_buf, dirp->dd_len, &dirp->dd_seek);
+			if (dirp->dd_size <= 0)
+				return (NULL);
+		}
+		dp = (struct dirent *)(dirp->dd_buf + dirp->dd_loc);
+		if ((int)dp & 03)	/* bogus pointer check */
+			return (NULL);
+		if (dp->d_reclen <= 0 ||
+		    dp->d_reclen > dirp->dd_len + 1 - dirp->dd_loc)
+			return (NULL);
+		dirp->dd_loc += dp->d_reclen;
+		if (dp->d_ino == 0)
+			continue;
+		if (dp->d_type == DT_WHT && (dirp->dd_flags & DTF_HIDEW))
+			continue;
+		return (dp);
+	}
+}
diff --git a/sys/miscfs/union/union.h b/sys/miscfs/union/union.h
index 463218a..2671985 100644
--- a/sys/miscfs/union/union.h
+++ b/sys/miscfs/union/union.h
@@ -34,7 +34,7 @@
  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  *
- *	@(#)union.h	8.2 (Berkeley) 2/17/94
+ *	@(#)union.h	8.9 (Berkeley) 12/10/94
  */
 
 struct union_args {
@@ -74,10 +74,14 @@ struct union_node {
 	struct vnode	        *un_uppervp;	/* overlaying object */
 	struct vnode	        *un_lowervp;	/* underlying object */
 	struct vnode		*un_dirvp;	/* Parent dir of uppervp */
+	struct vnode		*un_pvp;	/* Parent vnode */
 	char			*un_path;	/* saved component name */
 	int			un_hash;	/* saved un_path hash value */
 	int			un_openl;	/* # of opens on lowervp */
-	int			un_flags;
+	unsigned int		un_flags;
+	struct vnode		**un_dircache;	/* cached union stack */
+	off_t			un_uppersz;	/* size of upper object */
+	off_t			un_lowersz;	/* size of lower object */
 #ifdef DIAGNOSTIC
 	pid_t			un_pid;
 #endif
@@ -87,15 +91,22 @@ struct union_node {
 #define UN_LOCKED	0x02
 #define UN_ULOCK	0x04		/* Upper node is locked */
 #define UN_KLOCK	0x08		/* Keep upper node locked on vput */
+#define UN_CACHED	0x10		/* In union cache */
 
 extern int union_allocvp __P((struct vnode **, struct mount *,
 				struct vnode *, struct vnode *,
 				struct componentname *, struct vnode *,
-				struct vnode *));
-extern int union_copyfile __P((struct proc *, struct ucred *,
-				struct vnode *, struct vnode *));
+				struct vnode *, int));
+extern int union_copyfile __P((struct vnode *, struct vnode *,
+					struct ucred *, struct proc *));
+extern int union_copyup __P((struct union_node *, int, struct ucred *,
+				struct proc *));
+extern int union_dowhiteout __P((struct union_node *, struct ucred *,
+					struct proc *));
 extern int union_mkshadow __P((struct union_mount *, struct vnode *,
 				struct componentname *, struct vnode **));
+extern int union_mkwhiteout __P((struct union_mount *, struct vnode *,
+				struct componentname *, char *));
 extern int union_vn_create __P((struct vnode **, struct union_node *,
 				struct proc *));
 extern int union_cn_close __P((struct vnode *, int, struct ucred *,
@@ -104,6 +115,7 @@ extern void union_removed_upper __P((struct union_node *un));
 extern struct vnode *union_lowervp __P((struct vnode *));
 extern void union_newlower __P((struct union_node *, struct vnode *));
 extern void union_newupper __P((struct union_node *, struct vnode *));
+extern void union_newsize __P((struct vnode *, off_t, off_t));
 
 #define	MOUNTTOUNIONMOUNT(mp) ((struct union_mount *)((mp)->mnt_data))
 #define	VTOUNION(vp) ((struct union_node *)(vp)->v_data)
diff --git a/sys/miscfs/union/union_subr.c b/sys/miscfs/union/union_subr.c
index 77947d1..732a820 100644
--- a/sys/miscfs/union/union_subr.c
+++ b/sys/miscfs/union/union_subr.c
@@ -34,7 +34,7 @@
  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  *
- *	@(#)union_subr.c	8.4 (Berkeley) 2/17/94
+ *	@(#)union_subr.c	8.20 (Berkeley) 5/20/95
  */
 
 #include <sys/param.h>
@@ -47,6 +47,9 @@
 #include <sys/file.h>
 #include <sys/filedesc.h>
 #include <sys/queue.h>
+#include <sys/mount.h>
+#include <sys/stat.h>
+#include <vm/vm.h>		/* for vnode_pager_setsize */
 #include <miscfs/union/union.h>
 
 #ifdef DIAGNOSTIC
@@ -110,31 +113,38 @@ union_updatevp(un, uppervp, lowervp)
 {
 	int ohash = UNION_HASH(un->un_uppervp, un->un_lowervp);
 	int nhash = UNION_HASH(uppervp, lowervp);
+	int docache = (lowervp != NULLVP || uppervp != NULLVP);
+	int lhash, hhash, uhash;
 
-	if (ohash != nhash) {
-		/*
-		 * Ensure locking is ordered from lower to higher
-		 * to avoid deadlocks.
-		 */
-		if (nhash < ohash) {
-			int t = ohash;
-			ohash = nhash;
-			nhash = t;
-		}
+	/*
+	 * Ensure locking is ordered from lower to higher
+	 * to avoid deadlocks.
+	 */
+	if (nhash < ohash) {
+		lhash = nhash;
+		uhash = ohash;
+	} else {
+		lhash = ohash;
+		uhash = nhash;
+	}
 
-		while (union_list_lock(ohash))
+	if (lhash != uhash)
+		while (union_list_lock(lhash))
 			continue;
 
-		while (union_list_lock(nhash))
-			continue;
+	while (union_list_lock(uhash))
+		continue;
 
-		LIST_REMOVE(un, un_cache);
-		union_list_unlock(ohash);
-	} else {	
-		while (union_list_lock(nhash))
-			continue;
+	if (ohash != nhash || !docache) {
+		if (un->un_flags & UN_CACHED) {
+			un->un_flags &= ~UN_CACHED;
+			LIST_REMOVE(un, un_cache);
+		}
 	}
 
+	if (ohash != nhash)
+		union_list_unlock(ohash);
+
 	if (un->un_lowervp != lowervp) {
 		if (un->un_lowervp) {
 			vrele(un->un_lowervp);
@@ -148,6 +158,7 @@ union_updatevp(un, uppervp, lowervp)
 			}
 		}
 		un->un_lowervp = lowervp;
+		un->un_lowersz = VNOVAL;
 	}
 
 	if (un->un_uppervp != uppervp) {
@@ -155,10 +166,13 @@ union_updatevp(un, uppervp, lowervp)
 			vrele(un->un_uppervp);
 
 		un->un_uppervp = uppervp;
+		un->un_uppersz = VNOVAL;
 	}
 
-	if (ohash != nhash)
+	if (docache && (ohash != nhash)) {
 		LIST_INSERT_HEAD(&unhead[nhash], un, un_cache);
+		un->un_flags |= UN_CACHED;
+	}
 
 	union_list_unlock(nhash);
 }
@@ -182,6 +196,47 @@ union_newupper(un, uppervp)
 }
 
 /*
+ * Keep track of size changes in the underlying vnodes.
+ * If the size changes, then callback to the vm layer
+ * giving priority to the upper layer size.
+ */
+void
+union_newsize(vp, uppersz, lowersz)
+	struct vnode *vp;
+	off_t uppersz, lowersz;
+{
+	struct union_node *un;
+	off_t sz;
+
+	/* only interested in regular files */
+	if (vp->v_type != VREG)
+		return;
+
+	un = VTOUNION(vp);
+	sz = VNOVAL;
+
+	if ((uppersz != VNOVAL) && (un->un_uppersz != uppersz)) {
+		un->un_uppersz = uppersz;
+		if (sz == VNOVAL)
+			sz = un->un_uppersz;
+	}
+
+	if ((lowersz != VNOVAL) && (un->un_lowersz != lowersz)) {
+		un->un_lowersz = lowersz;
+		if (sz == VNOVAL)
+			sz = un->un_lowersz;
+	}
+
+	if (sz != VNOVAL) {
+#ifdef UNION_DIAGNOSTIC
+		printf("union: %s size now %ld\n",
+			uppersz != VNOVAL ? "upper" : "lower", (long) sz);
+#endif
+		vnode_pager_setsize(vp, sz);
+	}
+}
+
+/*
  * allocate a union_node/vnode pair.  the vnode is
  * referenced and locked.  the new vnode is returned
  * via (vpp).  (mp) is the mountpoint of the union filesystem,
@@ -213,20 +268,23 @@ union_newupper(un, uppervp)
  * the vnode free list.
  */
 int
-union_allocvp(vpp, mp, undvp, dvp, cnp, uppervp, lowervp)
+union_allocvp(vpp, mp, undvp, dvp, cnp, uppervp, lowervp, docache)
 	struct vnode **vpp;
 	struct mount *mp;
-	struct vnode *undvp;
+	struct vnode *undvp;		/* parent union vnode */
 	struct vnode *dvp;		/* may be null */
 	struct componentname *cnp;	/* may be null */
 	struct vnode *uppervp;		/* may be null */
 	struct vnode *lowervp;		/* may be null */
+	int docache;
 {
 	int error;
 	struct union_node *un;
 	struct union_node **pp;
 	struct vnode *xlowervp = NULLVP;
+	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
 	int hash;
+	int vflag;
 	int try;
 
 	if (uppervp == NULLVP && lowervp == NULLVP)
@@ -237,8 +295,22 @@ union_allocvp(vpp, mp, undvp, dvp, cnp, uppervp, lowervp)
 		lowervp = NULLVP;
 	}
 
+	/* detect the root vnode (and aliases) */
+	vflag = 0;
+	if ((uppervp == um->um_uppervp) &&
+	    ((lowervp == NULLVP) || lowervp == um->um_lowervp)) {
+		if (lowervp == NULLVP) {
+			lowervp = um->um_lowervp;
+			if (lowervp != NULLVP)
+				VREF(lowervp);
+		}
+		vflag = VROOT;
+	}
+
 loop:
-	for (try = 0; try < 3; try++) {
+	if (!docache) {
+		un = 0;
+	} else for (try = 0; try < 3; try++) {
 		switch (try) {
 		case 0:
 			if (lowervp == NULLVP)
@@ -269,7 +341,8 @@ loop:
 			    (un->un_uppervp == uppervp ||
 			     un->un_uppervp == NULLVP) &&
 			    (UNIONTOV(un)->v_mount == mp)) {
-				if (vget(UNIONTOV(un), 0)) {
+				if (vget(UNIONTOV(un), 0,
+				    cnp ? cnp->cn_proc : NULL)) {
 					union_list_unlock(hash);
 					goto loop;
 				}
@@ -352,8 +425,7 @@ loop:
 		 */
 		if (lowervp != un->un_lowervp) {
 			union_newlower(un, lowervp);
-			if (cnp && (lowervp != NULLVP) &&
-			    (lowervp->v_type == VREG)) {
+			if (cnp && (lowervp != NULLVP)) {
 				un->un_hash = cnp->cn_hash;
 				un->un_path = malloc(cnp->cn_namelen+1,
 						M_TEMP, M_WAITOK);
@@ -370,14 +442,16 @@ loop:
 		return (0);
 	}
 
-	/*
-	 * otherwise lock the vp list while we call getnewvnode
-	 * since that can block.
-	 */ 
-	hash = UNION_HASH(uppervp, lowervp);
+	if (docache) {
+		/*
+		 * otherwise lock the vp list while we call getnewvnode
+		 * since that can block.
+		 */ 
+		hash = UNION_HASH(uppervp, lowervp);
 
-	if (union_list_lock(hash))
-		goto loop;
+		if (union_list_lock(hash))
+			goto loop;
+	}
 
 	error = getnewvnode(VT_UNION, mp, union_vnodeop_p, vpp);
 	if (error) {
@@ -396,6 +470,7 @@ loop:
 	MALLOC((*vpp)->v_data, void *, sizeof(struct union_node),
 		M_TEMP, M_WAITOK);
 
+	(*vpp)->v_flag |= vflag;
 	if (uppervp)
 		(*vpp)->v_type = uppervp->v_type;
 	else
@@ -403,7 +478,13 @@ loop:
 	un = VTOUNION(*vpp);
 	un->un_vnode = *vpp;
 	un->un_uppervp = uppervp;
+	un->un_uppersz = VNOVAL;
 	un->un_lowervp = lowervp;
+	un->un_lowersz = VNOVAL;
+	un->un_pvp = undvp;
+	if (undvp != NULLVP)
+		VREF(undvp);
+	un->un_dircache = 0;
 	un->un_openl = 0;
 	un->un_flags = UN_LOCKED;
 	if (un->un_uppervp)
@@ -414,7 +495,7 @@ loop:
 	else
 		un->un_pid = -1;
 #endif
-	if (cnp && (lowervp != NULLVP) && (lowervp->v_type == VREG)) {
+	if (cnp && (lowervp != NULLVP)) {
 		un->un_hash = cnp->cn_hash;
 		un->un_path = malloc(cnp->cn_namelen+1, M_TEMP, M_WAITOK);
 		bcopy(cnp->cn_nameptr, un->un_path, cnp->cn_namelen);
@@ -427,13 +508,17 @@ loop:
 		un->un_dirvp = 0;
 	}
 
-	LIST_INSERT_HEAD(&unhead[hash], un, un_cache);
+	if (docache) {
+		LIST_INSERT_HEAD(&unhead[hash], un, un_cache);
+		un->un_flags |= UN_CACHED;
+	}
 
 	if (xlowervp)
 		vrele(xlowervp);
 
 out:
-	union_list_unlock(hash);
+	if (docache)
+		union_list_unlock(hash);
 
 	return (error);
 }
@@ -444,13 +529,18 @@ union_freevp(vp)
 {
 	struct union_node *un = VTOUNION(vp);
 
-	LIST_REMOVE(un, un_cache);
+	if (un->un_flags & UN_CACHED) {
+		un->un_flags &= ~UN_CACHED;
+		LIST_REMOVE(un, un_cache);
+	}
 
-	if (un->un_uppervp)
+	if (un->un_pvp != NULLVP)
+		vrele(un->un_pvp);
+	if (un->un_uppervp != NULLVP)
 		vrele(un->un_uppervp);
-	if (un->un_lowervp)
+	if (un->un_lowervp != NULLVP)
 		vrele(un->un_lowervp);
-	if (un->un_dirvp)
+	if (un->un_dirvp != NULLVP)
 		vrele(un->un_dirvp);
 	if (un->un_path)
 		free(un->un_path, M_TEMP);
@@ -467,11 +557,11 @@ union_freevp(vp)
  * and (tvp) are locked on entry and exit.
  */
 int
-union_copyfile(p, cred, fvp, tvp)
-	struct proc *p;
-	struct ucred *cred;
+union_copyfile(fvp, tvp, cred, p)
 	struct vnode *fvp;
 	struct vnode *tvp;
+	struct ucred *cred;
+	struct proc *p;
 {
 	char *buf;
 	struct uio uio;
@@ -490,12 +580,12 @@ union_copyfile(p, cred, fvp, tvp)
 	uio.uio_segflg = UIO_SYSSPACE;
 	uio.uio_offset = 0;
 
-	VOP_UNLOCK(fvp);				/* XXX */
-	LEASE_CHECK(fvp, p, cred, LEASE_READ);
-	VOP_LOCK(fvp);					/* XXX */
-	VOP_UNLOCK(tvp);				/* XXX */
-	LEASE_CHECK(tvp, p, cred, LEASE_WRITE);
-	VOP_LOCK(tvp);					/* XXX */
+	VOP_UNLOCK(fvp, 0, p);				/* XXX */
+	VOP_LEASE(fvp, p, cred, LEASE_READ);
+	vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY, p);	/* XXX */
+	VOP_UNLOCK(tvp, 0, p);				/* XXX */
+	VOP_LEASE(tvp, p, cred, LEASE_WRITE);
+	vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, p);	/* XXX */
 
 	buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK);
 
@@ -535,6 +625,123 @@ union_copyfile(p, cred, fvp, tvp)
 }
 
 /*
+ * (un) is assumed to be locked on entry and remains
+ * locked on exit.
+ */
+int
+union_copyup(un, docopy, cred, p)
+	struct union_node *un;
+	int docopy;
+	struct ucred *cred;
+	struct proc *p;
+{
+	int error;
+	struct vnode *lvp, *uvp;
+
+	error = union_vn_create(&uvp, un, p);
+	if (error)
+		return (error);
+
+	/* at this point, uppervp is locked */
+	union_newupper(un, uvp);
+	un->un_flags |= UN_ULOCK;
+
+	lvp = un->un_lowervp;
+
+	if (docopy) {
+		/*
+		 * XX - should not ignore errors
+		 * from VOP_CLOSE
+		 */
+		vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY, p);
+		error = VOP_OPEN(lvp, FREAD, cred, p);
+		if (error == 0) {
+			error = union_copyfile(lvp, uvp, cred, p);
+			VOP_UNLOCK(lvp, 0, p);
+			(void) VOP_CLOSE(lvp, FREAD, cred, p);
+		}
+#ifdef UNION_DIAGNOSTIC
+		if (error == 0)
+			uprintf("union: copied up %s\n", un->un_path);
+#endif
+
+	}
+	un->un_flags &= ~UN_ULOCK;
+	VOP_UNLOCK(uvp, 0, p);
+	union_vn_close(uvp, FWRITE, cred, p);
+	vn_lock(uvp, LK_EXCLUSIVE | LK_RETRY, p);
+	un->un_flags |= UN_ULOCK;
+
+	/*
+	 * Subsequent IOs will go to the top layer, so
+	 * call close on the lower vnode and open on the
+	 * upper vnode to ensure that the filesystem keeps
+	 * its references counts right.  This doesn't do
+	 * the right thing with (cred) and (FREAD) though.
+	 * Ignoring error returns is not right, either.
+	 */
+	if (error == 0) {
+		int i;
+
+		for (i = 0; i < un->un_openl; i++) {
+			(void) VOP_CLOSE(lvp, FREAD, cred, p);
+			(void) VOP_OPEN(uvp, FREAD, cred, p);
+		}
+		un->un_openl = 0;
+	}
+
+	return (error);
+
+}
+
+static int
+union_relookup(um, dvp, vpp, cnp, cn, path, pathlen)
+	struct union_mount *um;
+	struct vnode *dvp;
+	struct vnode **vpp;
+	struct componentname *cnp;
+	struct componentname *cn;
+	char *path;
+	int pathlen;
+{
+	int error;
+
+	/*
+	 * A new componentname structure must be faked up because
+	 * there is no way to know where the upper level cnp came
+	 * from or what it is being used for.  This must duplicate
+	 * some of the work done by NDINIT, some of the work done
+	 * by namei, some of the work done by lookup and some of
+	 * the work done by VOP_LOOKUP when given a CREATE flag.
+	 * Conclusion: Horrible.
+	 *
+	 * The pathname buffer will be FREEed by VOP_MKDIR.
+	 */
+	cn->cn_namelen = pathlen;
+	cn->cn_pnbuf = malloc(cn->cn_namelen+1, M_NAMEI, M_WAITOK);
+	bcopy(path, cn->cn_pnbuf, cn->cn_namelen);
+	cn->cn_pnbuf[cn->cn_namelen] = '\0';
+
+	cn->cn_nameiop = CREATE;
+	cn->cn_flags = (LOCKPARENT|HASBUF|SAVENAME|SAVESTART|ISLASTCN);
+	cn->cn_proc = cnp->cn_proc;
+	if (um->um_op == UNMNT_ABOVE)
+		cn->cn_cred = cnp->cn_cred;
+	else
+		cn->cn_cred = um->um_cred;
+	cn->cn_nameptr = cn->cn_pnbuf;
+	cn->cn_hash = cnp->cn_hash;
+	cn->cn_consume = cnp->cn_consume;
+
+	VREF(dvp);
+	error = relookup(dvp, vpp, cn);
+	if (!error)
+		vrele(dvp);
+
+	return (error);
+}
+
+/*
  * Create a shadow directory in the upper layer.
  * The new vnode is returned locked.
  *
@@ -558,6 +765,19 @@ union_mkshadow(um, dvp, cnp, vpp)
 	struct proc *p = cnp->cn_proc;
 	struct componentname cn;
 
+	error = union_relookup(um, dvp, vpp, cnp, &cn,
+			cnp->cn_nameptr, cnp->cn_namelen);
+	if (error)
+		return (error);
+
+	if (*vpp) {
+		VOP_ABORTOP(dvp, &cn);
+		VOP_UNLOCK(dvp, 0, p);
+		vrele(*vpp);
+		*vpp = NULLVP;
+		return (EEXIST);
+	}
+
 	/*
 	 * policy: when creating the shadow directory in the
 	 * upper layer, create it owned by the user who did
@@ -566,54 +786,62 @@ union_mkshadow(um, dvp, cnp, vpp)
 	 * mkdir syscall).  (jsp, kb)
 	 */
 
-	/*
-	 * A new componentname structure must be faked up because
-	 * there is no way to know where the upper level cnp came
-	 * from or what it is being used for.  This must duplicate
-	 * some of the work done by NDINIT, some of the work done
-	 * by namei, some of the work done by lookup and some of
-	 * the work done by VOP_LOOKUP when given a CREATE flag.
-	 * Conclusion: Horrible.
-	 *
-	 * The pathname buffer will be FREEed by VOP_MKDIR.
-	 */
-	cn.cn_pnbuf = malloc(cnp->cn_namelen+1, M_NAMEI, M_WAITOK);
-	bcopy(cnp->cn_nameptr, cn.cn_pnbuf, cnp->cn_namelen);
-	cn.cn_pnbuf[cnp->cn_namelen] = '\0';
+	VATTR_NULL(&va);
+	va.va_type = VDIR;
+	va.va_mode = um->um_cmode;
 
-	cn.cn_nameiop = CREATE;
-	cn.cn_flags = (LOCKPARENT|HASBUF|SAVENAME|SAVESTART|ISLASTCN);
-	cn.cn_proc = cnp->cn_proc;
-	if (um->um_op == UNMNT_ABOVE)
-		cn.cn_cred = cnp->cn_cred;
-	else
-		cn.cn_cred = um->um_cred;
-	cn.cn_nameptr = cn.cn_pnbuf;
-	cn.cn_namelen = cnp->cn_namelen;
-	cn.cn_hash = cnp->cn_hash;
-	cn.cn_consume = cnp->cn_consume;
+	/* VOP_LEASE: dvp is locked */
+	VOP_LEASE(dvp, p, cn.cn_cred, LEASE_WRITE);
 
-	VREF(dvp);
-	if (error = relookup(dvp, vpp, &cn))
+	error = VOP_MKDIR(dvp, vpp, &cn, &va);
+	return (error);
+}
+
+/*
+ * Create a whiteout entry in the upper layer.
+ *
+ * (um) points to the union mount structure for access to the
+ * the mounting process's credentials.
+ * (dvp) is the directory in which to create the whiteout.
+ * it is locked on entry and exit.
+ * (cnp) is the componentname to be created.
+ */
+int
+union_mkwhiteout(um, dvp, cnp, path)
+	struct union_mount *um;
+	struct vnode *dvp;
+	struct componentname *cnp;
+	char *path;
+{
+	int error;
+	struct vattr va;
+	struct proc *p = cnp->cn_proc;
+	struct vnode *wvp;
+	struct componentname cn;
+
+	VOP_UNLOCK(dvp, 0, p);
+	error = union_relookup(um, dvp, &wvp, cnp, &cn, path, strlen(path));
+	if (error) {
+		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, p);
 		return (error);
-	vrele(dvp);
+	}
 
-	if (*vpp) {
+	if (wvp) {
 		VOP_ABORTOP(dvp, &cn);
-		VOP_UNLOCK(dvp);
-		vrele(*vpp);
-		*vpp = NULLVP;
+		vrele(dvp);
+		vrele(wvp);
 		return (EEXIST);
 	}
 
-	VATTR_NULL(&va);
-	va.va_type = VDIR;
-	va.va_mode = um->um_cmode;
+	/* VOP_LEASE: dvp is locked */
+	VOP_LEASE(dvp, p, p->p_ucred, LEASE_WRITE);
 
-	/* LEASE_CHECK: dvp is locked */
-	LEASE_CHECK(dvp, p, p->p_ucred, LEASE_WRITE);
+	error = VOP_WHITEOUT(dvp, &cn, CREATE);
+	if (error)
+		VOP_ABORTOP(dvp, &cn);
+
+	vrele(dvp);
 
-	error = VOP_MKDIR(dvp, vpp, &cn, &va);
 	return (error);
 }
 
@@ -691,7 +919,7 @@ union_vn_create(vpp, un, p)
 	VATTR_NULL(vap);
 	vap->va_type = VREG;
 	vap->va_mode = cmode;
-	LEASE_CHECK(un->un_dirvp, p, cred, LEASE_WRITE);
+	VOP_LEASE(un->un_dirvp, p, cred, LEASE_WRITE);
 	if (error = VOP_CREATE(un->un_dirvp, &vp, &cn, vap))
 		return (error);
 
@@ -712,33 +940,147 @@ union_vn_close(vp, fmode, cred, p)
 	struct ucred *cred;
 	struct proc *p;
 {
+
 	if (fmode & FWRITE)
 		--vp->v_writecount;
-	return (VOP_CLOSE(vp, fmode));
+	return (VOP_CLOSE(vp, fmode, cred, p));
 }
 
 void
 union_removed_upper(un)
 	struct union_node *un;
 {
+	struct proc *p = curproc;	/* XXX */
+
+	union_newupper(un, NULLVP);
+	if (un->un_flags & UN_CACHED) {
+		un->un_flags &= ~UN_CACHED;
+		LIST_REMOVE(un, un_cache);
+	}
+
 	if (un->un_flags & UN_ULOCK) {
 		un->un_flags &= ~UN_ULOCK;
-		VOP_UNLOCK(un->un_uppervp);
+		VOP_UNLOCK(un->un_uppervp, 0, p);
 	}
-
-	union_newupper(un, NULLVP);
 }
 
+#if 0
 struct vnode *
 union_lowervp(vp)
 	struct vnode *vp;
 {
 	struct union_node *un = VTOUNION(vp);
 
-	if (un->un_lowervp && (vp->v_type == un->un_lowervp->v_type)) {
-		if (vget(un->un_lowervp, 0))
-			return (NULLVP);
+	if ((un->un_lowervp != NULLVP) &&
+	    (vp->v_type == un->un_lowervp->v_type)) {
+		if (vget(un->un_lowervp, 0) == 0)
+			return (un->un_lowervp);
 	}
 
-	return (un->un_lowervp);
+	return (NULLVP);
+}
+#endif
+
+/*
+ * determine whether a whiteout is needed
+ * during a remove/rmdir operation.
+ */
+int
+union_dowhiteout(un, cred, p)
+	struct union_node *un;
+	struct ucred *cred;
+	struct proc *p;
+{
+	struct vattr va;
+
+	if (un->un_lowervp != NULLVP)
+		return (1);
+
+	if (VOP_GETATTR(un->un_uppervp, &va, cred, p) == 0 &&
+	    (va.va_flags & OPAQUE))
+		return (1);
+
+	return (0);
+}
+
+static void
+union_dircache_r(vp, vppp, cntp)
+	struct vnode *vp;
+	struct vnode ***vppp;
+	int *cntp;
+{
+	struct union_node *un;
+
+	if (vp->v_op != union_vnodeop_p) {
+		if (vppp) {
+			VREF(vp);
+			*(*vppp)++ = vp;
+			if (--(*cntp) == 0)
+				panic("union: dircache table too small");
+		} else {
+			(*cntp)++;
+		}
+
+		return;
+	}
+
+	un = VTOUNION(vp);
+	if (un->un_uppervp != NULLVP)
+		union_dircache_r(un->un_uppervp, vppp, cntp);
+	if (un->un_lowervp != NULLVP)
+		union_dircache_r(un->un_lowervp, vppp, cntp);
+}
+
+struct vnode *
+union_dircache(vp, p)
+	struct vnode *vp;
+	struct proc *p;
+{
+	int cnt;
+	struct vnode *nvp;
+	struct vnode **vpp;
+	struct vnode **dircache;
+	struct union_node *un;
+	int error;
+
+	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
+	dircache = VTOUNION(vp)->un_dircache;
+
+	nvp = NULLVP;
+
+	if (dircache == 0) {
+		cnt = 0;
+		union_dircache_r(vp, 0, &cnt);
+		cnt++;
+		dircache = (struct vnode **)
+				malloc(cnt * sizeof(struct vnode *),
+					M_TEMP, M_WAITOK);
+		vpp = dircache;
+		union_dircache_r(vp, &vpp, &cnt);
+		*vpp = NULLVP;
+		vpp = dircache + 1;
+	} else {
+		vpp = dircache;
+		do {
+			if (*vpp++ == VTOUNION(vp)->un_uppervp)
+				break;
+		} while (*vpp != NULLVP);
+	}
+
+	if (*vpp == NULLVP)
+		goto out;
+
+	vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY, p);
+	VREF(*vpp);
+	error = union_allocvp(&nvp, vp->v_mount, NULLVP, NULLVP, 0, *vpp, NULLVP, 0);
+	if (error)
+		goto out;
+
+	VTOUNION(vp)->un_dircache = 0;
+	un = VTOUNION(nvp);
+	un->un_dircache = dircache;
+
+out:
+	VOP_UNLOCK(vp, 0, p);
+	return (nvp);
 }
diff --git a/sys/miscfs/union/union_vfsops.c b/sys/miscfs/union/union_vfsops.c
index 9fa2746..19a2f27 100644
--- a/sys/miscfs/union/union_vfsops.c
+++ b/sys/miscfs/union/union_vfsops.c
@@ -1,6 +1,6 @@
 /*
- * Copyright (c) 1994 The Regents of the University of California.
- * Copyright (c) 1994 Jan-Simon Pendry.
+ * Copyright (c) 1994, 1995 The Regents of the University of California.
+ * Copyright (c) 1994, 1995 Jan-Simon Pendry.
  * All rights reserved.
  *
  * This code is derived from software donated to Berkeley by
@@ -34,7 +34,7 @@
  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  *
- *	@(#)union_vfsops.c	8.7 (Berkeley) 3/5/94
+ *	@(#)union_vfsops.c	8.20 (Berkeley) 5/20/95
  */
 
 /*
@@ -69,7 +69,7 @@ union_mount(mp, path, data, ndp, p)
 	struct union_args args;
 	struct vnode *lowerrootvp = NULLVP;
 	struct vnode *upperrootvp = NULLVP;
-	struct union_mount *um;
+	struct union_mount *um = 0;
 	struct ucred *cred = 0;
 	struct ucred *scred;
 	struct vattr va;
@@ -95,34 +95,6 @@ union_mount(mp, path, data, ndp, p)
 	}
 
 	/*
-	 * Take a copy of the process's credentials.  This isn't
-	 * quite right since the euid will always be zero and we
-	 * want to get the "real" users credentials.  So fix up
-	 * the uid field after taking the copy.
-	 */
-	cred = crdup(p->p_ucred);
-	cred->cr_uid = p->p_cred->p_ruid;
-
-	/*
-	 * Ensure the *real* user has write permission on the
-	 * mounted-on directory.  This allows the mount_union
-	 * command to be made setuid root so allowing anyone
-	 * to do union mounts onto any directory on which they
-	 * have write permission and which they also own.
-	 */
-	error = VOP_GETATTR(mp->mnt_vnodecovered, &va, cred, p);
-	if (error)
-		goto bad;
-	if ((va.va_uid != cred->cr_uid) && 
-	    (cred->cr_uid != 0)) {
-		error = EACCES;
-		goto bad;
-	}
-	error = VOP_ACCESS(mp->mnt_vnodecovered, VWRITE, cred, p);
-	if (error)
-		goto bad;
-
-	/*
 	 * Get argument
 	 */
 	if (error = copyin(data, (caddr_t)&args, sizeof(struct union_args)))
@@ -132,18 +104,10 @@ union_mount(mp, path, data, ndp, p)
 	VREF(lowerrootvp);
 
 	/*
-	 * Find upper node.  Use the real process credentials,
-	 * not the effective ones since this will have come
-	 * through a setuid process (mount_union).  All this
-	 * messing around with permissions is entirely bogus
-	 * and should be removed by allowing any user straight
-	 * past the mount system call.
+	 * Find upper node.
 	 */
-	scred = p->p_ucred;
-	p->p_ucred = cred;
 	NDINIT(ndp, LOOKUP, FOLLOW|WANTPARENT,
 	       UIO_USERSPACE, args.target, p);
-	p->p_ucred = scred;
 
 	if (error = namei(ndp))
 		goto bad;
@@ -193,7 +157,18 @@ union_mount(mp, path, data, ndp, p)
 		goto bad;
 	}
 
-	um->um_cred = cred;
+	/*
+	 * Unless the mount is readonly, ensure that the top layer
+	 * supports whiteout operations
+	 */
+	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
+		error = VOP_WHITEOUT(um->um_uppervp, (struct componentname *) 0, LOOKUP);
+		if (error)
+			goto bad;
+	}
+
+	um->um_cred = p->p_ucred;
+	crhold(um->um_cred);
 	um->um_cmode = UN_DIRMODE &~ p->p_fd->fd_cmask;
 
 	/*
@@ -221,24 +196,18 @@ union_mount(mp, path, data, ndp, p)
 	 */
 	mp->mnt_flag |= (um->um_uppervp->v_mount->mnt_flag & MNT_RDONLY);
 
-	/*
-	 * This is a user mount.  Privilege check for unmount
-	 * will be done in union_unmount.
-	 */
-	mp->mnt_flag |= MNT_USER;
-
 	mp->mnt_data = (qaddr_t) um;
-	getnewfsid(mp, MOUNT_UNION);
+	vfs_getnewfsid(mp);
 
 	(void) copyinstr(path, mp->mnt_stat.f_mntonname, MNAMELEN - 1, &size);
 	bzero(mp->mnt_stat.f_mntonname + size, MNAMELEN - size);
 
 	switch (um->um_op) {
 	case UNMNT_ABOVE:
-		cp = "<above>";
+		cp = "<above>:";
 		break;
 	case UNMNT_BELOW:
-		cp = "<below>";
+		cp = "<below>:";
 		break;
 	case UNMNT_REPLACE:
 		cp = "";
@@ -260,6 +229,8 @@ union_mount(mp, path, data, ndp, p)
 	return (0);
 
 bad:
+	if (um)
+		free(um, M_UFSMNT);
 	if (cred)
 		crfree(cred);
 	if (upperrootvp)
@@ -296,38 +267,54 @@ union_unmount(mp, mntflags, p)
 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
 	struct vnode *um_rootvp;
 	int error;
+	int freeing;
 	int flags = 0;
-	extern int doforce;
 
 #ifdef UNION_DIAGNOSTIC
 	printf("union_unmount(mp = %x)\n", mp);
 #endif
 
-	/* only the mounter, or superuser can unmount */
-	if ((p->p_cred->p_ruid != um->um_cred->cr_uid) &&
-	    (error = suser(p->p_ucred, &p->p_acflag)))
-		return (error);
-
-	if (mntflags & MNT_FORCE) {
-		/* union can never be rootfs so don't check for it */
-		if (!doforce)
-			return (EINVAL);
+	if (mntflags & MNT_FORCE)
 		flags |= FORCECLOSE;
-	}
 
 	if (error = union_root(mp, &um_rootvp))
 		return (error);
+
+	/*
+	 * Keep flushing vnodes from the mount list.
+	 * This is needed because of the un_pvp held
+	 * reference to the parent vnode.
+	 * If more vnodes have been freed on a given pass,
+	 * the try again.  The loop will iterate at most
+	 * (d) times, where (d) is the maximum tree depth
+	 * in the filesystem.
+	 */
+	for (freeing = 0; vflush(mp, um_rootvp, flags) != 0;) {
+		struct vnode *vp;
+		int n;
+
+		/* count #vnodes held on mount list */
+		for (n = 0, vp = mp->mnt_vnodelist.lh_first;
+				vp != NULLVP;
+				vp = vp->v_mntvnodes.le_next)
+			n++;
+
+		/* if this is unchanged then stop */
+		if (n == freeing)
+			break;
+
+		/* otherwise try once more time */
+		freeing = n;
+	}
+
+	/* At this point the root vnode should have a single reference */
 	if (um_rootvp->v_usecount > 1) {
 		vput(um_rootvp);
 		return (EBUSY);
 	}
-	if (error = vflush(mp, um_rootvp, flags)) {
-		vput(um_rootvp);
-		return (error);
-	}
 
 #ifdef UNION_DIAGNOSTIC
-	vprint("alias root of lower", um_rootvp);
+	vprint("union root", um_rootvp);
 #endif	 
 	/*
 	 * Discard references to upper and lower target vnodes.
@@ -357,16 +344,11 @@ union_root(mp, vpp)
 	struct mount *mp;
 	struct vnode **vpp;
 {
+	struct proc *p = curproc;	/* XXX */
 	struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
 	int error;
 	int loselock;
 
-#ifdef UNION_DIAGNOSTIC
-	printf("union_root(mp = %x, lvp = %x, uvp = %x)\n", mp,
-			um->um_lowervp,
-			um->um_uppervp);
-#endif
-
 	/*
 	 * Return locked reference to root.
 	 */
@@ -375,7 +357,7 @@ union_root(mp, vpp)
 	     VOP_ISLOCKED(um->um_uppervp)) {
 		loselock = 1;
 	} else {
-		VOP_LOCK(um->um_uppervp);
+		vn_lock(um->um_uppervp, LK_EXCLUSIVE | LK_RETRY, p);
 		loselock = 0;
 	}
 	if (um->um_lowervp)
@@ -385,16 +367,17 @@ union_root(mp, vpp)
 			      (struct vnode *) 0,
 			      (struct componentname *) 0,
 			      um->um_uppervp,
-			      um->um_lowervp);
+			      um->um_lowervp,
+			      1);
 
 	if (error) {
-		if (!loselock)
-			VOP_UNLOCK(um->um_uppervp);
-		vrele(um->um_uppervp);
+		if (loselock)
+			vrele(um->um_uppervp);
+		else
+			vput(um->um_uppervp);
 		if (um->um_lowervp)
 			vrele(um->um_lowervp);
 	} else {
-		(*vpp)->v_flag |= VROOT;
 		if (loselock)
 			VTOUNION(*vpp)->un_flags &= ~UN_ULOCK;
 	}
@@ -403,18 +386,6 @@ union_root(mp, vpp)
 }
 
 int
-union_quotactl(mp, cmd, uid, arg, p)
-	struct mount *mp;
-	int cmd;
-	uid_t uid;
-	caddr_t arg;
-	struct proc *p;
-{
-
-	return (EOPNOTSUPP);
-}
-
-int
 union_statfs(mp, sbp, p)
 	struct mount *mp;
 	struct statfs *sbp;
@@ -457,7 +428,6 @@ union_statfs(mp, sbp, p)
 	if (error)
 		return (error);
 
-	sbp->f_type = MOUNT_UNION;
 	sbp->f_flags = mstat.f_flags;
 	sbp->f_bsize = mstat.f_bsize;
 	sbp->f_iosize = mstat.f_iosize;
@@ -468,18 +438,23 @@ union_statfs(mp, sbp, p)
 	 * kind of sense.  none of this makes sense though.
 	 */
 
-	if (mstat.f_bsize != lbsize) {
+	if (mstat.f_bsize != lbsize)
 		sbp->f_blocks = sbp->f_blocks * lbsize / mstat.f_bsize;
-		sbp->f_bfree = sbp->f_bfree * lbsize / mstat.f_bsize;
-		sbp->f_bavail = sbp->f_bavail * lbsize / mstat.f_bsize;
-	}
+
+	/*
+	 * The "total" fields count total resources in all layers,
+	 * the "free" fields count only those resources which are
+	 * free in the upper layer (since only the upper layer
+	 * is writeable).
+	 */
 	sbp->f_blocks += mstat.f_blocks;
-	sbp->f_bfree += mstat.f_bfree;
-	sbp->f_bavail += mstat.f_bavail;
+	sbp->f_bfree = mstat.f_bfree;
+	sbp->f_bavail = mstat.f_bavail;
 	sbp->f_files += mstat.f_files;
-	sbp->f_ffree += mstat.f_ffree;
+	sbp->f_ffree = mstat.f_ffree;
 
 	if (sbp != &mp->mnt_stat) {
+		sbp->f_type = mp->mnt_vfc->vfc_typenum;
 		bcopy(&mp->mnt_stat.f_fsid, &sbp->f_fsid, sizeof(sbp->f_fsid));
 		bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
 		bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
@@ -487,53 +462,22 @@ union_statfs(mp, sbp, p)
 	return (0);
 }
 
-int
-union_sync(mp, waitfor, cred, p)
-	struct mount *mp;
-	int waitfor;
-	struct ucred *cred;
-	struct proc *p;
-{
-
-	/*
-	 * XXX - Assumes no data cached at union layer.
-	 */
-	return (0);
-}
-
-int
-union_vget(mp, ino, vpp)
-	struct mount *mp;
-	ino_t ino;
-	struct vnode **vpp;
-{
-	
-	return (EOPNOTSUPP);
-}
-
-int
-union_fhtovp(mp, fidp, nam, vpp, exflagsp, credanonp)
-	struct mount *mp;
-	struct fid *fidp;
-	struct mbuf *nam;
-	struct vnode **vpp;
-	int *exflagsp;
-	struct ucred **credanonp;
-{
-
-	return (EOPNOTSUPP);
-}
-
-int
-union_vptofh(vp, fhp)
-	struct vnode *vp;
-	struct fid *fhp;
-{
-
-	return (EOPNOTSUPP);
-}
-
-int union_init __P((void));
+/*
+ * XXX - Assumes no data cached at union layer.
+ */
+#define union_sync ((int (*) __P((struct mount *, int, struct ucred *, \
+	    struct proc *)))nullop)
+
+#define union_fhtovp ((int (*) __P((struct mount *, struct fid *, \
+	    struct mbuf *, struct vnode **, int *, struct ucred **)))eopnotsupp)
+int union_init __P((struct vfsconf *));
+#define union_quotactl ((int (*) __P((struct mount *, int, uid_t, caddr_t, \
+	    struct proc *)))eopnotsupp)
+#define union_sysctl ((int (*) __P((int *, u_int, void *, size_t *, void *, \
+	    size_t, struct proc *)))eopnotsupp)
+#define union_vget ((int (*) __P((struct mount *, ino_t, struct vnode **))) \
+	    eopnotsupp)
+#define union_vptofh ((int (*) __P((struct vnode *, struct fid *)))eopnotsupp)
 
 struct vfsops union_vfsops = {
 	union_mount,
@@ -547,4 +491,5 @@ struct vfsops union_vfsops = {
 	union_fhtovp,
 	union_vptofh,
 	union_init,
+	union_sysctl,
 };
diff --git a/sys/miscfs/union/union_vnops.c b/sys/miscfs/union/union_vnops.c
index 96327b0..69fe12a 100644
--- a/sys/miscfs/union/union_vnops.c
+++ b/sys/miscfs/union/union_vnops.c
@@ -1,7 +1,7 @@
 /*
- * Copyright (c) 1992, 1993, 1994 The Regents of the University of California.
- * Copyright (c) 1992, 1993, 1994 Jan-Simon Pendry.
- * All rights reserved.
+ * Copyright (c) 1992, 1993, 1994, 1995 Jan-Simon Pendry.
+ * Copyright (c) 1992, 1993, 1994, 1995
+ *	The Regents of the University of California.  All rights reserved.
  *
  * This code is derived from software contributed to Berkeley by
  * Jan-Simon Pendry.
@@ -34,7 +34,7 @@
  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  *
- *	@(#)union_vnops.c	8.6 (Berkeley) 2/17/94
+ *	@(#)union_vnops.c	8.32 (Berkeley) 6/23/95
  */
 
 #include <sys/param.h>
@@ -42,6 +42,7 @@
 #include <sys/proc.h>
 #include <sys/file.h>
 #include <sys/time.h>
+#include <sys/stat.h>
 #include <sys/types.h>
 #include <sys/vnode.h>
 #include <sys/mount.h>
@@ -49,34 +50,40 @@
 #include <sys/malloc.h>
 #include <sys/buf.h>
 #include <sys/queue.h>
+#include <sys/lock.h>
 #include <miscfs/union/union.h>
 
-#define FIXUP(un) { \
+#define FIXUP(un, p) { \
 	if (((un)->un_flags & UN_ULOCK) == 0) { \
-		union_fixup(un); \
+		union_fixup(un, p); \
 	} \
 }
 
 static void
-union_fixup(un)
+union_fixup(un, p)
 	struct union_node *un;
+	struct proc *p;
 {
 
-	VOP_LOCK(un->un_uppervp);
+	vn_lock(un->un_uppervp, LK_EXCLUSIVE | LK_RETRY, p);
 	un->un_flags |= UN_ULOCK;
 }
 
 static int
-union_lookup1(udvp, dvp, vpp, cnp)
+union_lookup1(udvp, dvpp, vpp, cnp)
 	struct vnode *udvp;
-	struct vnode *dvp;
+	struct vnode **dvpp;
 	struct vnode **vpp;
 	struct componentname *cnp;
 {
 	int error;
+	struct proc *p = cnp->cn_proc;
 	struct vnode *tdvp;
+	struct vnode *dvp;
 	struct mount *mp;
 
+	dvp = *dvpp;
+
 	/*
 	 * If stepping up the directory tree, check for going
 	 * back across the mount point, in which case do what
@@ -84,21 +91,18 @@ union_lookup1(udvp, dvp, vpp, cnp)
 	 * hierarchy.
 	 */
 	if (cnp->cn_flags & ISDOTDOT) {
-		for (;;) {
+		while ((dvp != udvp) && (dvp->v_flag & VROOT)) {
 			/*
 			 * Don't do the NOCROSSMOUNT check
 			 * at this level.  By definition,
 			 * union fs deals with namespaces, not
 			 * filesystems.
 			 */
-			if ((dvp->v_flag & VROOT) == 0)
-				break;
-
 			tdvp = dvp;
-			dvp = dvp->v_mount->mnt_vnodecovered;
+			*dvpp = dvp = dvp->v_mount->mnt_vnodecovered;
 			vput(tdvp);
 			VREF(dvp);
-			VOP_LOCK(dvp);
+			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, p);
 		}
 	}
 
@@ -112,7 +116,7 @@ union_lookup1(udvp, dvp, vpp, cnp)
 	 * here to allow it to be unlocked again (phew) in union_lookup.
 	 */
 	if (dvp != tdvp && !(cnp->cn_flags & ISLASTCN))
-		VOP_LOCK(dvp);
+		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, p);
 
 	dvp = tdvp;
 
@@ -124,13 +128,12 @@ union_lookup1(udvp, dvp, vpp, cnp)
 	while (dvp != udvp && (dvp->v_type == VDIR) &&
 	       (mp = dvp->v_mountedhere)) {
 
-		if (mp->mnt_flag & MNT_MLOCK) {
-			mp->mnt_flag |= MNT_MWAIT;
-			sleep((caddr_t) mp, PVFS);
+		if (vfs_busy(mp, 0, 0, p))
 			continue;
-		}
 
-		if (error = VFS_ROOT(mp, &tdvp)) {
+		error = VFS_ROOT(mp, &tdvp);
+		vfs_unbusy(mp, p);
+		if (error) {
 			vput(dvp);
 			return (error);
 		}
@@ -159,10 +162,29 @@ union_lookup(ap)
 	struct vnode *dvp = ap->a_dvp;
 	struct union_node *dun = VTOUNION(dvp);
 	struct componentname *cnp = ap->a_cnp;
+	struct proc *p = cnp->cn_proc;
 	int lockparent = cnp->cn_flags & LOCKPARENT;
 	int rdonly = cnp->cn_flags & RDONLY;
 	struct union_mount *um = MOUNTTOUNIONMOUNT(dvp->v_mount);
 	struct ucred *saved_cred;
+	int iswhiteout;
+	struct vattr va;
+
+#ifdef notyet
+	if (cnp->cn_namelen == 3 &&
+			cnp->cn_nameptr[2] == '.' &&
+			cnp->cn_nameptr[1] == '.' &&
+			cnp->cn_nameptr[0] == '.') {
+		dvp = *ap->a_vpp = LOWERVP(ap->a_dvp);
+		if (dvp == NULLVP)
+			return (ENOENT);
+		VREF(dvp);
+		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, p);
+		if (!lockparent || !(cnp->cn_flags & ISLASTCN))
+			VOP_UNLOCK(ap->a_dvp, 0, p);
+		return (0);
+	}
+#endif
 
 	cnp->cn_flags |= LOCKPARENT;
 
@@ -170,6 +192,7 @@ union_lookup(ap)
 	lowerdvp = dun->un_lowervp;
 	uppervp = NULLVP;
 	lowervp = NULLVP;
+	iswhiteout = 0;
 
 	/*
 	 * do the lookup in the upper level.
@@ -177,9 +200,9 @@ union_lookup(ap)
 	 * then assume that something special is going
 	 * on and just return that vnode.
 	 */
-	if (upperdvp) {
-		FIXUP(dun);
-		uerror = union_lookup1(um->um_uppervp, upperdvp,
+	if (upperdvp != NULLVP) {
+		FIXUP(dun, p);
+		uerror = union_lookup1(um->um_uppervp, &upperdvp,
 					&uppervp, cnp);
 		/*if (uppervp == upperdvp)
 			dun->un_flags |= UN_KLOCK;*/
@@ -190,6 +213,16 @@ union_lookup(ap)
 				cnp->cn_flags &= ~LOCKPARENT;
 			return (uerror);
 		}
+		if (uerror == ENOENT || uerror == EJUSTRETURN) {
+			if (cnp->cn_flags & ISWHITEOUT) {
+				iswhiteout = 1;
+			} else if (lowerdvp != NULLVP) {
+				lerror = VOP_GETATTR(upperdvp, &va,
+					cnp->cn_cred, cnp->cn_proc);
+				if (lerror == 0 && (va.va_flags & OPAQUE))
+					iswhiteout = 1;
+			}
+		}
 	} else {
 		uerror = ENOENT;
 	}
@@ -201,10 +234,10 @@ union_lookup(ap)
 	 * back from the upper layer and return the lower vnode
 	 * instead.
 	 */
-	if (lowerdvp) {
+	if (lowerdvp != NULLVP && !iswhiteout) {
 		int nameiop;
 
-		VOP_LOCK(lowerdvp);
+		vn_lock(lowerdvp, LK_EXCLUSIVE | LK_RETRY, p);
 
 		/*
 		 * Only do a LOOKUP on the bottom node, since
@@ -216,17 +249,17 @@ union_lookup(ap)
 			saved_cred = cnp->cn_cred;
 			cnp->cn_cred = um->um_cred;
 		}
-		lerror = union_lookup1(um->um_lowervp, lowerdvp,
+		lerror = union_lookup1(um->um_lowervp, &lowerdvp,
 				&lowervp, cnp);
 		if (um->um_op == UNMNT_BELOW)
 			cnp->cn_cred = saved_cred;
 		cnp->cn_nameiop = nameiop;
 
 		if (lowervp != lowerdvp)
-			VOP_UNLOCK(lowerdvp);
+			VOP_UNLOCK(lowerdvp, 0, p);
 
 		if (cnp->cn_consume != 0) {
-			if (uppervp) {
+			if (uppervp != NULLVP) {
 				if (uppervp == upperdvp)
 					vrele(uppervp);
 				else
@@ -240,6 +273,14 @@ union_lookup(ap)
 		}
 	} else {
 		lerror = ENOENT;
+		if ((cnp->cn_flags & ISDOTDOT) && dun->un_pvp != NULLVP) {
+			lowervp = LOWERVP(dun->un_pvp);
+			if (lowervp != NULLVP) {
+				VREF(lowervp);
+				vn_lock(lowervp, LK_EXCLUSIVE | LK_RETRY, p);
+				lerror = 0;
+			}
+		}
 	}
 
 	if (!lockparent)
@@ -280,13 +321,13 @@ union_lookup(ap)
 	if (uerror != 0 /* && (lerror == 0) */ ) {
 		if (lowervp->v_type == VDIR) { /* case 2b. */
 			dun->un_flags &= ~UN_ULOCK;
-			VOP_UNLOCK(upperdvp);
+			VOP_UNLOCK(upperdvp, 0, p);
 			uerror = union_mkshadow(um, upperdvp, cnp, &uppervp);
-			VOP_LOCK(upperdvp);
+			vn_lock(upperdvp, LK_EXCLUSIVE | LK_RETRY, p);
 			dun->un_flags |= UN_ULOCK;
 
 			if (uerror) {
-				if (lowervp) {
+				if (lowervp != NULLVP) {
 					vput(lowervp);
 					lowervp = NULLVP;
 				}
@@ -295,21 +336,21 @@ union_lookup(ap)
 		}
 	}
 
-	if (lowervp)
-		VOP_UNLOCK(lowervp);
+	if (lowervp != NULLVP)
+		VOP_UNLOCK(lowervp, 0, p);
 
 	error = union_allocvp(ap->a_vpp, dvp->v_mount, dvp, upperdvp, cnp,
-			      uppervp, lowervp);
+			      uppervp, lowervp, 1);
 
 	if (error) {
-		if (uppervp)
+		if (uppervp != NULLVP)
 			vput(uppervp);
-		if (lowervp)
+		if (lowervp != NULLVP)
 			vrele(lowervp);
 	} else {
 		if (*ap->a_vpp != dvp)
 			if (!lockparent || !(cnp->cn_flags & ISLASTCN))
-				VOP_UNLOCK(dvp);
+				VOP_UNLOCK(dvp, 0, p);
 	}
 
 	return (error);
@@ -326,28 +367,26 @@ union_create(ap)
 {
 	struct union_node *un = VTOUNION(ap->a_dvp);
 	struct vnode *dvp = un->un_uppervp;
+	struct componentname *cnp = ap->a_cnp;
+	struct proc *p = cnp->cn_proc;
 
-	if (dvp) {
+	if (dvp != NULLVP) {
 		int error;
 		struct vnode *vp;
+		struct mount *mp;
 
-		FIXUP(un);
+		FIXUP(un, p);
 
 		VREF(dvp);
 		un->un_flags |= UN_KLOCK;
+		mp = ap->a_dvp->v_mount;
 		vput(ap->a_dvp);
-		error = VOP_CREATE(dvp, &vp, ap->a_cnp, ap->a_vap);
+		error = VOP_CREATE(dvp, &vp, cnp, ap->a_vap);
 		if (error)
 			return (error);
 
-		error = union_allocvp(
-				ap->a_vpp,
-				ap->a_dvp->v_mount,
-				ap->a_dvp,
-				NULLVP,
-				ap->a_cnp,
-				vp,
-				NULLVP);
+		error = union_allocvp(ap->a_vpp, mp, NULLVP, NULLVP, cnp, vp,
+				NULLVP, 1);
 		if (error)
 			vput(vp);
 		return (error);
@@ -358,6 +397,25 @@ union_create(ap)
 }
 
 int
+union_whiteout(ap)
+	struct vop_whiteout_args /* {
+		struct vnode *a_dvp;
+		struct componentname *a_cnp;
+		int a_flags;
+	} */ *ap;
+{
+	struct union_node *un = VTOUNION(ap->a_dvp);
+	struct componentname *cnp = ap->a_cnp;
+	struct proc *p = cnp->cn_proc;
+
+	if (un->un_uppervp == NULLVP)
+		return (EOPNOTSUPP);
+
+	FIXUP(un, p);
+	return (VOP_WHITEOUT(un->un_uppervp, cnp, ap->a_flags));
+}
+
+int
 union_mknod(ap)
 	struct vop_mknod_args /* {
 		struct vnode *a_dvp;
@@ -368,29 +426,27 @@ union_mknod(ap)
 {
 	struct union_node *un = VTOUNION(ap->a_dvp);
 	struct vnode *dvp = un->un_uppervp;
+	struct componentname *cnp = ap->a_cnp;
+	struct proc *p = cnp->cn_proc;
 
-	if (dvp) {
+	if (dvp != NULLVP) {
 		int error;
 		struct vnode *vp;
+		struct mount *mp;
 
-		FIXUP(un);
+		FIXUP(un, p);
 
 		VREF(dvp);
 		un->un_flags |= UN_KLOCK;
+		mp = ap->a_dvp->v_mount;
 		vput(ap->a_dvp);
-		error = VOP_MKNOD(dvp, &vp, ap->a_cnp, ap->a_vap);
+		error = VOP_MKNOD(dvp, &vp, cnp, ap->a_vap);
 		if (error)
 			return (error);
 
-		if (vp) {
-			error = union_allocvp(
-					ap->a_vpp,
-					ap->a_dvp->v_mount,
-					ap->a_dvp,
-					NULLVP,
-					ap->a_cnp,
-					vp,
-					NULLVP);
+		if (vp != NULLVP) {
+			error = union_allocvp(ap->a_vpp, mp, NULLVP, NULLVP,
+					cnp, vp, NULLVP, 1);
 			if (error)
 				vput(vp);
 		}
@@ -430,77 +486,7 @@ union_open(ap)
 		 */
 		tvp = un->un_lowervp;
 		if ((ap->a_mode & FWRITE) && (tvp->v_type == VREG)) {
-			struct vnode *vp;
-			int i;
-
-			/*
-			 * Open the named file in the upper layer.  Note that
-			 * the file may have come into existence *since* the
-			 * lookup was done, since the upper layer may really
-			 * be a loopback mount of some other filesystem...
-			 * so open the file with exclusive create and barf if
-			 * it already exists.
-			 * XXX - perhaps should re-lookup the node (once more
-			 * with feeling) and simply open that.  Who knows.
-			 */
-			error = union_vn_create(&vp, un, p);
-			if (error)
-				return (error);
-
-			/* at this point, uppervp is locked */
-			union_newupper(un, vp);
-			un->un_flags |= UN_ULOCK;
-
-			/*
-			 * Now, if the file is being opened with truncation,
-			 * then the (new) upper vnode is ready to fly,
-			 * otherwise the data from the lower vnode must be
-			 * copied to the upper layer first.  This only works
-			 * for regular files (check is made above).
-			 */
-			if ((mode & O_TRUNC) == 0) {
-				/*
-				 * XXX - should not ignore errors
-				 * from VOP_CLOSE
-				 */
-				VOP_LOCK(tvp);
-				error = VOP_OPEN(tvp, FREAD, cred, p);
-				if (error == 0) {
-					error = union_copyfile(p, cred,
-						       tvp, un->un_uppervp);
-					VOP_UNLOCK(tvp);
-					(void) VOP_CLOSE(tvp, FREAD);
-				} else {
-					VOP_UNLOCK(tvp);
-				}
-
-#ifdef UNION_DIAGNOSTIC
-				if (!error)
-					uprintf("union: copied up %s\n",
-								un->un_path);
-#endif
-			}
-
-			un->un_flags &= ~UN_ULOCK;
-			VOP_UNLOCK(un->un_uppervp);
-			union_vn_close(un->un_uppervp, FWRITE, cred, p);
-			VOP_LOCK(un->un_uppervp);
-			un->un_flags |= UN_ULOCK;
-
-			/*
-			 * Subsequent IOs will go to the top layer, so
-			 * call close on the lower vnode and open on the
-			 * upper vnode to ensure that the filesystem keeps
-			 * its references counts right.  This doesn't do
-			 * the right thing with (cred) and (FREAD) though.
-			 * Ignoring error returns is not righ, either.
-			 */
-			for (i = 0; i < un->un_openl; i++) {
-				(void) VOP_CLOSE(tvp, FREAD);
-				(void) VOP_OPEN(un->un_uppervp, FREAD, cred, p);
-			}
-			un->un_openl = 0;
-
+			error = union_copyup(un, (mode&O_TRUNC) == 0, cred, p);
 			if (error == 0)
 				error = VOP_OPEN(un->un_uppervp, mode, cred, p);
 			return (error);
@@ -510,14 +496,14 @@ union_open(ap)
 		 * Just open the lower vnode
 		 */
 		un->un_openl++;
-		VOP_LOCK(tvp);
+		vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, p);
 		error = VOP_OPEN(tvp, mode, cred, p);
-		VOP_UNLOCK(tvp);
+		VOP_UNLOCK(tvp, 0, p);
 
 		return (error);
 	}
 
-	FIXUP(un);
+	FIXUP(un, p);
 
 	error = VOP_OPEN(tvp, mode, cred, p);
 
@@ -536,9 +522,7 @@ union_close(ap)
 	struct union_node *un = VTOUNION(ap->a_vp);
 	struct vnode *vp;
 
-	if (un->un_uppervp) {
-		vp = un->un_uppervp;
-	} else {
+	if ((vp = un->un_uppervp) == NULLVP) {
 #ifdef UNION_DIAGNOSTIC
 		if (un->un_openl <= 0)
 			panic("union: un_openl cnt");
@@ -547,7 +531,8 @@ union_close(ap)
 		vp = un->un_lowervp;
 	}
 
-	return (VOP_CLOSE(vp, ap->a_fflag, ap->a_cred, ap->a_p));
+	ap->a_vp = vp;
+	return (VCALL(vp, VOFFSET(vop_close), ap));
 }
 
 /*
@@ -569,25 +554,29 @@ union_access(ap)
 	} */ *ap;
 {
 	struct union_node *un = VTOUNION(ap->a_vp);
+	struct proc *p = ap->a_p;
 	int error = EACCES;
 	struct vnode *vp;
 
-	if (vp = un->un_uppervp) {
-		FIXUP(un);
-		return (VOP_ACCESS(vp, ap->a_mode, ap->a_cred, ap->a_p));
+	if ((vp = un->un_uppervp) != NULLVP) {
+		FIXUP(un, p);
+		ap->a_vp = vp;
+		return (VCALL(vp, VOFFSET(vop_access), ap));
 	}
 
-	if (vp = un->un_lowervp) {
-		VOP_LOCK(vp);
-		error = VOP_ACCESS(vp, ap->a_mode, ap->a_cred, ap->a_p);
+	if ((vp = un->un_lowervp) != NULLVP) {
+		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
+		ap->a_vp = vp;
+		error = VCALL(vp, VOFFSET(vop_access), ap);
 		if (error == 0) {
 			struct union_mount *um = MOUNTTOUNIONMOUNT(vp->v_mount);
 
-			if (um->um_op == UNMNT_BELOW)
-				error = VOP_ACCESS(vp, ap->a_mode,
-						um->um_cred, ap->a_p);
+			if (um->um_op == UNMNT_BELOW) {
+				ap->a_cred = um->um_cred;
+				error = VCALL(vp, VOFFSET(vop_access), ap);
+			}
 		}
-		VOP_UNLOCK(vp);
+		VOP_UNLOCK(vp, 0, p);
 		if (error)
 			return (error);
 	}
@@ -596,7 +585,8 @@ union_access(ap)
 }
 
 /*
- *  We handle getattr only to change the fsid.
+ * We handle getattr only to change the fsid and
+ * track object sizes
  */
 int
 union_getattr(ap)
@@ -610,6 +600,7 @@ union_getattr(ap)
 	int error;
 	struct union_node *un = VTOUNION(ap->a_vp);
 	struct vnode *vp = un->un_uppervp;
+	struct proc *p = ap->a_p;
 	struct vattr *vap;
 	struct vattr va;
 
@@ -627,10 +618,21 @@ union_getattr(ap)
 
 	vp = un->un_uppervp;
 	if (vp != NULLVP) {
-		FIXUP(un);
+		/*
+		 * It's not clear whether VOP_GETATTR is to be
+		 * called with the vnode locked or not.  stat() calls
+		 * it with (vp) locked, and fstat calls it with
+		 * (vp) unlocked.
+		 * In the mean time, compensate here by checking
+		 * the union_node's lock flag.
+		 */
+		if (un->un_flags & UN_LOCKED)
+			FIXUP(un, p);
+
 		error = VOP_GETATTR(vp, vap, ap->a_cred, ap->a_p);
 		if (error)
 			return (error);
+		union_newsize(ap->a_vp, vap->va_size, VNOVAL);
 	}
 
 	if (vp == NULLVP) {
@@ -643,17 +645,16 @@ union_getattr(ap)
 	}
 
 	if (vp != NULLVP) {
-		VOP_LOCK(vp);
 		error = VOP_GETATTR(vp, vap, ap->a_cred, ap->a_p);
-		VOP_UNLOCK(vp);
 		if (error)
 			return (error);
+		union_newsize(ap->a_vp, VNOVAL, vap->va_size);
 	}
 
 	if ((vap != ap->a_vap) && (vap->va_type == VDIR))
 		ap->a_vap->va_nlink += vap->va_nlink;
 
-	vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
+	ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
 	return (0);
 }
 
@@ -667,6 +668,7 @@ union_setattr(ap)
 	} */ *ap;
 {
 	struct union_node *un = VTOUNION(ap->a_vp);
+	struct proc *p = ap->a_p;
 	int error;
 
 	/*
@@ -676,21 +678,11 @@ union_setattr(ap)
 	 */
 	if ((un->un_uppervp == NULLVP) &&
 	    /* assert(un->un_lowervp != NULLVP) */
-	    (un->un_lowervp->v_type == VREG) &&
-	    (ap->a_vap->va_size == 0)) {
-		struct vnode *vp;
-
-		error = union_vn_create(&vp, un, ap->a_p);
+	    (un->un_lowervp->v_type == VREG)) {
+		error = union_copyup(un, (ap->a_vap->va_size != 0),
+						ap->a_cred, ap->a_p);
 		if (error)
 			return (error);
-
-		/* at this point, uppervp is locked */
-		union_newupper(un, vp);
-
-		VOP_UNLOCK(vp);
-		union_vn_close(un->un_uppervp, FWRITE, ap->a_cred, ap->a_p);
-		VOP_LOCK(vp);
-		un->un_flags |= UN_ULOCK;
 	}
 
 	/*
@@ -698,9 +690,11 @@ union_setattr(ap)
 	 * otherwise return read-only filesystem error.
 	 */
 	if (un->un_uppervp != NULLVP) {
-		FIXUP(un);
+		FIXUP(un, p);
 		error = VOP_SETATTR(un->un_uppervp, ap->a_vap,
 					ap->a_cred, ap->a_p);
+		if ((error == 0) && (ap->a_vap->va_size != VNOVAL))
+			union_newsize(ap->a_vp, ap->a_vap->va_size, VNOVAL);
 	} else {
 		error = EROFS;
 	}
@@ -718,16 +712,36 @@ union_read(ap)
 	} */ *ap;
 {
 	int error;
+	struct proc *p = ap->a_uio->uio_procp;
 	struct vnode *vp = OTHERVP(ap->a_vp);
 	int dolock = (vp == LOWERVP(ap->a_vp));
 
 	if (dolock)
-		VOP_LOCK(vp);
+		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
 	else
-		FIXUP(VTOUNION(ap->a_vp));
+		FIXUP(VTOUNION(ap->a_vp), p);
 	error = VOP_READ(vp, ap->a_uio, ap->a_ioflag, ap->a_cred);
 	if (dolock)
-		VOP_UNLOCK(vp);
+		VOP_UNLOCK(vp, 0, p);
+
+	/*
+	 * XXX
+	 * perhaps the size of the underlying object has changed under
+	 * our feet.  take advantage of the offset information present
+	 * in the uio structure.
+	 */
+	if (error == 0) {
+		struct union_node *un = VTOUNION(ap->a_vp);
+		off_t cur = ap->a_uio->uio_offset;
+
+		if (vp == un->un_uppervp) {
+			if (cur > un->un_uppersz)
+				union_newsize(ap->a_vp, cur, VNOVAL);
+		} else {
+			if (cur > un->un_lowersz)
+				union_newsize(ap->a_vp, VNOVAL, cur);
+		}
+	}
 
 	return (error);
 }
@@ -742,20 +756,45 @@ union_write(ap)
 	} */ *ap;
 {
 	int error;
-	struct vnode *vp = OTHERVP(ap->a_vp);
-	int dolock = (vp == LOWERVP(ap->a_vp));
+	struct vnode *vp;
+	struct union_node *un = VTOUNION(ap->a_vp);
+	struct proc *p = ap->a_uio->uio_procp;
 
-	if (dolock)
-		VOP_LOCK(vp);
-	else
-		FIXUP(VTOUNION(ap->a_vp));
+	vp = UPPERVP(ap->a_vp);
+	if (vp == NULLVP)
+		panic("union: missing upper layer in write");
+
+	FIXUP(un, p);
 	error = VOP_WRITE(vp, ap->a_uio, ap->a_ioflag, ap->a_cred);
-	if (dolock)
-		VOP_UNLOCK(vp);
+
+	/*
+	 * the size of the underlying object may be changed by the
+	 * write.
+	 */
+	if (error == 0) {
+		off_t cur = ap->a_uio->uio_offset;
+
+		if (cur > un->un_uppersz)
+			union_newsize(ap->a_vp, cur, VNOVAL);
+	}
 
 	return (error);
 }
 
+union_lease(ap)
+	struct vop_lease_args /* {
+		struct vnode *a_vp;
+		struct proc *a_p;
+		struct ucred *a_cred;
+		int a_flag;
+	} */ *ap;
+{
+	register struct vnode *ovp = OTHERVP(ap->a_vp);
+
+	ap->a_vp = ovp;
+	return (VCALL(ovp, VOFFSET(vop_lease), ap));
+}
+
 int
 union_ioctl(ap)
 	struct vop_ioctl_args /* {
@@ -767,9 +806,10 @@ union_ioctl(ap)
 		struct proc *a_p;
 	} */ *ap;
 {
+	register struct vnode *ovp = OTHERVP(ap->a_vp);
 
-	return (VOP_IOCTL(OTHERVP(ap->a_vp), ap->a_command, ap->a_data,
-				ap->a_fflag, ap->a_cred, ap->a_p));
+	ap->a_vp = ovp;
+	return (VCALL(ovp, VOFFSET(vop_ioctl), ap));
 }
 
 int
@@ -782,9 +822,27 @@ union_select(ap)
 		struct proc *a_p;
 	} */ *ap;
 {
+	register struct vnode *ovp = OTHERVP(ap->a_vp);
+
+	ap->a_vp = ovp;
+	return (VCALL(ovp, VOFFSET(vop_select), ap));
+}
+
+int
+union_revoke(ap)
+	struct vop_revoke_args /* {
+		struct vnode *a_vp;
+		int a_flags;
+		struct proc *a_p;
+	} */ *ap;
+{
+	struct vnode *vp = ap->a_vp;
 
-	return (VOP_SELECT(OTHERVP(ap->a_vp), ap->a_which, ap->a_fflags,
-				ap->a_cred, ap->a_p));
+	if (UPPERVP(vp))
+		VOP_REVOKE(UPPERVP(vp), ap->a_flags);
+	if (LOWERVP(vp))
+		VOP_REVOKE(LOWERVP(vp), ap->a_flags);
+	vgone(vp);
 }
 
 int
@@ -796,9 +854,10 @@ union_mmap(ap)
 		struct proc *a_p;
 	} */ *ap;
 {
+	register struct vnode *ovp = OTHERVP(ap->a_vp);
 
-	return (VOP_MMAP(OTHERVP(ap->a_vp), ap->a_fflags,
-				ap->a_cred, ap->a_p));
+	ap->a_vp = ovp;
+	return (VCALL(ovp, VOFFSET(vop_mmap), ap));
 }
 
 int
@@ -811,19 +870,19 @@ union_fsync(ap)
 	} */ *ap;
 {
 	int error = 0;
+	struct proc *p = ap->a_p;
 	struct vnode *targetvp = OTHERVP(ap->a_vp);
 
-	if (targetvp) {
+	if (targetvp != NULLVP) {
 		int dolock = (targetvp == LOWERVP(ap->a_vp));
 
 		if (dolock)
-			VOP_LOCK(targetvp);
+			vn_lock(targetvp, LK_EXCLUSIVE | LK_RETRY, p);
 		else
-			FIXUP(VTOUNION(ap->a_vp));
-		error = VOP_FSYNC(targetvp, ap->a_cred,
-					ap->a_waitfor, ap->a_p);
+			FIXUP(VTOUNION(ap->a_vp), p);
+		error = VOP_FSYNC(targetvp, ap->a_cred, ap->a_waitfor, p);
 		if (dolock)
-			VOP_UNLOCK(targetvp);
+			VOP_UNLOCK(targetvp, 0, p);
 	}
 
 	return (error);
@@ -838,8 +897,10 @@ union_seek(ap)
 		struct ucred *a_cred;
 	} */ *ap;
 {
+	register struct vnode *ovp = OTHERVP(ap->a_vp);
 
-	return (VOP_SEEK(OTHERVP(ap->a_vp), ap->a_oldoff, ap->a_newoff, ap->a_cred));
+	ap->a_vp = ovp;
+	return (VCALL(ovp, VOFFSET(vop_seek), ap));
 }
 
 int
@@ -853,34 +914,37 @@ union_remove(ap)
 	int error;
 	struct union_node *dun = VTOUNION(ap->a_dvp);
 	struct union_node *un = VTOUNION(ap->a_vp);
+	struct componentname *cnp = ap->a_cnp;
+	struct proc *p = cnp->cn_proc;
+
+	if (dun->un_uppervp == NULLVP)
+		panic("union remove: null upper vnode");
 
-	if (dun->un_uppervp && un->un_uppervp) {
+	if (un->un_uppervp != NULLVP) {
 		struct vnode *dvp = dun->un_uppervp;
 		struct vnode *vp = un->un_uppervp;
 
-		FIXUP(dun);
+		FIXUP(dun, p);
 		VREF(dvp);
 		dun->un_flags |= UN_KLOCK;
 		vput(ap->a_dvp);
-		FIXUP(un);
+		FIXUP(un, p);
 		VREF(vp);
 		un->un_flags |= UN_KLOCK;
 		vput(ap->a_vp);
 
-		error = VOP_REMOVE(dvp, vp, ap->a_cnp);
+		if (union_dowhiteout(un, cnp->cn_cred, cnp->cn_proc))
+			cnp->cn_flags |= DOWHITEOUT;
+		error = VOP_REMOVE(dvp, vp, cnp);
 		if (!error)
 			union_removed_upper(un);
-
-		/*
-		 * XXX: should create a whiteout here
-		 */
 	} else {
-		/*
-		 * XXX: should create a whiteout here
-		 */
+		FIXUP(dun, p);
+		error = union_mkwhiteout(
+			MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount),
+			dun->un_uppervp, ap->a_cnp, un->un_path);
 		vput(ap->a_dvp);
 		vput(ap->a_vp);
-		error = EROFS;
 	}
 
 	return (error);
@@ -894,34 +958,51 @@ union_link(ap)
 		struct componentname *a_cnp;
 	} */ *ap;
 {
-	int error;
-	struct union_node *dun = VTOUNION(ap->a_vp);
-	struct union_node *un = VTOUNION(ap->a_tdvp);
-
-	if (dun->un_uppervp && un->un_uppervp) {
-		struct vnode *dvp = dun->un_uppervp;
-		struct vnode *vp = un->un_uppervp;
+	int error = 0;
+	struct componentname *cnp = ap->a_cnp;
+	struct proc *p = cnp->cn_proc;
+	struct union_node *un;
+	struct vnode *vp;
+	struct vnode *tdvp;
 
-		FIXUP(dun);
-		VREF(dvp);
-		dun->un_flags |= UN_KLOCK;
-		vput(ap->a_vp);
-		FIXUP(un);
-		VREF(vp);
-		vrele(ap->a_tdvp);
+	un = VTOUNION(ap->a_tdvp);
 
-		error = VOP_LINK(dvp, vp, ap->a_cnp);
+	if (ap->a_tdvp->v_op != ap->a_vp->v_op) {
+		vp = ap->a_vp;
 	} else {
-		/*
-		 * XXX: need to copy to upper layer
-		 * and do the link there.
-		 */
-		vput(ap->a_vp);
-		vrele(ap->a_tdvp);
+		struct union_node *tun = VTOUNION(ap->a_vp);
+		if (tun->un_uppervp == NULLVP) {
+			vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY, p);
+			if (un->un_uppervp == tun->un_dirvp) {
+				un->un_flags &= ~UN_ULOCK;
+				VOP_UNLOCK(un->un_uppervp, 0, p);
+			}
+			error = union_copyup(tun, 1, cnp->cn_cred, p);
+			if (un->un_uppervp == tun->un_dirvp) {
+				vn_lock(un->un_uppervp,
+						LK_EXCLUSIVE | LK_RETRY, p);
+				un->un_flags |= UN_ULOCK;
+			}
+			VOP_UNLOCK(ap->a_vp, 0, p);
+		}
+		vp = tun->un_uppervp;
+	}
+
+	tdvp = un->un_uppervp;
+	if (tdvp == NULLVP)
 		error = EROFS;
+
+	if (error) {
+		vput(ap->a_tdvp);
+		return (error);
 	}
 
-	return (error);
+	FIXUP(un, p);
+	VREF(tdvp);
+	un->un_flags |= UN_KLOCK;
+	vput(ap->a_tdvp);
+
+	return (VOP_LINK(vp, tdvp, cnp));
 }
 
 int
@@ -945,11 +1026,16 @@ union_rename(ap)
 	if (fdvp->v_op == union_vnodeop_p) {	/* always true */
 		struct union_node *un = VTOUNION(fdvp);
 		if (un->un_uppervp == NULLVP) {
-			error = EROFS;
+			/*
+			 * this should never happen in normal
+			 * operation but might if there was
+			 * a problem creating the top-level shadow
+			 * directory.
+			 */
+			error = EXDEV;
 			goto bad;
 		}
 
-		FIXUP(un);
 		fdvp = un->un_uppervp;
 		VREF(fdvp);
 		vrele(ap->a_fdvp);
@@ -958,11 +1044,14 @@ union_rename(ap)
 	if (fvp->v_op == union_vnodeop_p) {	/* always true */
 		struct union_node *un = VTOUNION(fvp);
 		if (un->un_uppervp == NULLVP) {
-			error = EROFS;
+			/* XXX: should do a copyup */
+			error = EXDEV;
 			goto bad;
 		}
 
-		FIXUP(un);
+		if (un->un_lowervp != NULLVP)
+			ap->a_fcnp->cn_flags |= DOWHITEOUT;
+
 		fvp = un->un_uppervp;
 		VREF(fvp);
 		vrele(ap->a_fvp);
@@ -971,7 +1060,13 @@ union_rename(ap)
 	if (tdvp->v_op == union_vnodeop_p) {
 		struct union_node *un = VTOUNION(tdvp);
 		if (un->un_uppervp == NULLVP) {
-			error = EROFS;
+			/*
+			 * this should never happen in normal
+			 * operation but might if there was
+			 * a problem creating the top-level shadow
+			 * directory.
+			 */
+			error = EXDEV;
 			goto bad;
 		}
 
@@ -981,16 +1076,14 @@ union_rename(ap)
 		vput(ap->a_tdvp);
 	}
 
-	if (tvp && tvp->v_op == union_vnodeop_p) {
+	if (tvp != NULLVP && tvp->v_op == union_vnodeop_p) {
 		struct union_node *un = VTOUNION(tvp);
-		if (un->un_uppervp == NULLVP) {
-			error = EROFS;
-			goto bad;
-		}
 
 		tvp = un->un_uppervp;
-		VREF(tvp);
-		un->un_flags |= UN_KLOCK;
+		if (tvp != NULLVP) {
+			VREF(tvp);
+			un->un_flags |= UN_KLOCK;
+		}
 		vput(ap->a_tvp);
 	}
 
@@ -1000,7 +1093,7 @@ bad:
 	vrele(fdvp);
 	vrele(fvp);
 	vput(tdvp);
-	if (tvp)
+	if (tvp != NULLVP)
 		vput(tvp);
 
 	return (error);
@@ -1017,27 +1110,26 @@ union_mkdir(ap)
 {
 	struct union_node *un = VTOUNION(ap->a_dvp);
 	struct vnode *dvp = un->un_uppervp;
+	struct componentname *cnp = ap->a_cnp;
+	struct proc *p = cnp->cn_proc;
 
-	if (dvp) {
+	if (dvp != NULLVP) {
 		int error;
 		struct vnode *vp;
 
-		FIXUP(un);
+		FIXUP(un, p);
 		VREF(dvp);
 		un->un_flags |= UN_KLOCK;
-		vput(ap->a_dvp);
-		error = VOP_MKDIR(dvp, &vp, ap->a_cnp, ap->a_vap);
-		if (error)
+		VOP_UNLOCK(ap->a_dvp, 0, p);
+		error = VOP_MKDIR(dvp, &vp, cnp, ap->a_vap);
+		if (error) {
+			vrele(ap->a_dvp);
 			return (error);
+		}
 
-		error = union_allocvp(
-				ap->a_vpp,
-				ap->a_dvp->v_mount,
-				ap->a_dvp,
-				NULLVP,
-				ap->a_cnp,
-				vp,
-				NULLVP);
+		error = union_allocvp(ap->a_vpp, ap->a_dvp->v_mount, ap->a_dvp,
+				NULLVP, cnp, vp, NULLVP, 1);
+		vrele(ap->a_dvp);
 		if (error)
 			vput(vp);
 		return (error);
@@ -1058,34 +1150,37 @@ union_rmdir(ap)
 	int error;
 	struct union_node *dun = VTOUNION(ap->a_dvp);
 	struct union_node *un = VTOUNION(ap->a_vp);
+	struct componentname *cnp = ap->a_cnp;
+	struct proc *p = cnp->cn_proc;
+
+	if (dun->un_uppervp == NULLVP)
+		panic("union rmdir: null upper vnode");
 
-	if (dun->un_uppervp && un->un_uppervp) {
+	if (un->un_uppervp != NULLVP) {
 		struct vnode *dvp = dun->un_uppervp;
 		struct vnode *vp = un->un_uppervp;
 
-		FIXUP(dun);
+		FIXUP(dun, p);
 		VREF(dvp);
 		dun->un_flags |= UN_KLOCK;
 		vput(ap->a_dvp);
-		FIXUP(un);
+		FIXUP(un, p);
 		VREF(vp);
 		un->un_flags |= UN_KLOCK;
 		vput(ap->a_vp);
 
+		if (union_dowhiteout(un, cnp->cn_cred, cnp->cn_proc))
+			cnp->cn_flags |= DOWHITEOUT;
 		error = VOP_RMDIR(dvp, vp, ap->a_cnp);
 		if (!error)
 			union_removed_upper(un);
-
-		/*
-		 * XXX: should create a whiteout here
-		 */
 	} else {
-		/*
-		 * XXX: should create a whiteout here
-		 */
+		FIXUP(dun, p);
+		error = union_mkwhiteout(
+			MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount),
+			dun->un_uppervp, ap->a_cnp, un->un_path);
 		vput(ap->a_dvp);
 		vput(ap->a_vp);
-		error = EROFS;
 	}
 
 	return (error);
@@ -1103,18 +1198,19 @@ union_symlink(ap)
 {
 	struct union_node *un = VTOUNION(ap->a_dvp);
 	struct vnode *dvp = un->un_uppervp;
+	struct componentname *cnp = ap->a_cnp;
+	struct proc *p = cnp->cn_proc;
 
-	if (dvp) {
+	if (dvp != NULLVP) {
 		int error;
 		struct vnode *vp;
 		struct mount *mp = ap->a_dvp->v_mount;
 
-		FIXUP(un);
+		FIXUP(un, p);
 		VREF(dvp);
 		un->un_flags |= UN_KLOCK;
 		vput(ap->a_dvp);
-		error = VOP_SYMLINK(dvp, &vp, ap->a_cnp,
-					ap->a_vap, ap->a_target);
+		error = VOP_SYMLINK(dvp, &vp, cnp, ap->a_vap, ap->a_target);
 		*ap->a_vpp = NULLVP;
 		return (error);
 	}
@@ -1137,17 +1233,21 @@ union_readdir(ap)
 		struct vnode *a_vp;
 		struct uio *a_uio;
 		struct ucred *a_cred;
+		int *a_eofflag;
+		u_long *a_cookies;
+		int a_ncookies;
 	} */ *ap;
 {
-	int error = 0;
 	struct union_node *un = VTOUNION(ap->a_vp);
+	struct vnode *uvp = un->un_uppervp;
+	struct proc *p = ap->a_uio->uio_procp;
 
-	if (un->un_uppervp) {
-		FIXUP(un);
-		error = VOP_READDIR(un->un_uppervp, ap->a_uio, ap->a_cred);
-	}
+	if (uvp == NULLVP)
+		return (0);
 
-	return (error);
+	FIXUP(un, p);
+	ap->a_vp = uvp;
+	return (VCALL(uvp, VOFFSET(vop_readdir), ap));
 }
 
 int
@@ -1159,16 +1259,19 @@ union_readlink(ap)
 	} */ *ap;
 {
 	int error;
+	struct uio *uio = ap->a_uio;
+	struct proc *p = uio->uio_procp;
 	struct vnode *vp = OTHERVP(ap->a_vp);
 	int dolock = (vp == LOWERVP(ap->a_vp));
 
 	if (dolock)
-		VOP_LOCK(vp);
+		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
 	else
-		FIXUP(VTOUNION(ap->a_vp));
-	error = VOP_READLINK(vp, ap->a_uio, ap->a_cred);
+		FIXUP(VTOUNION(ap->a_vp), p);
+	ap->a_vp = vp;
+	error = VCALL(vp, VOFFSET(vop_readlink), ap);
 	if (dolock)
-		VOP_UNLOCK(vp);
+		VOP_UNLOCK(vp, 0, p);
 
 	return (error);
 }
@@ -1181,6 +1284,8 @@ union_abortop(ap)
 	} */ *ap;
 {
 	int error;
+	struct componentname *cnp = ap->a_cnp;
+	struct proc *p = cnp->cn_proc;
 	struct vnode *vp = OTHERVP(ap->a_dvp);
 	struct union_node *un = VTOUNION(ap->a_dvp);
 	int islocked = un->un_flags & UN_LOCKED;
@@ -1188,13 +1293,14 @@ union_abortop(ap)
 
 	if (islocked) {
 		if (dolock)
-			VOP_LOCK(vp);
+			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
 		else
-			FIXUP(VTOUNION(ap->a_dvp));
+			FIXUP(VTOUNION(ap->a_dvp), p);
 	}
-	error = VOP_ABORTOP(vp, ap->a_cnp);
+	ap->a_dvp = vp;
+	error = VCALL(vp, VOFFSET(vop_abortop), ap);
 	if (islocked && dolock)
-		VOP_UNLOCK(vp);
+		VOP_UNLOCK(vp, 0, p);
 
 	return (error);
 }
@@ -1203,8 +1309,13 @@ int
 union_inactive(ap)
 	struct vop_inactive_args /* {
 		struct vnode *a_vp;
+		struct proc *a_p;
 	} */ *ap;
 {
+	struct vnode *vp = ap->a_vp;
+	struct proc *p = ap->a_p;
+	struct union_node *un = VTOUNION(vp);
+	struct vnode **vpp;
 
 	/*
 	 * Do nothing (and _don't_ bypass).
@@ -1219,12 +1330,17 @@ union_inactive(ap)
 	 * That's too much work for now.
 	 */
 
-#ifdef UNION_DIAGNOSTIC
-	struct union_node *un = VTOUNION(ap->a_vp);
+	if (un->un_dircache != 0) {
+		for (vpp = un->un_dircache; *vpp != NULLVP; vpp++)
+			vrele(*vpp);
+		free(un->un_dircache, M_TEMP);
+		un->un_dircache = 0;
+	}
 
-	if (un->un_flags & UN_LOCKED)
-		panic("union: inactivating locked node");
-#endif
+	VOP_UNLOCK(vp, 0, p);
+
+	if ((un->un_flags & UN_CACHED) == 0)
+		vgone(vp);
 
 	return (0);
 }
@@ -1246,24 +1362,40 @@ union_lock(ap)
 	struct vop_lock_args *ap;
 {
 	struct vnode *vp = ap->a_vp;
+	struct proc *p = ap->a_p;
+	int flags = ap->a_flags;
 	struct union_node *un;
+	int error;
 
-start:
-	while (vp->v_flag & VXLOCK) {
-		vp->v_flag |= VXWANT;
-		sleep((caddr_t)vp, PINOD);
-	}
 
+	vop_nolock(ap);
+	/*
+	 * Need to do real lockmgr-style locking here.
+	 * in the mean time, draining won't work quite right,
+	 * which could lead to a few race conditions.
+	 * the following test was here, but is not quite right, we
+	 * still need to take the lock:
+	if ((flags & LK_TYPE_MASK) == LK_DRAIN)
+		return (0);
+	 */
+	flags &= ~LK_INTERLOCK;
+
+start:
 	un = VTOUNION(vp);
 
-	if (un->un_uppervp) {
-		if ((un->un_flags & UN_ULOCK) == 0) {
+	if (un->un_uppervp != NULLVP) {
+		if (((un->un_flags & UN_ULOCK) == 0) &&
+		    (vp->v_usecount != 0)) {
+			error = vn_lock(un->un_uppervp, flags, p);
+			if (error)
+				return (error);
 			un->un_flags |= UN_ULOCK;
-			VOP_LOCK(un->un_uppervp);
 		}
 #ifdef DIAGNOSTIC
-		if (un->un_flags & UN_KLOCK)
-			panic("union: dangling upper lock");
+		if (un->un_flags & UN_KLOCK) {
+			vprint("union: dangling klock", vp);
+			panic("union: dangling upper lock (%lx)", vp);
+		}
 #endif
 	}
 
@@ -1274,7 +1406,7 @@ start:
 			panic("union: locking against myself");
 #endif
 		un->un_flags |= UN_WANT;
-		sleep((caddr_t) &un->un_flags, PINOD);
+		tsleep((caddr_t)&un->un_flags, PINOD, "unionlk2", 0);
 		goto start;
 	}
 
@@ -1289,11 +1421,27 @@ start:
 	return (0);
 }
 
+/*
+ * When operations want to vput() a union node yet retain a lock on
+ * the upper vnode (say, to do some further operations like link(),
+ * mkdir(), ...), they set UN_KLOCK on the union node, then call
+ * vput() which calls VOP_UNLOCK() and comes here.  union_unlock()
+ * unlocks the union node (leaving the upper vnode alone), clears the
+ * KLOCK flag, and then returns to vput().  The caller then does whatever
+ * is left to do with the upper vnode, and ensures that it gets unlocked.
+ *
+ * If UN_KLOCK isn't set, then the upper vnode is unlocked here.
+ */
 int
 union_unlock(ap)
-	struct vop_lock_args *ap;
+	struct vop_unlock_args /* {
+		struct vnode *a_vp;
+		int a_flags;
+		struct proc *a_p;
+	} */ *ap;
 {
 	struct union_node *un = VTOUNION(ap->a_vp);
+	struct proc *p = ap->a_p;
 
 #ifdef DIAGNOSTIC
 	if ((un->un_flags & UN_LOCKED) == 0)
@@ -1306,7 +1454,7 @@ union_unlock(ap)
 	un->un_flags &= ~UN_LOCKED;
 
 	if ((un->un_flags & (UN_ULOCK|UN_KLOCK)) == UN_ULOCK)
-		VOP_UNLOCK(un->un_uppervp);
+		VOP_UNLOCK(un->un_uppervp, 0, p);
 
 	un->un_flags &= ~(UN_ULOCK|UN_KLOCK);
 
@@ -1318,6 +1466,7 @@ union_unlock(ap)
 #ifdef DIAGNOSTIC
 	un->un_pid = 0;
 #endif
+	vop_nounlock(ap);
 
 	return (0);
 }
@@ -1333,16 +1482,18 @@ union_bmap(ap)
 	} */ *ap;
 {
 	int error;
+	struct proc *p = curproc;		/* XXX */
 	struct vnode *vp = OTHERVP(ap->a_vp);
 	int dolock = (vp == LOWERVP(ap->a_vp));
 
 	if (dolock)
-		VOP_LOCK(vp);
+		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
 	else
-		FIXUP(VTOUNION(ap->a_vp));
-	error = VOP_BMAP(vp, ap->a_bn, ap->a_vpp, ap->a_bnp, ap->a_runp);
+		FIXUP(VTOUNION(ap->a_vp), p);
+	ap->a_vp = vp;
+	error = VCALL(vp, VOFFSET(vop_bmap), ap);
 	if (dolock)
-		VOP_UNLOCK(vp);
+		VOP_UNLOCK(vp, 0, p);
 
 	return (error);
 }
@@ -1357,6 +1508,11 @@ union_print(ap)
 
 	printf("\ttag VT_UNION, vp=%x, uppervp=%x, lowervp=%x\n",
 			vp, UPPERVP(vp), LOWERVP(vp));
+	if (UPPERVP(vp) != NULLVP)
+		vprint("union: upper", UPPERVP(vp));
+	if (LOWERVP(vp) != NULLVP)
+		vprint("union: lower", LOWERVP(vp));
+
 	return (0);
 }
 
@@ -1379,16 +1535,18 @@ union_pathconf(ap)
 	} */ *ap;
 {
 	int error;
+	struct proc *p = curproc;		/* XXX */
 	struct vnode *vp = OTHERVP(ap->a_vp);
 	int dolock = (vp == LOWERVP(ap->a_vp));
 
 	if (dolock)
-		VOP_LOCK(vp);
+		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
 	else
-		FIXUP(VTOUNION(ap->a_vp));
-	error = VOP_PATHCONF(vp, ap->a_name, ap->a_retval);
+		FIXUP(VTOUNION(ap->a_vp), p);
+	ap->a_vp = vp;
+	error = VCALL(vp, VOFFSET(vop_pathconf), ap);
 	if (dolock)
-		VOP_UNLOCK(vp);
+		VOP_UNLOCK(vp, 0, p);
 
 	return (error);
 }
@@ -1403,9 +1561,10 @@ union_advlock(ap)
 		int  a_flags;
 	} */ *ap;
 {
+	register struct vnode *ovp = OTHERVP(ap->a_vp);
 
-	return (VOP_ADVLOCK(OTHERVP(ap->a_vp), ap->a_id, ap->a_op,
-				ap->a_fl, ap->a_flags));
+	ap->a_vp = ovp;
+	return (VCALL(ovp, VOFFSET(vop_advlock), ap));
 }
 
 
@@ -1449,6 +1608,7 @@ struct vnodeopv_entry_desc union_vnodeop_entries[] = {
 	{ &vop_default_desc, vn_default_error },
 	{ &vop_lookup_desc, union_lookup },		/* lookup */
 	{ &vop_create_desc, union_create },		/* create */
+	{ &vop_whiteout_desc, union_whiteout },		/* whiteout */
 	{ &vop_mknod_desc, union_mknod },		/* mknod */
 	{ &vop_open_desc, union_open },			/* open */
 	{ &vop_close_desc, union_close },		/* close */
@@ -1457,8 +1617,10 @@ struct vnodeopv_entry_desc union_vnodeop_entries[] = {
 	{ &vop_setattr_desc, union_setattr },		/* setattr */
 	{ &vop_read_desc, union_read },			/* read */
 	{ &vop_write_desc, union_write },		/* write */
+	{ &vop_lease_desc, union_lease },		/* lease */
 	{ &vop_ioctl_desc, union_ioctl },		/* ioctl */
 	{ &vop_select_desc, union_select },		/* select */
+	{ &vop_revoke_desc, union_revoke },		/* revoke */
 	{ &vop_mmap_desc, union_mmap },			/* mmap */
 	{ &vop_fsync_desc, union_fsync },		/* fsync */
 	{ &vop_seek_desc, union_seek },			/* seek */