diff --git a/sbin/mount_nullfs/mount_nullfs.8 b/sbin/mount_nullfs/mount_nullfs.8
--- a/sbin/mount_nullfs/mount_nullfs.8
+++ b/sbin/mount_nullfs/mount_nullfs.8
@@ -90,7 +90,7 @@
 .Xr mount 8
 man page for possible options and their meanings.
 Additionally the following option is supported:
-.Bl -tag -width nocache
+.Bl -tag -width nounixbypass
 .It Cm nocache
 Disable metadata caching in the null layer.
 Some lower-layer file systems may force this option.
@@ -98,6 +98,30 @@
 this may result in increased lock contention.
 .It Cm cache
 Force enable metadata caching.
+.It Cm nounixbypass
+Disable bypassing the unix socket file used for
+.Xr bind 2
+and
+.Xr connect 2 ,
+to the lower (mounted-from) filesystem layer.
+.Pp
+The effect is that lower and upper (bypassed) unix sockets
+are separate.
+.It Cm unixbypass
+Enable the bypass of unix socket file to lower filesystem layer.
+This is default.
+.Pp
+The effect is that
+.Xr bind 2
+and
+.Xr connect 2
+operations on unix socket done on either upper (nullfs) or lower
+layer file are performed on same unix socket.
+For instance, if a server
+.Xr bind 2
+is done on lower file, then
+.Xr connect 2
+on the socket file visible through nullfs mount, connects to the server.
 .El
 .El
 .Pp
diff --git a/sys/fs/nullfs/null.h b/sys/fs/nullfs/null.h
--- a/sys/fs/nullfs/null.h
+++ b/sys/fs/nullfs/null.h
@@ -35,11 +35,12 @@
 #ifndef	FS_NULL_H
 #define	FS_NULL_H
 
-#define	NULLM_CACHE	0x0001
-
 #include <sys/ck.h>
 #include <vm/uma.h>
 
+#define	NULLM_CACHE		0x0001
+#define	NULLM_NOUNPBYPASS	0x0002
+
 struct null_mount {
 	struct mount	*nullm_vfs;
 	struct vnode	*nullm_lowerrootvp;	/* Ref to lower root vnode */
@@ -82,6 +83,16 @@
 #endif
 
 extern struct vop_vector null_vnodeops;
+extern struct vop_vector null_vnodeops_no_unp_bypass;
+
+static inline bool
+null_is_nullfs_vnode(struct vnode *vp)
+{
+	const struct vop_vector *op;
+
+	op = vp->v_op;
+	return (op == &null_vnodeops || op == &null_vnodeops_no_unp_bypass);
+}
 
 extern uma_zone_t null_node_zone;
 
diff --git a/sys/fs/nullfs/null_subr.c b/sys/fs/nullfs/null_subr.c
--- a/sys/fs/nullfs/null_subr.c
+++ b/sys/fs/nullfs/null_subr.c
@@ -240,7 +240,9 @@
 	 */
 	xp = uma_zalloc_smr(null_node_zone, M_WAITOK);
 
-	error = getnewvnode("nullfs", mp, &null_vnodeops, &vp);
+	error = getnewvnode("nullfs", mp, (MOUNTTONULLMOUNT(mp)->nullm_flags &
+	    NULLM_NOUNPBYPASS) != 0 ? &null_vnodeops_no_unp_bypass :
+	    &null_vnodeops, &vp);
 	if (error) {
 		vput(lowervp);
 		uma_zfree_smr(null_node_zone, xp);
diff --git a/sys/fs/nullfs/null_vfsops.c b/sys/fs/nullfs/null_vfsops.c
--- a/sys/fs/nullfs/null_vfsops.c
+++ b/sys/fs/nullfs/null_vfsops.c
@@ -85,6 +85,10 @@
 	char *target;
 	int error, len;
 	bool isvnunlocked;
+	static const char cache_opt_name[] = "cache";
+	static const char nocache_opt_name[] = "nocache";
+	static const char unixbypass_opt_name[] = "unixbypass";
+	static const char nounixbypass_opt_name[] = "nounixbypass";
 
 	NULLFSDEBUG("nullfs_mount(mp = %p)\n", (void *)mp);
 
@@ -116,7 +120,7 @@
 	/*
 	 * Unlock lower node to avoid possible deadlock.
 	 */
-	if (mp->mnt_vnodecovered->v_op == &null_vnodeops &&
+	if (null_is_nullfs_vnode(mp->mnt_vnodecovered) &&
 	    VOP_ISLOCKED(mp->mnt_vnodecovered) == LK_EXCLUSIVE) {
 		VOP_UNLOCK(mp->mnt_vnodecovered);
 		isvnunlocked = true;
@@ -150,7 +154,7 @@
 	/*
 	 * Check multi null mount to avoid `lock against myself' panic.
 	 */
-	if (mp->mnt_vnodecovered->v_op == &null_vnodeops) {
+	if (null_is_nullfs_vnode(mp->mnt_vnodecovered)) {
 		nn = VTONULL(mp->mnt_vnodecovered);
 		if (nn == NULL || lowerrootvp == nn->null_lowervp) {
 			NULLFSDEBUG("nullfs_mount: multi null mount?\n");
@@ -205,9 +209,10 @@
 		MNT_IUNLOCK(mp);
 	}
 
-	if (vfs_getopt(mp->mnt_optnew, "cache", NULL, NULL) == 0) {
+	if (vfs_getopt(mp->mnt_optnew, cache_opt_name, NULL, NULL) == 0) {
 		xmp->nullm_flags |= NULLM_CACHE;
-	} else if (vfs_getopt(mp->mnt_optnew, "nocache", NULL, NULL) == 0) {
+	} else if (vfs_getopt(mp->mnt_optnew, nocache_opt_name, NULL,
+	    NULL) == 0) {
 		;
 	} else if (null_cache_vnodes &&
 	    (xmp->nullm_vfs->mnt_kern_flag & MNTK_NULL_NOCACHE) == 0) {
@@ -219,6 +224,13 @@
 		    &xmp->notify_node);
 	}
 
+	if (vfs_getopt(mp->mnt_optnew, unixbypass_opt_name, NULL, NULL) == 0) {
+		;
+	} else if (vfs_getopt(mp->mnt_optnew, nounixbypass_opt_name, NULL,
+	    NULL) == 0) {
+		xmp->nullm_flags |= NULLM_NOUNPBYPASS;
+	}
+
 	if (lowerrootvp == mp->mnt_vnodecovered) {
 		vn_lock(lowerrootvp, LK_EXCLUSIVE | LK_RETRY | LK_CANRECURSE);
 		lowerrootvp->v_vflag |= VV_CROSSLOCK;
diff --git a/sys/fs/nullfs/null_vnops.c b/sys/fs/nullfs/null_vnops.c
--- a/sys/fs/nullfs/null_vnops.c
+++ b/sys/fs/nullfs/null_vnops.c
@@ -278,7 +278,7 @@
 		 * that aren't.  (We must always map first vp or vclean fails.)
 		 */
 		if (i != 0 && (*this_vp_p == NULL ||
-			       (*this_vp_p)->v_op != &null_vnodeops)) {
+		    !null_is_nullfs_vnode(*this_vp_p))) {
 			old_vps[i] = NULL;
 		} else {
 			old_vps[i] = *this_vp_p;
@@ -1256,3 +1256,11 @@
 	.vop_copy_file_range =	VOP_PANIC,
 };
 VFS_VOP_VECTOR_REGISTER(null_vnodeops);
+
+struct vop_vector null_vnodeops_no_unp_bypass = {
+	.vop_default =		&null_vnodeops,
+	.vop_unp_bind =		vop_stdunp_bind,
+	.vop_unp_connect =	vop_stdunp_connect,
+	.vop_unp_detach =	vop_stdunp_detach,
+};
+VFS_VOP_VECTOR_REGISTER(null_vnodeops_no_unp_bypass);