diff --git a/sys/kern/vfs_lookup.c b/sys/kern/vfs_lookup.c
--- a/sys/kern/vfs_lookup.c
+++ b/sys/kern/vfs_lookup.c
@@ -1075,79 +1075,6 @@
 	return (EEXIST);
 }
 
-static int __noinline
-vfs_lookup_cross_mount(struct nameidata *ndp)
-{
-	struct componentname *cnp;
-	struct mount *mp;
-	struct vnode *dp, *tdp;
-	int error, crosslkflags;
-	bool crosslock;
-
-	cnp = &ndp->ni_cnd;
-	dp = ndp->ni_vp;
-
-	/*
-	 * The vnode has been mounted on, find the root of the mounted
-	 * filesystem.
-	 */
-	for (;;) {
-		mp = dp->v_mountedhere;
-		ASSERT_VOP_LOCKED(dp, __func__);
-		VNPASS((vn_irflag_read(dp) & VIRF_MOUNTPOINT) != 0 && mp != NULL, dp);
-
-		crosslock = (dp->v_vflag & VV_CROSSLOCK) != 0;
-		crosslkflags = enforce_lkflags(mp, cnp->cn_lkflags);
-		if (__predict_false(crosslock)) {
-			/*
-			 * We are going to be holding the vnode lock, which
-			 * in this case is shared by the root vnode of the
-			 * filesystem mounted at mp, across the call to
-			 * VFS_ROOT().  Make the situation clear to the
-			 * filesystem by passing LK_CANRECURSE if the
-			 * lock is held exclusive, or by clearinng
-			 * LK_NODDLKTREAT to allow recursion on the shared
-			 * lock in the presence of an exclusive waiter.
-			 */
-			if (VOP_ISLOCKED(dp) == LK_EXCLUSIVE) {
-				crosslkflags &= ~LK_SHARED;
-				crosslkflags |= LK_EXCLUSIVE | LK_CANRECURSE;
-			} else if ((crosslkflags & LK_EXCLUSIVE) != 0) {
-				error = vn_lock(dp, LK_UPGRADE);
-				if (error != 0)
-					break;
-				if (dp->v_mountedhere != mp) {
-					continue;
-				}
-			} else
-				crosslkflags &= ~LK_NODDLKTREAT;
-		}
-		if (vfs_busy(mp, 0) != 0)
-			continue;
-		if (__predict_true(!crosslock))
-			vput(dp);
-		if (dp != ndp->ni_dvp)
-			vput(ndp->ni_dvp);
-		else
-			vrele(ndp->ni_dvp);
-		vrefact(vp_crossmp);
-		ndp->ni_dvp = vp_crossmp;
-		error = VFS_ROOT(mp, crosslkflags, &tdp);
-		vfs_unbusy(mp);
-		if (__predict_false(crosslock))
-			vput(dp);
-		if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
-			panic("vp_crossmp exclusively locked or reclaimed");
-		if (error != 0)
-			break;
-		ndp->ni_vp = dp = tdp;
-		if ((vn_irflag_read(dp) & VIRF_MOUNTPOINT) == 0)
-			break;
-	}
-
-	return (error);
-}
-
 /*
  * Search a pathname.
  * This is a very central and rather complicated routine.
@@ -1559,17 +1486,54 @@
 		goto success;
 	}
 
-	if ((vn_irflag_read(dp) & VIRF_MOUNTPOINT) != 0 &&
-	    (cnp->cn_flags & NOCROSSMOUNT) == 0) {
-		error = vfs_lookup_cross_mount(ndp);
-		if (error != 0)
-			goto bad_unlocked;
+	/* From here: Not a symlink we have to follow. */
+
+	/*
+	 * Going to cross mounts, if any and if not ruled out.
+	 */
+	if (__predict_true((vn_irflag_read(dp) & VIRF_MOUNTPOINT) == 0) ||
+	    (cnp->cn_flags & NOCROSSMOUNT) != 0)
+		goto nextname;
+
+	/*
+	 * Unlock the parent 'ni_dvp' to avoid a problem with NFS where
+	 * VFS_ROOT() can hang forever on a lost server.  In this case, 'ni_dvp'
+	 * remains locked forever as well, gradually blocking all lookups (or
+	 * only modifications in the MNTK_LOOKUP_SHARED case) starting with
+	 * vnodes earlier in the path up to the root vnode.
+	 */
+	if (ndp->ni_dvp != dp)
+		VOP_UNLOCK(ndp->ni_dvp);
+
+	error = vn_cross_mounts_mounted(dp, cnp->cn_lkflags, &dp);
+	if (__predict_false(error != 0)) {
+		vrele(ndp->ni_dvp);
 		/*
-		 * FALLTHROUGH to nextname
+		 * If 'error' is ENOENT, during crossing, 'dp' was unlocked and
+		 * is now doomed, which, given the pre-existing active
+		 * reference, can only happen on a forced unmount.  We could
+		 * climb to the "parent" directory (in the resolution, i.e.,
+		 * 'ni_dvp'), but this is likely to be useless (if on the same
+		 * mount it can have been doomed already, and if not, it
+		 * currently is equal to 'vp_crossmp' anyway) and problematic
+		 * semantically if done standalone.  The most sensible
+		 * alternative would be to restart the whole resolution, but we
+		 * are not there yet.
 		 */
-		dp = ndp->ni_vp;
+		goto bad_unlocked;
 	}
 
+	/*
+	 * We effectively crossed some mounts.  Release 'ni_dvp', which
+	 * is going to be replaced by 'vp_crossmp'.
+	 */
+	ndp->ni_vp = dp;
+	vrele(ndp->ni_dvp);
+	vrefact(vp_crossmp);
+	if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
+		panic("vp_crossmp exclusively locked or reclaimed");
+	ndp->ni_dvp = vp_crossmp;
+
 nextname:
 	/*
 	 * Not a symbolic link that we will follow.  Continue with the