As stated, your problem doesn't have a unique solution. There are infinitely many pairs of numbers that yield the same correlation coefficient. For example, if x={1, -1}, then any y of the form y={t, -t} has perfect correlation (r=1) with x. Also, the correlation coefficient is invariant under shifts in the mean of the data, so you probably want additional assumptions on the distribution of the data.
A problem that makes sense to ask is "given a correlation (or covariance), can I generate multivariate normal vectors x and y of length N from a bivariate distribution with a given set of means and the given covariance?" The answer is yes, and you can use the RANDNORMAL module in SAS/IML to generate the data. See the RANDNORMAL module documentation for syntax and an example.
Obviously, the particular sample you generate is not guaranteed to have the given mean and covariance, even though it was generated from a distribution with those properties.