The CRT-calculation for an RSA secret key operation can only be used when the
prime factors and inverse value are available, i.e. (p && q && u). The logical
inverse of this condition is (!p || !q || !u). Seems De Morgan's wasn't
applied properly here.
Test case:
Create an RSA private key with a proper subset of {p, q, u} missing, but
otherwise complete. Call secret() (or the public interfaces that use secret)
with this key.