LibCrypto: Add optimized RSA decryption with CRT method

The textbook RSA decryption method of `c^d % n` is quite slow. If the
necessary parameters are present, the CRT variant will be used.
Performing RSA decryption this way is ~3 times faster.

Libraries/LibCrypto/PK/RSA.cpp

@@ -133,12 +133,24 @@ void RSA::encrypt(ReadonlyBytes in, Bytes& out)
 
 void RSA::decrypt(ReadonlyBytes in, Bytes& out)
 {
-    // FIXME: Actually use the private key properly
-
     auto in_integer = UnsignedBigInteger::import_data(in.data(), in.size());
-    auto exp = NumberTheory::ModularPower(in_integer, m_private_key.private_exponent(), m_private_key.modulus());
-    auto size = exp.export_data(out);
 
+    UnsignedBigInteger m;
+    if (m_private_key.prime1().is_zero() || m_private_key.prime2().is_zero()) {
+        m = NumberTheory::ModularPower(in_integer, m_private_key.private_exponent(), m_private_key.modulus());
+    } else {
+        auto m1 = NumberTheory::ModularPower(in_integer, m_private_key.exponent1(), m_private_key.prime1());
+        auto m2 = NumberTheory::ModularPower(in_integer, m_private_key.exponent2(), m_private_key.prime2());
+        if (m1 < m2)
+            m1 = m1.plus(m_private_key.prime1());
+
+        VERIFY(m1 >= m2);
+
+        auto h = NumberTheory::Mod(m1.minus(m2).multiplied_by(m_private_key.coefficient()), m_private_key.prime1());
+        m = m2.plus(h.multiplied_by(m_private_key.prime2()));
+    }
+
+    auto size = m.export_data(out);
     auto align = m_private_key.length();
     auto aligned_size = (size + align - 1) / align * align;