# Low Public Exponent Attack for RSA

I'm having trouble understanding the algorithm for finding the original message $m$, when there is a small public exponent. Here is the example I'm trying to follow (you can also read it in the 'Low exponent RSA paragraph' of this article- http://www.cims.nyu.edu/~regev/teaching/lattices_fall_2004/ln/rsa.pdf):

Someone sends a message, $m$ to 3 people (without any padding) with public keys $n_b$, $n_c$ and $n_d$ and public exponent of 3. The article says:

By using the Chinese Remainder Theorem, the eavesdropper computes a number $c$ such that $c = m^3 \mod n_b \cdot n_c \cdot n_d$.

Maybe I don't completely understand the Chinese Remainder Theorem, but I don't understand how you could choose a system of congruences to solve that would end up being equivalent to $c = m^3 \mod n_b \cdot n_c \cdot n_d$ when you don't know what $m$ is. Can you please help me understand what to use as the system of congruences to get to that point because I think I understand the rest.

-
The lesson from this attack is that RSA encryption MUST pad the message to be enciphered with randomness, distinct for each destination, as in PKCS#1 RSAES; a secondary lesson is that bad uses of RSA tend to get worse with low exponent; it should not be that RSA with low exponent is always weak. –  fgrieu Mar 17 '13 at 9:33
Glad you found the right site :) –  Maarten Bodewes Mar 17 '13 at 15:30

You don't need to know $m$. You know $m^3$ modulo each modulus, which is sufficient. You want to find:

$$c \equiv m^3 \pmod{n_b}$$

$$c \equiv m^3 \pmod{n_c}$$

$$c \equiv m^3 \pmod{n_d}$$

Because $n_b$, $n_c$, $n_d$ are pairwise coprime (assume they have no common factors) a solution must exist.

The Wikipedia page has a nice explanation of the algorithm to find $c$. The actual expression is:

$$c = c_b (n_c \cdot n_d) [ (n_c \cdot n_d)^{-1} ]_{n_b} + c_c (n_b \cdot n_d) [ (n_b \cdot n_d)^{-1} ]_{n_c} + c_d (n_b \cdot n_c) [ (n_b \cdot n_c)^{-1} ]_{n_d}$$

Where $[a^{-1}]_b$ is the multiplicative inverse of $a$ modulo $b$. Note $\gcd{(a, b)} = 1$ is always satisfied. Also, I used the notation $c_b = m^3 ~ \text{mod} ~ n_b$, $c_c = m^3 ~ \text{mod} ~ n_c$, $c_d = m^3 ~ \text{mod} ~ n_d$.

Let's try with some numbers. Suppose someone sends the message $m = 102$ to three different people with textbook RSA, with moduli $n_b = 377$, $n_c = 391$ and $n_d = 589$. So:

$$c_b = 102^3 ~ \text{mod} ~ 377 = 330$$ $$c_c = 102^3 ~ \text{mod} ~ 391 = 34$$ $$c_d = 102^3 ~ \text{mod} ~ 589 = 419$$

So the attacker wants to solve the following system of congruences:

$$c \equiv 330 \pmod{377}$$

$$c \equiv 34 \pmod{391}$$

$$c \equiv 419 \pmod{589}$$

Using the equation above, we obtain (compute each term separately for clarity):

$$t_b = c_b (n_c \cdot n_d) [ (n_c \cdot n_d)^{-1} ]_{n_b} = 330 (391 \times 589) [ (391 \times 589)^{-1}]_{377} = 24471571740$$

$$t_c = c_c (n_b \cdot n_d) [ (n_b \cdot n_d)^{-1} ]_{n_c} = 34 (377 \times 589) [ (377 \times 589)^{-1}]_{391} = 505836734$$

$$t_d = c_d (n_b \cdot n_c) [ (n_b \cdot n_c)^{-1} ]_{n_d} = 419 (377 \times 391) [ (377 \times 391)^{-1}]_{589} = 35452267942$$

$$\therefore c = t_b + t_c + t_d ~ \text{mod} ~ (n_b \cdot n_c \cdot n_d) = 1061208$$

And we get $m = \sqrt[3]{c} = \sqrt[3]{1061208} = 102 = m$.

Note this attack on textbook RSA would work on any exponent given a sufficiently large number of people, of course $e = 3$ is the most realistic setting (and easiest to demonstrate, for obvious reasons).

The idea, of course, being to use these relations with the CRT to manufacture a relation of the form $m^3 \equiv x \pmod{n'}$ where $n'$ is on the order of $n^3$ (and, more crucially, where $1 < m < n$) such that $m^3$ is not reduced (and so you can just take cube roots).

-
It seems like I need $e=3$ congruences in order to use this attack. How and why does e influence the number of congruences? If I am correct, CRT doesn't say anything about the number of congruences needed to find a solution - so why would it not just work with two congruences? –  CGFoX Jul 24 '14 at 7:24
@CGFoX If you had only two congruences then your $n'$ would be on the order of $n^2$ (product of two moduli) and so your $m^3$ would be larger than $n^2$ (by an order of $n$) and you couldn't easily take the cube root without knowing the factorization of the moduli, so it doesn't work. –  Thomas Jul 24 '14 at 7:40