Does AES_cbc_encrypt add padding?

Question

Consider the following snippet of C++ code:

#include <iostream>
#include <openssl/aes.h>

#define AES_KEY_LENGTH 32

using namespace std;

int main()
{
    AES_KEY encryption_key;
    AES_KEY decryption_key;

    unsigned char key[AES_KEY_LENGTH] = {'t', 'e', 's', 't', 't', 'e', 's', 't', 't', 'e', 's', 't', 't', 'e', 's', 't', 't', 'e', 's', 't', 't', 'e', 's', 't', 't', 'e', 's', 't', 't', 'e', 's', 't'};

    unsigned char iv[AES_BLOCK_SIZE] = {'t', 'e', 's', 't', 't', 'e', 's', 't', 't', 'e', 's', 't', 't', 'e', 's', 't'};

    unsigned char iv_enc[AES_BLOCK_SIZE];
    unsigned char iv_dec[AES_BLOCK_SIZE];

    memcpy(iv_enc, iv, AES_BLOCK_SIZE);
    memcpy(iv_dec, iv, AES_BLOCK_SIZE);

    AES_set_encrypt_key(key, AES_KEY_LENGTH * 8, &(encryption_key));
    AES_set_decrypt_key(key, AES_KEY_LENGTH * 8, &(decryption_key));

    char message[] = "Attack at dawn! Attack.";

    unsigned char * encryption_output = new unsigned char[32];
    encryption_output[31] = 3;

    AES_cbc_encrypt((unsigned char *) message, encryption_output, sizeof(message), &encryption_key, iv_enc, AES_ENCRYPT);

    unsigned char * decryption_output = new unsigned char[32];

    AES_cbc_encrypt(encryption_output, decryption_output, 32, &decryption_key, iv_dec, AES_DECRYPT);
}

What I do here is encrypt and then decrypt a message using openssl aes library. What I am concerned about is the length encryption_output. As far as my understanding goes, since AES encrypts in blocks of size AES_BLOCK_SIZE (aka 16 bytes) the number of output bytes should be equal to the size of the message, rounded up to the closest multiple of AES_BLOCK_SIZE. Is this correct? In particular, what happens if I extend the message to be exactly 32 bytes long? Will this still work, or will 16 empty padding bytes be added thus causing a segmentation fault when trying to write bytes 32 to 47 in encryption_output?

Solution

Proper PKCS#7 padding:

• rounds the length up to a multiple of the blocksize if it wasn´t a multiple before
• and it adds a whole block otherwise

Else, when decrypting, you couldn´t possibly know if the last ciperhtext block is "real" or only padding. (The actual byte values to pad with are specified too, but your real last block could contain these => again not possible to recognize it).

There are other schemes than PKCS#7, but this is not relevant here.

However, with AES_cbc_encrypt, you´ll have to implement this yourself, ie. pad before encrypting and remove the padding after decrypting. The encrypting itself will work with non-multiple lengths, but the used "padding" has the problem mentioned above. To answer your original question, AES_cbc_encrypt won´t add blocks, rounding up the length is the only thing it does.

For functions with proper padding (and without several other disadvantages of AES_cbc_encrypt, like missing AESNI support etc.etc.), look into the EVP part of OpenSSL. AES_cbc_encrypt is a more lowlevel part, depending on the situation it´s used by the highlevel function too.

Btw., something about C++: If you don´t get a segmentation fault,
it doesn´t mean that the code is correct.

Comments

[Matteo Monti]

Couldn't ask for a more complete answer, thank you. I was asking because I am exchanging fixed length messages between two endpoints. I already know what the length of the message is, so the whole padding thing is completely useless. In particular, I am sending and receiving those messages over the network and they are almost always multiples in length of the block size: I would have always been wasting those 16 bytes. Thanks again!

[Johannes Overmann]

Segfault comment: Yes! Even more, just because your program does what you expect (at the moment) does not mean it is correct.