On 01/04/2020 18:01, Andrew Felsher wrote:
> I'm aware of the symmetric/asymmetric differences. But the EVP_Enrypt...
> API takes the key as a bunch of bytes. It shouldn't care whether it's an
> RSA key or not, correct? (Though perhaps it would truncate to, in my
> case, the first 128 bits since I'm using AES-128-XTS.)
No - you must not do that.
On a side note XTS is a funny mode - it requires a double length key. So
AES-128-XTS requires you to supply a 256 bit key. See the description of
XTS on this page:
https://www.openssl.org/docs/man1.1.1/man3/EVP_aes_128_xts.html
On a 2nd side note - XTS has another quirk that you should be aware of.
It does not support streaming. That is you should only call
EVP_EncryptUpdate() once per EVP_EncryptInit_ex() call. The "IV" in the
EVP_EncryptInit_ex() parameter list is actually the XTS "tweak".
>
> But assuming that I can't somehow use the RSA private key's bits as my
> symmetric key, how should I generate a key without requiring user
> interaction? Simply calling RAND_bytes?
Preferrably symmetric keys are generated via RAND_priv_bytes() - but
RAND_bytes() would also work.
It's not clear from your description though how you intend to use the
RSA key in this case. Perhaps just to encrypt the symmetric key you
generated above?
Matt
>
> -Andrew
>
> On Wed, Apr 1, 2020 at 12:53 PM Matt Caswell <m...@openssl.org
> <mailto:m...@openssl.org>> wrote:
>
>
>
> On 01/04/2020 17:34, Andrew Felsher wrote:
> > Hi,
> >
> > I'm trying to do what I assumed would be a very common and typical use
> > of OpenSSL. I'm just encrypting and decrypting some data (in code; not
> > from command line). EVP_EncryptInit_ex (and decrypt, update, and final
> > variants) are the standard way to do this.
> >
> > However, the init functions take a char buffer. All the examples I can
> > find use hard-coded char buffer keys. But obviously I'm not going
> to be
> > hard-coding my keys. I'm generating them through a couple different
> > means, but ultimately I have EVP_PKEYs (in my case, containing RSA
> > private keys). And there doesn't seem to be a straightforward way
> to go
> > from EVP_PKEYs to a form consumable by the EVP init functions.
> >
> > EVP_PKEY_get_raw_private_key looks like it would be perfect, but
> it was
> > introduced in 1.1.1 and I'm limited to 1.1.0.
> >
> > This seems like it would be a very common use case, yet I can't
> seem to
> > find any examples or documentation anywhere.
> >
> > Am I doing something wrong or making some really off-base assumptions?
>
> EVP_EncryptInit_ex does *symmetric* encryption, i.e. both sides of the
> communication share the same private key. An RSA key in an EVP_PKEY does
> *asymmetric* encryption. Typically anyone can encrypt using the public
> RSA key, but only the owner of the private RSA key can decrypt.
>
> Asymmetric encryption using RSA is limited because it can only encrypt a
> very small amount of data. It is also very slow compared to symmetric
> encryption. Therefore, typically, if you want to encrypt data via an RSA
> key you generate a random symmetric key and encrypt that using your RSA
> key. Then you encrypt the data using the randomly generated
> symmetric key.
>
> OpenSSL has a built in high level API for doing this combined operation.
> To encrypt you can use the EVP_Seal*() functions. To decrypt use the
> EVP_Open*() functions. See:
>
> https://www.openssl.org/docs/man1.1.1/man3/EVP_SealInit.html
> https://www.openssl.org/docs/man1.1.1/man3/EVP_OpenInit.html
>
> It is also possible to encrypt directly in RSA using EVP_PKEY_encrypt():
>
> https://www.openssl.org/docs/man1.1.1/man3/EVP_PKEY_encrypt.html
>
>
> Matt
>