Engine updating a key's ex data in sign/encrypt method?

[Matt Caswell]

On 22/07/2019 17:44, Lynch, Andrew wrote:
> Hi,
> 
> I am wondering whether it is ok to update a key's ex data from within an engine method that is using the key.  This is with both OpenSSL 1.1.1a and 1.0.2j.
> 
> In my engine's load_privkey() method I connect to the HSM and obtain handles for a session and the key object.  Both handles are then stored in the key's ex data using RSA_set_ex_data() or EC_KEY_set_ex_data().  They will later be released by the registered free_func.  The sign and/or encrypt methods use these handles when making their calls to the HSM API.
> 
> This works fine for single operations like an openssl x509 command line, but not for long running applications such as stunnel.  It appears that stunnel loads all configured keys and certs at startup and then waits for clients to connect.  If there is no activity for a while the HSM will time out my session handle and the next sign operation using that session fails.  I can catch the error and reconnect, but this results in a new session handle.  The next time a client calls the same service, that key's ex data still contains the old outdated session handle resulting in the same error all over again.
> 
> So whenever I renew a session handle in the sign method I need to update the ex data of the affected key with the new value.
> 
> This appears to work fine for EC keys by calling EC_KEY_set_ex_data() from within the engine's ecdsa_do_sign() method.  The same goes for RSA keys in rsa_private_decrypt() and rsa_private_encrypt(), but with rsa_sign() I have a problem in that the key parameter is const RSA *.
> 
> Is the lack of const in rsa_private_encrypt() and ecdsa_do_sign() an oversight and I should not be manipulating ex data of the provided key in this way?
> Or is the const in rsa_sign() too strict and it is actually ok to modify the ex data of the key?

I can't see a problem with modifying the ex_data like that.

Matt