I have seen this done for hardware acceleration; where the crypto chip can do 
everything except the handshake.
(In fact, this mechanism protected at least one device that I know of from the 
Heartbleed debacle, since the hardware crypto did not understand the record 
type.)

Look at how the kernel handles TLS, and how the keys are extracted from OpenSSL:

https://github.com/torvalds/linux/blob/master/Documentation/networking/tls.txt
https://github.com/openssl/openssl/pull/5253

--
-Todd Short
// [email protected]<mailto:[email protected]>
// "One if by land, two if by sea, three if by the Internet."

On Nov 14, 2018, at 11:28 AM, Viktor Dukhovni 
<[email protected]<mailto:[email protected]>> wrote:



On Nov 14, 2018, at 6:54 AM, Hemant Ranvir 
<[email protected]<mailto:[email protected]>> wrote:

My main goal here is to use openssl for initial handshake sequence. Once the 
connection is established between server and client, decrypt the incoming 
message (this time not using the openssl api but rather by using the decrypt 
AES function implemented earlier)

This makes no sense, because TLS does not just emit a simple CBC encrypted 
stream
after performing the handshake.  So you can't do that.  Use 
SSL_read()/SSL_write,
and let the library do the message decryption/encryption for you.  When done use
SSL_shutdown() to cleanly terminate the stream, and depending on the application
protocol, make wait for the peer's SSL_shutdown() in turn to avoid truncation
attacks where completion of the stream is not implied by the higher level 
protocol.

--
Viktor.

--
openssl-users mailing list
To unsubscribe: https://mta.openssl.org/mailman/listinfo/openssl-users

-- 
openssl-users mailing list
To unsubscribe: https://mta.openssl.org/mailman/listinfo/openssl-users

Reply via email to