Hacker News new | ask | show | jobs
by CiPHPerCoder 3569 days ago
> The only whitepaper I could find for the whole system was for ShadowChat, which says they use AES-256-CBC to encrypt the chat.

Okay.

> AES is a symmetric encryption algorithm, and a strange choice for a chat protocol.

No, it's the standard choice.

Signal uses AES-256-CBC + HMAC-SHA-256, for example.

> Furthermore, CBC is not great at hiding patterns, and chat will often have a lot of patterns.

That's not true. CBC mode isn't great at hiding patterns with IV reuse. Solution: Don't reuse IVs.

The problem with CBC mode is that it's unauthenticated, and as a result is often susceptible to padding oracle attacks that allow messages to be decrypted byte-at-a-time.

https://paragonie.com/blog/2015/05/using-encryption-and-auth...

A better mode would be GCM, or maybe GCM-SIV, or scrap AES entirely in favor of ChaCha20-Poly1305.
1 comments
kewde 3569 days ago
It's not voiding the cryptographic doom principle, that's one thing. The HMAC is dealt by OpenSSL it seems and is the hash of the timestamp, destination and encrypted payload. It's appended outside of the encrypted payload. https://moxie.org/blog/the-cryptographic-doom-principle/ https://github.com/shadowproject/shadow/blob/master/src/smes...

On the other hand I agree to switching over to a different mode to eliminate padding oracle attacks.

The IV are 16 random bytes generated by OpenSSL's RAND_bytes. https://github.com/shadowproject/shadow/blob/master/src/smes...

link
CiPHPerCoder 3569 days ago
If one of the project devs is reading this thread:

https://github.com/shadowproject/shadow/blob/f3fa333f8377688...

That's really hard to read.

Also, don't use OpenSSL: https://paragonie.com/blog/2016/05/how-generate-secure-rando...

link
kewde 3569 days ago
Hi,

Yes we're always where the critics are, they are or best source of information.

I'm very happy to see our work being reviewed. I'm not an expert cryptographer but I am capable of understanding it. (fyi I didn't code it).

Our memcmp in constant time is not the prettiest, but it's short so we roll with it :P

This project started around 2014, LibSodium was still very small back then and OpenSSL, in ours view, remains the defacto standard. Is there any particular reason on why we should move away from RAND_bytes() ?

link
CiPHPerCoder 3569 days ago
Yes: Aside from being a userspace CSPRNG (which is an additional risk of failure over the kernel's CSPRNG and doesn't provide defense-in-depth), it isn't thread-safe.

https://github.com/ramsey/uuid/issues/80

https://github.com/nodejs/node/issues/5798

There are also some recent IACR papers (linked in the Node thread), but those are the two biggest concerns.

link
kewde 3569 days ago
Thank you CiPHPerCoder!

That link to NodeJS was a good read, I'm fairly convinced that we should ditch RAND_bytes from OpenSSL for something more secure, we'll look into LibSodium.

I've caught rumours of a possible RAND_sys_bytes which operates over the systems CSPRNG? We like to be conservative on the libs.

We'd like to tip you for your efforts, do you have a Bitcoin (or ShadowCash) address?

link
CiPHPerCoder 3569 days ago
I do, actually, but I haven't accessed it in almost two years. EDIT: And I forgot my password. Here's a new one:

  1D6RMsgnTAf2GLWpD91EaQvQ5ppuaZKkGT
link