Question

I’m working on creating small relay stations out of tiny embedded Linux boxes. They have some sensors connected to them and transport data back to a server via HTTP POST. Right now the server just accepts their message, along with a unique ID (the MAC address of eth0).

I want to expand this to include some type of security. I want to be able to deploy these little devices with minimal configuration. I’d like to copy a base firmware to the device, hook them up in the field, and they self-register. The first time they connect, I’d like the server and device to do some type of negotiation where I can store a fingerprint. Subsequent requests I could then authentication/verify the device using that fingerprint.

That way, once a device registers with its unique ID, I can be assured all data from that ID is from the same device. If a rouge device or set of devices does register, I’ll just delete them (I store IPs to so I can delete by unknown ranges and block them).

My question is what’s the best way to go about doing this? I think back to the idea of SSH fingerprints, where the first time you connect to a server you get a server fingerprint. If a future request yields a different fingerprint, you get a huge warning and have to manually delete the fingerprint out of your authorized_keys file if the server’s keys have actually regenerated (e.g. you did a reinstall without saving your old SSH keys).

Is something like this possibly with HTTP, possibly avoiding having to use preshared keys?

If it matters, the clients are running Python2 and the server they connect to is written mostly in Scala on Tomcat.

Answer 1

Basically, all you need to do is tell the server the public key, and then sign all of your messages with it. If you don’t want pre-shared keys, then the server cannot be assured that someone new who is registering is actually one of your devices. You can still validate that the message came from the same device that originally registered with that identifier, however.

The process basically goes like this:

1. Client generates a new key pair (e.g. an RSA public/private key pair).
2. Client registers with server, sending its public key. The server stores this public key.
3. When the client sends a message, it generates a signature of its message, which it attaches to the message. When the server receives the message, it validates the signature to ensure that the message was sent by someone holding the corresponding private key.

The code for this in PyCrypto goes something like this:

Generate key pair

from Crypto.PublicKey import RSA
key = RSA.generate(2048)
private_key = key.exportKey()
public_key = key.publickey().exportKey()
# private_key is a string suitable for storing on disk for retrieval later
# public_key is a string suitable for sending to the server
# The server should store this along with the client ID for verification

Generate signature

from Crypto.PublicKey import RSA
from Crypto.Hash import SHA
key = RSA.importKey(private_key)
# where private_key is read from wherever you stored it previously
digest = SHA.new(message).digest()
signature = key.sign(digest, None)
# attach signature to the message however you wish

The server should load the public key as it has previously stored, and use a “verify” method provided by the Scala/Java crypto API you use, and accept the message only if it succeeds.

It is important to understand the caveats of each approach, as various techniques only protect against certain types of attacks. For instance, the above approach does not protect against a “replay attack”, in which an attacker records a message with a certain meaning and then re-transmits it to the server at a later time. One way of protecting against this would be to include a timestamp in the message which is hashed; another would be to use an appropriately encrypted transport (e.g. SSL/TLS).