Protocol token¶
The following chapter describes the protocol token structures of the neuropil cybersecurity mesh.
Token in general¶
Within the neuropil library we us the aaatoken structure to fulfil authentication, authorization and accounting purposes. The usage and meaning of each token for/in each specific case is ambiguously and sometimes confusing (even for us).
This chapter will try to add the necessary details how tokens are used. Let us recall first how the token structure is composed before diving into the details:
// depicts the protocol version used by a node / identity
double version;
// who is the owner
char realm[64];
// who issued this token
char issuer[64];
// what is this token about
char subject[255];
// who is the intended audience
char audience[64];
// token creation time
double issued_at;
// to be used at
double not_before;
// not to be used after
double expires_at;
// internal state of the token
aaastate_type state;
// uuid of this token
char* uuid;
// public key of this token
unsigned char public_key[crypto_sign_PUBLICKEYBYTES];
// private key of this token (not to be send over the wire
unsigned char private_key[crypto_sign_SECRETKEYBYTES];
// signature of the token
unsigned char signature[crypto_sign_BYTES];
// key/value attribute list
np_tree_t* attributes;
One of the most important aspects is the use of the realm, issuer, subject and audience fields that interact with each other and reference each other as the library build up its network structure. Therefore we will concentrate on these four for the following chapter.
Protocol token format¶
Note
The protocol token format is work in progress. please be aware of possible changes
A draft schematic format with 32‑byte rows.
0 8 16 24
| | | | (Bytes)
+-----------------------------------------------------------+
| signature |
| |
+-----------------------------------------------------------+ ^
| public_key | :
+-----------------------------------------------------------+ :
| realm | :
+-----------------------------------------------------------+ :
| issuer | :
+-----------------------------------------------------------+ :
| subject | Signature
| ... |
+-----------------------------------------------------------+ covered
| audience | :
+--------------+--------------+--------------+--------------+ :
| issued_at | not_before | expires_at | attr_length | :
+--------------+--------------+--------------+--------------+ :
| version | tbd | | | :
+--------------+--------------+--------------+--------------+ :
| | :
~ attributes ~ :
| | :
+-----------------------------------------------------------+ v
In addition to specifying a layout we must define the format and semantics of the individual fields, for example:
- signature
64 bytes of signature, computed over all following bytes.
- public_key
32 bytes of public key material.
- realm, issuer, audience
32 bytes of digest, see np_get_id or np_get_fingerprint
- subject
the subject this toke is about. Should use urn: syntax style. (e.g. urn:np:node:… or urn:np:id:…)
- issued_at, not_before, expires_at
IEEE 754 double-precision floating-point numbers in big endian format (network byte order).
- version
The version of the neuropil library being used.
- attribute_length
Unsigned 64‑bit integer in big endian format.
- attributes
A variable length, MessagePack encoded map. No requirements on alignment.
Handshaking token¶
For the handshake we need to send some core information to the other node, so the above mentioned fields will contain:
realm := <empty> | <fingerprint(realm)> 64
issuer := <empty> | <fingerprint(issuer)> 64
subject := 'urn:np:node:<protocol>:<hostname>:<port>' 255
audience := <empty> | <fingerprint(realm)> | <fingerprint(issuer)> 64
attributes := { <session key for DHKE> } 64
public_key := <pk(node)> 32
signature := <signature of above fields excluding attributes> 64
signature_ext := <signature of all above fields> 64
-----
max 666 bytes
Please remember that the main purpose here is to establish a secure conversation channel between any two nodes. The clear text hostname and port could also be found by doing a network scan. Furthermore we have to keep the token size small to fit into the 1024 bytes bounds of our selected paket size. Only one paket as an initial handshake message is allowed.
From the above structure you can create a node fingerprint (nfp), which is unique to this specific token. This fingerprint again is used as the visible part of the DHT which can be addressed.
nfp = hash(nodetoken, signature)
similar we could create a handshake fingerprint (which we do not need, it is just here to complete the picture):
hfp = hash(nodetoken, signature_ext)
A separate node token will be supplied in the join message to verify the use of a potential identity.
Join token¶
The join message contains the token of the identity which is using a node. Identity token can be exported and imported and are available in the userspace.
realm := <empty> | <fingerprint(realm)> 64
issuer := <empty> | <fingerprint(issuer)> 64
subject := 'urn:np:id:'<hash(userdata)> 255
audience := <empty> | <fingerprint(realm)> | <fingerprint(issuer)> 64
attributes := { _np.partner_fp: nfp, <?user supplied data> } min 64
public_key := <pk(identity)> 32
signature := <signature of above fields excluding attributes> 64
signature_ext := <signature of all above fields> 64
----
min 666
Again we can create a fingerprint of this token (‘info’). This fingerprint is not the same as the fingerprint of a pure identity (ifp), as we do not know in advance which ‘nfp’ this idenity will use. A pure identity token of does not contain the ‘nfp’. But we can still calculate the fingerprint afterwards, because:
ifp = hash(idtoken, signature) infp = hash(idtoken, signature_ext, nfp)
all signatures can be validated using the public keys of tokens that have been received.
‘nfp’ potentially contains the issuing fingerprint in the issuer field again. But if a technical node hosts more than one identity, then the join message will also contain again the node token, this time in full length and containing the required identity fingerprint:
realm := <empty> | <fingerprint(realm)>
issuer := <empty> | <fingerprint(issuer)>
subject := 'urn:np:node:<protocol>:<hostname>:<port>'
audience := <empty> | <fingerprint(realm)> | <fingerprint(issuer)>
attributes := { <?identity: ifp>, <?user supplied data> }
public_key := <pk(node)>
signature := <signature of above fields excluding attributes>
signature_ext := <signature of all above fields>
The second transmit of the node token is needed to certify that this identity is really running on this specific node, a kind of automated cross-signing between node and identity. We could add the handshake fingerprint to this token to make it really foolproof, but currently we do not think that it would be necessary.
Please note that in the case of a pure technical node that should support the network we will only transmit the node token again in the join message. The reason for doing so is the authentication callback, which is only triggered when sending a join message.
Message intent token¶
If an identity would like to exchange information with another identity in the network, it sends out its message intents, where we use token again.:
realm := <empty> | <fingerprint(realm)>
issuer := <ifp>
subject := 'urn:np:sub:'<hash(subject)>
audience := <empty> | <fingerprint(realm)> | <fingerprint(issuer)>
attributes := { _np.partner_fp: nfp, <mx properties>, <?user supplied data> }
public_key := <pk(identity)>
signature := <signature of above fields excluding attributes>
signature_ext := <signature of all above fields>
Please note that a message intent is somehow different, as you may get a message intent of an identity that your node may not have any connection to. So first you need to authenticate the issuer of this message intent. you can accomplish this by doing one of the three steps:
you implement a callback that is able to properly authenticate peers (e.g. using MerkleTree / Secure Remote Password / Shamirs shared secret schemes / …)
you forward the received token to do the authn work for your node: either to your own realm, or to the realm set in the message intent, or you ask the partner fingerprint contained in the token whether the identity is really known
you do some sort of out-of-band deployment for know public identity tokens. you could even use neuropil itself to inject a trusted public identity token into a device.
Once you know, that the received peer is the correct one, you do the second step and authorize the message exchange. Again you have the three options above with the following restriction to the second choice:
you forward the received token to do the authz work for your node to your own realm
PKI / Web of trust / zero knowledge setups¶
Sometimes it is desirable to choose a pki setup for the tokens that you use. For this case the issuer field of the token structure can be used. It indicates whether a token has been signed by another party. There is no pre-defined setup for this kind of , but the usual setup as you know it from certificates is required. Especially you will have to add your signature token to the attributes of an identity token.
One interesting feature of the neuropil message layer is the use of fingerprints as hash values, which are addressable via the DHT. Without any further configuration we can exchange message intents with a realm or an issuer, and there can be only one identity in the whole DHT which is able to create such identity or message intent tokens.
Therefore we (ourselves) favor the use of realms, because it lets you create ‘online’ registration instances without pre-issuing and deploying public tokens. A fingerprint of an identity token is enough to identify the right partner or to find a third party (realm) who is willing to prove the authenticity of a device, application or person. In a similar way you can remote control your devices, because for authorization requests each device, application or person is able to contact your realm for allowance.
The missing accounting tokens¶
The chapters above have described the measures how you can authenticate and authorize token, but we have not yet covered how you can use tokens for accounting purposes. But basically it is very easy.
An identity e.g. could create and send an accounting token for the messages and message intents it has received, just by copying its own message intent
realm := <empty> | <fingerprint(realm)>
issuer := <ifp>
subject := 'urn:np:sub:'<hash(subject)>
audience := <empty> | <fingerprint(realm)> | <fingerprint(issuer)>
attributes := { _np.partner_fp: nfp, <mx properties>, <?user supplied data> }
public_key := <pk(identity)>
signature := <signature of above fields excluding attributes>
signature_ext := <signature of all above fields>
Under ‘user supplied data’ you can add any content, for example the message intents that you have received from your peers, plus the actual usage of your/their token (it’s a json structure). The main difference towards your initial message intent token is the address that you’re sending this token to.
Similar each node on the network can record received messages and
realm := <empty> | <fingerprint(realm)>
issuer := <nfp>
subject := 'urn:np:sub:'<hash(subject)>
audience := <empty> | <fingerprint(realm)> | <fingerprint(issuer)>
attributes := { _np.partner_fp: nfp, <mx properties>, <?user supplied data> }
public_key := <pk(identity)>
signature := <signature of above fields excluding attributes>
signature_ext := <signature of all above fields>
in this case the section ‘user supplied data’ would contain the uuid of each message(part) that a single node has received and forwarded. Most important are the nodes which do the message intent matching ! These nodes act as a technical attesting notary that confirms the exchange of message intents. plus it could also confirm the abuse of message intents.
Once an accounting token is ready it will be send to your own accounting realm (and this could be a different one than your authn/authz realm), the token and it’s contents can be analyzed and store in a database i.e. for monitoring purposes. Once you put all distributed accounting tokens together, you will be able to see how your messages have travelled through the DHT (via the uuid).
Conclusion¶
You can create arbitrary complex hierarchical token constructs and facilitate them in the way we have described them. Please do not overdo it! Setting up a new realm and rejoining your devices and applications is easier than creating complex pki hierarchies, and it can be done online ! Try this with certificates/pki and you know that you have fallen into a trap …