Certificate-Based Security

this section contains an advanced discussion of ditto's underlying certificate, identity, and encryption implementation most readers can skip this section however, if you are deploying an enterprise on premises deployment of ditto, you may be required to reference the following material all communications are consistently protected by modern and robust encryption for all of ditto's communication methods cryptographically signed business rules ensure users can only sync data that they are permitted to access the app developer is in complete control of the keys, certificates, and rules the ditto big peer provisions a signed x509 certificate this certificate may then be presented to other ditto peers to mutually establish trust and create encrypted communications channels the following details information about the certificates used to encrypt the communication channels capability type encryption tls 1 3 authentication ec key pairs with signed certificates trust infrastructure x 509 with a developer controlled certificate authority access rules query patterns on document id's describing read and or write access identities an identity is a bundle of the device and app specific information production development site id allocated by central authority defaults to a random number; can be customized app id set by central authority for example, "5322afcd 5a70 43a3 bc2d 85d98ccf5ac0" access rules set by central authority all devices may read/write all documents private key either generated on device or distributed by central authority hard coded and shared by all devices identity certificate unique and signed by a central authority; contains this device's public key hard coded and shared by all devices ca certificate shared by all users of the same app hard coded and shared by all devices certificates ditto identities and public keys are distributed in the standard x 509 certificate format they do not directly contain potentially sensitive data such as access rules, but these can be defined by the app's authentication webhook with the , or by the developer through the manual identity when you are ready to use production identities, feel free to contact us through the ditto portal and we will help you set up the right ca tooling for your use case or provide specifications so you can build your own discovering peers devices need to have the same appid to discover other peers on the network, as well as matching certificates to connect over tls 1 3 peer to peer connections use mtls (client certificates) with tls 1 3 connections to the big peer use a tls secured websocket connection, with authentication by jwt once the certificates match, then the embedded authorization information inside each certificate is used to authorize any incoming requests by that peer this ensures that those access control rules are enforced syncing with big peer the description below covers internal details of the big peer implementation ditto’s authentication module handles this implementation for you under the hood as part of the onlinewithauthentication and onlineplayground identities overview online authentication can be used when a ditto application is hosted on a big peer the big peer hosts an https identity service, which handles login requests in online authentication mode, a small peer must log in with credentials before it can communicate with any peers the login flow populates the small peer with valid authentication material that identifies the user and defines their level of access once the small peer receives this data after a successful login, the transports which depend on it will start automatically complete a peer key challenge the peer’s public key will be included in the certificates returned by the identity service the identity service needs proof that the authenticating small peer actually holds the corresponding private key small peer downloads a challenge token from / ditto/auth/cert this is a time limited jwt which the client treats as opaque data small peer uses their peer key to sign it log in with credentials when a client attempts to authenticate, it will make an https request to the identity service containing the following payload signed challenge app id provider name credentials to be forwarded to the app's once the identity service has a response from the auth webhook, it will build and sign a jwt in addition to the information we would expect to find in a jwt (issuer, issued at, expiry, etc), we also include an encoded representation of the ditto identity the ditto identity is built from the information provided by the app's authentication webhook response upgrade to an x 509 certificate once a peer has a jwt, it will want to use this to get a device certificate the peer will now make an https certificate request to the identity service the “request certificate” endpoint requires a valid jwt the contents of the new certificate is closely tied to the jwt the certificate expiration time is the same as the jwt, and the ditto identity embedded in the certificate is pulled from the jwt the identity service generates a private key for the device it uses this private key to generate a csr on behalf of the client and then the identity service’s ca keys to sign it once this information is prepared, the identity service is ready to respond to the request the response will include the client's generated certificate and private key, a list of ca certificates the client should trust, and a datetime for expiration the client will persist the jwt, device private key, device certificate, and ca certificates on disk and only update them when they get near expiration when two peers authenticate with each other, they can use either the jwt or the device certificate via mtls the decision depends on the client's capabilities and connection type rationale why does x 509 return both a key and a certificate instead of locally generating a key and sending a csr? it would be possible to use a standard csr flow it was chosen to issue keys directly for a few reasons there is no security benefit as our certificate request is in a secured tunnel, and the identity service is presumed to be completely trustworthy validating and signing csrs is more complex than simply creating one with the correct format and fields this is a convenient workflow if using hashicorp vault or similar to manage your pki and issue certificates on demand why is x 509 a separate request? the idea is that jwts could be quite short lived but the x 509 certs could be long lived to permit long periods of working offline, so the small peer could call them to refresh at different intervals different expiry periods between jwt and x 509 isn’t implemented yet additionally, browser peers have no use for x 509 certificates so they don’t request them of course, the small peer could have a single request that lets you request some combination of identities at once there is no reason this couldn’t be added in the future to reduce latency