[[!meta title="Architecture"]] [[!toc ]] # Assumptions The architecture has been designed under the following assumptions: * Privacy is important, and should be the default. People should only have access to the information they are authorized to access. * Members should be able to see their own information, without having to go through the Effi membership register admin. * It's not practical to have every member have a password. Authentication can be done by sending the member a unique, single-use link when they want to log in. # Components [[!img arch.png]] Effireg consists of four main components: * **effiapi** provides a RESTful HTTP API for managing the membership data, and for doing things with or to the data. All operations on the data go via the API. * **effiweb** provides the frontend for the web application to use the membership register. It renders HTML to the user, is accesses via a web browser, and generally is the user-visible part of Effireg. Everything effiweb does, it does by using the effiapi API. If any other user interfaces are added (say, mobile or command line), they will be parallel to effiweb and will also use the effiapi API in the same way as effiweb. * **Qvisqve** provides authentication of end-users (the members, and admins). It lets users log in, and keeps track of what each user is allowed to do. * **Muck-POC** stores the actual membership register data, and controls access to it, based on access tokens from Qvisqve. # Authentication [OpenID Connect]: https://openid.net/specs/openid-connect-core-1_0.html [OAuth2]: https://oauth.net/2/ End-users are authenticated using the [OpenID Connect][] protocol, specifically the authorization code flow. In this flow, Qvisqve provides cryptographically signed access tokens, which identify the user and specify a list of things the user may do. The signature guarantees the token comes from Qvisqve. To prevent the access token from leaking to the browser, effiweb keeps the token, and also manages user sessions. Since it is impractical to give each member a password, in a secure manner, Effireg will instead allow a member to request a unique, single-use login link to be emailed to them. The link will take the member's web browser to effiweb, which will extract a unique token from the link, and use that to authenticate the member for Qvisqve, and start a new session. As long as the session exist (until the member logs out), they can use the effiweb application. (A valied-until-logout session, without timeouts, is not ideal from a security perspective, but given the harm of an un-privileged member's hijacked session is minor, it's good enough. When Effireg develops further and the potential harm become bigger, session timeout will be implemented.) Non-interactive API clients are authenticated using the [OAuth2][] protocol, specifically using client credential grants. This also provides an access token, similar to the one from end-user authentication. API clients will not be available to normal members at this time, only members trusted by the board. Regardless of trust level, each member and API client will have a set of allowed operations that constrain what they can do. # Data model The membership register stores data as "resources" in Muck-POC. Each resource is a JSON object. The following types of objects are supported: * **subject** represents a person who is allowed to use the register; it it used to identify the user for authentication * **password** stores the encrypted password for a subject * **memb** represets a subject's membership in Effi ## Subject resource { "username": "007@mi6.example.com" } A subject resource has the following fields: * `username` — the username of the subject; this can change, the subject is identified by the resource identifier in the register, not by the username The subject resource does not have any data that isn't needed for end-user identification. Effiapi manages the subject resource via the Qvisqve API. Effiapi does not expose it via its own API. To allow members to log using their email address, the username is set to their primary email address. ## Password resource { "subject-id": "abcdef0123456789#, "version": 1, "hash": "xxxxx", "salt": "yyyyy", "key_len": 128, "p": 1, "r": 8, "N": 16384 } A password resource has the following fields: * `subject_id` — resource id of the subject whose password this is * `version` — version of the password resource (identifies algorithm); must be 1 * `hash` — the password, encrypted with the scrypt algorithm * `salt` — a random string to prevent dictionary attacks * `key_len` — parameter for scrypt * `N` — parameter for scrypt * `r` — parameter for scrypt * `p` — parameter for scrypt Effiapi manages the password resource via the Qvisqve API. Effiapi does not expose it via its own API. ## Member resource (memb) { "subject-id": "abcdef0123456789", "fullname": "James Bond", "primary-email": "007@mi6.example.com", "years-paid": [2010, 2011, 2012, 2018] } A membership resource has the following fields: * `subject-id` — the resource id of the subject resource for the member; this is assigned by effiapi (really Muck), when the resource is created, and can't be changed via the API * `fullname` — the full name of the member * `primary-email` — the primary email address for the member (and currently the only one) * `years-paid` — list of integers for the years for which the member has paid their membership Effiapi manages and uses the memb resource, and exposes it via its own API. Effiweb renders it for the user. # API The effiapi API is a RESTful, HTTP-based API using JSON. The only resource exposed is `memb`, via the `/memb` endpoint. The following operations are defined: * `POST /memb` — create a new member resource; this implicitly creates the subject resource as well, allowing the member to log in; the `subject-id` field is ignored, if set * `PUT /memb` — update an existing member resource; the whole resource is updated, not just selected fields * `GET /memb` — retrieve a member resource; the whole resource is returned In all responses, the following headers convey metadata about the resource: * `Muck-Id` — the internal, random identifier for the resource, assigned by the API * `Muck-Revision` — the current revision of the resource `GET` requests must use the `Muck-Id` header to specify which member resource is to be retrieved. `PUT` requests must use the `Muck-Id` header, and also the the `Muck-Revision` header, with the value showing the **current** revision. This prevents concurrent updates from updating with conflicting data. `PUT` is not allowed to change the `subject-id` field of a member resource. ## Search language for members In addition to managing individual member resources, the API provides a search function. Note that there is no way to list member resources; a search that matches all can be used instead. * `GET /search` — search for member resources matching the required criteria; the search language is described below The search request body should look like this: { "where": "data", "field": "email", "pattern": "007@mi6.example.com", "op": "==" } A search may return any number of results: { "resources": ["abcdef0123456789", "cafef00d"] } It is the API client's (effiweb's) responsibility to retrieve the resources given the id, and to do any more detailed selecting among the resources. It is only possible to search based on fields that are at the top level, and have a string value. The search will only match, and return, resources that the user is allowed to see.