Luzifer / nginx-sso

This program is intended to be used within the ngx_http_auth_request_module of nginx to provide a single-sign-on for a domain using one central authentication directory.

Usage

You can use the luzifer/nginx-sso docker image to start your SSO service. On first start an example configuration will be generated and after you've changed that configuration you can start the container:

# docker run -d -p 127.0.0.1:8082:8082 -v /data/sso-config:/data luzifer/nginx-sso

After you did this you need to configure your nginx to use the SSO service:

server {
  listen        443 ssl;
  server_name   kibana.hub.luzifer.io;

  ssl_certificate     /data/ssl/certs/luzifer.io.pem;
  ssl_certificate_key /data/ssl/certs/luzifer.io.key;

  # Redirect the user to the login page when they are not logged in
  error_page 401 = @error401;

  location / {
    # Protect this location using the auth_request
    auth_request /sso-auth;

    ## Optionally set a header to pass through the username
    #auth_request_set $username $upstream_http_x_username;
    #proxy_set_header X-User $username;

    # Automatically renew SSO cookie on request
    auth_request_set $cookie $upstream_http_set_cookie;
    add_header Set-Cookie $cookie;

    proxy_pass http://127.0.0.1:1720/;
  }

  # If the user is lead to /logout redirect them to the logout endpoint
  # of ngninx-sso which then will redirect the user to / on the current host
  location /logout {
    # Another server{} directive also proxying to http://127.0.0.1:8082
    return 302 https://login.luzifer.io/logout?go=$scheme://$http_host/;
  }

  location /sso-auth {
    # Do not allow requests from outside
    internal;
    # Access /auth endpoint to query login state
    proxy_pass http://127.0.0.1:8082/auth;
    # Do not forward the request body (nginx-sso does not care about it)
    proxy_pass_request_body off;
    proxy_set_header Content-Length "";
    # Set custom information for ACL matching: Each one is available as
    # a field for matching: X-Host = x-host, ...
    proxy_set_header X-Origin-URI $request_uri;
    proxy_set_header X-Host $http_host;
    proxy_set_header X-Real-IP $remote_addr;
    proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
    proxy_set_header X-Forwarded-Proto $scheme;
    proxy_set_header X-Application "kibana";
  }

  # Define where to send the user to login and specify how to get back
  location @error401 {
    # Another server{} directive also proxying to http://127.0.0.1:8082
    return 302 https://login.luzifer.io/login?go=$scheme://$http_host$request_uri;
  }
}

To implement a logout you can send the user to the /logout?go=<url> endpoint which will ensure the cookie-stored login will be erased.

Configuration

The configuration is mainly done using a YAML configuration file. Some options are configurable through command line flags and can be looked up using --help flag.

For an example configuration see the config.yaml file in this repository. Within the next sections the options are explained in more detail:

Main configuration: Login form

The login form can be customized with its wording and the default login method.

login:
  title: "luzifer.io - Login"
  default_method: "simple"
  hide_mfa_field: false
  names:
    simple: "Username / Password"
    yubikey: "Yubikey"

Most options should explain themselves, the names dictionary maps IDs of the authentication methods (shown in the title of their config section below) to human readable strings. You can set any string you need and your user recognizes.

In case you don't want to show up the "MFA Token" fields even though the providers being used does support them (for example if you are not using MFA and don't want to confuse your users) you can set the hide_mfa_field flag to hide them.

Main configuration: Cookie Settings

Most of the cookie settings are pre-set to sane defaults but you definitly need to configure some.

cookie:
  domain: ".example.com"
  authentication_key: "Ff1uWJcLouKu9kwxgbnKcU3ps47gps72sxEz79TGHFCpJNCPtiZAFDisM4MWbstH"
  expire: 3600        # Optional, default: 3600
  prefix: "nginx-sso" # Optional, default: nginx-sso
  secure: true        # Optional, default: false

Adjust the domain to your service. So if all of your services live under *.luzifer.io you want to set the domain to .luzifer.io. The authentication_key needs to be set to some unique string not known to others. It is used to validate nobody messed with your session cookies. If this is leaked (or you just used the default) attackers can just set any username inside the corresponding cookie and are able to access your services!

If you are accessing your services through HTTPs you want to enable secure cookies. Also you should think about customizing the cookie prefix and the expire time of the cookie.

Main configuration: HTTP Listener

This section configures where you can reach the program using HTTP and where you will point your nginx to. The example below shows the defaults and you don't need to change them.

listen:
  addr: "127.0.0.1"
  port: 8082

Pay attention if you are running the docker container you need to change the IP to 0.0.0.0 to expose the port in the container. If you miss this the service will not be available.

Main configuration: Audit Logging

nginx-sso can be configured to write an audit log which for example can be used to detect brute-force attacks on passwords. By default the audit logging is disabled and gets enabled by providing targets in the audit_log section of the config.

Due to the fact we do have several credential providers the login and logout events do not contain an username. The username is only known in requests the user is doing after their login so the access_denied and validate events contains it.

audit_log:
  targets:
    - fd://stdout
    - file:///var/log/nginx-sso/audit.jsonl
  events: ['access_denied', 'login_success', 'login_failure', 'logout', 'validate']
  headers: ['x-origin-uri']
  trusted_ip_headers: ["X-Forwarded-For", "RemoteAddr", "X-Real-IP"]

• targets - required - Supported targets are fd://stdout, fd://stderr or any file://... URI
• events - required - All supported events are listed above in the example. Pay attention validate is a quite verbose event
• headers - optional - List of headers to include into the log entry (for details about the headers see the ACL section below)
• trusted_ip_headers - optional - List of headers to use for reading the real IP the request is coming from (defaults see example above)

Main configuration: ACL

The rules of the ACL are the most complex part of the configuration and you should take your time to make this bullet-proof. If you mess up you're probably are getting complaints from your users because the default policy applied is to deny all access. So in the end you are configuring a white-list here.

acl:
  rule_sets:
  - rules:
    - field: "host"
      equals: "test.example.com"
    - field: "x-origin-uri"
      regexp: "^/api"
    allow: ["luzifer", "@admins"]

Each rule_sets entry consists of three parts: rules, allow and deny directives. You can supply as many rules as you need, they are connected using AND logic per rule-set.

Each rules entry has two mandantory and three optional fields of which at least one must be set:

• field - required - Selector of the header your nginx is sending to the /auth endpoint (e.g. Host, X-Origin-URI, ...)
• invert - required - Boolean used to invert the matching: What was true will be false. Useful for "does not match this regexp" rules (default: false)
• present - optional - Boolean stating a certain header must exist or must not exist
• regexp - optional - String containing a regexp which must match the contents of the header selected by field
• equals - optional - String which must fully match the contents of the header selected by field

The allow and deny directives are arrays of users and groups. Groups are prefixed using an @ sign. There is a simple logic: Users before groups, denies before allows. So if you allow the group @test containing the user mike but deny the user mike, mike will not be able to access the matching sites.

MFA Configuration

Each provider supporting MFA does have some kind of configuration for the MFA providers. As there are multiple MFA providers the configuration sadly isn't that simple and needs to have the following format:

provider: <provider name>
attributes:
  <mapping of attributes>

Duo

This provider needs a configuration to function correctly:

mfa:
  duo:
    # Get your ikey / skey / host from  https://duo.com/docs/duoweb#first-steps
    ikey: "<IKEY>"
    skey: "<SKEY>"
    host: "<API HOST>"
    user_agent: "nginx-sso"

The corresponding expected MFA configuration is as following:

provider: duo

Google Authenticator

The provider name here is google while the only supported argument at the moment is secret. The secret is what you need to provide to your users for them to add the config to their authenticator. (It MUST be base32 encoded!)

Here is an example of the URI to provide in a QRCode:

provider: google
attributes:
  secret: MZXW6YTBOIFA

otpauth://totp/Example:myusername?secret=myverysecretsecret (Docs)

Yubikey

This provider needs a configuration to function correctly:

mfa:
  yubikey:
    # Get your client / secret from https://upgrade.yubico.com/getapikey/
    client_id: "12345"
    secret_key: "foobar"

The corresponding expected MFA configuration is as following:

provider: yubikey
attributes:
  device: ccccccfcvuul

Provider configuration: Atlassian Crowd (crowd)

The crowd auth provider connects nginx-sso with an Atlassian Crowd directory server. The SSO authentication cookie used by Jira and Confluence is also used by nginx-sso which means a login in Jira will also perform a login on nginx-sso and vice versa.

providers:
  crowd:
    url: "https://crowd.example.com/crowd/"
    app_name: ""
    app_pass: ""

The configuration is quite simple: Create an application in Crowd, enter the Crowd URL and the application credentials into the config and you're done.

Provider configuration: LDAP Auth (ldap)

The LDAP provider connects to a (remote) LDAP directory server and authenticates users against and reads groups from it.

providers:
  ldap:
    enable_basic_auth: false
    manager_dn: "cn=admin,dc=example,dc=com"
    manager_password: ""
    root_dn: "dc=example,dc=com"
    server: "ldap://ldap.example.com"
    # Optional, defaults to root_dn
    user_search_base: ou=users,dc=example,dc=com
    # Optional, defaults to '(uid={0})'
    user_search_filter: ""
    # Optional, defaults to root_dn
    group_search_base: "ou=groups,dc=example,dc=com"
    # Optional, defaults to '(|(member={0})(uniqueMember={0}))'
    group_membership_filter: ""
    # Replace DN as the username with another attribute
    # Optional, defaults to "dn"
    username_attribute: "uid"
    # Configure TLS parameters for LDAPs connections
    # Optional, defaults to null
    tls_config:
      # Set the hostname for certificate validation
      # Optional, defaults to host from the connection URI
      validate_hostname: ldap.example.com
      # Disable certificate validation
      # Optional, defaults to false
      allow_insecure: false

To use this provider you need to have a LDAP server set up and filled with users. The example (and default) config above assumes each of your users carries an uid attribute and groups does contains member or uniqueMember attributes. Inside the groups full DNs are expected. For the ACL also full DNs are used.

• enable_basic_auth - optional - Allows automated clients to pass credentials using basic auth instead of using the login form
• manager_dn - required - A LDAP account which is allowed to list users and groups (it needs no access to the password!)
• manager_password - required - The password for the manager_dn
• root_dn - required - The base of your directory
• server - required - Connection string to the LDAP server in format ldap[s]://<host>[:<port>]
• user_search_base - optional - Using this parameter you can limit the user search to a certain sub-tree. Within this sub-tree the uid must be unique (as the name already states). If unset the root_dn is used here
• user_search_filter - optional - The query to issue to find the user from its uid ({0} is replaced with the uid). If unset the query (uid={0}) is used
• group_search_base - optional - Like the user_search_base this limits the sub-tree where to search for groups, also defaults to root_dn
• group_membership_filter - optional - The query to issue to list all groups the user is a member of. The DN of each group is used as the group name. If unset the query (|(member={0})(uniqueMember={0})) is used ({0} is replaced with the users DN, {1} is replaced with the content of the username_attribute)
• username_attribute - optional - The attribute containing the username returned to nginx instead of the dn. If unset the dn is used
• tls_config - optional - Configures TLS parameters for LDAPs connections
  • validate_hostname - optional - Set the hostname for certificate validation, when unset the hostname from the server URI is used
  • allow_insecure - optional - Disable certificate validation. Setting this is not recommended for production setups

When using the LDAP provider you need to pay attention when writing your ACL. As DNs are used as names for users and groups you also need to specify those in the ACL:

acl:
  rule_sets:
  - rules:
    - field: "host"
      equals: "test.example.com"
    allow:
    - "cn=myuser,ou=users,dc=example,dc=com"
    - "@cn=mygroup,ou=groups,dc=example,dc=com"

Provider configuration: Simple Auth (simple)

The simple auth provider consists of a static mapping between users and passwords and groups and users. This can be seen as the replacement of htpasswd files.

providers:
  simple:
    enable_basic_auth: false

    # Unique username mapped to bcrypt hashed password
    users:
      luzifer: "$2a$10$FSGAF8qDWX52aBID8.WpxOyCvfSQ3JIUVFiwyd1jolb4jM3BzJmNu"
      mike: "$2a$10$/0nrpYkdVhAifCLCI1DTz.4CkbCkc8CsvYhfvBRIhTTQDfBrkJ8Re"

    # Groupname to users mapping
    groups:
      admins: ["luzifer"]
      users: ["mike"]

    # MFA configs: Username to configs mapping
    mfa:
      luzifer:
        - provider: google
          attributes:
            secret: asdgsdfhgshf

You can see how to configure the provider the example above: No surprises, just ensure you are using bcrypt hashes for the passwords, no other hash functions are supported.

If enable_basic_auth is set to true the credentials can also be submitted through basic auth. This is useful for services whose clients does not support other types of authentication.

When there is at least one MFA configuration provided for the user inside the mfa block the user will be forced to enter a MFA token during login or otherwise the login will fail.

Provider configuration: Token Auth (token)

The token auth provider is intended to give machines access to endpoints. Users will not be able to "login" using tokens when they see the login form.

providers:
  token:
    # Mapping of unique token names to the token
    tokens:
      tokenname: "MYTOKEN"
      mycli: "kQHjQLuQdkSPwdJ1mueniLMPSjCc6GVt"

    # Groupname to token mapping
    groups:
      mytokengroup: ["tokenname"]

When accessing the sites using a token this header is expected:

Authorization: Token MYTOKEN

Provider configuration: Yubikey One-Factor-Auth (yubikey)

The Yubikey auth provider is a one-factor-authentication mechanism. Not to be confused by U2F or HOTP two-factor methods. Your users only need to press the button to fully login. (Be sure you know what you're doing here!)

providers:
  yubikey:
    # Get your client / secret from https://upgrade.yubico.com/getapikey/
    client_id: "12345"
    secret_key: "foobar"

    # First 12 characters of the OTP string mapped to the username
    devices:
      ccccccfcvuul: "luzifer"

    # Groupname to users mapping
    groups:
      admins: ["luzifer"]

You need to configure the client_id and the secret_key for the Yubico online validation service and the Yubikeys need to comply the specifications of that API (do not put random values into the device ID). Afterwards just take the first 12 characters of the keys OTP and map it to an user.