Goldilocks

The node.js netserver that's just right.

• HTTPS Web Server with Automatic TLS (SSL) via ACME (Let's Encrypt)
  • Static Web Server
  • URL Redirects
  • SSL on localhost (with bundled localhost.daplie.me certificates)
  • Uses node cluster to take advantage of multiple CPUs (in progress)
• TLS name-based (SNI) proxy
• TCP port-based proxy
• WS Tunnel Server (i.e. run on Digital Ocean and expose a home-firewalled Raspberry Pi to the Internet)
• WS Tunnel Client (i.e. run on a Raspberry Pi and connect to a Daplie Tunnel)
• UPnP / NAT-PMP forwarding and loopback testing (in progress)
• Configurable via API
• mDNS Discoverable (configure in home or office with mobile and desktop apps)
• OAuth3 Authentication

Install Standalone

# v1 in npm
npm install -g goldilocks

# v1 in git (via ssh)
npm install -g git+ssh://git@git.daplie.com:Daplie/goldilocks.js#v1

# v1 in git (unauthenticated)
npm install -g git+https://git@git.daplie.com:Daplie/goldilocks.js#v1

goldilocks

Serving /Users/foo/ at https://localhost.daplie.me:8443

Install as a System Service (daemon-mode)

We have service support for

• systemd (Linux, Ubuntu)
• launchd (macOS)

curl https://git.daplie.com/Daplie/goldilocks.js/raw/master/install.sh | bash

Modules & Configuration

Goldilocks has several core systems, which all have their own configuration and some of which have modules:

• http
  • proxy (reverse proxy)
  • static
  • redirect
• tls
  • proxy (reverse proxy)
  • acme
• tcp
  • proxy
  • forward
• udp
  • forward
• domains
• tunnel_server
• DDNS
• tunnel_client
• mDNS
• socks5
• api

All modules require a type and an id, and any modules not defined inside the domains system also require a domains field (with the exception of the forward modules that require the ports field).

http

The HTTP system handles plain http (TLS / SSL is handled by the tls system)

Example config:

http:
  trust_proxy: true                 # allow localhost, 192.x, 10.x, 172.x, etc to set headers
  allow_insecure: false             # allow non-https even without proxy https headers
  primary_domain: example.com       # attempts to access via IP address will redirect here

  # An array of modules that define how to handle incoming HTTP requests
  modules:
    - type: static
      domains:
        - example.com
      root: /srv/www/:hostname

http.proxy - how to reverse proxy (ruby, python, etc)

The proxy module is for reverse proxying, typically to an application on the same machine. (Though it can also reverse proxy to other devices on the local network.)

It has the following options:

address     The DNS-resolvable hostname (or IP address) and port connected by `:` to proxy the request to.
            Takes priority over host and port if they are also specified.
            ex: locahost:3000
            ex: 192.168.1.100:80

host        The DNS-resolvable hostname (or IP address) of the system to which the request will be proxied.
            Defaults to localhost if only the port is specified.
            ex: localhost
            ex: 192.168.1.100

port        The port on said system to which the request will be proxied
            ex: 3000
            ex: 80

Example config:

http:
  modules:
    - type: proxy
      domains:
        - api.example.com
      host: 192.168.1.100
      port: 80
    - type: proxy
      domains:
        - www.example.com
      address: 192.168.1.16:80
    - type: proxy
      domains:
        - '*'
      port: 3000

http.static - how to serve a web page

The static module is for serving static web pages and assets and has the following options:

root        The path to serve as a string.
            The template variable `:hostname` represents the HTTP Host header without port information
            ex: `root: /srv/www/example.com` would load the example.com folder for any domain listed
            ex: `root: /srv/www/:hostname` would load `/srv/www/example.com` if so indicated by the Host header

index       Set to `false` to disable the default behavior of loading `index.html` in directories
            ex: `false`

dotfiles    Set to `allow` to load dotfiles rather than ignoring them
            ex: `"allow"`

redirect    Set to `false` to disable the default behavior of ensuring that directory paths end in '/'
            ex: `false`

indexes     An array of directories which should be have indexes served rather than blocked
            ex: `[ '/' ]` will allow all directories indexes to be served

Example config:

http:
  modules:
    - type: static
      domains:
        - example.com
      root: /srv/www/:hostname

http.redirect - how to redirect URLs

The redirect module is for, you guessed it, redirecting URLs.

It has the following options:

status      The HTTP status code to issue (301 is usual permanent redirect, 302 is temporary)
            ex: 301

from        The URL path that was used in the request.
            The `*` wildcard character can be used for matching a full segment of the path
            ex: /photos/
            ex: /photos/*/*/

to          The new URL path which should be used.
            If wildcards matches were used they will be available as `:1`, `:2`, etc.
            ex: /pics/
            ex: /pics/:1/:2/
            ex: https://mydomain.com/photos/:1/:2/

Example config:

http:
  modules:
    - type: proxy
      domains:
        - example.com
      status: 301
      from: /archives/*/*/*/
      to: https://example.net/year/:1/month/:2/day/:3/

tls

The tls system handles encrypted connections, including fetching certificates, and uses ServerName Indication (SNI) to determine if the connection should be handled by the http system, a tls system module, or rejected.

Example config:

tls:
  modules:
    - type: proxy
      domains:
        - example.com
        - example.net
      address: '127.0.0.1:6443'

Certificates are saved to ~/acme, which may be /var/www/acme if Goldilocks is run as the www-data user.

tls.proxy

The proxy module routes the traffic based on the ServerName Indication (SNI) without decrypting it.

It has the same options as the HTTP proxy module.

Example config:

tls:
  modules:
    - type: proxy
      domains:
        - example.com
      address: '127.0.0.1:5443'

tls.acme

The acme module defines the setting used when getting new certificates.

It has the following options:

email              The email address for ACME certificate issuance
                   ex: john.doe@example.com

server             The ACME server to use
                   ex: https://acme-v01.api.letsencrypt.org/directory
                   ex: https://acme-staging.api.letsencrypt.org/directory

challenge_type     The ACME challenge to request
                   ex: http-01, dns-01, tls-01

Example config:

tls:
  modules:
    - type: acme
      domains:
        - example.com
        - example.net
      email: 'joe.shmoe@example.com'
      server: 'https://acme-staging.api.letsencrypt.org/directory'
      challenge_type: 'http-01'

NOTE: If you specify dns-01 as the challenge type there must also be a DDNS module defined for all of the relevant domains (though not all domains handled by a single TLS module need to be handled by the same DDNS module). The DDNS module provides all of the information needed to actually set the DNS records needed to verify ownership.

tcp

The tcp system handles both raw and tls-terminated tcp network traffic (see the Note section below the example). It may use port numbers or traffic sniffing to determine how the connection should be handled.

It has the following options:

bind      An array of numeric ports on which to bind
          ex: 80

Example Config:

tcp:
  bind:
    - 22
    - 80
    - 443
  modules:
    - type: forward
      ports:
        - 22
      address: '127.0.0.1:2222'

Note: When tcp traffic comes into goldilocks it will be tested against the tcp modules. The connection may be handed to the TLS module if it appears to be a TLS/SSL/HTTPS connection and if the tls module terminates the traffic, the connection will be sent back to the TLS module. Due to the complexity of node.js' networking stack it is not currently possible to tell which port tls-terminated traffic came from, so only the SNI header (serername / domain name) may be used for modules matching terminated TLS.

tcp.proxy

The proxy module routes traffic after tls-termination based on the servername (domain name) contained in a SNI header. As such this only works to route TCP connections wrapped in a TLS stream.

It has the same options as the HTTP proxy module.

This is particularly useful for routing ssh and vpn traffic over tcp port 443 as wrapped TLS connections in order to access one of your servers even when connecting from a harsh or potentially misconfigured network environment (i.e. hotspots in public libraries and shopping malls).

Example config:

tcp:
  modules:
    - type: proxy
      domains:
        - ssh.example.com      # Note: this domain would also listed in tls.acme.domains
      host: localhost
      port: 22
    - type: proxy
      domains:
        - vpn.example.com      # Note: this domain would also listed in tls.acme.domains
      host: localhost
      port: 1194

Note: In same cases network administrators purposefully block ssh and vpn connections using Application Firewalls with DPI (deep packet inspection) enabled. You should read the ToS of the network you are connected to to ensure that you aren't subverting policies that are purposefully in place on such networks.

Using with ssh

In order to use this to route SSH connections you will need to use ssh's ProxyCommand option. For example to use the TLS certificate for ssh.example.com to wrap an ssh connection you could use the following command:

ssh user@example.com -o ProxyCommand='openssl s_client -quiet -connect example.com:443 -servername ssh.example.com'

Alternatively you could add the following lines to your ssh config file.

Host example.com
  ProxyCommand openssl s_client -quiet -connect example.com:443 -servername ssh.example.com

Using with OpenVPN

There are two strategies that will work well for you:

1. Use ssh with the config above to reverse proxy tcp port 1194 to you.

ssh -L 1194:localhost:1194 example.com

2. [Use stunnel]https://serverfault.com/questions/675553/stunnel-vpn-traffic-and-ensure-it-looks-like-ssl-traffic-on-port-443/681497)

[openvpn-over-goldilocks]
client = yes
accept = 127.0.0.1:1194
sni = vpn.example.com
connect = example.com:443

3. Use stunnel.js as described in the "tunnel_server" section below.

tcp.forward

The forward module routes traffic based on port number without decrypting it.

In addition to the same options as the HTTP proxy module, the TCP forward modules also has the following options:

ports       A numeric array of source ports
            ex: 22

Example Config:

tcp:
  bind:
    - 22
    - 80
    - 443
  modules:
    - type: forward
      ports:
        - 22
      port: 2222

udp

The udp system handles all udp network traffic. It currently only supports forwarding the messages without any examination.

It has the following options:

bind      An array of numeric ports on which to bind
          ex: 53

Example Config:

udp:
  bind:
    - 53
  modules:
    - type: forward
      ports:
        - 53
      address: '127.0.0.1:8053'

udp.forward

The forward module routes traffic based on port number without decrypting it.

It has the same options as the TCP forward module.

Example Config:

udp:
  bind:
    - 53
  modules:
    - type: forward
      ports:
        - 53
      address: '127.0.0.1:8053'

domains

To reduce repetition defining multiple modules that operate on the same domain name the domains field can define multiple modules of multiple types for a single list of names. The modules defined this way do not need to have their own domains field. Note that the tcp.forward module is not allowed in a domains group since its routing is not based on domains.

Example Config

domains:
  - names:
      - example.com
      - www.example.com
      - api.example.com
    modules:
      tls:
        - type: acme
          email: joe.schmoe@example.com
          challenge_type: 'http-01'
      http:
        - type: redirect
          from: /deprecated/path
          to: /new/path
        - type: proxy
          port: 3000
      dns:
        - type: 'dns@oauth3.org'
          token_id: user_token_id

  - names:
      - ssh.example.com
    modules:
      tls:
        - type: acme
          email: john.smith@example.com
          challenge_type: 'http-01'
      tcp:
        - type: proxy
          port: 22
      dns:
        - type: 'dns@oauth3.org'
          token_id: user_token_id

tunnel_server

The tunnel server system is meant to be run on a publicly accessible IP address to server tunnel clients which are behind firewalls, carrier-grade NAT, or otherwise Internet-connect but inaccessible devices.

It has the following options:

secret          A 128-bit or greater string to use for signing tokens (HMAC JWT)
                ex: abc123

servernames     An array of string servernames that should be captured as the
                tunnel server, ignoring the TLS forward module
                ex: api.tunnel.example.com

Example config:

tunnel_server:
  secret: abc123def456ghi789
  servernames:
    - 'api.tunnel.example.com'

DDNS

The DDNS module watches the network environment of the unit and makes sure the device is always accessible on the internet using the domains listed in the config. If the device has a public address or if it can automatically set up port forwarding the device will periodically check its public address to ensure the DNS records always point to it. Otherwise it will to connect to a tunnel server and set the DNS records to point to that server.

The loopback setting specifies how the unit will check its public IP address and whether connections can reach it. Currently only tunnel@oauth3.org is supported. If the loopback setting is not defined it will default to using oauth3.org.

The tunnel setting can be used to specify how to connect to the tunnel. Currently only tunnel@oauth3.org is supported. The token specified in the tunnel setting will be used to acquire the tokens that are used directly with the tunnel server. If the tunnel setting is not defined it will default to try using the tokens in the modules for the relevant domains.

If a particular DDNS module has been disabled the device will still try to set up port forwarding (and connect to a tunnel if that doesn't work), but the DNS records will not be updated to point to the device. This is to allow a setup to be tested before transitioning services between devices.

ddns:
  disabled: false
  loopback:
    type: 'tunnel@oauth3.org'
    domain: oauth3.org
  tunnel:
    type: 'tunnel@oauth3.org'
    token_id: user_token_id
  modules:
    - type: 'dns@oauth3.org'
      token_id: user_token_id
      domains:
        - www.example.com
        - api.example.com
        - test.example.com

mDNS

enabled by default

Although it does not announce itself, Goldilocks is discoverable via mDNS with the special query _cloud._tcp.local. This is so that it can be easily configured via Desktop and Mobile apps when run on devices such as a Raspberry Pi or SOHO servers.

mdns:
  disabled: false
  port: 5353
  broadcast: '224.0.0.251'
  ttl: 300

You can discover goldilocks with mdig.

npm install -g git+https://git.daplie.com/Daplie/mdig.git

mdig _cloud._tcp.local

socks5

Run a Socks5 proxy server.

socks5:
  enable: true
  port: 1080

api

See API.md

@tigerbot: How are the APIs used (in terms of URL, Method, Headers, etc)?

TODO

• http - nowww module
• http - Allow match styles of www.*, *, and *.example.com equally
• http - redirect based on domain name (not just path)
• tcp - bind should be able to specify localhost, uniquelocal, private, or ip
• tcp - if destination host is omitted default to localhost, if dst port is missing, default to src
• sys - curl https://daplie.me/goldilocks | bash -s example.com
• oauth3 - example.com/.well-known/domains@oauth3.org/directives.json
• oauth3 - commandline questionnaire
• modules - use consistent conventions (i.e. address vs host + port)
  • tls - tls.acme vs tls.modules.acme
• tls - forward should be able to match on source port to reach different destination ports