sproxy2

Secure HTTP proxy for authenticating users via OAuth2

Latest on Hackage:1.96.0

This package is not currently in any snapshots. If you're interested in using it, we recommend adding it to Stackage Nightly. Doing so will make builds more reliable, and allow stackage.org to host generated Haddocks.

MIT licensed and maintained by Igor Pashev

Sproxy2

HTTP proxy for authenticating users via OAuth2.

Motivation

This is overhaul of original Sproxy. See ChangeLog.md for the differences.

Why use a proxy for doing OAuth2? Isn't that up to the application?

  • sproxy is secure by default. No requests make it to the web server if they haven't been explicitly whitelisted.
  • sproxy is independent. Any web application written in any language can use it.

Use cases

  • Existing web applications with concept of roles. For example, Mediawiki, Jenkins, Icinga Web 2. In this case you configure Sproxy to allow unrestricted access to the application for some groups defined by Sproxy. These groups are mapped to the application roles. There is a plugin for Jenkins which can be used for this. Mediawiki and Icinga Web 2 were also successfully deployed in this way, though it required changes to their source code.

  • New web applications designed to work specifically behind Sproxy. In this case you define Sproxy rules to control access to the application's API. It would likely be a single-page application. Examples are MyWatch and Juan de la Cosa.

  • Replace HTTP Basic authentication.

How it works

When an HTTP client makes a request, Sproxy checks for a session cookie. If it doesn't exist (or it's invalid, expired), it responses with HTTP status 511 with the page, where the user can choose an OAuth2 provider to authenticate with. Finally, we store the the email address in a session cookie: signed with a hash to prevent tampering, set for HTTP only (to prevent malicious JavaScript from reading it), and set it for secure (since we don't want it traveling over plaintext HTTP connections).

From that point on, when sproxy detects a valid session cookie it extracts the email, checks it against the access rules, and relays the request to the back-end server (if allowed).

Permissions system

Permissions are stored in internal SQLite3 database and imported from data sources, which can be a PostgreSQL database or a file. See sproxy.sql and datafile.example.yml for details.

Do note that Sproxy2 fetches only group_member, group_privilege and privilege_rule tables, because only these tables are used for authorization. The other tables in PostgreSQL schema serve for data integrity. Data integrity of the data file is not verfied, though import may fail due to primary key restrictions.

Only one data source can be used. The data in internal database, if any, is fully overwritten by the data from a data source. If no data source is specified, the data in internal database remains unchanged, even between restarts. Broken data source is not fatal. Sproxy will keep using existing internal database, or create a new empty one if missed. Broken data source means inability to connect to PostgreSQL database, missed datafile, etc.

The data from a PostgreSQL database are periodically fetched into the internal database, while the data file is read once at startup.

Here are the main concepts:

  • A group is identified by a name. Every group has - members (identified by email address, through group_member) and - associated privileges (through group_privilege).
  • A privilege is identified by a name and a domain. It has associated rules (through privilege_rule) that define what the privilege gives access to.
  • A rule is a combination of sql patterns for a domain, a path and an HTTP method. A rule matches an HTTP request, if all of these components match the respective attributes of the request. However of all the matching rules only the rule with the longest path pattern will be used to determine whether a user is allowed to perform a request. This is often a bit surprising, please see the following example:

Privileges example

Consider this group_privilege and privilege_rule relations:

group | privilege | domain ---------------- | --------- | ----------------- readers | basic | wiki.example.com readers | read | wiki.example.com editors | basic | wiki.example.com editors | read | wiki.example.com editors | edit | wiki.example.com administrators | basic | wiki.example.com administrators | read | wiki.example.com administrators | edit | wiki.example.com administrators | admin | wiki.example.com

privilege | domain | path | method ----------- | ------------------ | -------------- | ------ basic | wiki.example.com | /% | GET read | wiki.example.com | /wiki/% | GET edit | wiki.example.com | /wiki/edit/% | GET edit | wiki.example.com | /wiki/edit/% | POST admin | wiki.example.com | /admin/% | GET admin | wiki.example.com | /admin/% | POST admin | wiki.example.com | /admin/% | DELETE

With this setup, everybody (that is readers, editors and administratorss) will have access to e.g. /imgs/logo.png and /favicon.ico, but only administrators will have access to /admin/index.php, because the longest matching path pattern is /admin/% and only administrators have the admin privilege.

Likewise readers have no access to e.g. /wiki/edit/delete_everything.php.

Keep in mind that:

  • Domains are converted into lower case (coming from a data source or HTTP requests).
  • Emails are converted into lower case (coming from a data source or OAuth2 providers).
  • Groups are case-sensitive and treated as is.
  • HTTP methods are case-sensitive.
  • HTTP query parameters are ignored when matching a request against the rules.
  • Privileges are case-sensitive and treated as is.
  • SQL wildcards (_ and %) are supported for emails, paths (this will change in future versions).

Checking access in a bunch

There is an API end-point for checking access rights in a single POST query: /.sproxy/access. Users should be authenticated to use this end-point, otherwise the respond will be HTTP 511.

The request body shall be a JSON object like this:

{
  "tag1": {"path": "/foo", "method": "GET"},
  "tag2": {"path": "/bar", "method": "GET"}
}

And the respond will contain a JSON array with tag matching path and method pairs allowed to the user. For example:

$ curl -d '{"foo": {"path":"/get", "method":"GET"}, "bar": {"path":"/post", "method":"POST"}}' -XPOST -k 'https://example.ru:8443/.sproxy/access' ...
["foo","bar"]

$ curl -d '{"foo": {"path":"/get", "method":"POST"}, "bar": {"path":"/post", "method":"POST"}}' -XPOST -k 'https://example.ru:8443/.sproxy/access' ...
["bar"]

$ curl -d '{"foo": {"path":"/", "method":"POST"}, "bar": {"path":"/post", "method":"GET"}}' -XPOST -k 'https://example.ru:8443/.sproxy/access' ...
[]

Logout

Hitting the endpoint /.sproxy/logout will invalidate the session cookie. The user will be redirected to / after logout.

Robots

Since all sproxied resources are private, it doesn't make sense for web crawlers to try to index them. In fact, crawlers will index only the login page. To prevent this, sproxy returns the following for /robots.txt:

User-agent: *
Disallow: /

HTTP headers passed to the back-end server

All Sproxy headers are UTF8-encoded.

header | value -------------------- | ----- From: | visitor's email address, lower case X-Groups: | all groups that granted access to this resource, separated by commas (see the note below) X-Given-Name: | the visitor's given (first) name X-Family-Name: | the visitor's family (last) name X-Forwarded-Proto: | the visitor's protocol of an HTTP request, always https X-Forwarded-For | the visitor's IP address (added to the end of the list if header is already present in client request)

X-Groups denotes an intersection of the groups the visitor belongs to and the groups that granted access:

Visitor's groups | Granted groups | X-Groups ---------------- | -------------- | --------- all | all, devops | all all, devops | all | all all, devops | all, devops | all,devops all, devops | devops | devops devops | all, devops | devops devops | all | Access denied

Requirements

Sproxy2 is written in Haskell with GHC. All required Haskell libraries are listed in sproxy2.cabal. Use cabal-install to fetch and build all pre-requisites automatically.

Configuration

By default sproxy2 will read its configuration from sproxy.yml. There is example file with documentation sproxy.example.yml. You can specify a custom path with:

sproxy2 --config /path/to/sproxy.yml

Changes

For differences with the original Sproxy scroll down.

1.96.0

  • Added support for Yandex (https://tech.yandex.com/oauth/).

  • Encode full URL (including protocol) into the state parameter, not just path. This makes it possible to work with OAuth2 providers that do not support multiple callback URL, like Yandex.

  • Fixed POST requests for tokens with Google and LinkedIn. They were mistakenly using URL paramaters instead of URL-encoded bodies.

1.95.0

  • Add end-point for checking access in a bunch (/.sproxy/access).

  • Respond with 502 (Bad Gateway) on any backend error. Previously it was 500 (Internal Server Error).

1.94.1

  • Fixed a typo introduced in version 1.94.0 in SQL query: ... WHERE domain = domain ... -> ... WHERE domain = :domain ...

1.94.0

  • BREAKING: Disregard possible port in the Host HTTP header. Previously, Sproxy took possible port number into account when looking for backend and privileges. Now it ignores port and considers domain name only. This also gets Sproxy in line with browsers and SSL certificates: certificates do not include port numbers, browsers ignore ports when sending cookies.

  • BREAKING: no SQL wildcards (% or _) in domain names when looking up for privileges. This feature was ambiguous (in the same way as paths are) and never used anyway.

1.93.0

  • BREAKING: Allow !include in config file. This changes semantics of options key and oauth2.<provider>.client_secret. They are no longer files, but strings. To read content from files, use !include. The point of being files or read from files is to segregate secrets from non-sensitive easily discoverable settings. With !include it is much more simple and flexible.

1.92.0

  • Allow running in plain HTTP mode (no SSL). Useful when Sproxy is behind some other proxy or load-balancer. Added two more options: ssl (defaults to true) and https_port (defaults to like listen). Options ssl_key and ssl_cert are required only if ssl == true. SSL-terminations is still required at upstream proxies, because the cookie is set for HTTPS only.

  • Added "user" table into sproxy.sql. No action is required, but PostgreSQL database built after this file will be incompatible with Sproxy Web ( <= 0.4.1 at least).

1.91.0

  • In addition to good old PostgreSQL data source, made it possible to import permission data from a YAML file. This means that Sproxy2 can work without any PostgreSQL database, just using file-only configuration. Useful for development or trivial deployments. Added new datafile option in configuration file.

1.90.2

  • Make sure all Sproxy-specific HTTP headers are UTF8-encoded.

  • /.sproxy/logout just redirects if no cookie. Previously it was returning HTTP 404 to unauthenticated users, and redirecting authenticated users with removal of the cookie. The point is not to reveal cookie name.

  • Made Warp stop printing exceptions, mostly "client closed connection", which happens outside of our traps.

1.90.1

  • Fixed headers processing. Wrong headers were making Chromium drop connection in HTTP/2. Firefox sometimes couldn't handle gzipped and chunked responses in HTTP/1.1.

  • After authenticating, redirect to original path with query parameters if method was GET. Otherwise redirect to "/". Previously, when unauthenticated users click on "https://example.net/foo?bar", they are redirected to "https://example.net/foo" regardless of the method.

1.90.0 (Preview Release)

Sproxy2 is overhaul of original Sproxy (see also Hackage). Here are the key differences (with Sproxy 0.9.8):

  • Sproxy2 can work with remote PostgreSQL database. Quick access to the database is essential as sproxy does it on every HTTP request. Sproxy2 pulls data into local SQLite3 database.

  • At this release Sproxy2 is compatible with Sproxy database with one exception: SQL wildcards are not supported for HTTP methods. I. e. you have to change '%' in the database to specific methods like GET, POST, etc.

  • OAuth2 callback URLs changed: Sproxy2 uses /.sproxy/oauth2/:provider, e. g. /.sproxy/oauth2/google. Sproxy used /sproxy/oauth2callback for Google and /sproxy/oauth2callback/linkedin for LinkedIn.

  • Sproxy2 does not allow login with email addresses not known to it.

  • Sproxy2: OAuth2 callback state is serialized, signed and passed base64-encoded. Of course it's used to verify the request is legit.

  • Sproxy2: session cookie is serialized, signed and sent base64-encoded.

  • Path /.sproxy belongs to Sproxy2 completely. Anything under this path is never passed to backends.

  • Sproxy2 supports multiple backends. Routing is based on the Host HTTP header.

  • Sproxy2 uses WAI / Warp for incoming connections. As a result Sproxy2 supports HTTP2.

  • Sproxy2 uses HTTP Client to talk to backends. As a result Sproxy2 reuses backend connections instead of closing them after each request to the backend.

  • Sproxy2 optionally supports persistent key again (removed in Sproxy 0.9.2). This can be used in load-balancing multiple Sproxy2 instances.

  • Configuration file has changed. It's still YAML, but some options are renamed, removed or added. Have a look at well-documented sproxy.example.yml

comments powered byDisqus