---
layout: default
navsection: installguide
title: Install Keep-web server
...
{% comment %}
Copyright (C) The Arvados Authors. All rights reserved.

SPDX-License-Identifier: CC-BY-SA-3.0
{% endcomment %}

# "Introduction":#introduction
# "Configure DNS":#introduction
# "Configure anonymous user token.yml":#update-config
# "Update nginx configuration":#update-nginx
# "Install keep-web package":#install-packages
# "Start the service":#start-service
# "Restart the API server and controller":#restart-api
# "Confirm working installation":#confirm-working

h2(#introduction). Introduction

The Keep-web server provides read/write access to files stored in Keep using WebDAV and S3 protocols.  This makes it easy to access files in Keep from a browser, or mount Keep as a network folder using WebDAV support in various operating systems. It serves public data to unauthenticated clients, and serves private data to clients that supply Arvados API tokens. It can be installed anywhere with access to Keep services, typically behind a web proxy that provides TLS support. See the "godoc page":http://godoc.org/github.com/curoverse/arvados/services/keep-web for more detail.

h2(#dns). Configure DNS

It is important to properly configure the keep-web service to so it does not open up cross-site-scripting (XSS) attacks.  A HTML file can be stored in collection.  If an attacker causes a victim to visit that HTML file through Workbench, it will be rendered by the browser.  If all collections are served at the same domain, the browser will consider collections as coming from the same origin and thus have access to the same browsing data (such as API token), enabling malicious Javascript in the HTML file to access Arvados as the victim.

There are two approaches to mitigate this.

# The service can tell the browser that all files should go to download instead of in-browser preview, except in situations where an attacker is unlikely to be able to gain access to anything they didn't already have access to.
# Each collection served by @keep-web@ is served on its own virtual host.  This allows for file with executable content to be displayed in-browser securely.  The virtual host embeds the collection uuid or portable data hash in the hostname.  For example, a collection with uuid @xxxxx-4zz18-tci4vn4fa95w0zx@ could be served as @xxxxx-4zz18-tci4vn4fa95w0zx.collections.ClusterID.example.com@ .  The portable data hash @dd755dbc8d49a67f4fe7dc843e4f10a6+54@ could be served at @dd755dbc8d49a67f4fe7dc843e4f10a6-54.collections.ClusterID.example.com@ .  This requires "wildcard DNS record":https://en.wikipedia.org/wiki/Wildcard_DNS_record and "wildcard TLS certificate.":https://en.wikipedia.org/wiki/Wildcard_certificate

h3. Collections download URL

Downloads links will served from the URL in @Services.WebDAVDownload.ExternalURL@ .  The collection uuid or PDH is put in the URL path.

If blank, serve links to WebDAV with @disposition=attachment@ query param.  Unlike preview links, browsers do not render attachments, so there is no risk of XSS.

If @WebDAVDownload@ is blank, and @WebDAV@ has a single origin (not wildcard, see below), then Workbench will show an error page

<notextile>
<pre><code>    Services:
      WebDAVDownload:
        ExternalURL: https://<span class="userinput">download.ClusterID.example.com</span>
</code></pre>
</notextile>

h3. Collections preview URL

Collections will be served using the URL pattern in @Services.WebDAV.ExternalURL@ .  If blank, use @Services.WebDAVDownload.ExternalURL@ instead, and disable inline preview.  If both are empty, downloading collections from workbench will be impossible.  When wildcard domains configured, credentials are still required to access non-public data.

h4. In their own subdomain

Collections can be served from their own subdomain:

<notextile>
<pre><code>    Services:
      WebDAV:
        ExternalURL: https://<span class="userinput">*.collections.ClusterID.example.com/</span>
</code></pre>
</notextile>

This option is preferred if you plan to access Keep using third-party S3 client software, because it accommodates S3 virtual host-style requests and path-style requests without any special client configuration.

h4. Under the main domain

Alternately, they can go under the main domain by including @--@:

<notextile>
<pre><code>    Services:
      WebDAV:
        ExternalURL: https://<span class="userinput">*--collections.ClusterID.example.com/</span>
</code></pre>
</notextile>

h4. From a single domain

Serve preview links from a single domain, setting uuid or pdh in the path (similar to downloads).  This configuration only allows previews of public data (data accessible by the anonymous user) and collection-sharing links (where the token is already embedded in the URL); it will ignore authorization headers, so a request for non-public data may return "404 Not Found" even if normally valid credentials were provided.

<notextile>
<pre><code>    Services:
      WebDAV:
        ExternalURL: https://<span class="userinput">collections.ClusterID.example.com/</span>
</code></pre>
</notextile>

Note the trailing slash.

{% include 'notebox_begin' %}
Whether you choose to serve collections from their own subdomain or from a single domain, it's important to keep in mind that they should be served from me same _site_ as Workbench for the inline previews to work.

Please check "keep-web's URL pattern guide":../api/keep-web-urls.html#same-site to learn more.
{% include 'notebox_end' %}

h2. Set InternalURLs

<notextile>
<pre><code>    Services:
      WebDAV:
        InternalURLs:
          http://<span class="userinput">localhost:9002</span>: {}
</code></pre>
</notextile>

h2(#update-config). Configure anonymous user token

{% assign railscmd = "bin/bundle exec ./script/get_anonymous_user_token.rb --get" %}
{% assign railsout = "zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz" %}
If you intend to use Keep-web to serve public data to anonymous clients, configure it with an anonymous token.

# First, generate a long random string and put it in the @config.yml@ file, in the @AnonymousUserToken@ field.
# Then, use the following command on the <strong>API server</strong> to register the anonymous user token in the database. {% include 'install_rails_command' %}

<notextile>
<pre><code>    Users:
      AnonymousUserToken: <span class="userinput">"{{railsout}}"</span>
</code></pre>
</notextile>

Set @Users.AnonymousUserToken: ""@ (empty string) or leave it out if you do not want to serve public data.

h3. Update nginx configuration

Put a reverse proxy with SSL support in front of keep-web.  Keep-web itself runs on the port 9002 (or whatever is specified in @Services.WebDAV.InternalURL@) while the reverse proxy runs on port 443 and forwards requests to Keep-web.

Use a text editor to create a new file @/etc/nginx/conf.d/keep-web.conf@ with the following configuration. Options that need attention are marked in <span class="userinput">red</span>.

<notextile><pre>
upstream keep-web {
  server                127.0.0.1:<span class="userinput">9002</span>;
}

server {
  listen                443 ssl;
  server_name           <span class="userinput">download.ClusterID.example.com</span>
                        <span class="userinput">collections.ClusterID.example.com</span>
                        <span class="userinput">*.collections.ClusterID.example.com</span>
                        <span class="userinput">~.*--collections.ClusterID.example.com</span>;

  proxy_connect_timeout 90s;
  proxy_read_timeout    300s;

  ssl                   on;
  ssl_certificate       <span class="userinput">/YOUR/PATH/TO/cert.pem</span>;
  ssl_certificate_key   <span class="userinput">/YOUR/PATH/TO/cert.key</span>;

  location / {
    proxy_pass          http://keep-web;
    proxy_set_header    Host            $host;
    proxy_set_header    X-Forwarded-For $proxy_add_x_forwarded_for;

    client_max_body_size    0;
    proxy_http_version      1.1;
    proxy_request_buffering off;
    proxy_max_temp_file_size 0;
  }
}
</pre></notextile>

{% include 'notebox_begin' %}
If you restrict access to your Arvados services based on network topology -- for example, your proxy server is not reachable from the public internet -- additional proxy configuration might be needed to thwart cross-site scripting attacks that would circumvent your restrictions.

Normally, Keep-web accepts requests for multiple collections using the same host name, provided the client's credentials are not being used. This provides insufficient XSS protection in an installation where the "anonymously accessible" data is not truly public, but merely protected by network topology.

In such cases -- for example, a site which is not reachable from the internet, where some data is world-readable from Arvados's perspective but is intended to be available only to users within the local network -- the downstream proxy should configured to return 401 for all paths beginning with "/c="
{% include 'notebox_end' %}

{% assign arvados_component = 'keep-web' %}

{% include 'install_packages' %}

{% include 'start_service' %}

{% include 'restart_api' %}

h2(#confirm-working). Confirm working installation

<notextile><code><pre>
$ curl -H "Authorization: Bearer $system_root_token" https://<span class="userinput">download.ClusterID.example.com</span>/c=59389a8f9ee9d399be35462a0f92541c-53/_/hello.txt
</code></pre></notextile>

If wildcard collections domains are configured:

<notextile><code><pre>
$ curl -H "Authorization: Bearer $system_root_token" https://<span class="userinput">59389a8f9ee9d399be35462a0f92541c-53.collections.ClusterID.example.com</span>/hello.txt
</code></pre></notextile>

If using a single collections preview domain:

<notextile><code><pre>
$ curl https://<span class="userinput">collections.ClusterID.example.com</span>/c=59389a8f9ee9d399be35462a0f92541c-53/t=$system_root_token/_/hello.txt
</code></pre></notextile>