Improve html in your Rails app.

This gem replaces the normal ERB parsing with an HTML-aware ERB parsing. This makes your templates smarter by adding runtime checks around the data interpolated from Ruby into HTML.

Add better-html to your Gemfile with its dependency:

gem "better_html"

If you want to use html_attributes helper as described further down, add it to your app/helpers/application_helper.rb,

module ApplicationHelper
  include BetterHtml::Helpers

  ...

A global configuration for the app is stored at BetterHtml.config. The default configuration can be changed like this:

# config/initializers/better_html.rb
BetterHtml.configure do |config|
  config.allow_single_quoted_attributes = false
end

or if you prefer storing the config elsewhere, in a yml file for example:

# config/initializers/better_html.rb
BetterHtml.config = BetterHtml::Config.new(YAML.load_file(file_path, permitted_classes: [Regexp]))

Available configuration options are:

• partial_tag_name_pattern: Regex to validate foo in <foo>. Defaults to /\A[a-z0-9\-\:]+\z/.
• partial_attribute_name_pattern: Regex to validate bar in <foo bar=1>. Defaults to /\A[a-zA-Z0-9\-\:]+\z/.
• allow_single_quoted_attributes: When true, <foo bar='1'> is valid syntax. Defaults to true.
• allow_unquoted_attributes: When true, <foo bar=1> is valid syntax. Defaults to false.
• javascript_safe_methods: List of methods that return javascript-safe strings. This list is used by SafeErbTester when determining whether ruby interpolation is safe for a given attribute. Defaults to ['to_json'].
• lodash_safe_javascript_expression: Same as javascript_safe_methods, but for lodash templates. Defaults to [/\AJSON\.stringify\(/].
• javascript_attribute_names: List of all attribute names that contain javascript code. This list is used by SafeErbTester when determining whether or not a given attribute value will be eval'ed as javascript. Defaults to [/\Aon/i] (matches onclick for example).
• template_exclusion_filter: This is called when determining whether to apply runtime checks on a .erb template. When this Proc returns false, no safety checks are applied and parsing is done using the default Rails erubi engine. For example, to exclude erb templates provided by libraries, use: Proc.new { |filename| !filename.start_with?(Rails.root.to_s) }. Defaults to nil (all html.erb templates are parsed).

By default, only files named .html.erb are parsed at runtime using BetterHtml's erubi implementation. To change this behavior and parse other file types, assign the erubi implementation into BetterHtml::BetterErb.content_types like this:

# config/initializers/better_html.rb
impl = BetterHtml::BetterErb.content_types['html.erb']
BetterHtml::BetterErb.content_types['htm.erb'] = impl
BetterHtml::BetterErb.content_types['atom.erb'] = impl
BetterHtml::BetterErb.content_types['html+variant.erb'] = impl

In order to apply effective runtime checks, it is necessary to enforce the validity of all HTML contained in an application's templates. This comes with an opinionated approach to what ERB syntax is allowed given any HTML context. The next section describes the allowed syntax.

Use ruby expressions inside quoted html attributes.

Allowed ✅
<img class="<%= value %>">

Not allowed ❌
<img <%= value %>>

Not allowed ❌
<img class=<%= value %>>

Use interpolation into tag or attribute names.

Allowed ✅
<img data-<%= value %>="true">

Allowed ✅
<ns:<%= value %>>

Not allowed ❌ (missing space after closing quote)
<img class="hidden"<%= value %>>

Not allowed ❌
<img <%= value %>="true">

Insert conditional attributes using html_attributes helper.

Allowed ✅
<img <%= html_attributes(class: 'hidden') if condition? %>>

Not allowed ❌
<img <% if condition? %>class="hidden"<% end %>>

Only insert expressions (<%= or <%==) inside script tags, never statements (<%)

<script>
  // Allowed ✅
  var myValue = <%== value.to_json %>;
  if(myValue)
    doSomething();

  // Not allowed ❌
  <% if value %>
    doSomething();
  <% end %>
</script>

Looking only at a ERB file, it's impossible to determine if a given Ruby value is safe to interpolate. For example, consider:

<img class="<%= value %>">

Assuming value may not be escaped properly and could contain a double-quote character (") at runtime, then the resulting HTML would be invalid, and the application would be vulnerable to XSS when value is user-controlled.

With HTML-aware ERB parsing, we wrap value into a runtime safety check that raises and exception when value contains a double-quote character that would terminate the html attribute. The safety check is performed after normal ERB escaping rules are applied, so the standard html_safe helper can be used.

The html_attributes helper works the same way, it will raise when attribute values are escaped improperly.

Consider the following ERB template

<img data-<%= value %>="true">

When value is user-controlled, an attacker may achieve XSS quite easily in this situation. We wrap value in a runtime check that ensures it only contains characters that are valid in an attribute name. This excludes =, / or space, which should prevent any risk of injection.

The html_attributes helper works the same way, it will raise when attribute names contain dangerous characters.

Consider the following ERB template:

<textarea>
  <%== value %>
</textarea>

In circumstances where value may contain input such as </textarea><script>alert(1)</script>, an attacker can easily achieve XSS. We make best-effort runtime validations on this value in order to make it safe against some obvious attacks.

We check for any interpolation containing </textarea and raise an exception if this substring occurs. Note that this won't catch cases where an end tag is split across multiple adjacent interpolations.

The same strategy is applied to other tags which contain non-html data, such as <script>, html comments and CDATA tags.

In addition to runtime validation, this gem provides test helpers that makes it easy to write a test to assert .to_json is used in every script tag and every html attribute which end up being executed as javascript (onclick and similar). The main goal of this helper is to assert that Ruby data translates into Javascript data, but never becomes javascript code.

Simply create test/unit/erb_safety_test.rb and add code like this:

# frozen_string_literal: true

require 'test_helper'
require 'better_html/test_helper/safe_erb_tester'

class ErbSafetyTest < ActiveSupport::TestCase
  include BetterHtml::TestHelper::SafeErbTester
  ERB_GLOB = Rails.root.join(
    'app', 'views', '**', '{*.htm,*.html,*.htm.erb,*.html.erb,*.html+*.erb}'
  )

  Dir[ERB_GLOB].each do |filename|
    pathname = Pathname.new(filename).relative_path_from(Rails.root)
    test "missing javascript escapes in #{pathname}" do
      assert_erb_safety(File.read(filename), filename:)
    end
  end
end

You may also want to assert that all .html.erb templates are parseable, to avoid deploying broken templates to production. Add this code in test/unit/erb_implementation_test.rb

# frozen_string_literal: true

require 'test_helper'

class ErbImplementationTest < ActiveSupport::TestCase
  ERB_GLOB = Rails.root.join(
    'app', 'views', '**', '{*.htm,*.html,*.htm.erb,*.html.erb,*.html+*.erb}'
  )

  Dir[ERB_GLOB].each do |filename|
    pathname = Pathname.new(filename).relative_path_from(Rails.root)
    test "html errors in #{pathname}" do
      data = File.read(filename)
      BetterHtml::BetterErb::ErubiImplementation.new(data, filename:).validate!
    end
  end
end

If you're using RSpec you can add the following code to spec/better_html_spec.rb

# frozen_string_literal: true

require "rails_helper"

RSpec.describe "BetterHtml" do
  it "does assert that all .html.erb templates are parseable" do
    erb_glob = Rails.root.join(
      "app", "views", "**", "{*.htm,*.html,*.htm.erb,*.html.erb,*.html+*.erb}"
    )

    Dir[erb_glob].each do |filename|
      data = File.read(filename)
      expect {
        BetterHtml::BetterErb::ErubiImplementation.new(data, filename:).validate!
      }.not_to raise_exception
    end
  end
end

This gem provides an ERB parser that builds an AST from HTML+ERB templates. Unlike higher-level libraries like Nokogiri, this parser does not make assumptions about the validity of HTML documents (for example, opening tags being matched with closing tags). The parser also handles ERB tags as first class nodes in the syntax tree.

require 'better_html/parser'

buffer = Parser::Source::Buffer.new('(buffer)')
buffer.source = '<div><%= value -%></div>'
parser = BetterHtml::Parser.new(buffer)

puts parser.inspect
# => #<BetterHtml::Parser ast=s(:document,
#   s(:tag, nil,
#     s(:tag_name, "div"), nil, nil),
#   s(:text,
#     s(:erb,
#       s(:indicator, "="), nil,
#       s(:code, " value "),
#       s(:trim))),
#   s(:tag,
#     s(:solidus),
#     s(:tag_name, "div"), nil, nil))>

The syntax tree exposed by this parser is not to be confused with the nested nature of HTML elements. At this stage, the parser does not build html elements, only tags which mark the beginning and end of elements.