sanitize-html-plus

v1.13.0-2

Published

2 years ago

Clean up user-submitted HTML, preserving whitelisted elements and whitelisted attributes on a per-element basis

Downloads

443

0High
0Medium
0Low

eramdam

html parser sanitizer html sanitizer apostrophe

fork of sanitize-html to include options.styleFilter and pass the stack to textFilter

sanitize-html "plus"

sanitize-html provides a simple HTML sanitizer with a clear API.

sanitize-html is tolerant. It is well suited for cleaning up HTML fragments such as those created by ckeditor and other rich text editors. It is especially handy for removing unwanted CSS when copying and pasting from Word.

sanitize-html allows you to specify the tags you want to permit, and the permitted attributes for each of those tags.

If a tag is not permitted, the contents of the tag are still kept, except for script, style and textarea tags.

The syntax of poorly closed p and img elements is cleaned up.

href attributes are validated to ensure they only contain http, https, ftp and mailto URLs. Relative URLs are also allowed. Ditto for src attributes.

HTML comments are not preserved.

Requirements

sanitize-html is intended for use with Node. That's pretty much it. All of its npm dependencies are pure JavaScript. sanitize-html is built on the excellent htmlparser2 module.

How to use

Browser

Think first: why do you want to use it in the browser? Remember, servers must never trust browsers. You can't sanitize HTML for saving on the server anywhere else but on the server.

But, perhaps you'd like to display sanitized HTML immediately in the browser for preview. Or ask the browser to do the sanitization work on every page load. You can if you want to!

Clone repository
Run npm install and build / minify:

npm install
npm run minify

You'll find the minified and unminified versions of sanitize-html (with all its dependencies included) in the dist/ directory.

Use it in the browser:

<html>
    <body>
        <script type="text/javascript"  src="dist/sanitize-html.js"></script>
        <script type="text/javascript" src="demo.js"></script>
    </body>
</html>

var html = "<strong>hello world</strong>";
console.log(sanitizeHtml(html));
console.log(sanitizeHtml("<img src=x onerror=alert('img') />"));
console.log(sanitizeHtml("console.log('hello world')"));
console.log(sanitizeHtml("<script>alert('hello world')</script>"));

Node (Recommended)

Install module from console:

npm install sanitize-html

Use it in your node app:

var sanitizeHtml = require('sanitize-html');

var dirty = 'some really tacky HTML';
var clean = sanitizeHtml(dirty);

That will allow our default list of allowed tags and attributes through. It's a nice set, but probably not quite what you want. So:

// Allow only a super restricted set of tags and attributes
clean = sanitizeHtml(dirty, {
  allowedTags: [ 'b', 'i', 'em', 'strong', 'a' ],
  allowedAttributes: {
    'a': [ 'href' ]
  }
});

Boom!

"I like your set but I want to add one more tag. Is there a convenient way?" Sure:

clean = sanitizeHtml(dirty, {
  allowedTags: sanitizeHtml.defaults.allowedTags.concat([ 'img' ])
});

If you do not specify allowedTags or allowedAttributes our default list is applied. So if you really want an empty list, specify one.

"What are the default options?"

allowedTags: [ 'h3', 'h4', 'h5', 'h6', 'blockquote', 'p', 'a', 'ul', 'ol',
  'nl', 'li', 'b', 'i', 'strong', 'em', 'strike', 'code', 'hr', 'br', 'div',
  'table', 'thead', 'caption', 'tbody', 'tr', 'th', 'td', 'pre' ],
allowedAttributes: {
  a: [ 'href', 'name', 'target' ],
  // We don't currently allow img itself by default, but this
  // would make sense if we did
  img: [ 'src' ]
},
// Lots of these won't come up by default because we don't allow them
selfClosing: [ 'img', 'br', 'hr', 'area', 'base', 'basefont', 'input', 'link', 'meta' ],
// URL schemes we permit
allowedSchemes: [ 'http', 'https', 'ftp', 'mailto' ],
allowedSchemesByTag: {}

"What if I want to allow all tags or all attributes?"

Simple! instead of leaving allowedTags or allowedAttributes out of the options, set either one or both to false:

allowedTags: false,
allowedAttributes: false

"What if I don't want to allow any tags?"

Also simple! Set your allowedTag and allowedAttributes to empty arrays ([]).

allowedTags: [],
allowedAttributes: []

Wildcards for attributes

You can use the * wildcard to allow all attributes with a certain prefix:

allowedAttributes: {
  a: [ 'href', 'data-*' ]
}

Also you can use the * as name for a tag, to allow listed attributes to be valid for any tag:

allowedAttributes: {
  '*': [ 'href', 'align', 'alt', 'center', 'bgcolor' ]
}

htmlparser2 Options

santizeHtml is built on htmlparser2. By default the only option passed down is decodeEntities: true You can set the options to pass by using the parser option.

clean = sanitizeHtml(dirty, {
  allowedTags: ['a'],
  parser: {
    lowerCaseTags: true
  }
});

See the [htmlparser2 wiki] (https://github.com/fb55/htmlparser2/wiki/Parser-options) for the full list of possible options.

Transformations

What if you want to add or change an attribute? What if you want to transform one tag to another? No problem, it's simple!

The easiest way (will change all ol tags to ul tags):

clean = sanitizeHtml(dirty, {
  transformTags: {
    'ol': 'ul',
  }
});

The most advanced usage:

clean = sanitizeHtml(dirty, {
  transformTags: {
    'ol': function(tagName, attribs) {
        // My own custom magic goes here

        return {
            tagName: 'ul',
            attribs: {
                class: 'foo'
            }
        };
    }
  }
});

You can specify the * wildcard instead of a tag name to transform all tags.

There is also a helper method which should be enough for simple cases in which you want to change the tag and/or add some attributes:

clean = sanitizeHtml(dirty, {
  transformTags: {
    'ol': sanitizeHtml.simpleTransform('ul', {class: 'foo'}),
  }
});

The simpleTransform helper method has 3 parameters:

simpleTransform(newTag, newAttributes, shouldMerge)

The last parameter (shouldMerge) is set to true by default. When true, simpleTransform will merge the current attributes with the new ones (newAttributes). When false, all existing attributes are discarded.

You can also add or modify the text contents of a tag:

clean = sanitizeHtml(dirty, {
  transformTags: {
    'a': function(tagName, attribs) {
        return {
            tagName: 'a',
            text: 'Some text'
        };
    }
  }
});

For example, you could transform a link element with missing anchor text:

<a href="http://somelink.com"></a>

To a link with anchor text:

<a href="http://somelink.com">Some text</a>

Filters

You can provide a filter function to remove unwanted tags. Let's suppose we need to remove empty a tags like:

<a href="page.html"></a>

We can do that with the following filter:

sanitizeHtml(
  '<p>This is <a href="http://www.linux.org"></a><br/>Linux</p>',
  {
    exclusiveFilter: function(frame) {
        return frame.tag === 'a' && !frame.text.trim();
    }
  }
);

The frame object supplied to the callback provides the following attributes:

tag: The tag name, i.e. 'img'.
attribs: The tag's attributes, i.e. { src: "/path/to/tux.png" }.
text: The text content of the tag.
tagPosition: The index of the tag's position in the result string.

You can also process all text content with a provided filter function. Let's say we want an ellipsis instead of three dots.

<p>some text...</p>

We can do that with the following filter:

sanitizeHtml(
  '<p>some text...</p>',
  {
    textFilter: function(text) {
      return text.replace(/\.\.\./, '&hellip;');
    }
  }
);

Note that the text passed to the textFilter method is already escaped for safe display as HTML. You may add markup and use entity escape sequences in your textFilter.

Allowed CSS Classes

If you wish to allow specific CSS classes on a particular element, you can do so with the allowedClasses option. Any other CSS classes are discarded.

This implies that the class attribute is allowed on that element.

// Allow only a restricted set of CSS classes and only on the p tag
clean = sanitizeHtml(dirty, {
  allowedTags: [ 'p', 'em', 'strong' ],
  allowedClasses: {
    'p': [ 'fancy', 'simple' ]
  }
});

Allowed URL schemes

By default we allow the following URL schemes in cases where href, src, etc. are allowed:

[ 'http', 'https', 'ftp', 'mailto' ]

You can override this if you want to:

sanitizeHtml(
  // teeny-tiny valid transparent GIF in a data URL
  '<img src="data:image/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==" />',
  {
    allowedTags: [ 'img', 'p' ],
    allowedSchemes: [ 'data', 'http' ]
  }
);

You can also allow a scheme for a particular tag only:

allowedSchemes: [ 'http', 'https' ],
allowedSchemesByTag: {
  img: [ 'data' ]
}

Discarding the entire contents of a disallowed tag

Normally, with a few exceptions, if a tag is not allowed, all of the text within it is preserved, and so are any allowed tags within it.

The exceptions are:

style, script, textarea

If you wish to expand this list, for instance to discard whatever is found inside a noscript tag, use the nonTextTags option:

nonTextTags: [ 'style', 'script', 'textarea', 'noscript' ]

Note that if you use this option you are responsible for stating the entire list. This gives you the power to retain the content of textarea, if you want to.

The content still gets escaped properly, with the exception of the script and style tags. Allowing either script or style leaves you open to XSS attacks. Don't do that unless you have good reason to trust their origin.

Changelog

1.13.0: transformTags can now add text to an element that initially had none. Thanks to Dushyant Singh.

1.12.0: option to build for browser-side use. Thanks to Michael Blum.

1.11.4: fixed crash when __proto__ is a tag name. Now using a safe check for the existence of properties in all cases. Thanks to Andrew Krasichkov.

Fixed XSS attack vector via textarea tags (when explicitly allowed). Decided that script (obviously) and style (due to its own XSS vectors) cannot realistically be afforded any XSS protection if allowed, unless we add a full CSS parser. Thanks again to Andrew Krasichkov.

1.11.3: bumped htmlparser2 version to address crashing bug in older version. Thanks to e-jigsaw.

1.11.2: fixed README typo that interfered with readability due to markdown issues. No code changes. Thanks to Mikael Korpela. Also improved code block highlighting in README. Thanks to Alex Siman.

1.11.1: fixed a regression introduced in 1.11.0 which caused the closing tag of the parent of a textarea tag to be lost. Thanks to Stefano Sala, who contributed the missing test.

1.11.0: added the nonTextTags option, with tests.

1.10.1: documentation cleanup. No code changes. Thanks to Rex Schrader.

1.10.0: allowedAttributes now allows you to allow attributes for all tags by specifying * as the tag name. Thanks to Zdravko Georgiev.

1.9.0: parser option allows options to be passed directly to htmlparser. Thanks to Danny Scott.

1.8.0:

transformTags now accepts the * wildcard to transform all tags. Thanks to Jamy Timmermans.
Text that has been modified by transformTags is then passed through textFilter. Thanks to Pavlo Yurichuk.
Content inside textarea is discarded if textarea is not allowed. I don't know why it took me this long to see that this is just common sense. Thanks to David Frank.

1.7.2: removed array-includes dependency in favor of indexOf, which is a little more verbose but slightly faster and doesn't require a shim. Thanks again to Joseph Dykstra.

1.7.1: removed lodash dependency, adding lighter dependencies and polyfills in its place. Thanks to Joseph Dykstra.

1.7.0: introduced allowedSchemesByTag option. Thanks to Cameron Will.

1.6.1: the string 'undefined' (as opposed to undefined) is perfectly valid text and shouldn't be expressly converted to the empty string.

1.6.0: added textFilter option. Thanks to Csaba Palfi.

1.5.3: do not escape special characters inside a script or style element, if they are allowed. This is consistent with the way browsers parse them; nothing closes them except the appropriate closing tag for the entire element. Of course, this only comes into play if you actually choose to allow those tags. Thanks to aletorrado.

1.5.2: guard checks for allowed attributes correctly to avoid an undefined property error. Thanks to Zeke.

1.5.1: updated to htmlparser2 1.8.x. Started using the decodeEntities option, which allows us to pass our filter evasion tests without the need to recursively invoke the filter.

1.5.0: support for * wildcards in allowedAttributes. With tests. Thanks to Calvin Montgomery.

1.4.3: invokes itself recursively until the markup stops changing to guard against this issue. Bump to htmlparser2 version 3.7.x.

1.4.1, 1.4.2: more tests.

1.4.0: ability to allow all attributes or tags through by setting allowedAttributes and/or allowedTags to false. Thanks to Anand Thakker.

1.3.0: attribs now available on frames passed to exclusive filter.

1.2.3: fixed another possible XSS attack vector; no definitive exploit was found but it looks possible. See this issue. Thanks to Jim O'Brien.

1.2.2: reject javascript: URLs when disguised with an internal comment. This is probably not respected by browsers anyway except when inside an XML data island element, which you almost certainly are not allowing in your allowedTags, but we aim to be thorough. Thanks to Jim O'Brien.

1.2.1: fixed crashing bug when presented with bad markup. The bug was in the exclusiveFilter mechanism. Unit test added. Thanks to Ilya Kantor for catching it.

1.2.0:

The allowedClasses option now allows you to permit CSS classes in a fine-grained way.
Text passed to your exclusiveFilter function now includes the text of child elements, making it more useful for identifying elements that truly lack any inner text.

1.1.7: use he for entity decoding, because it is more actively maintained.

1.1.6: allowedSchemes option for those who want to permit data URLs and such.

1.1.5: just a packaging thing.

1.1.4: custom exclusion filter.

1.1.3: moved to lodash. 1.1.2 pointed to the wrong version of lodash.

1.1.0: the transformTags option was added. Thanks to kl3ryk.

1.0.3: fixed several more javascript URL attack vectors after studying the XSS filter evasion cheat sheet to better understand my enemy. Whitespace characters (codes from 0 to 32), which browsers ignore in URLs in certain cases allowing the "javascript" scheme to be snuck in, are now stripped out when checking for naughty URLs. Thanks again to pinpickle.

1.0.2: fixed a javascript URL attack vector. naughtyHref must entity-decode URLs and also check for mixed-case scheme names. Thanks to pinpickle.

1.0.1: Doc tweaks.

1.0.0: If the style tag is disallowed, then its content should be dumped, so that it doesn't appear as text. We were already doing this for script tags, however in both cases the content is now preserved if the tag is explicitly allowed.

We're rocking our tests and have been working great in production for months, so: declared 1.0.0 stable.

0.1.3: do not double-escape entities in attributes or text. Turns out the "text" provided by htmlparser2 is already escaped.

0.1.2: packaging error meant it wouldn't install properly.

0.1.1: discard the text of script tags.

0.1.0: initial release.

About P'unk Avenue and Apostrophe

sanitize-html was created at P'unk Avenue for use in Apostrophe, an open-source content management system built on node.js. If you like sanitize-html you should definitely check out apostrophenow.org. Also be sure to visit us on github.

Support

Feel free to open issues on github.