"Let strings be the result of strictly splitting the string payload on U+0026 AMPERSAND characters"
From: https://www.w3.org/TR/2014/REC-html5-20141028/forms.html#url...
They removed the old 1999 guidance that said "in particular, CGI implementors support the use of ";" in place of "&" to save authors the trouble of escaping "&" characters in this manner."
As it turns out, the issue was in urllib.parse, a library that ships with python.
That's what I thought the article was about.
> If any of the name-value pairs in pairs have a name component consisting of the string "_charset_" encoded in US-ASCII, and the value component of the first such pair, when decoded as US-ASCII, is the name of a supported character encoding, then let encoding be that character encoding (replacing the default passed to the algorithm).
Ouch. I hope nobody actually implements that.
But googling a bit does show it's supported in a lot of software.
For example: https://bugzilla.mozilla.org/show_bug.cgi?id=18643
Flask (or rather Werkzeug what it's based on) removed implicit semicolon support for URL 13 years ago: https://github.com/pallets/werkzeug/commit/0ea28bbc6f5f05eef...
Bottle does use parts of urllib but appears to do it's own splitting on ampersands: https://github.com/bottlepy/bottle/blob/d1413a81ead7a6f130a0...
This is really about caching proxies. If the proxy attempts to cache based on some attributes of the request, then it has a lot of request normalizing it needs to do. Some more examples of things that some languages (in this case Go) may parse differently than some web proxies: https://github.com/golang/go/issues/25192#issuecomment-80625...
And a more overall argument that putting this "vulnerability" on the language library (in Python and Go) does not make the most sense: https://github.com/golang/go/issues/25192#issuecomment-79685...
As we’ve seen with multiple other vulnerabilities this can be very fragile. Especially when the complexity of the protocol grows, the recent saml signature mixups being a good example of such https://news.ycombinator.com/item?id=28064835
If your auth/proxy parses some parameters used by both the proxy and the app it should forward what it parsed, not the original unparsed data. Only exception is a cache or a best-effort-firewall, but then it’s no longer a vulnerability, just an ordinary bug.
https://www.w3.org/TR/REC-html40/interact/forms.html#form-co...
Thankfully that did not happen (widespread), and now people understand they have to escape.
I do remember jsessionid being after the semi-colon..