Persistent XSS on Twitter.com | Hacker News Reader

Persistent XSS on Twitter.com

praetorianprefect.com

62 pointsforkqueue16 comments

This has been demo'd a long time ago already [1], and it seems they haven't done anything yet ? Wtf.

[1] http://www.davidnaylor.co.uk/massive-twitter-cross-site-scri...

EDIT: nevermind.. you're right. WTF is right.

It was fixed, but now it's back again...

> The problem is similar to one described last August by James Slater. That time around the issue was with the application URL, this time it appears the application name is the issue.

"appears to be due to a lack of input validation of the application name field"

They should just be sure that they _render_ the application name field appropriately. Angle brackets should be escaped, minimally. It's really not so difficult, Ruby does it with three calls to gsub: http://rdoc.sourceforge.net/rd/doc/classes/CGI.src/M000003.h...

At least this script in particular seems pretty harmless. I glossed over the "rainbow links" code, so maybe there was something vicious in there.

Either way, XSS sucks. Surprised that they haven't plugged this one yet.

Twitter is probably still using Rails 2.3, where you have to explicitly tell the framework to html escape every time you're outputting a string.

Rails 3 changes this by always html escaping strings.

Security shouldn't be a matter of the framework, especially if it belongs to well known problems like XSS.

With programmers being human, there's a lot to be said for the framework providing a secure default. Even so, it's surprising how often this particular mistake occurs.

You seriously think developers will manually HTML encode every time user input is rendered in the response? It's not just HTML they have to worry about, but Javascript, URL, HTML attributes, etc. If the framework doesn't automatically do it, nobody does it. That is, until they get hit by XSS.

Of course. But there's no reason not to make security easier and more natural (pit of success vs. struggling uphill).

Depends on what you mean by "framework". I would interpret that as "the language in which you write your application", and in that case a language that treats text and HTML as different datatypes does provide more security.

Rails' conflation of these types guarantees that whatever the default for escaping, there will be bugs in applications written in/on rails.

I think twitter is using Lift, not Rails.

Nope

This seems a good time to mention interpolique: http://recursion.com/interpolique.html

I'm curious what people here think of that idea, ie, preventing string injection attacks at the language level.

Twitter sure does have issues with stuff like this. I noticed a while back that they were double encoding some strings on output, too - I had an ampersand in my location and it was showing as & on the page.

None of the code looks malicious, but I would suggest that if you have a Twitter account and/or are logged into it, don't visit the page because he might be stealing cookies.

Reply on news.ycombinator.com