Monday, February 15, 2010

How weak is Facebook's XMPP security?

As I mentioned earlier, Facebook does not use SSL on their new XMPP server. All of their security is based on an algorithm called Digest-Md5. As the name says this is based on the popular MD5 hashing algorithm that is no longer considered secure.  From Facebook:

No. At this time, Facebook Chat does not support SSL.
Does Facebook Chat use plaintext authentication?
No. Facebook Chat uses DIGEST-MD5 during authentication.

Are my Chat messages encrypted?
No. However, authentication information is secured using DIGEST-MD5.

Regarding MD5: 
US-CERT, the US Department of Homeland Security's Computer Emergency Readiness Team, says:
  • "Do not use the MD5 algorithm"
    "Software developers, Certification Authorities, website owners, and users should avoid using the MD5 algorithm in any capacity. As previous research has demonstrated, it should be considered cryptographically broken and unsuitable for further use."
  • "Scrutinize SSL certificates signed by certificates using the MD5 algorithm"
    "Users may wish to manually analyze the properties of web site certificates (...) Certificates listed as md5RSA or similar are affected. Such certificates that include strange or suspicious fields or other anomalies may be fraudulent. Because there are no reliable signs of tampering it must be noted that this workaround is error-prone and impractical for most users."

What happens when you don't use SSL for your chat? People sniffing the network can see what you are saying.  This is very simple to do.
Either use tcpdump  from the command line (works on Mac,linux etc) or grab a packet sniffer like wireshark . I would recommend wireshark anyway because it makes the next part even easier.  If you used tcp dump, this you would have something like this

sudo tcpdump -i en0  -s500 -w ~/Desktop/DumpFile01.pcap -vv 

Lets say you have a network dump somehow and you open up wire shark. set it to filter jabber and you will start to see something like this. I've highlighted a message packet so you can see that I was able to read this message someone sent.

Now everything sent on the network can be read and the password "security" is based on MD5 right? behold a captured password sequence:

this decodes from  base64 to:

Excellent we have some information. The client replied with:


this decodes from base64 to:


Nonce and cnonce are used to calculate the encrypted "response". This is where the password is. nonce and cnonce are really there for one time use to prevent a replay attack.

What do we know now?

password hash: 520a27b9dc9195dc12c55cdf81529b62
nonce: 67ED7BA876807C28B3AB6FA634FA1605
cnonc: d33c8862126626563771097b70
qpop: auth
nc: 00000001

password hash is based on a series of md5 hashes:
first hash (X) is of username:realm:password
we know this to be = (password)
Next hash (Y) is X:nonce:cnonce
we know this to be: (password hash): 67ED7BA876807C28B3AB6FA634FA1605: d33c8862126626563771097b70

Next hash (Z) fully known to be of AUTHENTICATE:xmpp/

Final hash is the response value, which is also known and we know it is in the form:

we fully know Z, nonce, cnonce, qpop,  nc

The only thing missing here is Y, which is dependent on X,  which is just an unsalted  MD5 hash based on a password. How would you crack this? There are a few ways,  but the simplest in most cases is just to use rainbow tables  but compute the hash 2 times more using the other parts of the string from above and  see if it matches the response. I'm going to stop the explanation here for obvious reasons.  More info on rainbow tables here:

There is also the possibility of finding  another string that will create the same password hash. This is called a hash collision. In this case, the attacker doesn't even need to know the actual password. They can just use this other string in its place and access an account just the same.

If the person is using a short or insecure password a dedicated  attacker (identity thieves, spammers etc) should have it cracked easily and since the packet was captured, the  attacker can test against the password as much as he likes on his own machine.


  1. Well not really. MD5 is broken but in other ways not applicable here (e.g. collisions). Also, your 'Y' is heavily salted by nonces.

  2. This comment has been removed by the author.

  3. the nonce are both known. they are not there as salts, they are there to prevent replay attacks.


Note: Only a member of this blog may post a comment.