Critical Bug Last Year Allowed Bypassing Authentication On HPE ILO4 Servers With 29 'A' Characters (bleepingcomputer.com)

Public exploit code has been published for a severe vulnerability which last year affected Hewlett Packard Integrated Lights-Out 4 (HP iLO 4), a tool for remotely managing the company's servers.

HPE "silently released" patches last August, an anonymous reader reports, adding "details only emerged this spring after researchers started presenting their work at security conferences." The vulnerability is an authentication bypass that allows attackers access to HP iLO consoles. Researchers say this access can later be used to extract cleartext passwords, execute malicious code, and even replace iLO firmware. But besides being a remotely exploitable flaw, this vulnerability is also as easy as it gets when it comes to exploitation, requiring a cURL request and 29 letter "A" characters, as below:

curl -H "Connection: AAAAAAAAAAAAAAAAAAAAAAAAAAAAA"

Because of its simplicity and remote exploitation factor, the vulnerability — tracked as CVE-2017-12542 — received a severity score of 9.8 out of 10.

Comments

Re:Cleartext passwords

[sjames]

They used a shellcode exploit to return the contents of a file on the ILO processor that has the passwords in cleartext! They didn't publish that as far as I can see, but there is a published python program to add a new user with admin privileges and a password of your choice.

Bad HP! Go stand in the corner.

Is it just me or have HP servers been a bit flaky for the last 5 years or so?

This is what happens

[dgatwood]

... when your network infrastructure was certified secure by the Fonz.

Aaaaaaaaaaaaaaaaaaaay.

Programmers think about making things work right

[raymorris]

You may have hears the phrase "garbage in, garbage out".
That's how programmers used to think. The design and test code try to make it work right, when the user uses it right, of course. If the user mashes keys at random, random things might happen. That used to be an okay way of thinking.

The internet has changed that. Now the user (connecting over the internet) WILL mash keys at random. Well, their script will send random bytes. It's no longer okay for software to respond in random ways when it receives random input. Any software accessible via the network MUST be designed thinking about how things can go wrong, not just about how it should work correctly.

Many programmers, especially those who learned writing desktop applications, still think in terms of the program doing the right thing when it receives sane input. Insane input isn't handled securely. The programmers who wrote the ilo software made this mistake.

Specifically, the input is up to 16 characters long, so they failed to handle the case of very long input. Network. Software should be tested with these inputs, at least:

Empty input
Zero
The null character (ASCII 0)
Very long input

Re:Cleartext passwords

[Junta]

The password is used directly as a shared secret for HMAC in IPMI. Therefore to support the ipmi protocol, the server must be able to know the plaintext of the password to a) prove to the client that they know the secret and b) to validate the HMAC sent by the client.

Another potentially tricky one is SNMP. It's a shared secret, but at least you can localize the key. Of course it is localized to an snmp engine id, so while you may not directly have the cleartext password, you can spoof a matching snmp engine id to use the localized key as if you knew the password, at least to impersonate an snmp agent.

Even on the TLS side of things, in practice things are not rosy because the vast majority of this class of devices have a self-signed cert, with all automation disabling cert validation and all users blindly clicked 'continue' at the warnings (there's no harsher message for "we have a conflicting stored cert" than "this is a self signed cert we haven't seen before")