[VoiceOps] Mitigating SIP threats with SBC policies, configuration settings

[Mark R Lindsey]

Robert -- These are good questions.

On Sep 3, 2014, at 11:49 , Robert Nystrom <ronystrom at gmail.com> wrote:
> First:  Customer's ACME is visible from public Internet on udp/5060, and SIP trunk is only being used to interconnect to SIP trunk carrier for inbound and outbound dialing.  I've tested the SBC from the Internet and it actually responds to INVITE and REGISTER messages (with 403 Forbidden).  They are alsonot supposed to be allowing any REGISTER for remote user MD5 Digest Authentication - but it does respond.  Question:  Is there any operational need or business usage case that you would see that would make this setup a good idea? 

Yes; you often do want to be able to REGISTER with the SBC from random places on the Internet. Does your specific customer need to allow registration from anywhere on the Internet? Maybe not. One popular place to blacklist in advance is APNIC IP space.

The Acme Packet has auto-blacklisting features that can be setup so that if a specific source IP sends several SIP messages without successfully registering or completing a phone call, then the SBC can blacklist the source for a while. E.g., if you don't successfully register within your first three SIP messages, then blacklist you for an hour.

If you do know in advance all the right places from which you 'd be sending SIP to the SBC, then it *is* best to setup the SBC for default-deny so that traffic only from those sources is permitted. That's straightforward to do by matching access-control-trust-levels between the realm-config and the access-control objects.

> Second:  I have written some SIP software that sends malformed message headers, and have noticed that the SBC responds with different errors other than 403 Forbidden when headers have unexpected values.  For example, when I send an INVITE with extra CRLFs, I get a 400 CSeq missing header.  When I send a Contact header of "None", I get 400 Invalid Contact.  This leads me to believe that the SBC sip parser is parsing all of the SIP message rather than always sending a 403 Forbidden to an IP address sourced from the untrusted public Internet...this also seems to be very risky.  Is there a specific security configuration with the policies that you would recommend?  It seems like this introduces the risk of DoS and fuzzing attacks if the SBC is parsing more of the SIP message rather than just dropping the message based on invalid source IP.  Could lead to cpu and memory issues if the queues are filled from invalid and fuzzed traffic.

The auto-blacklisting functions can help mitigate this risk. In addition to blacklisting based on failure to successfully REGISTER (or do anything else allowed by policy), you can auto-blacklist sources of "invalid signaling". I haven't found a good written definition of "invalid signaling", but I have definitely seen devices that sent slightly-malformed SIP trigger it and get blacklisted. Non-RFC3261-compliant CR's and LF's are definitely a popular way to do it. Bria/EyeBeam/X-Lite's "UDP Keepalives" (0-byte UDP datagrams) have been another way.

Further, there are internal resources (primarily, CPU capacity) allocated differently for trusted sources versus untrusted sources. An untrusted source could be a device that hasn't yet successfully registered , for example, and we might only want to give 2% of total system capacity to all untrusted sources. 

To your bigger point, though, Oracle/Acme Packet does try to be a security device, and I know they test it against fuzzers. Their release notes show when they fixed a security bug or a SIPd crash that was found through fuzzer testing. Yes, it means they have to be extremely careful with how they parse the SIP in order to do so safely.

>>> mark at ecg.co +1-229-316-0013 http://ecg.co/lindsey

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://puck.nether.net/pipermail/voiceops/attachments/20140903/87fd9259/attachment.html>