Anatomy of SIP Attacks

Transcription

1 Anatomy of SIP Attacks João M. Ceron, Klas Steding-Jessen, and Cristine Hoepers João Marcelo Ceron is a Secrity Analyst at CERT.br/NIC.br. He holds a master s degree from Federal University of Rio Grande do Sl, Brazil, where he worked with honeypots for botnet detection. He crrently works with incident handling, and his research interests inclde honeypot data analysis. ceron@cert.br Klas Steding-Jessen is CERT. br/nic.br Technical Manager. He holds a PhD in applied compting from the Brazilian National Institte for Space Research, where he worked with honeypots for spam detection. His research area is the se of honeypots for detecting Internet infrastrctre abse by attackers and spammers. He is also one of the athors of the chkrootkit tool. jessen@cert.br Cristine Hoepers is CERT. br/nic.br General Manager. She holds a PhD in applied compting from the Brazilian National Institte for Space Research in the area of honeypot data analysis. Her research interests inclde the se of honeypots for incident detection and trend analysis. She has spoken at several secrity conferences, inclding FIRST, APWG, MAAWG, LACNIC, and AsCERT. cristine@cert.br In the past few years we have seen a steady increase in the poplarity of VoIP (Voice over IP) services. Scans for SIP (Session Initiation Protocol [4]) servers have been reported for many years, and to gather more details abot these activities we emlated SIP servers in a network of 50 low-interaction honeypots, and collected data abot these attacks for 358 days. What will follow is a description of or observations and advice on how to prevent these attacks from being sccessfl. Tracking SIP Servers Abse For qite some time, the secrity commnity has been reporting an increase in scans for the SIP defalt port 5060/UDP, as well as some anecdotal evidence of other types of abse. Similarly, at the CERT.br honeytarg Honeynet Project [3] (a chapter of The Honeynet Project), port 5060/UDP was consistently among the top-10 targeted ports. Bearing that in mind, we have been tracking the abse of SIP servers more closely since last year. This project consists of 50 low-interaction honeypots, based on Honeyd [7], deployed in the Brazilian Internet space. In order to enable Honeyd to collect SIP attack information, we implemented a listener that emlates Asterisk Server [2] and allows the definition of which extensions are available, as well as their defalt responses and passwords. This software allows s to collect the initial stages of a SIP session, logging information sch as attack origins and the phone nmbers the attackers attempted to call. For privacy reasons, we chose not to record adio sessions, limiting the implementation only to the SIP signaling. Figre 1 presents a SIP conversation fragment logged by or listener. There are two SIP methods: REGISTER and INVITE. The first part is a REGISTER reqest. This is sed by a ser agent (UA) for registering contact information, sch as its crrent IP address. The second part illstrates the INVITE method, which is sed to establish a media session between UAs. In this log, a UA places a call from the extension 100 to the external phone nmber 201*****274 (sanitized nmber). Additionally, the ser-agent field shows that this UA has provided the identification string X-Lite release 1006e stamp 34025, a common softphone. ;login: december

2 Figre 1: Honeypot log showing the attacker s IP, the phone nmber being reqested, and the ser agent identification string Making Sense of the Data The traffic targeted to port 5060/UDP in or honeypots was related to the following attack steps: 1. Scanning: searching for SIP servers. 2. Enmeration: once a SIP server is identified, the attackers try to enmerate the server configration, available extensions, and so on. 3. Brte force: attackers try to access extensions that are protected with weak passwords. 4. Abse: after gaining access to a PABX extension, the attackers will try to call external PSTN (Pblic Switched Telephone Network) nmbers, sally to place international calls. In a preliminary analysis of the collected data, we were able to identify that the attackers wold try to call a given nmber by sing several prefixes to increase the attack sccess (see Figre 2). This occrs becase a SIP server can be configred in different ways for example 0 or 9 to access PSTN lines. In some contries, sch as Brazil, one mst also specify the telecommnication operator to be sed for long distance calls. Figre 2: A phone nmber reqested in different ways in order to identify the correct prefix for dialing long distance or international calls Figre 2 illstrates the many variations one attacker was sing for the phone nmber 9725*****586. To deal with this sitation, which we called redndancy, we implemented a heristic to identify it and to store only a nified SIP session related to this nmber in the database. Besides redcing the size of the database, this heristic also helped s to identify the phone nmber s contry code and to 26 ;login: Vol. 37, No. 6

3 correlate calls placed at different times and coming from different sorces, to a niqe phone nmber. Table 1 smmarizes the data that reached or honeypot infrastrctre from September 2011 to September Data Cont REGISTER messages 64,249,923 INVITE messages 1,007,697 Unified INVITE messages 153,773 Uniqe IPs 7,752 Uniqe Atonomos System Nmbers - ASNs 858 Total nmber of days 358 Uniqe sorce contry codes - CCs 83 Table 1: Smmary of the data collected from September 2011 to September 2012 The majority of the REGISTER messages are from atomated scans. Most of them have the signatre of the SipVicios toolkit [5], a collection of tools for aditing SIP-based VoIP systems. The INVITE messages are actal abse attempts directed to the listeners, i.e., phone call attempts. The nified INVITE messages are the INVITE messages after redndancies were identified. Note that the nmber of niqe ASNs and CCs demonstrate a high dispersion of the origin of the attacks. In the following sections, we will focs on the analysis of the nified INVITE messages, inclding the phone nmbers that were called the most and the abse sorces. User Agents and IDS Evasion An important piece of information logged is the ser agent identification string provided by the SIP clients that connected to the honeypots and tried to place a call. The most freqent ser agents provided are presented in Figre 3. Figre 3: Top ser agent identification strings provided by the SIP clients that tried to place calls ;login: december 2012 Anatomy of SIP Attacks 27

4 Note that 61.23% of the connections came from SIP clients that didn t provide any ser agent string. We groped all these clients nder an identification string we called ndefined. This behavior is not expected from SIP clients and may sggest that many attackers are sing cstomized tools to abse SIP servers. Also note that the third most freqent ser agent is the VaxSIPUserAgent, which is sed by a software development kit, also sggesting cstomized tools. Additionally, there was a grop of attempts where the ser agent almost never repeated. In every new session, the client provided a random 20-character ser agent, as shown in Table 2. This behavior was the second most freqent and was observed even in seqential reqests coming from the same IP address. Or best gess is that this is being sed to hide attack fingerprints or to evade IDS detection. Timestamp IP User Agent String T04:02:15Z 194.X.X.131 DmQCAsNRKZYayfosaXES T04:02:17Z 194.X.X.131 yy3bhtwncbpco3knmrqg T04:02:19Z 194.X.X.131 KdUhQNVVxaZYfHg0rXFD T04:02:21Z 194.X.X.131 otyvaff8mpzvis2cff6m T04:02:23Z 194.X.X.131 5y5ttWMXPbFIeyHb4l4D T04:02:25Z 194.X.X.131 YDjb3Q8Wiw6442YCXMnE Table 2: Examples of random ser agent identification strings captred by the honeypots We have also observed ser agents commonly sed by SIP servers, sch as Asterisk. These ser agents cold be fake (set by the attacker) or cold represent compromised SIP servers sed to abse other servers. The remaining ser agents presented in Figre 3 refer to poplar softphones. As we can see, almost 85% of all connections came from cstomized or potentially malicios software. Where Is It Coming From? When looking into the sorce of the abse attempts, we can try to identify specific patterns in the geographical origin and try to identify other characteristics that cold give some insights abot possible motivations. Based on the sorce IP addresses of the attempted calls, we were able to estimate the sorce contry code (CC) for the attacks. The contry code allocation is based on information provided by the Regional Internet Registries (RIRs). Figre 4 shows these top CCs. 28 ;login: Vol. 37, No. 6

5 Figre 4: Top contry codes of call reqests (based on sorce IPs), aggregated by reqested phone nmbers In Table 3 we list the top 10 IP addresses. For each IP, there is also information abot how many different IDD (International Direct Dialing) codes it tried to call and the ser agent string provided. # Cont IP CC IDDs User Agent String 01 19, X.X.205 CN 142 ndefined 02 11, X.X.16 NL 62 ndefined 03 7, X.X.9 US 67 VaxSIPUserAgent/ , X.X.99 US 27 ndefined 05 6, X.X.93 TH 38 ndefined 06 5, X.X.212 FR 2 ndefined 07 5, X.X.83 TH 37 ndefined 08 4, X.X.37 CA 6 ndefined 09 4, X.X.36 MD 48 random 10 4, X.X.204 ID 27 ndefined Table 3: Top sorce IPs, contry codes, nmber of contries called, and the ser agent provided The most freqent CCs observed are US and CN, which are also the ones for three of the niqe IPs that tried to place most of the calls. Note that the top 10 IPs were responsible for 44% of all call attempts. Additionally, the first IP address is responsible for 67% of all attempts coming from Chinese IPs. Likewise, the third and forth IPs were responsible for 47% of all call attempts coming from IPs allocated to the US. Another interesting fact is that the ser agents provided by the top sorce IPs are not those of poplar softphones bt, instead, are possibly from cstomized attack tools. And, most interestingly, all the ser agents provided by the ninth IP were 20-character random strings, as discssed in the previos section. This combination of few IPs with distinctive ser agents points to the possibility of these being roge VoIP servers or proxies sed as hbs to place phone calls. ;login: december 2012 Anatomy of SIP Attacks 29

6 Considering that one of the expected behaviors of a roge VoIP server is high geographic dispersion of the destination phone calls, we tried to corroborate this hypothesis with additional analysis. We sed AfterGlow [1] to explore the relationship among the top IP (113.X.X.205) and the destination of all calls. Figre 5 presents this IP address and the contry codes it attempted to call. The CC was determined sing the Perl library Nmber::Phone::Contry, that associates an IDD to a contry code. We can see that this IP, a possible VoIP server, places calls to 142 different contries. Figre 5: Destination contry code for all calls placed by the IP 113.X.X.205 Who They Are Calling, and Why... To gain more nderstanding of the abse, we have also stdied the natre of the phone nmbers the attackers attempted to call. The most reqested phone nmbers fall into the following categories: Cell phones: identified by the nmber prefix Financial services: cstomer services from financial instittions (mainly Bank of America and Citibank) Pre-paid phones: pre-paid card services for international phone calls (Net2Phone) Cstomer relations: e-commerce cstomer relation services The most called phone nmber was Bank of America s Credit Card Cstomer Service, totaling 5,090 attempts. Only seven IPs reqested this phone nmber, and for of them have the roge VoIP server behavior that was discssed earlier. Actally, 64% of all calls to Bank of America s Credit Card Cstomer Service came from the eighth IP listed in Table ;login: Vol. 37, No. 6

7 Sorce IP Destination IDD Cont (%) PS IL % EG EG % MD CZ % US CZ % FR RU % CA 800 (Free) % US IL % US ZW % CA CZ % DE CZ % Table 4: Most freqent combinations of sorce IPs and destination IDD contry codes The IP addresses that originated the calls were, for the most part, not the same as the IDD destination contry. Table 4 shows the most freqent pairs, consisting of IPs allocated to a contry code that are calling nmbers in a given destination IDD contry code. Based on the data analyzed so far, we can present some hypotheses abot the attackers motivations: 1. Absing SIP servers in order to place free phone calls or to gain anonymity; 2. Absing the premim-rate telephone nmbers bsiness model; 3. Reselling VoIP services by absing poorly configred SIP servers; and 4. Validating personal identifiable information, sch as credit cards and bank accont details. Secring Yor SIP Server The types of activities observed reinforce the importance of implementing the crrent SIP secrity best practices [6]. Most attacks wold have been prevented or mitigated by following one or more of these recommendations: Protect the SIP server from the Internet: be more restrictive in terms of which extensions can be reached from external IP addresses. Use strong passwords: se long, hard-to-gess passwords. Most SIP clients reqire the password to be entered only once, so there is no need to create easy-toremember passwords. The crrent recommendation is to se at least 12-character passwords, inclding nmbers, symbols, and lower and ppercase letters. Create sernames different from extensions: most brte force attempts try sernames that match the extension nmbers. Monitor the SIP se in yor organization: monitor yor SIP server logs for abse attempts, bt also keep an eye on yor PSTN billing information, looking for nsal long distance and international calls. ;login: december 2012 Anatomy of SIP Attacks 31

8 Conclsion As the adoption of SIP services grows, being aware of the characteristics of the abse against them is increasingly important. As or analysis showed, almost 85% of all call reqests came from cstomized or potentially malicios software, and some of the calls may be related to nlawfl activities. Also, becase there are attackers crrently taking advantage of poorly configred servers, the need to increase monitoring is clear. The good news is that the implementation of basic VoIP secrity best practices will prevent most of the attacks seen in the wild. References [1] AfterGlow: Link Graph Visalization: [2] Asterisk: The Open Sorce Telephony Projects: [3] CERT.br: honeytarg Honeynet Project: [4] RFC 3261 SIP: Session Initiation Protocol: rfc3261.txt. [5] SIPVicios Tools for aditing SIP-based VoIP systems: [6] John Todd, Seven Steps to Better SIP Secrity with Asterisk, 2009: [7] Niels Provos and Thorsten Holz, Virtal Honeypots: From Botnet Tracking to Intrsion Detection, Addison-Wesley, BECOME A USENIX SUPPORTER AND REACH YOUR TARGET AUDIENCE The USENIX Association welcomes indstrial sponsorship and offers cstom packages to help yo promote yor organization, programs, and prodcts to or membership and conference attendees. Whether yo are interested in sales, recriting top talent, or branding to a highly targeted adience, we offer key otreach for or sponsors. To learn more abot becoming a USENIX Spporter, as well as or mltiple conference sponsorship packages, please contact sponsorship@ senix.org. Yor spport of the USENIX Association frthers or goal of fostering technical excellence and innovation in netral forms. Sponsorship of USENIX keeps or conferences affordable for all and spports scholarships for stdents, eqal representation of women and minorities in the compting research commnity, and the development of open sorce technology ;login: Vol. 37, No. 6