The Security Newsletter
Transcription
The Security Newsletter
N°12/ Winter 2009 The Security Newsletter In this issue INTRODUCTION The news - DNSSEC starts to deploy - A new partnership for asynchronous chips - 3D chips stacking 2 2 Latest attack on WPA 3 Attack on Intel TXT 4 Attack on BGP 4 2 Forging SSL certificates 5 Published Quarterly By: Thomson’s Corporate Research - part of the Licensing, Research & Innovation Division Technical Editor: Eric Diehl Editors: Sharon Ayalde Natalie Hamrick Contributors: Patrice Auffret Olivier Courtay Mohamed Karroumi Sylvain Lelievre Nicolas Prigent VP and Head of Corporate Research: Gary Donnan LR&I Head: Beatrix de Russé Email and to subscribe: [email protected] Copyright Thomson 2008 “Is DRM dead?” I have answered this question countless times, but this quarter it was much more frequent. Once more, Apple bears the hit. In February 2007, the CEO of Apple, Steve Jobs, shared his thoughts on music1. To solve the problem of interoperability of digital music, he foresaw three alternatives: the status-quo, widely license Apple’s DRM FairPlay, or sell DRM free music. He clearly favored the third solution. At the beginning of 2009, Apple started the movement - iTunes selling its catalog as DRM free music. Will other merchants follow? Warner France announced their two sites, Fnac Music and Virgin Media, would sell DRM free songs. Nevertheless, it is only a trial. The final decision will be made in 2010. Is the game over? For music, it is probably true, as the market share of iTunes is so dominant. Furthermore, the iTunes catalog exceeds 8 million titles and it will be difficult for other merchants not to follow this trend. Also, the audio industry is looking for alternative solutions, such as music access or sponsored listening2. Can we extrapolate this trend to video DRM? The answer is no - music and video are different. • Making a blockbuster is far more expensive than recording an album and these investments need to be protected. • Musicians have additional sources of revenue, such as concerts. Many years ago, concerts were the main revenue stream and discs were promotional. However, actors do not have such alternate revenue streams. • The release of a music album is worldwide and internationalization simply requires a new cover. Worldwide releasing of movies is rare, as subtitling or dubbing can be expensive and lengthier. • Music sells with basic versioning: a unique song, album, or collector. Movies use more complex versioning. Furthermore, the window release system is defined by a complex set of legal agreements that require enforcement. These differences highlight that video still may need DRM for some time. Eric Diehl Domain Director, Security 1 The Security Newsletter N°12/ Winter 2009 reduces power consumption. It also reduces noise and gives more resistance to power fluctuation. Therefore, it will lead to better resistance to known side channel attacks such as DPA or DFA. The News DNSSEC starts to deploy The DNS (Domain Name Service) translates human-manageable addresses (e.g. www.thomson.net) into internet-manageable IP addresses. A few months ago, security researcher Dan Kaminsky presented an attack against DNS3 that allowed the corruption of DNS servers, for instance, to impersonate servers on the Internet. Although Dan Kaminsky and most of the DNS software companies quickly provided a solution prior to the official disclosure of the attack, DNSSEC4 is the only long-term efficient solution. By cryptographically extending DNS, DNSSEC allows the authentication of DNS information and thus prevents unauthorized modification. A few weeks after Dan Kaminsky’s presentation, the US Office of E-Government and Information Technologies decided that DNSSEC should be deployed in the full .gov top-level domain (TLD) before the end of 2009. This will be the first time that DNSSEC is fully deployed in a TLD. The first benefit of this deployment is of course to authenticate the addresses of the servers whose names end in .gov. However, a longer-term objective is to encourage Internet Service Providers (which DNS cache servers are considered as the principal target for the attacks) to deploy DNSSEC. Indeed, once the .gov TLD uses DNSSEC, ISPs will have an incentive in using DNSSEC-compliant cache servers. Similarly, once ISPs have DNSSEC compliant cache servers, other top-level domains will have incentives to deploy DNSSEC. Deploying DNSSEC in the .gov domain is a very positive initiative. Nevertheless, extending it to other domains may be more complex. Indeed, other TLDs (such as.com for instance) contain even more domains and are more dynamic. Furthermore, domains that belong to the .gov TLD are strongly controlled. Only official US agencies can register a name in the .gov domain. This control is weaker in the .com domain, for instance, where almost anyone can obtain a name. Consequently, complete and efficient DNSSEC deployment may take much more time. > N. PRIGENT A new partnership for asynchronous chips A partnership between contactless chipmaker Inside Contactless and Tiempo, a French company specializing in asynchronous ICs, is looking at designing the next generation chip that incorporates asynchronous technology. The new chip will use Tiempo’s clockless and delay insensitive technology. This technology provides significant gains in performance and • Power (or electromagnetic) signature of the chip is strongly reduced. Power traces are thus more difficult to resynchronize. • No glitch attack can be applied to the clock signal. Manipulating the power supply has little effect or may lead to deadlocks that are not exploitable. This inherent protection can be improved by adding extra countermeasures during the chip design5. Tiempo already proposes core IPs for the implementation of microcontrollers (16-bit) and cryptoprocessors (DES) with asynchronous logic. Last November, they announced a prototype chip that implements its TAM16 microcontroller. > S. LELIEVRE 3D chips stacking 3D chip stacking is a new trend that extends the functionality of a chip while keeping or reducing its size. It allows the splitting of a large system-on-chip into a stacked-die system. Products already exist (memory chip in particular) that use wire bonding to interconnect identical stacked dies. The capability to produce thinner and thinner dies enables the stacking up of approximately 20 dies. IMEC, a European leading independent nanoelectronics research institute recently demonstrated the first functional 3D integrated circuits using its 3D stacked IC technology (3D-SIC). The dies interconnections are 5µm copper through-silicon via (TSV) passing completely through a silicon wafer or die. This reduces the size of the chip compared to traditional wire bonding. Furthermore, this technology leads to better resistance of the chip against a physical attack by chip observation or micro Figure 1: Stacked IC probing. Each stacked die uses smaller and smaller dimension processes and may use several metallization layers. Reverse engineering this type of component will require more sophisticated equipment and skills. Unfortunately, this ever-shrinking and complex process technology requires new testing equipment. Once available on the secondhand market, the equipment will likely be used by the most motivated and funded attackers! > S. LELIEVRE 2 The Security Newsletter N°12/ Winter 2009 Latest attack on WPA During the PacSec 2008 conference held last November in Tokyo, Japan, researchers Martin Beck and Erik Tews presented the first practical attack6 against WPA-TKIP, one of the two protocols proposed to replace WEP to protect WiFi networks. WPA-TKIP7 is a modified version of WEP designed for legacy Wi-Fi hardware and it mitigates the attacks against WEP by using: 1. A key-scheduling algorithm to diversify the keys provided to RC4 to generate the key-stream. 2. Michael, a new Message Integrity Code (MIC), is appended to the end of the message before the CRC computation and used in addition to the original WEP CRC value. Although it is more secure than CRC, it is still reversible. 3. An anti-replay mechanism that discards messages arriving out of order and resets the communication channel (requiring rekeying) if more than two messages having a correct CRC but an incorrect MIC are received. Despite these counter-measures, Beck and Tews have adapted the “chopchop attack”8 that worked against WEP in order to work against WPA-TKIP. In the traditional chopchop attack, the attacker eavesdrops on an encrypted frame protected by WEP, but the content is easily guessable (ARP messages are targets of choice in this case) and uses the access point as an oracle to obtain the clear content of the frame, and thus the key-stream encrypting the frame. The attacker first removes the last byte of the content in the message and makes a hypothesis on what its clear value could be, and then recomputes a new CRC based on the original encrypted message, the encrypted CRC and the supposed value (details can be found here9). The attacker then generates a new frame by concatenating the encrypted shortened message and the recomputed CRC and sends this frame to the access point. According to the accesspoint reaction, the attacker knows if their hypothesis was valid. If the hypothesis was not valid, the attacker makes another hypothesis and redoes the operation. If the hypothesis is correct, the attacker goes on with the other bytes of the message. In the worst case, the attacker needs to send 256 messages to obtain a byte of clear data, and 128 messages on average. At the end of the process, the attacker knows the clear-text message as well as the key-stream in which it was encrypted. WPA-TKIP’s mitigation features 2 and 3 previously described should prevent the chopchop attack. Indeed, feature 2 requires that messages arriving out of order are dropped. Consequently, the attacker could not reuse the eavesdropped message. However, Beck and Tews have taken benefits of 802.11e, recent Quality of Service (QoS) features mandatory in any 802.11n access points. 802.11e uses different channels to offer QoS, and each of these channels has its own counter. The idea is then to eavesdrop on a message on a busy channel since its counter will be high, and to make the trial-and-error operations on a less busy channel where its counter is lower. Nonetheless, feature 3 should block the chopchop attack that will indeed send many packets to the access point with incorrect MIC. Once again, Beck and Tews circumvented the protection. They consider the MIC as being a normal part of the message and apply the chopchop mechanism to it. The only difference is that if the guess is correct, then the CRC is correct, and the MIC is wrong. The access point then sends a MIC failure message, informing the attacker that indeed they were right. To prevent the rekeying (feature 3), the attacker simply waits one minute before the next guess. Due to the way a WPA-TKIP packet is organized, the attacker only needs 12 rounds of the chopchop algorithm to obtain enough decrypted information to easily brute-force offline the information that is still encrypted in an ARP packet. At the end of the process, the attacker knows the clear-text message, the key stream, and the MIC value. Because Michael is reversible, it is possible to discover the MIC key. Using the keystream and this MIC key, the attacker can inject a single message (the length does not exceed the length of the eavesdropped message) on each of the 802.11e channels, where counter value is still less than the eavesdropped message. In other words, the attacker can, at most, inject seven short messages on the networks. Currently, there is no recognized way to inject arbitrary traffic on WPA-TKIP-protected networks, nor there is a way to obtain the WPA key. This attack is important since it is the first practical attack against WPA-TKIP and it may lead to other attacks. However, WPA-TKIP should not be considered broken. Since this attack does not affect WPA-CCMP, the version of WPA that uses the AES encryption algorithm, we also believe that it would be more secure to switch to this protection method. As a conclusion, two lessons can be learned: First, tweaking an insecure protocol does not make it secure. Second, even for the slightest modifications made to a system (in this case, using more channels to ensure QoS), one should always consider the security implications. It is so easy to open a back door. > N. PRIGENT 3 The Security Newsletter N°12/ Winter 2009 Attack on Intel TXT Many actors in computer data security are aiming at hardware-based security. The most known and used hardware component is the so-called Trusted Platform Module (TPM). TPM is the heir of the contested TCPA/ Palladium project initiated by Microsoft. The TPM specifications are defined by the Trusted Computing Group (TCG). The most known application using TPM is BitLocker™. BitLocker encrypts hard disk under Vista™. The secret key is securely stored by the TPM. However, hardware-based security is never 100% guaranteed. It is possible to reset the TPM1.1 without resetting the Operating System. TPM 1.2 solves this issue but is only available on new computers. Manufacturers such as Intel or AMD are now integrating hardware security in the heart of their architectures (i.e. in the processor and in the motherboard). Recent Intel computers dedicated to the professional market include Active Management Technology (AMT) and Trusted eXecution Technology (TXT). These technologies are a combination of hardware-based components and software. Intel has built a complete security solution that allows the deployment of security updates on computers - even when powered off. This solution also enables the detection of a virus in the OS without any detection software running on this OS. The anti-virus software runs in parallel with the OS. The hardware guarantees the integrity of the anti-virus. Yet, the deployment is slow: If AMT is already being used to manage large computer parks by some companies, TXT is not really used today. Nevertheless, Intel already ships its computer with this technology. TXT may create a secure environment for operating system execution, especially at boot and for virtualization purposes. Intel provides Tboot, (for Trusted Boot) a software based on TXT functionality. Joanna Rutkowska, a recognized expert in trusted computing, recently announced that she would explain how to divert Tboot at the next Black Hat conference10. Details are not yet known, however we are confident about the reality of the flaw. Tboot developers argue that the software is still under development and that they are aware of some potential flaws – and perhaps Rutkowska has discovered one of them. Some hypotheses have circulated regarding the discovered flaw. Tboot seems to use virtual addresses instead of physical addresses. The attack may exploit a flaw in the management of this mode. Tboot cannot be directly attacked, as the attacker should first find a flaw in the xen™ hypervisor (Rutkowska seems to have found a least one). It is important to emphasize that even if Rutkowska has virtually broken Tboot, she is still confident that TXT will be a key technology for trusted computing in the future. > O. COURTAY Attack on BGP At the last DEFCON Las Vegas, researchers Alex Pilosov and Tony Kapela11 demonstrated the ease of implementing an Internet-scale man in the middle attack. This was shown with a live demo, by redirecting traffic for all DEFCON attendees to their own network, all of that in a stealth manner. Border Gateway Protocol (BGP) is a route exchange protocol between Autonomous Systems (AS). Every large company has its own AS along with their own routing architecture. To make an Internet Protocol (IP) frame travel from source to destination, the route needs to be communicated across the entire Internet. This is the purpose of BGP. Each AS exchanges IP network routes with its neighbors using BGP. Those neighbors then exchange those routes with their own neighbors, recursively. A route is exchanged and bound to an AS number, linking an IP prefix to a specific AS. This information chain is known as the AS-PATH. Injecting routes is trivial when you have a BGP router. To hijack an IP network prefix you must announce an IP prefix more specific than the one already announced. For example, if the 1.0.0.0/24 prefix is already announced, you must announce the 1.0.0.0/25 to hijack it. Spammers use this to send their junk emails and avoid complaints being sent to their hosting provider. Although, hijacking in this manner unveils your AS number. Every skilled Internet user may identify the AS announcing this more specific route from the ASPATH, and it will be flagged as potentially malicious. Even worse, legitimate users of the victim IP network prefix will no longer be able to reach it, thus, the victim network will be under a Denial of Service (DoS) condition12. Using a tool such as the traceroute program, legitimate users may be able to trace back the attacker’s AS. 4 The Security Newsletter N°12/ Winter 2009 Pilosov and Kapela demonstrated a stealthier technique to execute the attack. First, they showed a way to mount a true man in the middle. They insert themselves between legitimate users and the victim network while the service continues to work. No more DoS conditions for the victim. The attacker has access to users’ traffic. To ensure success, the attacker must inject routes from the attacker’s network to the victim’s network and from legitimate users to the attacker network. The exact details can be found in their presentation13. 1. A Certification Authority distributes its CA root certificate (the red one in the figure) via browser vendors to users. This root certificate is added in a “trust list” on the user’s PC. This means that all certificates issued by this CA will be trusted by default by the users. Another technique used by these researchers is incrementing the TTL (Time to Live), in order for a traceroute program to be unable to see hops (IP addresses) used by the attacker. Now, the attack is almost perfectly invisible. The only way to know if an attack occurs is by observing BGP route announcements. 3. When a user visits the secure website, the browser asks the certificate to the web server. If its signature can be verified with the certificate of a CA in the trust list, the website certificate will be accepted. The browser then loads the website and all traffic between the browser and the website will be secured using SSL. > P. AUFFRET 2. The website owner purchases a website certificate at the CA (the white one on the figure). This certificate is signed by the CA and guarantees the identity of the website to the users. Forging Certificates In a previous security newsletter14, an attack that exploits collisions in MD5 hash function was presented. The attack was announced by researchers Marc Stevens, Arjen Lenstra, and Benne de Weger. As an illustration, they predicted the outcome of the 2008 US Presidential elections15. The same team recently struck again. With the help of other researchers: Alex Sotriov, Jacob Appelbaum, David Molnar, Dag Arne Osvik, a better attack was designed16. Using the same weakness of MD5, they were able to impersonate any secure website on the Internet, including banking and e-commerce sites To do so, they trick an official Certificate Authority (CA) and forge a rogue intermediary CA certificate that was trusted by most browsers. To better understand the new attack, Figure 2 illustrates how a SSL website works: Figure 2: Certificate Issuing Process Figure 3: The Attack The attack scenario illustrated by Figure 3 is described below. 1. A legitimate website certificate is obtained by a rogue CA (the attacker) from a trusted CA (the blue one in the figure). 2. A fake intermediary CA certificate is constructed (the black one in the figure). It contains the exact signature as the blue website certificate, thus it appears that it is issued by a trusted CA. Then, a fake website certificate (the green one in the figure) containing the genuine website’s identity, but another public key, is created and signed by the rogue CA. Forging the fake intermediary CA certificate is the most interesting part in the attack scenario. Indeed, rogue CA can create unlimited valid website certificates. It exploits collisions in MD5 hash function. Computation used about 200 Sony Playstation®3 (PS3) game consoles. 3. A copy of the secure website is constructed and receives the two fake certificates. Next, known techniques such as phishing redirect users at this rogue website whose look and feel is identical to the legitimate one. 4. Finally, the rogue website presents the two fake certificates to the browser. The signature in the fake website certificate is 5 The Security Newsletter N°12/ Winter 2009 verified with the fake intermediary CA certificate. This fake CA certificate is accepted by the browser, as its signature is verified with the CA root certificate (the red one in the figure) and the user sees a genuine SSL website! This attack is possible because some Certificate Authorities are still signing certificates using MD5. Amongst them are RapidSSL, RSA Data Security and Verisign. The researchers targeted RapidSSL because they could predict some of the fields (serial number and time) of RapidSSL certificates. The attack is not due to a weakness in SSL. This may affect any security application that uses MD5 as a collision free hash function. Our recommendation is to check which Certification Authority issued a certificate as well as the root certificate fields. If the root certificate using MD5 is recent, then it may fall in this attack scenario. In this case, do not trust the site. A more efficient measure would be to remove all the certificates that use MD5 for signing in the trusted list of your browser. Unfortunately, this operation is not straightforward. For some time, many researchers have recommended to stop using MD5. However, despite these warnings, MD5 is still used. Why does it take so long for some organizations to improve their security? Unfortunately, in the case of the Internet, negligence of some may affect every user. Having a more secure Internet requires the collaboration of all actors and users. Remember Law 7: Security is not stronger than its weakest link. > M. KARROUMI Authors What if your public key was not some random-looking bit string, but simply your name or email address? This idea, put forward by Adi Shamir back in 1984, still keeps cryptographers busy today. Some cryptographic primitives, like signatures, were easily adapted to this new “identity-based” setting, but for others, including encryption, it was not until recently that the first practical solutions were found. The advent of pairings to cryptography caused a boom in the creation of new identitybased schemes. A recent book (shown in the image above), co-edited by Marc Joye, summarizes the current state-of-the-art research in this active subfield of cryptographic research. It covers a broad range of aspects, ranging from the mathematical background of pairings and the main cryptographic constructions to software and hardware implementation issues. This self-contained volume bundles fourteen contributed chapters written by experts in the field, and is suitable for a wide audience of scientists, graduate students, and implementers alike. Where will we be? * 5th Information Security Practice and Experience Conference (ISPEC 2009), Xian, China, April 13-15, 2009 Paper presentation: Hash-based key management schemes for MPEG4-FGS, by Mohamed Karroumi and Ayoub Massoudi * NAB Show, Las Vegas, USA, April 23, 2009 Paper presentation: Image and video fingerprinting: forensic applications, by Frédéric Lefebvre, Bertrand Chupeau, Ayoub Massoudi and Eric Diehl References Steve Jobs, “Apple - Thoughts on Music,” February 6, 2007, http:// www.apple.com/hotnews/thoughtsonmusic/. 2 “Digital music Report 2009, New business models for a changing environment” (IFPI, January 2009), http://www.ifpi.org/content/library/ DMR2009.pdf. 3 Patrice AUFFRET, “DNS weakness,” Security Newsletter, no. 11 (September 2008). 4 R. Arends et al., DNS Security Introduction and Requirements (RFC 4033, March 2005), Google Scholar. 5 Y. Monnet et al., “Practical Evaluation of Fault Countermeasures on an Asynchronous DES Crypto Processor,” in Proceedings of the 12th IEEE International Symposium on On-Line Testing (IEEE Computer Society, 2006), 125-130, http://portal.acm.org/citation. cfm?id=1157732.1157776. 6 M. Beck and E. Tews, “Practical attacks against WEP and WPA,” in Proceedings of PACSEC 2008 (presented at the PacSec 08, Tokyo, Japan, 2008). 7 “Temporal Key Integrity Protocol,” in Wikipedia, http://en.wikipedia. org/wiki/Temporal_Key_Integrity_Protocol. 8 “chopchoptheory [Aircrack-ng],” http://www.aircrack-ng.org/doku. php?id=chopchoptheory. 9 Ibid. 10 Joanna Rutkowska, “Attacking Intel Trusted Execution Technology,” invisible things, January 5, 2009, http://theinvisiblethings.blogspot. com/2009/01/attacking-intel-trusted-execution.html. 11 Anton Kapela and Alex Pilosov, “Stealing The Internet - A Routed, Wide-area, Man in the Middle Attack,” in DEFCON 16, 2008, https:// www.defcon.org/html/defcon-16/dc-16-speakers.html#Kapela 12 “Pakistan lifts the ban on YouTube,” BBC NEWS, February 26, 2008, Online edition, sec. Technology, http://news.bbc.co.uk/1/hi/ technology/7262071.stm. 13 Anton Kapela and Alex Pilosov, Defcon presentation, 2008, http:// media.defcon.org/dc-16/video/dc16_kapela-pilosov_stealing/dc16_ kapela-pilosov.m4v . 14 Mohamed Karroumi, “Nostradamus predicts next US President,” Security Newsletter, no. 9 (Spring 2008). 15 Marc Stevens, Arjen Lenstra, and Benne de Weger, “ Predicting the winner of the 2008 US Presidential Elections using a Sony PlayStation 3,” November 30, 2007, http://www.win.tue.nl/hashclash/ Nostradamus/. 16 Alexander Sotirov et al., “Creating a rogue CA certificate,” December 30, 2008, http://www.win.tue.nl/hashclash/rogue-ca/. 1 6