1 / 59

Cryptography And Secure Channels

Cryptography And Secure Channels. Kenny Paterson Information Security Group Royal Holloway, University of London 05/22/09 | Cryp-203. What is a secure channel?. Why ad hoc approaches to building secure channels may be dangerous.

gerodi
Download Presentation

Cryptography And Secure Channels

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Cryptography And Secure Channels Kenny Paterson Information Security Group Royal Holloway, University of London 05/22/09 | Cryp-203

  2. What is a secure channel? Why ad hoc approaches to building secure channels may be dangerous Why the order of encryption and integrity protection matters Conclusions Short theoretical interlude Why encryption alone is not enough Agenda

  3. What is a secure channel?

  4. The Secure Channel Concept • We often want to guarantee the confidentiality and integrity of data travelling over insecure networks. • Applications: • Branch office connectivity. • Connecting to business partners at remote site. • Remote access for employees. • Remote administration of network devices and servers. • E-commerce: protecting credit card numbers in transactions. • Secure file transfers. • …

  5. The Secure Channel Concept • Achieved by building a secure channel. • Typically this secure channel will offer: • Data origin authentication • Data integrity • Confidentiality • Anti-replay • But usually not: • Non-repudiation • Any security services once data has been received

  6. The Secure Channel Concept Secure channels are usually constructed as follows: • An authenticated key establishment protocol. • During this one or both parties is authenticated and a fresh, shared secret is established. • Using asymmetric (public key) or symmetric cryptography, or a combination of the two. • A key derivation phase. • MAC and encryption keys are derived from the shared secret established during protocol. • And then further traffic protected using derived keys. • MAC gives data integrity mechanism and data origin authentication. • Encryption gives confidentiality. • Using symmetric cryptography for speed.

  7. IPsec, SSL/TLS and SSH • Three commonly used protocols for building secure channels • All three use an authenticated key establishment protocol followed by key derivation. • IKE and IKEv2 in IPsec. • Handshake Protocol in SSL/TLS. • Authentication Protocol in SSH. • All three then use these keys with symmetric algorithms to build a secure channel. • With subtle but important differences in the approaches taken.

  8. This Talk In this talk, we will: • Examine how each of IPsec, SSL/TLS and SSH use symmetric key cryptography to provide secure channels. • Key establishment and key derivation are out of scope for this talk. • Discuss each in relation to our current theoretical knowledge. • Highlight some of the pitfalls in the design and implementation of these protocol families. • Identify open research problems for the design of secure channels.

  9. Short Theoretical Interlude

  10. Short Theoretical Interlude • Security models for symmetric encryption are well established. • IND-CPA security: • Attacker has no access to decryption oracle, but has access to encryption oracle. • Attacker has to decide which one of two selected messages was encrypted by the challenger. • Easy to achieve with, e.g. block cipher in CBC mode with random IVs, or block cipher in CTR mode. • See Bellare et al., FOCS’97. • IND-CCA security: • Attacker now has access to encryption and decryption oracles. • All basic modes of operation are insecure in this model.

  11. Short Theoretical Interlude • IND-CCA security from generic constructions (Bellare and Namprempre): • Encrypt-then-MAC: SECURE • As used by IPsec ESP enc+auth. • Encrypt-and-MAC: INSECURE • MAC can leak plaintext information. • But specific instantiations may be provably secure, e.g. as used in SSH (Bellare, Kohno and Namprempre) or Kerberos (Boldyreva and Kumar). • MAC-then-Encrypt: INSECURE • But MAC followed by CBC mode encryption or stream cipher is IND-CCA secure (Krawczyk). • As used by SSL/TLS. • Also available: dedicated authenticated encryption algorithms – e.g. CCM, EAX, GCM, OCB.

  12. Why encryption alone is not enough – IPsec

  13. Encryption in IPsec • IPsec provides security at the IP layer. • Cryptographic protection of IP packets and their payloads. • Widely used in VPNs and remote access systems. • ESP = Encapsulating Security Protocol. • v1, v2, v3 in IETF RFCs 1827, 2406, 4303. • IPsec’s “encryption workhorse”. • ESP provides one or both of: • Confidentiality for packet/payload (v1, v2, v3). • Integrity protection for packet/payload (v2, v3). • ESP uses symmetric encryption and MACs. • Usually CBC mode of block cipher for encryption. • HMAC-SHA1 and HMAC-MD5 for integrity protection.

  14. ESP header ESP trailer Encrypt Encrypted inner packet MAC Outer IP header MAC tag ESP header Encrypted inner packet ESP in Tunnel Mode Original IP packet Inner IP header Inner payload • Encrypt-then-MAC construction, generically secure. • ESP trailer contains padding, pad length and next header field. • ESP commonly uses a block cipher in CBC mode for encryption. CINS/F1-01

  15. Ci-1 Ci dK dK Pi-1 Pi CBC Mode – Reminder Pi-1 Pi Typical block size n: 64 bits (DES, triple DES) or 128 bits (AES). Typical key size: 56 bits (DES), 168 bits (triple DES), 128, 192 or 256 bits (AES). eK eK Ci-1 Ci

  16. ESP header ESP trailer Encrypt Encrypted inner packet Outer IP header Integrity Protection and ESPv1 • ESPv1 (RFC 1827) provided no integrity protection. • Reliant on separate AH protocol to provide this. Original IP packet Inner IP header Inner payload

  17. Integrity Protection and ESPv1 • Bellovin-Wagner attack on ESPv1 without AH: • Limited plaintext recovery: • One byte of plaintext from each TCP segment. • Requires ciphertexts matching 224 chosen plaintexts. • Requires receiver to ignore encryption padding format. • Attack fails if padding check carried out upon decryption. • Requirements can be reduced to 28 chosen plaintexts if variable length padding acceptable.

  18. Integrity Protection and ESPv2 • IETF response to Bellovin-Wagner: • ESPv2 (RFC 2406) recommends receiver should check format of encryption padding. • Sufficient to stop Bellovin-Wagner attack. • Also includes integrity protection as an option. • But implementations MUST still support “encryption-only” mode. • ESPv2 represents a compromise between improving security and maintaining backwards-compatibility.

  19. Integrity Protection and ESPv3 • ESPv3 (RFC 4303): • Still allows encryption-only ESP: “ESP allows encryption-only … because this may offer considerably better performance and still provide adequate security, e.g., when higher layer authentication/integrity protection is offered independently.” • Implying a MAC-then-encrypt construction. • But no longer requires support for encryption-only modes. • Gives strong warnings about Bellovin-Wagner attack and refers to theoretical cryptography literature to motivate need to use integrity protection.

  20. Encryption-only ESP in the Real World • The theoretical cryptography community is well aware of the need to carefully combine integrity protection with encryption. • To achieve IND-CCA security and so prevent active attacks against confidentiality. • It is also well-known amongst IPsec experts that encryption-only configurations should be avoided. • Clear warnings against their use in the RFCs. • So is there really any problem if encryption-only modes continue to be allowed in the RFCs?

  21. Encryption-only ESP in the Real World • Developers are required by RFC 2406 to support encryption-only ESP. • Developers rarely pass RFC warnings to end users. • Developers don’t properly implement RFCs. • End users don’t read RFCs or technical papers. • End users might reasonably assume that encryption on its own gives confidentiality. • On-line tutorials sometimes do not highlight the dangers of encryption-only IPsec…

  22. Encryption-only ESP in the Real World From the IPsec Tunnel Implementation administrator's guide of a well-known vendor (until very recently): “If you require data confidentiality only in your IPSec tunnel implementation, you should use ESP without authentication. By leaving off the authentication service, you gain some performance speed but lose the authentication service.”

  23. Attacking Encryption-only ESP • Paterson and Yau (Eurocrypt 2006): • Efficient ciphertext-only, active attacks against encryption-only ESP. • Implemented in a realistic lab setting against the Linux implementation. • Attacks do not work for implementation strictly following the RFCs. • Degabriele and Paterson (IEEE S&P 2007) • Ciphertext-only attacks against encryption-only ESP. • Less efficient than Linux attacks. • But attacks should work for any implementation strictly following the RFCs. • Implemented against OpenSolaris implementation.

  24. Bit Flipping in CBC Mode • Flipping bits in ciphertext block Ci-1 leads to controlled changes in plaintext block Pi upon decryption. • But block Pi-1 is randomised. Flipping bits here Ci-1 Ci dK dK Pi-1 Pi And randomised block here Leads to bit flips here

  25. Attacking Linux ESP Paterson and Yau (Eurocrypt 2006): • Exploit bit flipping weakness of CBC mode encryption. • Modify headers for inner packets so they produce error messages when processed by IP. • Error messages are carried by ICMP and reveal partial plaintext data. • Modify headers so that error messages are sent outside of IPsec tunnel. • In some variants, straight to attacker’s machine.

  26. Example: Protocol Field Manipulation for 128-bit Blocks Encrypted inner packet IV C1 C2 C3 dK dK dK Flip bits here Dest addr Payload Payload PF Csum Src addr To change protocol field and source address here Inner packet Correction of checksum via further IV bit flips

  27. Example (ctd.) • End-host generates ICMP “protocol unreachable” message in response to modified protocol field in header. • In Linux, ICMP payload carries inner packet header and up to 528 bytes of inner packet’s payload. • Payload now in plaintext form. • ICMP message is sent to host indicated in source address. • And we have modified this so ICMP message does not pass through IPsec tunnel. • Attacker intercepts ICMP message to get plaintext bytes.

  28. Characteristics of Linux Attacks The attacks: • Recover complete contents of IPsec-protected datagrams. • But not encryption keys. • Are efficient. • Can be run in near real-time against an IPsec tunnel. • Do not require special operating conditions. • Attacker needs to capture packets from network, inject packets into network, and monitor gateway for ICMP messages. • Work in practice against Linux implementation of IPsec. • But not against implementations strictly compliant with the RFCs.

  29. Attacking Linux ESP – Reactions Some reactions to the Linux ESP attacks: • “…the possibility of active attacks on unauthenticated but encrypted ESP packets is well known, and we advise against such use in the most recent set of IPsec documents. These documents have been approved for publication by the IESG and are in the queue to be published as RFCs. As a result, no further, substantiative changes will be made.” • “This is all very well understood among the IPSec community, and is not news.” • “I think the spec is clear about the dangers of encryption without authentication. If anyone built implementations that negotiate encryption without authentication, then maybe they weren't paying attention closely enough.”

  30. Attacking the RFCs – Main Ideas Degabriele and Paterson (IEEE S&P 2007): • Extension of Vaudenay’s padding oracle attacks on CBC mode (Eurocrypt 2002) combined with Paterson-Yau techniques. • Attack should work if (and only if) implementor has followed all the advice in IPsec RFCs: • Check correctness of padding format, as per RFCs 2406, 4303. • Post-processing policy checks (omitted in Linux). • Resulting attack is less efficient than Linux attack, but still recovers all plaintext. • About 216 packet injections to decrypt each block. • So are any implementations strict enough?

  31. Implementing the RFC Attack • We examined 10 different open-source IPsec implementations: • Linux kernel, KAME, OpenBSD, FreeBSD, NetBSD, MacOS X, Openswan, strongSwan, FreeS/WAN, OpenSolaris. • All except one (OpenSolaris) failed to follow some of the advice to implementors given in the RFCs. • But variants of our attacks could still extract some plaintext data in each case. • OpenSolaris closely followed the RFCs but had a bug in the padding check implementation. • Once the bug was fixed (Release 55) we were able to break this implementation.

  32. Summary of ESP Attacks • We produced a range of attacks against encryption-only ESP that work: • Against any implementation strictly following the RFCs, e.g. OpenSolaris. • Against many implementations not following the RFCs, e.g. Linux. • The attacks that work in practice shouldn’t work against the RFCs. • The attacks that work against the RFCs often don’t work in practice.

  33. Final Remarks on IPsec • The attacks show that encryption-only ESP is dangerously weak in a practical sense. • There are not so many real-world examples where the lack of integrity protection has such devastating consequences. • This kind of attack seems necessary to convince practitioners. • No security is gained from the provision of upper layer integrity protection, despite claims to the contrary in ESPv3. • An appropriate combination of encryption and integrity protection is needed. • As provided, for example, by ESP with encryption and integrity protection (Encrypt-then-MAC construction).

  34. Final Remarks on IPsec • We cannot view cryptography in isolation. • Our ESP attacks arose from interaction between cryptographic processing and network protocols. • Sometimes called reaction attacks. • Attacks in theory versus attacks in practice. • Implementations often deviate from standards (RFCs) in small ways that turn out to be important for security. • These deviations can result in big differences between attacks on paper and attacks that work in practice. • Our attacker is incomparable with the IND-CPA attacker that is considered in theoretical work.

  35. Why ad hoc approaches to building secure channels may be dangerous – SSH

  36. SSH Secure Shell or SSH is a network protocol that allows data to be exchanged using a secure channel between two networked devices. Used primarily on Linux and Unix based systems to access shell accounts, SSH was designed as a replacement for TELNET and other insecure remote shells, which send information, notably passwords, in plaintext, leaving them open for interception. The encryption used by SSH provides confidentiality and integrity of data over an insecure network, such as the Internet. – Wikipedia CINS/F1-01

  37. SSH • SSHv1 had several security flaws. • Worst ones arising from use of CRC algorithm to provide integrity. • Enabling, for example, traffic injection attacks. • SSHv2 was standardised in 2006 by the IETF in RFCs 4251-4254. • RFC 4253 specifies the SSH Binary Packet Protocol (BPP). • SSHv2 is widely regarded as providing strong security. • It is widely used to secure remote administration of sensitive systems. • But one minor flaw in the BPP that in theory allows distinguishing attacks (Dai; Bellare, Kohno and Namprempre). • Several minor variants of the SSH BPP were proven secure by Bellare, Kohno and Namprempre. • E.g. SSH-$NPC: SSH using randomised padding and CBC mode encryption with explicit IVs. • This formal security analysis was not integrated with the design phase. CINS/F1-01

  38. Sequence Number Packet Length Pad Len Padding 1 4 4 ≥4 The SSH BPP Payload Encrypt MAC Ciphertext MAC tag • Encode-then-Encrypt&MAC construction, not generically secure. • Packet length field measures the total size of the packet on the wire in bytes and is encrypted to hide the true length of SSH packets. CINS/F1-01

  39. Ci-1 Ci dK dK Pi-1 Pi CBC Mode in SSH Pi-1 Pi • RFC 4253 mandates 3DES-CBC and recommends AES-CBC. • In fact, all originally specified optional configurations involve CBC mode, and ARCFOUR was the only optional stream cipher. • SSH uses a chained IV in CBC mode: • IV for current packet is the last ciphertext block from the previous packet. • Effectively creates a single stream of data from multiple SSH packets. eK eK Ci-1 Ci

  40. Attacking the SSH BPP • We (Albrecht, Paterson and Watson, IEEE S&P 2009) have recently developed plaintext recovering attacks against SSH. • These new attacks exploit the interaction of the following BPP features: • The packet length field encodes how much data needs to be received before the MAC is checked. • So this field must be decrypted before the MAC is checked, and the amount of data needed to trigger the MAC check depends on the content of this field. • The attacker can send data on an SSH connection in small chunks (TCP). • CBC mode is mandated. • A MAC failure is visible on the network.

  41. Attacking the SSH BPP (Theory) • The attacker intercepts a target ciphertext block Ci* from the SSH connection. Ci* = eK(Ci-1*Pi*), i.e. Pi* = Ci-1* dK(Ci*) • The attacker injects Ci*as the first block of a new SSH packet on the connection. • The recipient will treat the first 32 bits of the corresponding plaintext block P0’as the packet length field. Here: P0’ = IVdK(Ci*). • The attacker then feeds random ciphertext blocks into the SSH connection. • One block at a time, waiting to see what happens with each new block.

  42. Attacking the SSH BPP (Theory) • Once enough data has arrived, the recipient checks the MAC. • This check will fail with overwhelming probability. • Consequently the connection is terminated with an error message. • How much is enough? • The number of random blocks required to trigger this behaviour reveals the content of the 32-bit packet length field, i.e. the first 32 bits of: P0’ = IV dK(Ci*). • Because of CBC mode, this in turn reveals 32 bits of the target plaintext block: Pi* = Ci-1*  dK(Ci*) = Ci-1*  IV P0’

  43. Attack Performance • As described, the attack would succeed in recovering 32 bits of plaintext with probability 1. • But would require the injection of about 227 random blocks to trigger the MAC check. • And would lead to an SSH connection tear-down. • The attack still works if a fresh IV is used for each new SSH packet. • Breaking SSH-$NPC that was proven secure by Bellare, Kohno and Namprempre. • In practice, things are a bit more complicated….

  44. Attacking OpenSSH • OpenSSH is the most popular implementation of the SSH RFCs. • Open-source, distributed as part of OpenBSD. • OpenSSH webpages state that OpenSSH accounts for more than 80% of all deployed SSH servers. • www.openssh.org/usage/index.html • We worked with OpenSSH 5.1. • Version 5.2 released 23/2/09 partly as a consequence of our work. • Main file of interest is packet.c.

  45. Attacking OpenSSH • In the OpenSSH implementation, two sanity checks are carried out on the length field after first block is decrypted. • When each of the checks fails, the SSH connection is terminated in subtly different ways. • If the length checks pass, then when the MAC check fails, a third type of connection termination is seen.

  46. Attacking OpenSSH • The manner in which OpenSSH behaves on failure allows: • A first attack verifiably recovering 14 bits of plaintext with probability 2-14. • A second attack verifiably recovering 32 bits of plaintext with probability 2-18 (for a 128-bit block cipher). • Both attacks result in termination of the SSH connection with high probability. • The attacks can be iterated to boost success probability if a fixed plaintext is repeated at a fixed position in SSH packets over multiple connections. • We implemented the attacks. • And they worked.

  47. Impact of the Attacks • The attacks recover small amounts of plaintext and have low success probabilities. • The attacks are easily circumvented by switching to CTR mode or by modifying error handling in CBC mode. • Countermeasures now adopted in OpenSSH release 5.2. • So the attacks are of limited practical significance. • Yet SSH is meant to be a “bullet-proof” protocol. • The fact that quite simple attacks are possible against such an important protocol is troubling.

  48. Theoretical Impact • Various flavours of SSH were proven secure by Bellare, Kohno and Namprempre. • Including SSH-$NPC – randomised, per-packet IV with random padding field – to which our attacks apply. • So what went wrong with the theory? • The security model does model errors during the BPP decryption process. • Even so, there is a probability 1 distinguishing attack against SSH-$NPC that exploits these errors! • But it fails to reflect the fact that the decryption process depends on the content of the length field. • In fact, the length field does not appear at all in the security model, and all operations on data are atomic rather than block-wise.

  49. Theoretical Impact From Bellare, Kohno and Namprempre: “…in the modern era of strong cryptography it would seem counterintuitive to voluntarily use a protocol with low security when it is possible to fix the security of SSH at low cost.” • Provable security is one of the best tools we currently have. • But how do we know that “fixing” SSH actually improves security? • We need more comprehensive security models. • Capturing the on-line nature of decryption process and the potential for error-based side channels. • We need to integrate security analysis with the design process. • And we need to persuade designers that this is worthwhile.

  50. Why the order of encryption and integrity protection matters – SSL/TLS

More Related