Network Working Group J. Vilhuber INTERNET DRAFT Cisco Systems Inc. Expire in July, 2003 January, 2003 IP header compression in IPsec ESP Status of this Memo This document is an Internet-Draft and is subject to all provisions of Section 10 of RFC2026. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet- Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet- Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/1id-abstracts.html The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html Distribution of this memo is unlimited. Abstract This draft outlines how to use IP Header compression over IP tunnels [HCOIP] inside IPsec ESP [ESP]. Vilhuber [Page 1] INTERNET DRAFT September, 2002 Expires March, 2003 1. Introduction [HCOIP] defines a new IP protocol number (and IPv6 Next Header) value for IP Header Compressed payloads for use in tunneling protocols over IP. This draft outlines exactly how to encapsulate IP Header Compressed packets into an ESP packet. In this document, the key words "MAY", "MUST", "optional", "recommended", "required", "SHOULD", and "SHOULD NOT", are to be interpreted as described in [RFC2119]. 2. IP header compressed packets in ESP [HCOIP] defines a number-space for the "Header Compression Payload Type" as well as a new IP protocol number, which can be used to indicate that a packet inside ESP has been previously header compressed. Note that not all packets that fall under a certain ESP SA may be header compressed. Whether a packet is header compressed or not depends on whether the compressor has an empty slot for a flow, and whether the packet is deemed compressible by the compressor. Hence, we can not simply assume that all packets under an ESP SA with header compression will be compressed. We need an explicit indication, hence the need for the new IP protocol number in [HCOIP]. 2.1. Order of processing of outbound packets On outbound processing, the relevant SA bundle is found in whatever manner the implementation uses. The SA bundle MUST indicate that header compression needs to be attempted for this packet. The SA should contain enough information to retrieve the relevant compression context for this flow. Header compression MUST be done before any other ESP or IPCOMP processing. Any fragmentation decisions MUST be made on the result of the header-compressed packet, rather than on the original (un-header- compressed packet). Original (pre-IPsec packet): +-------------+ | IP | Data | +-------------+ Header Compression is done, which removes the IP (and possibly other) headers, replacing it with the appropriate compression context as defined by [IPHC], [CRTP], and/or [ECRTP]: +----------------------------+ | Comp ID and context | Data | +----------------------------+ [HCOIP] header is prepended: Vilhuber [Page 2] INTERNET DRAFT September, 2002 Expires March, 2003 +---------------------------------+ | Comp Type | Comp context | Data | +---------------------------------+ [ESP] is performed (including fragmentation decisions and possibly [IPCOMP]) as usual, with next header set to IPHC: +---------------------------------------------------------------------+ | IP | ESP-SPI | SEQ | IV | Comp Type | Comp context | Data | Trailer | +---------------------------------------------------------------------+ An example of an ESP packet carrying a header compressed packet is as follows: 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Security Parameters Index (SPI) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sequence Number | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ------ | IV (depends on the cipher used; variable) | ^ ESP +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Payload | Comp Type | Header Compressed Data (variable) | | Data +-+-+-+-+-+-+-+-+ + | ~ ~ | | | | + +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | | Padding (0-255 bytes) | v +-+-+-+-+-+-+-+-+ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ------ | | Pad Length | IPHC Proto | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Authentication Data (variable) | ~ ~ | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Depending on the cipher used, the start of the ESP Payload data is used as the IV. The length (and presence or absence) of the IV is implicit knowledge, known to both sides. "Comp Type" is the [HCOIP] header, and will be the first byte of the plaintext payload. The rest of the Payload Data is the compressed packet, the format of which conforms to the relevant RFC that covers the type indicated in Comp Type (see [HCOIP] table 1). 2.2. Order of processing of inbound packets On receiving an IPsec packet, the regular SA-lookup is used to determine the SA bundle needed for decryption (decapsulation and decompression). The SA bundle should carry enough information to retrieve the decompression context. Vilhuber [Page 3] INTERNET DRAFT September, 2002 Expires March, 2003 If the receiver gets a packet with IP protocol IPHC, but the SA bundle does not indicate that compression has been negotiated for this SA, the packet MUST be dropped. Since Header compression is the first thing done during encryption/encapsulation, decompression is the last thing done. After decapsulation and decryption (and maybe IPCOMP decompression), if the resulting packet has a protocol type of IPHC, then the [HCOIP] header is removed, and the packet along with the [HCOIP] type (from the header), along with the decompression context stored with the IPsec SA, MUST be handed to the decompressor. After decompression, the decompressed packet MUST be checked against the SADB as normal, and dropped, if the packet does not match the SADB. 2.3. Decompressor Error handling In the event that the decompressor has no appropriate compression slot (as identified by the compression ID; see [IPHC]) for the packet handed to it, the packet MUST be dropped. There is no error indication that can be communicated to the peer. In the event that the decompressor is out of sync with the compressor (i.e. a decompression context for this Compression ID exists, but packet loss has occurred), the decompressor may need to communicate a CONTEXT_STATE packet back to the compressor. This packet MUST be sent back through the IPsec Tunnel, i.e. must be encrypted and encapsulated using the correct SA, i.e. the SA we would use to send ESP packets to the peer. The out-of-sync packet MUST be dropped. Should the Compressor receive a CONTEXT_STATE packet that has not been authenticated via IPsec, then, as per [SECARCH], the packet MUST be dropped and ignored. On receipt of a valid CONTEXT_STATE packet, the receiver, who was the sender of the packet that failed to decompress, will invalidate any contexts listed in the CONTEXT_STATE packet, as per [RFC2507] (and various addenda in [RFC2508] and [ECRTP]). 3. Security Considerations 3.1. Removing plain-text Omitting the IP (as well as TCP or UDP/RTP) header removes a large amount of known (or guess-able) plaintext from the to-be-encrypted payload. While this can benefit security it still should not be relied upon as a replacement for a strong cryptographic mechanism. Vilhuber [Page 4] INTERNET DRAFT September, 2002 Expires March, 2003 3.2. Active Attack Analysis There is some concern about what an attacker can or can not do to the decompressor, even when protected with ESP. We assume that an attacker can not modify packets, because of ESP protection. For this reason, it is recommended that ESP not be run without authentication, even when esp-null is used. Whether an attacker can look at the packets (i.e. a passive attacker) has no immediate relevance to header compression. An active attacker can drop packets, or insert fake ones. Fake ones will be discarded by IPsec, but dropped packets can have an influence on the decompressor as outlined in the following sections. 3.2.1 COMPRESSED_TCP, COMPRESSED_UDP, and COMPRESSED_RTP packets DELTA fields depend on the number of packets we receive and send, i.e. DELTA fields depend on the value of sent in the preceding FULL_HEADER packet, and we increment the field value by a fixed, known delta for each packet received. There are well-defined algorithms that try to help detect dropped packets and correct in those situations. However, to be conservative, we should assume that dropped packets MAY influence DELTA fields (however, see below). Likewise, any INFERRED fields that depend on DELTA fields could be decompressed wrong (but most INFERRED fields do not, in fact, depend on DELTA fields). An attacker can NOT influence any NOCHANGE fields, as those are explicitely copied from the compression context set up (and refreshed) via FULL_HEADER packets, which can not be tampered with. Examples of NOCHANGE fields are IP addresses (src and dst), protocol, and src and dst ports. RANDOM fields are carried explicitely in the compressed packet and thus can not be tampered with, either. Examples of RANDOM fields are checksums. An attacker can also not change any of the data in the packet by selectively dropping packets, as the header compression mechanisms do not affect the data. So the best an attacker could do is influence DELTA fields, which are generally sequence numbers, by selectively and intelligently dropping packets. If the higher level protocol uses checksums (as TCP does, and, mostly, UDP as well), then mis-decompression due to dropped packets will be detected by the recipient of the mis-decompressed packet. For TCP packets, it is presumed that the end-host will detect and discard any "off-by-k" sequence numbers via the TCP checksum. Neither TCP nor UDP checksums cover anything in the IP header other than the pseudo-header, which doesn't cover parts of the IPv4 header, nor Vilhuber [Page 5] INTERNET DRAFT September, 2002 Expires March, 2003 large parts of the IPv6 headers. [CRTP] contains a 4 bit sequence number, to detect dropped packets within a 16 packet window. As long as the affected DELTA fields are covered by a higher level checksum (i.e. UDP or TCP checksum), then attacking the data-stream by selectively dropping some packets amounts to a denial of service attack, which the attacker could perpetrate anyway, if he can drop packets at will. If the DELTA fields are not covered by a higher level i.e. UDP or TCP) checksum, then these fields could be wrong after decompression and the recipient may not notice. A new kind of HDRCKSUM similar to the one defined in [ECRTP] should be devised to counteract this. 3.2.2 COMPRESSED_NON_TCP and COMPRESSED_TCP_NODELTA As per [RFC2507], COMPRESSED_NON_TCP packets do not use differential coding, and all fields are assumed to be NOCHANGE. If a NOCHANGE field changes, a FULL_HEADER packet is sent, instead. Thus dropping packets in this case has no effect on the values of the decompressed packets. COMPRESSED_TCP_NODELTA "is only sent in response to a header request from the decompressor" [RFC2507]. Since there are no DELTA fields in this packet type, dropping this packet (which, in any case is not sent during normal operation) has no effect (except causing more drops, i.e. more denial of service). 3.2.3 Future compressed headers This analysis covers only known header-compression types known at the time of this writing (see section 6. References). Any future new compressed types of additional compressed headers should consider the impact separately, following a similar analysis as in the previous sections. NOCHANGE and RANDOM fields can be safely ignored. They are safe. INFERRED fields are safe as long as they do not depend on DELTA fields. DELTA fields need to be considered on a case by case basis, and should be covered by some checksum. Checksums should never be optional. Alternatively, a scheme like [ECRTP]'s HDRCKSUM should be used (see 3.2.3 as well). 3.2.3 Future work As eluded to earlier, a simple way to fix this entire dilemma of DELTA headers and decompression, is to define a checksum similar to [ECRTP] HDRCKSUM, which covers the entire header that was compressed, Vilhuber [Page 6] INTERNET DRAFT September, 2002 Expires March, 2003 but none of the data, define the TCP and UDP checksums as INFERRED, and carry ONLY the HDRCKSUM in the compressed packet. Since ESP packets, when used with authentication, already verify that the data hasn't been tampered with, we can re-calculate the TCP and UDP checksums during decompression, as long as we have a way to verify that the decompressed headers are exactly the same as they were prior to compression. The HDRCKSUM gives us this assurance, and thus the mechanism is secure. The cost is extra processing at the decompressor (who needs to calculate TCP or UDP checksums, which include the data of the packet). 4. IANA Considerations There are no IANA Considerations. 5. Acknowledgments This document is derived in part from discussions with Cheryl Madson, David McGrew, Mark Gillott, Patrick Ruddy, and Dan Wing. 6. References [IPHC] Degermark, M., Nordgren, B. and S. Pink, "Header Compression for IP", RFC 2507, February 1999. [CRTP] Casner, S. and V. Jacobson, "Compressing IP/UDP/RTP Headers for Low-Speed Serial Links", RFC 2508, February 1999. [ECRTP] Koren, Casner, Geevarghese, Thompson, Ruddy, "Compressing IP/UDP/RTP headers on links with high delay, packet loss and reordering", draft-ietf-avt-crtp-enhance-04.txt, work in progress, February 2002 [ESP] Kent, S., Atkinson, R., "IP Encapsulating Security Payload", RFC 2406, November 1998 [SECARCH] Kent, S., Atkinson, R., "Security Architecture for the Internet Protocol", RFC 2401, November 1998 [HCOIP] Vilhuber, "IP header compression in IP tunneling protocols", draft-vilhuber-hcoip-00.txt, work in progress, September, 2002 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", RFC 2119, March 1997 [IPCOMP] Shacham, Monsour, Pereira, Thomas, "IP Payload Compression Protocol (IPComp)", RFC 3173, September 2001 Vilhuber [Page 7] INTERNET DRAFT September, 2002 Expires March, 2003 7. Editor's Address Jan Vilhuber Cisco Systems, Inc. Vilhuber [Page 8]