Re: [IPsec] [ipsecme] #112: Truncation of SHA-1 ICVs

[<Black_David at emc.com>]

Tero and Sheila,

> The AUTH_HMAC_MD5_128 and
> AUTH_HMAC_SHA1_160 are really only specified when used with Fibre
> Channel CT_Authentication format instead of ESP format. Even when
> using fibre channel if ESP format is used then I think you must use
> the truncated versions.

That is correct for Fibre Channel.  If someone wants to use the
non-truncated versions of these with conventional IPsec, there's
no specification barrier to doing so, but I don't think this has
or is likely to be implemented.  FWIW, I concur with Tero's
suggestion to look to the HMAC-SHA2 family for longer IPsec MACs.

> So AUTH_HMAC_MD5_128 and AUTH_HMAC_SHA1_160 cannot be used in IPsec,
> but they have the numbers in the IKEv2 registry, as they are
> negotiated for their CT_Authentication use using IKEv2.

I wouldn't say "cannot be used" - "are not used" is more accurate.

The potentially related discussion about separate identifiers:

>    This document specifies identifiers for IKEv2 over FC in a
>    fashion that ensures that any mistaken usage of IKEv2/FC over IP
will
>    result in a negotiation failure due to the absence of an acceptable
>    proposal (and likewise for IKEv2/IP over FC).

is primarily about the Security Protocol Identifiers and the
Traffic Selector Type - RFC 4595 defines FC-only values for these
that any IPsec implementation will quickly reject; see the IANA
Considerations (Section 6) of RFC 4595.

> > NOTE to Tero, Paul, Yaron: do we want to expand the IKEv2 IANA
> > registry to include non-truncated AES-XCBC-MAC,
> > HMAC-SHA-256/384/512, AES-CMAC and HMAC-RIPEMD? 
>
> Not for IPsec use. I do not know if the Fibre Channel people want to
> use non-truncated versions of them in their CT_Authentication format,
> but for IPsec if you want to have longer MAC, use longer HMAC-SHA-2
> variant... 

I would expect to see at least a non-truncated versions of
HMAC-SHA2-256 show up when the Fibre Channel specifications are
updated, but I think we (IETF) can wait for that (i.e., no registry
changes needed now).

Thanks,
--David (RFC 4595 co-author)

> -----Original Message-----
> From: ipsec-bounces at ietf.org [mailto:ipsec-bounces at ietf.org] 
> On Behalf Of Tero Kivinen
> Sent: Wednesday, October 28, 2009 7:43 AM
> To: Frankel, Sheila E.
> Cc: ipsec at ietf.org; Paul Hoffman; suresh.krishnan at ericsson.com
> Subject: Re: [IPsec] [ipsecme] #112: Truncation of SHA-1 ICVs
> 
> Frankel, Sheila E. writes:
> > Additional text:
> >    Some of these algorithms generate a fixed-length ICV, which is
truncated 
> >    when it is included in an IPsec-protected packet. For example,
standard 
> >    HMAC-SHA-1 generates a 160-bit ICV, which is truncated to 96 bits
when it 
> >    is used to provide integrity-protection to an ESP or AH packet.
The 
> >    individual RFC descriptions mention those algorithms that are
truncated. 
> >    When these algorithms are used to protect IKEv1 SAs, they are not

> >    truncated. For HMAC-SHA-1 and HMAC-MD5, the IKEv2 IANA registry
contains 
> >    values for both the truncated version and the standard
non-truncated 
> >    version; thus, IKEv2 has the capability to negotiate either
version to 
> >    protect IKEv2 and/or IPsec-v3 SAs.
> 
> This is not completely correct. The non-truncated versions are not
> meant to be used with normal IPsec-v2/v3. They are meant to be used
> with Fibre Channel Security (RFC4595). The AUTH_HMAC_MD5_128 and
> AUTH_HMAC_MSHA1_160 are really only specified when used with Fibre
> Channel CT_Authentication format instead of ESP format. Even when
> using fibre channel if ESP format is used then I think you must use
> the truncated versions.
> 
> Here is some cut & paste parts of RFC4595 to explain situation:
> 
> ----------------------------------------------------------------------
> 2. Overview
> 
>    Fibre Channel defines two security protocols that provide security
>    services for different portions of Fibre Channel traffic: the
>    ESP_Header defined in [FC-FS] and CT_Authentication defined in
>    [FC-GS-4].
> 
>    The ESP_Header protocol is a transform applied to FC-2 Fibre
Channel
>    frames.  It is based on the IP Encapsulation Security Payload
>    [RFC4303] to provide origin authentication, integrity, anti-replay
>    protection, and optional confidentiality to generic fibre channel
>    frames.  The CT_Authentication protocol is a transform that
provides
>    the same set of security services for Common Transport Information
>    Units, which are used to convey control information.  As a result
of
>    the separation of Fibre Channel data traffic from control traffic,
>    only one protocol (either ESP_Header or CT_Authentication) is
>    applicable to any FC Security Association (SA).
> ...
>    Since IP is transported over Fibre Channel [RFC4338] and Fibre
>    Channel/SCSI are transported over IP [RFC3643], [RFC3821] there is
>    the potential for confusion when IKEv2 is used for both IP and FC
>    traffic.  This document specifies identifiers for IKEv2 over FC in
a
>    fashion that ensures that any mistaken usage of IKEv2/FC over IP
will
>    result in a negotiation failure due to the absence of an acceptable
>    proposal (and likewise for IKEv2/IP over FC).  This document gives
an
>    overview of the security architecture defined by the FC-SP
standard,
>    including the security protocols used to protect frames and to
>    negotiate SAs, and it specifies the entities for which new
>    identifiers have been assigned.
> ...
> 3.2. CT_Authentication Protocol
> 
> 
>    CT_Authentication is a security protocol for Common Transport FC-4
>    Information Units that provides origin authentication, integrity,
and
>    anti-replay protection.  The CT_Authentication protocol is carried
in
>    the optional extended CT_IU preamble
> ...
>    The Authentication Hash Block is computed as an HMAC keyed hash of
>    the CT_IU, as defined in [RFC2104].  The entire output of the HMAC
>    computation is included in the Authentication Hash Block, without
any
>    truncation.  Two transforms are defined: HMAC-SHA1-160 that is
based
>    on the cryptographic hash function SHA1 [NIST.180-1.1995], and
>    HMAC-MD5-128 that is based on the cryptographic hash function MD5
>    [RFC1321].
> ...
> 4.3. CT_Authentication Protocol Transform Identifiers
> 
> 
>    The CT_Authentication Transform IDs defined for Transform Type 3
>    (Integrity Algorithm) are:
> 
>            Name                   Number                    Defined in
>            ----                   ------                    ----------
>            AUTH_HMAC_MD5_128      6                         FC-SP
> 
>            AUTH_HMAC_SHA1_160     7                         FC-SP
> 
>    These transforms differ from the corresponding _96 transforms used
in
>    IPsec solely in the omission of the truncation of the HMAC output
to
>    96 bits; instead, the entire output (128 bits for MD5, 160 bits for
>    SHA-1) is transmitted.  MD5 support is required due to existing
usage
>    of MD5 in CT_Authentication; SHA-1 is RECOMMENDED in all new
>    implementations.
> ...
> ----------------------------------------------------------------------
> 
> So AUTH_HMAC_MD5_128 and AUTH_HMAC_SHA1_160 cannot be used in IPsec,
> but they have the numbers in the IKEv2 registry, as they are
> negotiated for their CT_Authentication use using IKEv2. 
> 
> > For the other algorithms (AES-XCBC, 
> >    HMAC-SHA-256/384/512, AES-CMAC and HMAC-RIPEMD), only the
truncated 
> >    version can be used for both IKEv2 and IPsec-v3 SAs.
> >  
> > NOTE to Tero, Paul, Yaron: do we want to expand the IKEv2 IANA
> > registry to include non-truncated AES-XCBC-MAC,
> > HMAC-SHA-256/384/512, AES-CMAC and HMAC-RIPEMD? 
> 
> Not for IPsec use. I do not know if the Fibre Channel people want to
> use non-truncated versions of them in their CT_Authentication format,
> but for IPsec if you want to have longer MAC, use longer HMAC-SHA-2
> variant... 
> -- 
> kivinen at iki.fi
> _______________________________________________
> IPsec mailing list
> IPsec at ietf.org
> https://www.ietf.org/mailman/listinfo/ipsec
> 
>