FW: CDR review ROHCoIPSec

Ertekin, Emre [USA] writes:
> > There is an inherent friction between bandwidth and security
> concerning
> > the ROHC ICV. Since one of primary goals of ROHC is to reduce
> > bandwidth,
> > perhaps it would be worthwhile if the IKEv2 extension to ROHC provided
> > some flexibility to negotiate the length of the ROHC ICV (perhaps by
> > truncation), agnostic of the algorithm used.
> 
> Although this may provide efficiency benefits, there are some
> considerations to this approach.  One of them includes interoperability,
> where one vendor may end up supporting ICVs of length 32, another vendor
> would support ICVs of 64, another 96.  One might propose providing the
> flexibility to support variable length ICVs in a ROHCoIPsec
> implementation--but this flexibility comes at the cost of increased
> complexity when implementing.

As the macs normally used in the IPsec are already truncated, and we
would here only want to truncate them even more, I think it would
quite easy to support any ICV length between 0-12 bytes. There is also
completely valid reasons to use 0 bytes of ICV, for example in case
the environment is known to be such that attackers cannot remove
packets. Also even though the actual algorithm used should be same for
both directions (as is normal in IPsec SAs), there is no need for the
truncation length (if it is negotiated separately as I think would be
best) to be same in both directions, i.e for high bandwidth upload
link it might be longer, than in the very low bandwidth backward link
which is used only for TCP ACKs anyways.

> Although the current text for negotiation of ROHC channel parameters in
> the IKEv2 extensions draft aligns with the operation of IKEv2 (e.g.,
> negotiation of IPcomp algorithms, where the initiator lists the
> algorithms supported, and the responder selects the algorithm selected
> for the SA), there appears to be no technical motivation for the
> "negotiation"-based approach.

I actually think it is unfortunate that we made the IPcomp to use that
kind of negotiation in the IKEv2. I think much better method for the
IPcomp negotiation in the IKEv2 would have been so that both ends just
send list of the supported IPcomp algorithms they are willing to
receive on, and sender can use any of those to compress packet when
sending packets to him. The different IPcomp algorims are already
distinguished from each other by the CPI value send in the
notification.

Unfortunately the IKEv2 specification decided to explictly forbid that
kind of negotiation (and I do not know any good reason why we decided
that way)...

In the ROHC, I think it makes much more sense to advertise what kind
of parameters peer can accept when receiving packet, and sender peer
will then limit its parameters to those that the responder can accept.

I do not think there is any need to agree symmetric parameters, and
agreeing on specific parameters is always hard if you have multiple
different type of parameters (i.e. some are lists where other end
needs to pick one, other are numeric parameters where other end needs
to pick number smaller or equal the number sent by other end etc).

It gets even more complicated if not all values are allowed (for
example you have paremeter which must be multiple of 4 or similar in
your end, and the other end does not have this limitation). 

Of course it does require that parameters are such that they can
either been seen from the packet or the mismatch does not matter. I.e.
the MAX_CID I guess it does not matter what maximum number each other
is configured, as long as the sender keeps its context ID numbers
below the MAX_CID announced by the recipient.

On the other hand I do not know if the PROFILE of the ROHC can be
seen from the packet. If the recipient of compressed packet can see
from the packet to which profile it is using, then there is no need to
agree on one profile, but it is better to just announce what can be
received. If the profile cannot be seen from the packet, then we must
agree on one profile only.

On the other hand I would suggest the notification data structure to
have list of attributes instead of the current fixed format, i.e. make
the notification data of the ROHC_SUPPORTED to be list of transform
attributes using same format as specified in the section 3.3.5 of
RFC4306, i.e.:

----------------------------------------------------------------------
                           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
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      !A!     ROHC Attribute Type     !  AF=0  ROHC Attribute Length  !
      !F!                             !  AF=1  ROHC Attribute Value   !
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
      !                   AF=0  ROHC Attribute Value                  !
      !                   AF=1  Not Transmitted                       !
      +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

Where the "Attribute Type" values would be IANA registry having
following initial values:

  ROHC Attribute Type Values
							      Type
      RESERVED							0
      Maximum value of context identifier (MAX_CID)		1
      ROHC Profile (PROFILE)					2
      ROHC Integrity Algorithm (ROHC_INTEG)			3
      ROHC ICV Length in bytes (ROHC_ICV_LEN)			4

If unknown ROHC Attribute Type Value is received in the request then
it is silently ignored, and no such attribute is added to the reply
sent back to the other end. If unknown ROHC Attribute Type Value is
received in the reply then that is silently ignored. Using this rules
it means there is way to add new attributes which can be of either
negotiated attributes (like ROHC_INTEG) or announced attributes like
(MAX_CID, PROFILE or ROHC_ICV_LEN). For negotiated attributes they
must be supported by both initiator and responder to be enabled, thus
if initiator does not add support such attribute it will not include
it to request, then responder cannot select one, and if initiator
supports it but responder does not, then responder does not select any
of the attributes, and the feture is not enabled. For the announcement
type of attributes both ends can simply send them at will, and they
are only announcing what other end can do, so there is no danger of
ignoring them.

There MUST be exactly one MAX_CID attribute and its ROHC Attribute
Value field must number between 0-16383 (and is always using AF=1
format). The MAX_CID value specifies the maximum value the sender of
this message can support, i.e. the recipient of this message can use
up to that many context identifiers when sending data to other end.
Both ends simply announce their own capabilities, and they does not
need to negotiate or agree on single value. This is mandatory
attribute.

There can be multiple PROFILE attributes, and in those the ROHC
Attribute Value specifies the supported 16-bit profile identifiers
(using AF=1 format). The list of profile identifiers specifies which
ROHC profiles are supported by the sender for incoming traffic, i.e.
the recipient of this message can use any of those profiles when
sending data to the other end. Both ends simply announce their own
capabilities, and they does nto need to negotiate or agree on using
any of the profile. There must be at least one of the profiles if ROHC
is supported at all.

There can be multiple ROHC_INTEG attributes when the initiator is
sending its list of supported ROHC integrity algorithms, and responder
needs to select exactly one of those and send that back. The integrity
values are taken from the IKEv2 Transform Type 3 (Integrity Algorithm)
IANA registry (either by copying the current values to new registry or
directly referencing to the existing registry) and are encoded using
AF=1 format. This same integrity algorithm is used in both directions.
This is optional attribute, but if initiator sent any of the
ROHC_INTEG attributes, then responder MUST select one of them to be
used. The selected one can be value 0 (NONE) if that was proposed by
the initiator.

The ROHC_ICV_LEN attribute specifies the number of octects the sender
of this message wants to receive on incoming ROHC packets. This is
always encoded using AF=1 format. I.e. the ICV field of the negotiated
ROHC_INTEG algorithms are truncated to this many bytes by taking the
first n bytes of the output. Both ends simply announce their own
requirement, and they does not need to negotiate or agree on single
value. The sender end MUST truncate the ICV to length required by the
other end before sending ICV. If ROHC_ICV_LEN length is zero then no
ICV is calculated or sent. If no ROHC_ICV_LEN attribute is sent at all
or the ROHC_ICV_LEN is larger than the length of the ICV of selected
algorithm, then the full ICV length as specified by the ROHC_INTEG
algorithm is sent.
----------------------------------------------------------------------

The reason for using the attribute format already specified in the RFC
4306 is that there is already code to parse and process that kind of
lists of attributes, thus parsing and generating such lists is easy,
and this also makes it easy to extend parameter negotiation in future
in case there are new ROHC profiles created which need more
parameters. 
-- 
[email protected]