On 07/30/2014 01:39 PM, Martin Thomson wrote:
> I see two potential failure paths here: in one, hosts adaptively
> respond to damage by reducing what is protected.  That is in line with
> the opportunistic nature of the mechanism, but it allows an attacker
> to erode any optional protections.

What does this give to the user?  opportunistically encrypting the
session protects against a passive eavesdropper.  opportunistically
providing integrity protection to parts of the packet headers doesn't
protect against any attackers i'm aware of.

Would this require some other API, whereby a tcpinc-aware client could
query and detect that this protection had failed and then choose on its
side to deal with this change in some way?

> The other failure mode is where
> you kill the connection in response to modifications,

Or you could ignore the malformed packets without killing the connection
in the hope that the protected packets will arrive shortly -- this would
provide a defense against attackers able to inject packets but not able
to block packets.

(e.g. search for "race condition" in
https://www.schneier.com/blog/archives/2013/10/how_the_nsa_att.html)

> and we thereby create a disincentive to deploy tcpinc.

Right, this is an argument against making these header protections
mandatory -- but i'm trying to clarify what people think it should mean
to make these protections optional.

        --dkg

Attachment: signature.asc
Description: OpenPGP digital signature

_______________________________________________
Tcpinc mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/tcpinc

Reply via email to