we've put up a new draft of tcpcrypt:

  http://www.scs.stanford.edu/~dbg/tcpcrypt/

the major change is to eliminate vulnerability to
header-manipulation and resegmentation by moving all but the
initial protocol negotiation into the TCP datastream.

because most of the protocol is now logically built atop a
reliable stream instead of unreliable segments, this leads
to some simplifications; for example, the previous, boutique
authenticated-encryption algorithm has been replaced with
standard AE modes.  also it seems that extensions to the
key-exchange or application subprotocols should be easier.

nevertheless, we've made some effort to allow for
implementations that prefer to intervene not at the socket
level, but rather at the packet level, "underneath" TCP.
here, it will be possible to reframe and protect the
contents of outgoing segments, and similarly to unframe
incoming ones before passing them to the local TCP.  in the
perhaps-rare case of resegmentation, incomplete frames would
have to be buffered until they can be coalesced and
decrypted.

some small details remain to be specified: the protocol
allows arbitrarily-sized messages in several places, and
we'll need to limit these to permit finite-sized buffers.

we'd be grateful for comments.

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

Reply via email to