Hi Yaron, This minimal set is a description of the surface that any generic API must have to cover the common *transport* features of protocols only. This document is not meant in itself to describe an API that applications use directly, but instead describe the distilled feature set from transport services (such as reliable/unreliable sending, etc). To that end, it essentially describes what you need to be able to use basic things like TCP, UDP, etc. Even without security considered, this distillation is a non-trivial task, thus it required this document. The security document that is referenced (draft-ietf-taps-transport-security) provides a similar survey and distillation for how transport security interfaces add more surface to control and interact with handshake and record layers used on top of/in conjunction with the base transport. The set of documents that actually describe the API surface that we want applications to use (draft-ietf-taps-arch, draft-ietf-taps-api, draft-ietf-taps-interface, draft-ietf-taps-impl) does indeed provide a unified abstraction for a secure transport connection. However, this surface depends upon both the distillation from draft-ietf-taps-minset and draft-ietf-taps-transport-security. There is no intention of promoting insecure connections for applications. Thanks, Tommy > On Aug 26, 2018, at 10:03 PM, Yaron Sheffer <yaronf.ietf@xxxxxxxxx> wrote: > > Reviewer: Yaron Sheffer > Review result: Not Ready > > The whole notion of TAPS is new to me, so I may be missing the point here. This > document defines a minimal set of network APIs that should be available to > applications, in order to allow multiple different transport protocols to be > used as interchangeable plug-ins with minimal or no change to applications. > > However the document does not cover security, and instead refers readers to a > security protocol survey (draft-ietf-taps-transport-security). > > There's a disconnect here: in many cases we want applications to be aware of > security features. For example, a typical TCP-using application should choose > whether to enable TLS encryption of the connection (or as a receiver, whether > to require encryption), and if TLS is selected, should at the very least > receive access to the authenticated address of the connection's peer. In other > words, a meaningful minimal set of APIs cannot be defined without considering > the effects and requirements of security protocols. > > Put differently, the application normally treats the transport protocol and the > security protocol layered on it as one protocol. Hence the old name of TLS: > Secure Socket Layer. The application sees a single socket, not one socket for > transport and another for security. This has been the case for TCP for the last > 20-odd years, and is unlikely to change any time soon. > > I have not surveyed the protocols discussed in this draft, and I don't know > whether a viable transport-level security protocol exists for each of them. If > this is not the case, then I guess the industry is not yet ready for the kind > of solution proposed here. > > _______________________________________________ > Taps mailing list > Taps@xxxxxxxx > https://www.ietf.org/mailman/listinfo/taps
