[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: users and connections

> > Why "connection oriented"?  I didn't see anything in RFC 1826 about
> > connection oriented integrity.
> RFC-1826 provides integrity:
>    The Authentication Header is a mechanism for providing strong
>    integrity and authentication for IP datagrams.  ...
> Perhaps you could become more familiar with the Bellovin paper?
> The attack described involves mutually suspicious users.  To prevent the
> attack, the SPI keying should be on a per user (or for proxying
> firewalls, per connection) basis.  Otherwise, the attacking user has
> access to the SPI, and can trivially decode the traffic.

Bill raises a fine point that, in particular, host vendors should take to
heart.  It is not only possible to implement, but REQUIRED that
unique-per-endpoint SA assignment be allowed to happen.

(Small hair to split:	A single socket can be in use by two users, so
			per-user might be harder than per-endpoint, but like
			I said, it's a small hair to split, since most
			endpoints are unique-per-user anyway.)

There is freely available source out there (NRL for one, and perhaps more...)
that implements per-socket IPsec keying.  Anyone requiring hints on
implementing it should check it out.  I suspect we have enough people who've
figured it out on their own, though.  And yes, it's in my future
implementation plans as well.

Just wanted to chime in along with Bill.


Follow-Ups: References: