- From: Paul Leach <paulle@microsoft.com>
- Date: Wed, 28 Feb 96 11:31:03 PST
- To: john@math.nwu.edu
- Cc: http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
] ] I have changed <message-body> to <entity-body> and added the two sentences: ] ] The <entity-body> is the "entity body" as ] prescribed in the Hypertext Transfer Protocol. It consists of the ] data transferred after the <CRLF><CRLF> signaling the end of the ] entity headers. I think there's a good argument that the <message-digest> should include at least the entity-headers and Date: as well as the <entity-body>, and maybe the other headers, too. This would prevent mucking with the Last-Modified, or Content-Type, etc, and Date: would prevent substituting an old reply for a new one. (This was another of Allan's points, BTW, that seems to have been left off of Larry's list. Sorry for not mentioning it earlier, but I coudn't tell until getting the <message-body> thing clarified. Actually it was two of his points -- that the total request wan't authenticated, and that there was no freshness information.) If this is a backwards compatibility problem, then a new optional parameter "header=" could be used. This approach could also permit the separation of the entity-headers from the rest of the headers -- a cache would need to cough up entity-related digest that it got from the origin server, but construct a digest of the other headers using its own secret that it shares with the client. If this isn't too out of line, I'll write up specific proposed text. ] > ] > 2. In the security considerations section, the rationale for including ] > client IP in the recommended nonce needs to be given, over just ] > checking the IP address of a later request containing a nonce against ] > the IP address to which the nonce was originally given. Is it to ] > reduce the amount of state that the server needs to hold? ] > ] ] It is done so the server can be stateless. As far as I know there ] are no stateful implementations of Digest Authentication. I have ] added the following sentence to section 3.2 ] ] Digesting the client ] IP and timestamp in the nonce permits an implementation which does ] not maintain state between transactions. ] ] ] On a related topic, I don't want to move the recommended nonce ] construction material to an appendix. This might make sense from ] an editorial point of view, but we were explicitly charged with ] expanding the nonce section and I want it to be very clear we have ] met this charge, not just tacked on an appendix. Makes sense to me. Then how about putting it into it's own subsection of section 2? It's a bit of a long digression in the middle of the parameter descriptions for the WWW-Authenticate header, and if a description of how to construct a nonce when trying to avoid replays were added, that would only get worse. (Again, I'll write something on a replay protection nonce if you agree.) Paul
Received on Wednesday, 28 February 1996 11:29:06 UTC