- From: Mark Nottingham <mnot@mnot.net>
- Date: Sat, 15 Feb 2025 11:05:18 +1100
- To: Petr Špaček <pspacek@isc.org>
- Cc: draft-ietf-httpbis-rfc6265bis.all@ietf.org, "dnsdir@ietf.org" <dnsdir@ietf.org>, ietf-http-wg@w3.org, Geoff Huston <gih@apnic.net>
Hi Petr, I think you're looking for this: https://url.spec.whatwg.org/ Cheers, > On 15 Feb 2025, at 4:44 am, Petr Špaček <pspacek@isc.org> wrote: > > Hello. > > I have to say the more I dig into this area the more I'm confused. > > Could authors clarify for me, a HTTP noob, what are possible inputs for algorithm described in > > 5.1.2. Canonicalized Host Names > > In other words, who are callers and consumers of this algorithm? > > > Second question: > Is it standardized somewhere wow is an user input (e.g. typing in a URL manually, or clicking to a link) get converted into Host portion of Authority field specified in RFC 3986 section 3.2/RFC 9110 section 4.3.1? > > > I'm trying to shed light on relation between this generic conversion mechanism (which presumably exists), this specific draft under review, RFC 6055, RFC 8222, and the real world. > > > A simple test on a modern Linux installation indicates this might be broken already in the HTTP-world, but I want to understand this for real before hand-waving the problem away. > > (beware, non-ASCII stuff follows) > > $ grep hosts /etc/nsswitch.conf > hosts: files dns > > $ tail -n1 /etc/hosts > 2001:db8::5 háčkyčárk.cz > > $ getent hosts háčkyčárk.cz > 2001:db8::5 háčkyčárk.cz > > So far so good. Name service switch did its job. > > $ curl http://háčkyčárk.cz/ > curl: (6) Could not resolve host: xn--hkyrk-xqac36ac.cz > > Not good at all. Judging by quick GDB excursion into guts of curl, it has converted URL into A-labels prematurely (before reaching dns module in the NSS). > > That is a nice example of a problem explained in RFC 6055 page 6, i.e. doing format conversions at a wrong layer. > > Now the main question: > Should this problem be baked into the Cookies spec? Should we try to fix it? Or perhaps only document there _is_ a problem? > > Thank you for your thoughts. > > Petr Špaček > > > > On 25. 01. 25 4:27, Mark Nottingham wrote: >> This may be helpful for context: >> https://httpwg.org/specs/rfc9110.html#authoritative.access >> Cheers, >>> On 25 Jan 2025, at 1:48 am, Petr Špaček <pspacek@isc.org> wrote: >>> >>> Hello everyone. >>> >>> I was assigned as the dnsdir reviewer for draft-ietf-httpbis-rfc6265bis. >>> For more information about the DNS Directorate, please see >>> https://wiki.ietf.org/en/group/dnsdir >>> >>> I have a question before I start with real review: >>> >>> What is the intended interaction with non-DNS naming systems? >>> >>> I'm not an HTTP expert, but I would guess that anything in Name Service Switch equivalent on a given operating system will become entangled in the cookie Domain attribute business. >>> >>> Just from top of my head >>> - DNS >>> - /etc/hosts equivalent >>> - Tor onion. >>> - GNUnet >>> - NetBIOS >>> - LLMNR >>> - mDNS >>> - (Let's not go any further...) >>> >>> Yes, it is a mine field. I'm trying to find out how generic the name formatting text + security considerations should be because right now it mentions 'DNS' in couple places. >>> >>> Thank you for your time. > > -- > Petr Špaček -- Mark Nottingham https://www.mnot.net/
Received on Saturday, 15 February 2025 00:05:27 UTC