Hi.
Apologizes for new thread.
I got lost in threading for version -12 review and back and forth on the
list, so it feels cleaner to start over.
Good news is vast majority of defects is now fixed. I think I found
couple more but I consider these to be text clarifications and fixes do
not change the protocol on the wire.
---------------------
4. Name Server Requirements
When the value of the EDNS(0) DE flag is 0, the server behaves as a server that
does not implement this specification, i.e., Delegation Types are processed as
ordinary Data Types.
This wording does not work without further clarification: 'server
behaves as a server that does not implement this specification' implies
different NSEC(3) chain bitmaps and has far-reaching consequences.
I propose exchanging this paragraph to use a different angle:
"When the value of the EDNS(0) DE flag is 0, server responses MUST be
compatible with clients that do not implement this specification. The DE
flag MUST NOT change which data are occluded within a zone. In
particular, NS-Omitting Delegation Types still trigger the occlusion
rules described in RFC 2136 section 7.18. In signed zones, NSEC and
NSEC3 type bitmaps MUST include the Delegation Types present at their
corresponding names, independently of the DE flag. Existing
zone-selection and response-processing rules apply, except where this
document specifies otherwise."
Why - an example (taken from DELEG-13 section A.1)
DELEG-aware auth zone content:
test. DELEG server-ipv6=3fff::33
test. RRSIG DELEG 13 1 300 20260101000000 (
20250101000000 33333 . SigTestDELEG )
test. NSEC . RRSIG NSEC DELEG
test. RRSIG NSEC 13 1 300 20260101000000 (
20250101000000 33333 . SigTestNSEC/ )
; a forgotten glue record from legacy (NS) delegation
; it is NOT present in NSEC chain and it is occluded
;
; this can happen e.g. after dynamic UPDATE
a.test. A 192.0.2.1
If we say
'as a server that does not implement this specification'
or
'ordinary Data Types'
then query for DO=1 DE=0 QNAME=a.test. QTYPE=A would treat
'test. DELEG' RRset the same as if it were
'test. TXT' RRset
But TXT does not cause occlusion and consequently 'a.test. A' should be
returned.
That's impossible to do correctly without extra:
- a.test. RRSIG(A)
- the A bit set in 'test. NSEC' type bitmap
- extra a.test. NSEC
... so an implementation would have to have two versions of the NSEC
chain and extra RRSIGs.
If we establish a rule there's only one NSEC chain and the auth must use
it to generate response, then this is the ~ only possible response for
DELEG-unaware client:
;; Authority
. SOA ...
. RRSIG SOA ...
test. NSEC . RRSIG NSEC DELEG
test. RRSIG NSEC 13 1 300 20260101000000 (
20250101000000 33333 . SigTestNSEC/ )
which is listed in DELEG A.4.2.3.
---------------------
4.1.1. Compact Denial of Existence
This document updates Compact Denial of Existence (CDOE) [RFC9824]. For queries
with DE=0 and a QNAME below a delegation point containing an NS-Omitting
Delegation Type RRset but no NS RRset, authoritative servers implementing
Compact Denial of Existence (CDOE) [RFC9824] MUST NOT use the CDOE method when
the DE flag is clear. Instead, they MUST return a conventional negative
response with the corresponding DNSSEC denial-of-existence proof as specified
in [RFC4035] or [RFC5155], regardless of whether the Compact Answers OK (CO)
flag is set.
For an NSEC zone, a single NSEC record whose owner name matches the delegation
point and whose Next Domain Name is formed by prepending the label \043 to the
delegation point satisfies both components of the Name Error proof. It covers
both the queried name and the wildcard at the closest encloser, while its Type
Bit Maps field indicates the types present at the delegation point. The label
\043 sorts after the wildcard label (\042) in canonical ordering.
CDoE makes my head spin... Apologies if I got this wrong.
I probably misunderstood what this is saying. Unless I'm mistaken, the
canonical name order is:
test. < *.test. < +.test. < foo.test.
If the QNAME=foo.test. then the Next Owner cannot be \043 (+) because it
would not cover foo.test.
What did I get wrong?
All three NSEC3 records use a Next Hashed Owner Name formed by incrementing the
corresponding hash value by one.
Honestly I would not know how to do it. "corresponding" makes my head
explode.
---------------------
4.2. Explicit Queries for Delegation Types
When the EDNS(0) DE flag is set to 1 and the QNAME matches a delegation point,
a query for a Delegation Type MUST result in an authoritative answer if the
queried Delegation Type exists, or a NODATA response (AA flag set, RCODE=0,
empty answer section).
This has insufficient detail and over-specification at the same time :-)
I suggest:
- omitting the final () completely
- replace s/authoritative answer/positive answer/
- replace s/QNAME/effective QNAME/
I.e.
"When the EDNS(0) DE flag is set to 1 and the effective QNAME matches a
delegation point, a query for a Delegation Type MUST result in an
positive answer if the queried Delegation Type exists, or a NODATA
response."
A typical trap of re-specifying things. (AA flag set, RCODE=0, empty
answer section) is wrong, e.g. CNAME chains thwart this and I don't want
to go into the silly AA semantics debate vs. implementation practice.
Perhaps just add pointer to RFC 2308 for NODATA spec?
---------------------
5.2. Referrals
The presence of one or more NS-Omitting Delegation Type RRsets or an NS RRset
in the Authority section identifies the response as a referral.
Same story, also over-specified and wrong.
I suggest something more generic:
"Referral responses are identified according to the existing DNS
response-processing rules. For referral identification, an NS-Omitting
Delegation Type RRset in the Authority section serves the same role as
an NS RRset."
If more words are desired, something like this can be added:
"The presence of either kind of RRset alone is insufficient to identify
a response as a referral. Partial answers followed by referrals remain
possible. See RFC 9499 section 4, RFC 2308 section 2, RFC 1034 section
4.3.2."
There is multiple implementations which add NS RRs into authority
because they feel like it without creating referral responses.
As an example, the authoritative response from com auth servers looks
like this:
;; ANSWER SECTION:
example.com. 86400 IN DS 2371 13 2 [omitted]
example.com. 86400 IN RRSIG DS 13 2 86400 20261109013359 20261102002359
65344 com. [omitted]
;; AUTHORITY SECTION:
com. 172800 IN NS e.gtld-servers.net.
com. 172800 IN NS f.gtld-servers.net.
com. 172800 IN NS k.gtld-servers.net.
...
It does not feel like referral to me (or current BIND codebase!)
8.2.1. Stripping of Delegation Types from Referrals
A referral lacking this proof MUST be treated as tampered with and MUST be
ignored.
s/ and MUST be ignored//
How to handle 'tampered with' data is implementation decision. Some
cache it as failure, some retry. Mandating ignoring it is not within
scope of this document.
---------------------
8.4. Denial-of-Service via DE Flag Manipulation
They MUST instead return a conventional Name Error proof as described in
Section 4.1.1.
I think Name Error smells like over-specification again. I would prefer
'negative response' as it is less likely to cause problems in corner cases.
---------------------
Examples for DELEXT.
I agree with Libor that lack of examples makes it hard to follow, but I
propose to tackle it the other way around: DELEG is now required part of
DELEXT, so we can extend DELEG examples and point from DELEXT to DELEG
examples.
Keeping the examples together avoids duplication and lets the same
delegation demonstrate the interaction between DELEG and the other
Delegation Type segments.
I would much rather throw in additional NS-Preserving and On-Demand
Delegation Types into DELEG examples, which should be easy to do to and
it would demonstrate how it behaves in 'real world', and avoid
duplicating all the examples in DELEXT.
Having said that, the idea of exercising
(NS-preserving/omitting/on-demand) types along DELEG creates a new
conundrum:
What type numbers should be used for this?
Now I think I can see where Libor might have been going with his private
range commentary: If each sub-range (NS-preserving/omitting/on-demand)
had tiny sub-sub-range dedicated as private, we could throw these in
examples safely.
I think this allocation policy would accomplish it:
9.3. Changes to the DNS Parameters RR Types Registry
Decimal Hex Registration Procedures Note
61440- 0xF000- Expert Review or NS-Omitting
61551 0xF06F Standards Action Delegation TYPEs
61552- 0xF070- Private Use NS-Omitting
61567 0xF07F Delegation TYPEs
61568- 0xF080- Expert Review or NS-Preserving
61679 0xF0EF Standards Action Delegation TYPEs
61680- 0xF0F0- Private Use NS-Preserving
61695 0xF0FF Delegation TYPEs
61696- 0xF100- Expert Review or On-Demand
61935 0xF1EF Standards Action Delegation TYPEs
61936- 0xF1F0- Private Use On-Demand
61951 0xF1FF Delegation TYPEs
It affects only IANA policy and is simple for mplementations: They can
keep single numeric range per behavior class.
--
Petr Špaček
Internet Systems Consortium
_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]