[DNSOP] UDP fragmentation vs multiple-responses and multi-qtypes

Ondřej Surý <ondrej.sury@nic.cz> Thu, 20 July 2017 16:41 UTC

Date: Thu, 20 Jul 2017 18:41:33 +0200
From: Ondřej Surý <ondrej.sury@nic.cz>
To: dnsop <dnsop@ietf.org>
Message-ID: <1686897744.6421.1500568893614.JavaMail.zimbra@nic.cz>
MIME-Version: 1.0
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: quoted-printable
Thread-Index: drTJLRO/4nkR2JIAujzyHoaQf8cQWQ==
Thread-Topic: UDP fragmentation vs multiple-responses and multi-qtypes
Archived-At: <https://mailarchive.ietf.org/arch/msg/dnsop/orVbmYDlqkKxz5luv7xST-yI8RM>
Subject: [DNSOP] UDP fragmentation vs multiple-responses and multi-qtypes
Precedence: list

multi-qtypes Security Considerations says:
>    The method documented here does not change any of the security
>    properties of the DNS protocol itself.

I don't think this statement is true.  Why?

a) DNS DDoS threats are real and there was a shift towards minimizing
   answers.  This goes in the reverse direction.  But you address this
   in both security considerations.

multiple-responses Security Considerations says:
> 
>    Additional records will make DNS responses even larger than they are
>    currently, leading to larger records that can be used in DNS
>    reflection attacks.  One could mitigate this by only serving
>    responses to EXTRA requests over TCP or when using Cookies [RFC5395],
>    although there is no easy way to signal this to a client other than
>    through the use of the truncate bit.

multi-qtypes Security Considerations says:
>    It should however be noted that this method does increase the
>    potential amplification factor when the DNS protocol is used as a
>    vector for a denial of service attack.


b) UDP fragmentations - it strongly increases the risk of UDP fragmentation
   which is strongly discouraged (SHOULD NOT) in BCP 145.

also multiple-responses Security Considerations says:

>    A malicious authoritative server could include a large number of
>    extra records (and associated DNSSEC information) and attempt to DoS
>    the recursive by making it do lots of DNSSEC validation.  However,
>    this is not considered a realistic threat; CPU for validation is
>    cheap compared to bandwidth.  This can be mitigated by allowing the
>    recursive resolver to ignore Additional records whenever it considers
>    itself under attack or its CPU resources are otherwise over-
>    committed.

It should be noted, that ECC validation is more CPU intensive than RSA, as
as such I find "CPU for validation is cheap compared to bandwidth" quite
bold claim that should come with some data.

Cheers,
--
 Ondřej Surý -- Technical Fellow
 --------------------------------------------
 CZ.NIC, z.s.p.o.    --     Laboratoře CZ.NIC
 Milesovska 5, 130 00 Praha 3, Czech Republic
 mailto:ondrej.sury@nic.cz    https://nic.cz/
 --------------------------------------------

[DNSOP] UDP fragmentation vs multiple-responses a… Ondřej Surý
Re: [DNSOP] UDP fragmentation vs multiple-respons… Lanlan Pan