On Mon, 16 Aug 2021, Hollenbeck, Scott wrote:

The iterative nature of recursive resolution gives an on-path monitor multiple 
opportunities to observe query traffic between a recursive resolver and an 
authoritative name server. Even with encryption, the name server IP addresses 
can be used to draw accurate conclusions about qnames by matching IP addresses 
to name servers identified in publicly available zones. This is something that 
needs to be addressed in the Security Considerations section of any draft that 
describes a recursive-to-authoritative encryption solution.

It can be pointed out, would I would not say it can be "addressed".

BEGIN
Encryption is one of several controls that can reduce the risk of disclosure of 
sensitive information. Resolver-to-authoritative DNS encryption will protect 
the DNS traffic exchanged between the resolver and the authoritative name 
server from being directly observed and/or modified by an adversary. However, 
as in other systems where encryption is deployed, implementers should also 
consider other ways in which the same or related information may also be 
exposed, and how to mitigate those risks. For example, encryption provides 
confidentiality protection for DNS query and response payloads between 
recursive resolvers and authoritative name servers, but iterative resolution 
makes it possible to perform traffic analysis using other, non-encrypted 
information that can be observed in subsequent iteration steps. The names and 
IP addresses of the name servers for most DNS zones are publicly available, so 
an attacker that can monitor traffic exchanged between a resolver and an 
authoritative
 n

DNS padding (RFC 7830) could be mentioned to hide the exact packet size of DNS 
queries/responses.

ame server will often be able to identify the zone of interest based on the IP 
address of the authoritative name server. Iterative resolution using 
destination IP addresses that were encrypted in a previous query makes it 
possible to identify the set of domains that might be associated with a query 
by observing the destination IP address and comparing that to the addresses and 
names found in publicly-available zone files. This risk can be reduced by 
limiting the number of queries sent to authoritative name servers using 
techniques such as hyperlocal root processing [RFC7706], NXDOMAIN cut 
processing [RFC8020], and aggressive DNSSEC caching [RFC8198].

I'm not sure I agree with "often" and "make it possible". That really
depends on the number of zones hosted by the particular nameserver
and the particular service contacted after the lookup. If the nameserver
hosts 1M domains, and the target webserver hosts 1K domains, it might
still be possible to somewhat hide what you were talking to.

The size of encrypted query and response packets can also be used to detect 
query patterns. The set of name servers and IP addresses returned in a response 
to a query changes infrequently, so the size of the DNS response for a specific 
set of name servers tends to stay the same during a given time period. The size 
of responses can be monitored to observe patterns associated with responses 
from specific name servers, and this information can be used to identify 
queries for specific domain names. This risk can be reduced by padding the 
record payload of TLS-protected responses to eliminate response size 
variability.

As I said, DNS padding can be used for that.

But in general, a little bit of this discussion is suitable in the
Security Considerations. But clearly, we can't add a multitude of
research papers in this section. Perhaps the goals can be stated
more realistically, eg "use encryption to not make it trivially
easy for any on-path attaker to see your DNS" and ensure there is
no expectation of "total anonymity".

Paul

_______________________________________________
dns-privacy mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dns-privacy

Reply via email to