[dns-operations] DNSSEC and FIPS-140

[James Stevens]

Viktor, you're a hero - this is insightful & comprehensive response.

Thank you so much.


James


On 01/12/2018 19:09, Viktor Dukhovni wrote:
> On Sat, Dec 01, 2018 at 03:23:10PM +0000, James Stevens wrote:
> 
>> We're running a PowerDNS Master on RHEL 7 installed with FIPS compliance,
> 
> Presumably OpenSSL 1.0.2 with the OpenSSL FIPS module version 2.0.9.
> There is not yet a FIPS module for the new OpenSSL 1.1.0 and 1.1.1
> releases, design discussions for that are under way now, and if all
> goes well a FIPS validated OpenSSL 3.0 will be available (for some
> validated platforms) some time in 2019.
> 
> In DNSSEC, MD5 is only "needed" for the long deprecated algorithm 1
> (RSAMD5):
> 
>    https://www.iana.org/assignments/dns-sec-alg-numbers/dns-sec-alg-numbers.xhtml#dns-sec-alg-numbers-1
> 
> Among the ~9 million domains covered in my DANE/DNSSEC survey, only
> 31 list RSAMD5 (algorithm 1) in their DS records.  Of these 11 also
> list some of the usual algorithms (5, 7, 8 or 13), and validate
> with one of those.  None publish RSAMD5 keys.
> 
> Two domains are up and running, but always fail validation (no DNSKEY
> RRset returned):
> 
>      http://dnsviz.net/d/0jl.com/XALVSA/dnssec/
>      http://dnsviz.net/d/sdi-gh.com/XALXmQ/dnssec/
> 
> the remaining 18 are simply out of service, with or without validation.
> 
> Bottom line, RSAMD5 is not only deprecated, but is also at this
> point *never* used.  Server and resolver implementations can and
> should remove the code point from the library, and treat RSAMD5 as
> "unknown", with any domains that publish only RSAMD5 treated as
> unsigned.
> 
>> ... but this query will crash PowerDNS ...
>>
>> $ dig +dnssec @127.0.0.1 example.com soa
>>
>> By "crash" I mean PowerDNS core-dumps - interestingly with signal 6,
>> SIGABRT - so sounds like some kind of assertion has failed - like you
>> said, looks like this needs to be taken up elsewhere.
> 
> A lot depends on how PowerDNS "uses" MD5, which in DNSSEC is only
> If it directly uses MD5_Init, MD5_Update(), MD5_Final() or EVP_md5()
> (you should be able to see the symbols used via "nm" or "objdump
> -T") and the OpenSSL library is built with MD5 disabled, late
> binding of dynamic symbols might leed to a failure when these
> functions are first accessed.
> 
> The recommended way to access EVP digest algorithms is via:
> 
> 	EVP_get_digestbyname()
> 	EVP_get_digestbyobj()
> 
> which won't crash when an expected algorithm is not present in the
> library, but will return NULL handles, that the application can
> check for, and continue without the algorithm in question, treating
> it as unsupported.  PowerDNS should avoid direct access to specific
> agorithms, and should use the algorithm-independent lookup functions.
>