Hi, 

I'm currently working on a hopefully-quick-and-not-so-dirty pipebackend to feed 
powerdns from an openstack nova database.


Biggest challenge: Different tenants MUST NOT see any data from within other 
tenants' environments.

Tenants can (by nova-db means) be separated by their IP; I can handle & cache 
that processing in the backend, so no (performance) issue here (yet).


Now, there are two issues arising, each of which can be subsumed as 
"edns-subnet-partial-processing", hence the subject. The main problem is pdns 
not-yet-fully-edns-subnet-processing, the sideshow would be a feature to have 
pdns willingly only doing one-armed-edns-subnet-processing.



Main Problem: Even with edns-subnet-processing, responses are cached and 
re-used beyond the scope-bits' allowance:

Querying from host A, I see the backend being called, answering authoritative 
with scopebits=32.
Same query from host B, I do NOT see the backend being called again, but get 
the same response. 
So at this point, the backend isn't even given the chance to find out which 
tenant the IP belongs to, leading to any tenant being able to query any cached 
responses from any other tenant.

Very, very ugly.

Currently using:

root@crns:~# dpkg -l | grep pdns
ii  pdns-backend-pipe                3.0-1.1ubuntu1               
pipe/coprocess backend for PowerDNS
ii  pdns-server                      3.0-1.1ubuntu1               extremely 
powerful and versatile nameserver
root@crns:~#

Is there a timeline when pdns will start respecting scope bits, or is it 
already active in current codebase and I only need to get the -static- packages?



Sideshow: 

Regarding potential "information privilege escalation" by edns-subnet as shown 
by Florian Streibelt at Denog5*), a configuration stanza as mentioned in the 
subject would be great, telling pdns to ignore/discard any edns-subnet 
information provided by the client whilse still working with edns towards its 
backends (and the recursor).



Kind regards,

Sebastian
*): http://www.denog.de/meetings/denog5/pdf/08_Streibelt_DNS_clientip.pdf
--
Sebastian Posner
Unix-Systemspezialist
Deutsche Telekom AG, Products & Innovation 
"Es hat einmal einer gesagt, das geht nicht. Dann kam einer, der wusste das 
nicht und hat es einfach gemacht"



_______________________________________________
Pdns-dev mailing list
Pdns-dev@mailman.powerdns.com
http://mailman.powerdns.com/mailman/listinfo/pdns-dev

Reply via email to