mailing list archives
Re: CVE Request New-djbdns: dnscache: potential cache poisoning
From: P J P <ppandit () redhat com>
Date: Mon, 10 Feb 2014 17:47:14 +0530 (IST)
+-- On Mon, 10 Feb 2014, Florian Weimer wrote --+
| How it is possible to poison the cache if the response is not cached?
IIUC, response is cached, and cached in the same location. Because it
'hashes' to the same bucket always, an attacker is able to overwrite entries
in that bucket by flooding a resolver with queries involving other domains
whose resource records also 'hash' to the same bucket.
As 'dnscache' does not go beyond 100 entries in this bucket, it is made to
contact TLD servers for new requests. If this query pattern of 'dnscache' is
predictable, it could be possible to poison it with usual response flood
technique (of-course that's easier said than done).
With 'SipHash' function, that 'bucket' selection is randomised. IOW, multiple
queries with a same domain/key might 'hash' to different buckets.
That's my understanding of the post. I'll check with the upstream author for
Prasad J Pandit / Red Hat Security Response Team