I'm put in an awkward position of having to respond to a message which
wasn't sent to me in the first place. But still...
"This bug was reported over and over again" - I find this statement
confusing. The bug class of "DNS transaction ID not being random enough"
was sure reported for several DNS server, including BIND. My paper
clearly references e.g.
http://www.openbsd.org/advisories/res_random.txt (as reference [7]).
However, I'm not familiar with public reports that outline the
seriousness of the non-randomness of BIND *9*, to the extent my report
did. So the way I see it is that this particular bug, in BIND 9, was not
explicitly reported before.
"it's not like this hasn't been reported, and fixed, many times by many
others" - so if it's fixed so many times, how come it was still
vulnerable, and ISC had to issue their patches?
-Amit
Gadi Evron wrote:
This is Paul Vixie's response on this, when I asked him for verification:
-----
this bug has been reported over and over again for a dozen years. it's
odd to have to keep fixing it-- i fixed it in bind4 and bind8 when theo
de raadt offered me his random number generator to use. bind9 should've
used that same one but apparently didn't. note that with this fix, the
difficulty in poisoning someone's cache rises from "a few tens of
seconds"
to "a few minutes". it's a 16-bit field. not a lot of room for
randomness or unpredictability. only DNSSEC, a protocol change, fixes
this problem, which is fundamentally a protocol problem. but since folks
just won't leave it alone and keep on reporting it year after decade, we
will keep on improving our random number generator for this dinky little
16-bit field. i just wish the reporters wouldn't be so smarmy and self
congradulatory about it. it's not like this hasn't been reported, and
fixed, many times by many others.
-----
Gadi.
