Paul Vixie wrote: > On Saturday, January 09, 2016 11:26:16 AM Mukund Sivaraman wrote: > > > > If a DNS message is received on the wire, that has a compressed name in > > some RR's RDATA which it should not have (going by its type), is it fair > > to still accept it as a valid message and process it if the > > implementation is able to do so? i.e., can Postel's law be followed > > here, or must the implementation strictly reject such messages? > > > > i followed postel's law with regard to receipt of compressed names anywhere > in any RDATA that i knew the format of, in both BIND4 and BIND8. the result > was that implementations who wrongly compressed non-well-known RDATA's > (including BIND4 and BIND8) were able to break that rule without pain. > > it's my strongly held belief that postel's law is wrong for RDATA > interpretation, and that the first implementation to compress something they > should not have compressed, should feel pain.
There is an analogous case with compression pointers themselves, which 1035 requires point to a "prior occurance [sic] of the same name". But BIND allowed pointers to point to later occurrences, and later implementations had to make the same allowance for compatibility reasons. -- Robert Edmonds _______________________________________________ DNSOP mailing list DNSOP@ietf.org https://www.ietf.org/mailman/listinfo/dnsop
