[idn] Specifying leakage effects

[Paul Hoffman / IMC <phoffman@imc.org>]

The term "leakage" has been used on this list in the past few weeks. It 
would be good to look at, because it is a very real concern for any 
protocol design. Any change or addition that idn makes to the DNS will leak 
to other parts of the Internet. This has been seen over and over in the 
history of protocol development on the Internet. The new encoding for idn 
names will appear in places other than on wire between resolution clients 
and resolvers; the question is what the effect of that leakage will be.

Leakage of text-based encodings will be visually annoying. An 
internationalized name part that meant something will mean nothing and will 
take up more characters than the original characters. If the idn protocol 
has part tagging, a user seeing this will at least know that it is a leaked 
name part, but still will have no idea what the contents of the part is.

Leakage of a UTF-8-based encoding to protocols that do not handle non-ASCII 
domain names, such as SMTP, will hopefully cause the leaked data to be 
rejected (it would be worse from a security standpoint if the data were 
mis-directed because of misinterpretation of the domain name). If the idn 
protocol for using UTF-8 has an ASCII fallback mechanism, that mechanism 
has to be installed in all implementations of all protocols that do not 
handle non-ASCII domain names. Leakage to any implementation that has not 
implemented the fallback mechanism will still fail.

If the idn is text-based, we have to live with and mitigate the visual 
annoyance for users. If the idn is UTF-8-based, we have to be really, 
really, really sure that it is designed to mitigate any leakage because of 
the much more serious consequences (such as large amounts of lost mail and 
failure of PKIX certificates to validate, just to name two protocols). I 
propose that the latter is very difficult and is likely to cause much more 
damage to the Internet than the former.

Just as it is required for anyone writing a text-based idn proposal to deal 
with the effects of leakage, it is incumbent on whoever proposes a 
UTF-8-based idn to deal with leakage as well. I think it would be good to 
add to the requirements document:

All idn protocol documents must fully detail the expected effects of 
leaking of the specified encoding to protocols other than the DNS 
resolution protocol.


--Paul Hoffman, Director
--Internet Mail Consortium