IDNs in TLS and Kerberos

[Simon Josefsson]

Erik van der Poel <erikv at google.com> writes:

> On Fri, Jan 23, 2009 at 4:21 AM, Simon Josefsson <simon at josefsson.org> wrote:
>> Harald Tveit Alvestrand <harald at alvestrand.no> writes:
>>> I haven't seen anyone claiming that they want to register a TLD in
>>> Dhivehi or Yiddish (the two BIDI cases where it matters that 2003 is
>>> more restrictive than 2008). But there are people arguing for
>>> registration of a TLD with a ZWNJ in it.
>>
>> There seems to be security problems with such a zone, if you consider
>> IDN strings in TLS certificates and Kerberos realms etc.
>
> Do the major implementations only accept Punycode in TLS certificates
> and Kerberos realms? Or do they also accept UTF-8, ISO-8859-1,
> upper-case, etc?

Only punycode in practice, as far as I know.  I recall some unclarity
for the TLS server name extension (UTF-8 vs Punycode) though.  I don't
see how this makes the situation different though: most web browsers
accepts UTF-8 or even Latin-1 etc from the user, and this is the string
that will be used to compare against the TLS certificate or sent in a
TLS server name extension field.

Problems would occur, for example, when comparing the punycode string in
a certificate or in the TLS server name field with what the
user/administrator typed locally.  If the preparation of these two
different strings were done using different IDNA algorithms and their
output differ depending on the algorithm used, it might trigger both
unwanted matches and unwanted rejections.

/Simon