Standards tree image/ktx mime registration for review

Ned Freed ned.freed at mrochek.com
Tue Sep 7 02:24:30 CEST 2010



> Hi Ned,

> Thank you for your reply.

> First let me apologize for a typo in the registration request. The URL
> of the spec. was incorrect. A corrected registration template is
> attached. Other changes to this version are:

>     * Added URL of the main Khronos Group web site under "Change
>       Controller."
>     * Added the name of an application that uses (reads) this file format.
>     * Added additional info. to Security Considerations.
>     * Fixed some typos.

> See below for other comments.

> On 27/08/2010 05:27, Ned Freed wrote:
> > ...
> > As for your actual registration:
> >
> > ...
> > Given that this is a standards tree registration, the IESG is going to ask if
> > this comes from a recognized standards body. It sounds to me like The Khronos
> > Group qualifies, but this will be for the IESG to determine.

> I could find no information on the IANA web site that provides a
> definitive definition of "recognized" in this context.

Um, exactly what part of "the IESG makes the determination" did you fail to
understand? IANA != IESG. It would be frankly astonishing if the rules for how
the IESG determines standards body status could be found on the IANA site.

Having served on the IESG in the past, I also doubt very much the IESG has
bothered to create formalized rules for what consistitutes a recognized
standards body - maybe they should, but I doubt they have. So IMO it's also
pretty unlikely you'll find such rules on the IESG's site either, but at least
you'd be looking in the right place...

> However I believe
> Khronos should qualify as it is a widely supported industry consortium
> with responsibility for several well known standards.

That's my reading as well, but again, this is up to the IESG to determine.

> > ...
> >> The ktx type is a binary data stream which contains no executable code that
> >> could disrupt a client processor. There is no provision in the type
> >> specification that would allow authors to insert executable code that would
> >> present any security risk to a client machine.
> > IMO these security considerations are nowhere near sufficient for a standards
> > tree type. At an absolute minimum you need to add a discussion of possible
> > integrity/confidentiality concerns: Does data of this type require such
> > protection, and if it does, does the type provide it internally (and if so,
> > how) or must it be provided externally (and again, how)?
> It is not clear to me exactly what you are asking for. I looked at
> Security Considerations (Section 8.5) in the image/png registration (RFC
> 2083 <http://tools.ietf.org/html/rfc2083>) and could see nothing along
> the lines you seem to be suggesting.

RFC 2083 was published in March 1997. The current security requirement rules
for media types first appeared in a slightly different form in RFC 2048, which
was published in November 1996 - only four months before. This means that the
image/png specification was mostly developed under the previous set of rules
laid out in RFC 1590, which impose almost no requirements on what needs to be
discussed in the security considerations. (It's also clear from the failure to
use RFC 2048 terminology and the proper registration template that RFC 2083 was
only cursorily updated to refer to RFC 2048 and not RFC 1590.)

In other words, the failure of RFC 2083 to meet our current requirements means
almost nothing. Indeed, there was a period in the 90s where IANA cheerfully
registered anything anybody sent to them, no matter how sketchy. The fact that
IANA wasn't even managing to follow RFC 1521/RFC 1590 rules doesn't give you
license not to follow the current rules now.

RFC 2083 actually ends up doing a pretty good job of listing various
type-specific security issues, especially considering how lax RFC 1590 was
about this stuff.

In any case, RFC 4288 is the current specification for media type registration
procedures, and bullet point 4 in section 4.6 specifically notes that
unintentional information disclosure is a concern that should be covered. My
interpretation of that is that, as a practical matter, the simplest way to
address this is to note if encryption is ever needed and if it is, whether the
type provides the necesary protection or if it has to be done externally.

> Since this is an image format,
> naturally any image could be sent including images which one may wish to
> keep confidential.

That may be the case for a general image format, but there are many highly
specialized formats, including image formats, that are intended for specific
purposes where confidentiality is a complete nonissue. And at the opposite
extreme, there are formats where confidentiality are absolute requirements for
any sort of usage.

> There is no encryption in the format so users wishing
> to keep their images confidential should overlay their own encryption.
> Is this the kind of discussion you are requesting?

Yes, if correct, that's the sort of statement that needs to be made for a
general-purpose image type. Really, this is not a big deal.

> > ...
> >> The KTX file format specification can be found at
> >>     http://www.khronos.org/opengles/sdk/sdk/tools/KTX.
> >> This is a temporary location. The spec. will be announced next week 7/28
> >> and the permanent location will settled then.
> > A stable specification location is very important for types in the standards
> > tree. This will need to be updated once a permanent location is decided on.
> It is stable. The specification was ratified and a permanent home
> created during the more than a month that has passed since I wrote the
> above for my initial message to ietf-types. The permanent home is

>     http://www.khronos.org/opengles/sdk/tools/KTX/file_format_spec/

> as noted in the revised registration that I have attached.

First, I was talking about the stability of the specification's *location*, in
response to you having said the location was temporary. Having a stable
location now addresses this concern.

Second, it actually isn't a requirement that the specification be stable.
Features and capabilities are added to and removed from media types all the
time. text/html is an especially good example of this, but there are plenty of
others. And yes, this means that the boundary between when you're modifying an
existing type and creating a new one is hard tp nail down, but when you get
down to specifics it's usually pretty clear what the right answer is.

				Ned



More information about the Ietf-types mailing list