Re: [secdir] review draft-ietf-httpbis-content-disp-06

[Julian Reschke <[e-mail]>]

On 14.03.2011 18:58, Hilarie Orman wrote:
> Security Review of
>
> Use of the Content-Disposition Header Field in the Hypertext Transfer Protocol
>    (HTTP)
>
> Do not be alarmed.  I have reviewed this document as part of the
> security directorate's ongoing effort to review all IETF documents
> being processed by the IESG. These comments were written primarily for
> the benefit of the security area directors. Document editors and WG
> chairs should treat these comments just like any other last call
> comments.

Hilarie, thanks for the feedback.

> The document specifies the syntax of the content-disposition field and
> its parameters for http responses.  It is based on the MIME
> specification for the same field.
>
> The Content-Disposition header supports a parameter called "filename",
> the recommended name for saving the content.  That feature is just
> plain dangerous.  A good many exploits are based it.  But, it's not
> really the fault of the protocol.  Protocols don't kill people, people
> clicking "yes" kill themselves.

Indeed.

Furthermore, it's saving web content, not the header field itself. 
Recall that if the C-D header field is not present, the UA will derive 
one from the URI, which can have exactly the same problems (except for 
the path-related ones).

> The document is clearly written, and it does point out many of the
> dangers inherent in trusting arbitrary filenames and file extensions.
> However, I think that this feature will continue to be a major source
> of vulnerabilities.

Out of curiosity: do you believe it's a vulnerability today? Anything 
specific? UA bug reports that haven't been fixed?

>  From a security viewpoint, I think the protocol should restrict
> filenames to ascii alphabetic characters (no "." or "/"), and that if

We can't rule out "."; it just occurs frequently in filenames (think 
"foo.tar.gz").

We *could* forbid path separators, but that wouldn't mean that we 
wouldn't have to say what to do when you see them.

> the filename does not conform, the receiver SHOULD reject the message
> and send an error code back to the sender.  The sender should not be

As Alexey said, there's no back channel. Also, if sending a 
non-conforming name was actually intentional, complaining about it will 
not achieve anything.

> allowed to specify a file extension.  The receiver SHOULD write the
> files into a quarantined disk area using a temporary filename, the
> file to be released to the recommended name pending manual review.
> But, that would break the world, as would so many security
> recommendations.

Indeed. I'd rather not try to impose normative requirements here when we 
know that UAs won't do it anyway. It's much more important to call out 
the risks.

Best regards, Julian