Re: [TLS] TLS and middleboxes again

[Daniel Kahn Gillmor <[e-mail]>]

On 08/26/2011 04:04 PM, Yaron Sheffer wrote:
> two reasons for the middlebox to use the MAC (which may or may not be
> "justifiable" to you) are:
> 
> - Many security applications add a textual signature similar to "This
> mail was scanned by The Best Antivirus".

Just to be clear, this approach has no security value unless they also
actively filter *out* any and all content which might be rendered as
"This mail was scanned by The Best Antivirus".  Otherwise, it's trivial
for an attacker to include the string in their initial mail.

And frankly, with HTML e-mail these days, i suspect it would be rather
difficult to match all strings that might render to something visually
indistinguishable from this.  and difficult to excise them cleanly
without damaging other content.  Once an antivirus starts actively
tampering with content, i think it's beyond its useful scope, regardless
of the TLS implementations.  Keep the communication channels separate,
thanks.

> - If something bad is detected in the traffic, the middlebox might want 
> to issue an HTTP redirect, e.g. to display a copy of the company's 
> security policy.

Since HTTP redirects can only be done in an HTTP header, the middlebox
would need to buffer all HTTP content before relaying it to the User
Agent, in case "bad" material was found at the end.

I don't think either of these use cases are valuable, certainly not
valuable enough to warrant allowing middleboxes to violate the integrity
guarantees of TLS.

	--dkg