Re: Pseudo end-to-end connections considered harmful

On 25/07/2014 1:22 a.m., Roland Zink wrote:
> Accessing web sites through TLS gives the feeling of just talking to
> this site.

Such feeling is an illusion. Human estimations about security and safety
are notoriously inaccurate.

It is often the case that http:// traffic going to a local proxy which
encrypts using only the latest most secure TLS 1.2 ciphers is far better
for safety than the browser itself connecting https:// directly with
silent fallback to outdated TLS or even SSL encryption.

> The retrieved HTML content however cause the browser to open
> more connections for subresources of the displayed page, e.g. there are
> multiple endpoints and third parties are involved. It is known that in
> some countries it is possible for intelligence agencies to get access to
> the data after decryption has been done. If encryption is done to
> provide real end to end security then the use of any third party
> subresource must be avoided in order to not violate the users privacies
> concerns. For example an intelligence agency can surveil who is browsing
> where by just using some tracking companies data including the referer
> header data, ever cookies and other tracking data.
> When a http2 browser is using TLS then it should use a single end-to-end
> connection and refrain from open any further connections. The server is
> the endpoint and is therefore not allowed to forward the request. Any
> proxy / gateway must mark responses with a via header and http2 clients
> using a TLS connection must close the connection if they discover such a
> via header.

Which requirement to add Via exists in RFC2616 and is already soundly
ignored by the intelligence community middelware causing risk.

The only thing this advice will do is break end-user middleware
providing useful and non-harmful protection. Users AV or adware
protection, corporate TLS tunnel proxies, ISP based AV proxy, content
provider CDN TLS gateway, and such like.

Also, please do not confuse TLS and HTTPS.
 - Any agent using *TLS* should expect the connection to the server it
is connecting to be secure. Hops beyond that server are not relevant and
offer no guarantee of security.
 - Any agent using *HTTPS* should expect end-to-end security even if
that connection goes via several proxy hops.


Received on Thursday, 24 July 2014 14:48:12 UTC