[TLS] Read-only vs. read-write proxies (resent as plain text)

[Yaron Sheffer <[e-mail]>]

First, I strongly dislike this type of corporate proxies, and would take 
the German constitutional regime any time. But in other free countries, 
this scenario is perfectly OK (legal and socially acceptable), and we 
need to cater for this use case.

If we break the TLS architecture as proposed here, we will have killed 
forever any chance of TLS-level mutual authentication (TLS-PSK, 
TLS-SRP). This is IMHO too high a price for the benefits of R/W proxies.

I have not been convinced by David's explanation of the need for R/W 
proxies, nor by Yoav's use case. A malware detection proxy can drop the 
connection precisely at the time malware is detected: the Word document 
would not be received in its entirety and would be rejected by the 
client. Even if a partial document is acceptable (e.g. JPEG), a partial 
malware is (hopefully) not effective.

What are the reasons proxies need to modify the stream?

- Purely informational app-level insertion, of the "sent from my iPhone" 
kind. The *client* has a trusted channel to the proxy, and if this is 
needed, the client can insert this stuff.

- Security-specific signaling, of the "no spam detected here" kind. The 
server wouldn't trust the proxy with that anyway (this is a client-side 
proxy!) so this is not useful.

- Malware detection: just ax the connection.

- Other reasons?

Lastly, it is important that the server should know of the proxy's 
presence, even if the indication relies on the client's and/or the 
proxy's good will. I can think of banks who still use passwords for 
client auth, but would not allow logins where that password is clearly 
intercepted by a third party.

Thanks,

     Yaron