Re: [Alldispatch] WebSocket Extension to disable masking

[Eric Rescorla <[e-mail]>]

On Fri, Mar 1, 2024 at 6:01 AM Dragana Damjanovic <[e-mail]>
wrote:

>
>
>> I am aware that TLS encryption does not protect from the original attack.
>>>
>>
>> But then I don't understand what the point of requiring encryption in
>> order to enable this feature is, because the attack in question involves
>> priming the cache on one connection and then having it affect another
>> connection, and it's *that* connection which has to be unencrypted for the
>> attack to succeed. So, it's not clear to me what the restriction to
>> encrypted traffic does.
>>
>>
> The reason I am restricting this to TLS is that it is a bit harder to
> create the attack, not impossible. It also would mean that the vulnerable
> cashing proxies will interpret data coming on port 443 as plain text data
> and cache it. Then later the proxies would serve the data on port 80. I do
> not know how common that is, I would expect it is not very common, because
> I expect the caching would be limited to port 80.
>

I'm not sure that this has the impact you suggest, because the attacker
controls the port on which the data is sent, so they should be able to run
HTTPS on port 80, no?


About the benefits vs. complexity:
>>> The proposal uses an already existing mechanism for negotiating an
>>> extension, therefore adding this extension is fairly simple because the
>>> negotiation already exists. I do not have performance data and the masking
>>> operation, XOR, will not cause a lot of CPU usage, but it is still a
>>> unnecessary operation.
>>>
>>
>>  I agree that the negotiation is not very complicated, but what's even
>> less complicated is to do nothing, and given that you haven't shown any
>> significant cost to the existing design, this doesn't seem like much in the
>> way of motivation.
>>
>
> Even if CPU usage is not affected a lot, on servers that serve a large
> number of connections that will be multiplied by a lot.
>

It would be nice if you were able to quantify this as a fraction of the
cost of the connection as a whole. I suspect the number is extremely low,
given the cost of the cryptography itself.

-Ekr


>
> Dragana
>
>
>>
>> -Ekr
>>
>>
>> -Ekr
>>>>
>>>>
>>>> [0]
>>>> https://www.adambarth.com/papers/2011/huang-chen-barth-rescorla-jackson.pdf
>>>>
>>>> On Wed, Feb 28, 2024 at 4:37 AM Dragana Damjanovic <
>>>> [e-mail]> wrote:
>>>>
>>>>>
>>>>> Hi,
>>>>> I am looking for the right working group for my draft "WebSocket
>>>>> Extension to disable masking"
>>>>>
>>>>> https://datatracker.ietf.org/doc/draft-damjanovic-websockets-nomasking/
>>>>>
>>>>> There was some discussion about the previous version of the draft in
>>>>> the HTTPbis working group:
>>>>> https://lists.w3.org/Archives/Public/ietf-http-wg/2023AprJun/0135.htmlv
>>>>>
>>>>>
>>>>> Thanks,
>>>>> Dragana
>>>>> --
>>>>> Alldispatch mailing list
>>>>> [e-mail]
>>>>> https://www.ietf.org/mailman/listinfo/alldispatch
>>>>>
>>>>