Re: SETTINGS error handling

[Tatsuhiro Tsujikawa <[e-mail]>]

On Wed, Jul 17, 2013 at 1:31 AM, Roberto Peon <[e-mail]> wrote:

> Clarification:
> If you were meaning to say that the server/client should close the
> connection, etc. upon receipt of bad settings, then I wholly agree.
>

I think that is what I was meant to say.


>  The client shouldn't bother to validate data that it won't interpret,
> however, which is what I was saying very poorly in the last email.
>
Yes, the current spec says this about unknown settings. The point is flow
control window size affects the interactions between client and server,
clearly misinterpreting it will lead to deadlock situation. So it should be
validated and if it is bad, then tear down the connection.

Best regards,

Tatsuhiro Tsujikawa

-=R
> On Jul 16, 2013 9:26 AM, "Roberto Peon" <[e-mail]> wrote:
>
>> I think that knowledge client-side of things like that is unwarranted
>> complexity -- the server must sanity check that data anyway to deal with
>> malicious clients, or, in the case of port-80 checksum "failures" which
>> have allowed corrupted bits through.
>>
>> -=R
>> On Jul 16, 2013 8:35 AM, "Tatsuhiro Tsujikawa" <[e-mail]>
>> wrote:
>>
>>> +1 for strict validation.
>>>
>>> Regarding validation for SETTINGS, how about validating the values in
>>> SETTINGS frame?
>>>
>>> For example, the value in SETTINGS has unsigned 32 bit and it means
>>> SETTINGS_INITIAL_WINDOW_SIZE
>>> could have, say, 2^31, which is invalid for flow control.
>>> http://tools.ietf.org/html/draft-ietf-httpbis-http2-04#section-6.9.1 says
>>> if such value is received in
>>> WINDOW_UPDATE frame, it must be responded with FLOW_CONTROL_ERROR.
>>> But it does not say about SETTINGS frame for invalid window size (it may
>>> infer that but still).
>>> I think it would be good to add some error handling of values on
>>> SETTINGS frame reception.
>>>
>>> Best regards,
>>>
>>> Tatsuhiro Tsujikawa
>>>
>>>
>>> On Tue, Jul 16, 2013 at 9:22 PM, Amos Jeffries <[e-mail]>wrote:
>>>
>>>> On 16/07/2013 10:20 a.m., Martin Thomson wrote:
>>>>
>>>>> On 15 July 2013 12:44, Patrick McManus <[e-mail]> wrote:
>>>>>
>>>>>> I'm wondering why the text proscribes error handling of MUST ignore in
>>>>>> response to violation of the MUST NOT send provision.
>>>>>>
>>>>> I think that someone either caught Postel's DIsease, or is in
>>>>> remission.
>>>>>
>>>>>  Can we either change it to PROTOCOL ERROR (preferred) or just be
>>>>>> silent on
>>>>>> handling of the error?
>>>>>>
>>>>> PROTOCOL_ERROR seems appropriate.
>>>>>
>>>>> Opened:
>>>>> https://github.com/http2/**http2-spec/issues/174<https://github.com/http2/http2-spec/issues/174>
>>>>>
>>>>
>>>> +1. And double that for more strictness everywhere.
>>>>
>>>>
>>>> Amos
>>>>
>>>>
>>>