Re: svn commit: r1801594 - in /httpd/httpd/trunk: docs/manual/mod/mod_proxy_wstunnel.xml modules/proxy/mod_proxy_wstunnel.c

[Yann Ylavic <yl...@gmail.com>]

Hi Jean-Frédéric,

sorry to dig up an old commit, I'm currently working on integrating
wstunnel into mod_proxy_http and wondering what is the use case for
upgrade=NONE. Please see below..

On Thu, Jul 13, 2017 at 2:58 PM jean-frederic clere <jf...@gmail.com> wrote:
>
> On 07/12/2017 07:25 PM, Jacob Champion wrote:
> > On 07/11/2017 05:36 AM, Yann Ylavic wrote:
> >> I think it's quite hazardous to use/allow ANY and would prefer the
> >> upgrade_method (worker->s->upgrade) to be a list of acceptable protocols.
> >
> > I think both ANY *and* NONE are dangerous. Both of them turn
> > proxy_wstunnel into a generic TCP forwarder (and NONE does so without
> > any opt-in on the client's part).

I see some possible use of upgrade=ANY (tunnel any protocol after the
upgrade, no restriction), but NONE is bypassing the check for the
client Upgrade header and act as if it were Websocket. Why would we
upgrade a connection if the client does not ask for it?

>
> So how I have the following to proxy:
> <request>
> GET /jboss-websocket-hello/websocket/helloName HTTP/1.1
[snip]
> Connection: keep-alive, Upgrade
[snip]
> Upgrade: websocket
[snip]
> </request>
>
> <response>
> HTTP/1.1 101
> Upgrade: websocket
> Connection: upgrade
[snip]
> </response>
>
> ...,..9e...,.$.H...W(I-.QH+..U(OM*.O.N-QH.K)...+..
> Tomcat web socket stuff...
>
> So yes the HTTP/1.1 really needs to upgrade and NONE is just a work-around.

The above is some usual Websocket upgrade AFAICT, don't
upgrade=websocket or upgrade=any or even no upgrade at all work in
this case?

Regards,
Yann.

Re: svn commit: r1801594 - in /httpd/httpd/trunk: docs/manual/mod/mod_proxy_wstunnel.xml modules/proxy/mod_proxy_wstunnel.c

[jean-frederic clere <jf...@gmail.com>]

On 01/11/2019 21:15, Yann Ylavic wrote:
> Hi Jean-Frédéric,
> 
> sorry to dig up an old commit, I'm currently working on integrating
> wstunnel into mod_proxy_http and wondering what is the use case for
> upgrade=NONE. Please see below...

The upgrade=NONE was added to work around 
https://issues.jboss.org/browse/JBCS-291, I will dig again in the 
problem later today.

> 
> On Thu, Jul 13, 2017 at 2:58 PM jean-frederic clere <jf...@gmail.com> wrote:
>>
>> On 07/12/2017 07:25 PM, Jacob Champion wrote:
>>> On 07/11/2017 05:36 AM, Yann Ylavic wrote:
>>>> I think it's quite hazardous to use/allow ANY and would prefer the
>>>> upgrade_method (worker->s->upgrade) to be a list of acceptable protocols.
>>>
>>> I think both ANY *and* NONE are dangerous. Both of them turn
>>> proxy_wstunnel into a generic TCP forwarder (and NONE does so without
>>> any opt-in on the client's part).
> 
> I see some possible use of upgrade=ANY (tunnel any protocol after the
> upgrade, no restriction), but NONE is bypassing the check for the
> client Upgrade header and act as if it were Websocket. Why would we
> upgrade a connection if the client does not ask for it?
> 
>>
>> So how I have the following to proxy:
>> <request>
>> GET /jboss-websocket-hello/websocket/helloName HTTP/1.1
> [snip]
>> Connection: keep-alive, Upgrade
> [snip]
>> Upgrade: websocket
> [snip]
>> </request>
>>
>> <response>
>> HTTP/1.1 101
>> Upgrade: websocket
>> Connection: upgrade
> [snip]
>> </response>
>>
>> ...,..9e...,.$.H...W(I-.QH+..U(OM*.O.N-QH.K)...+..
>> Tomcat web socket stuff...
>>
>> So yes the HTTP/1.1 really needs to upgrade and NONE is just a work-around.
> 
> The above is some usual Websocket upgrade AFAICT, don't
> upgrade=websocket or upgrade=any or even no upgrade at all work in
> this case?
> 
> Regards,
> Yann.
> 


-- 
Cheers

Jean-Frederic