Re: [Pound Mailing List] Location comparision with backend address & host

On 03/25/2013 04:40 PM, Joe Gooch wrote:
> Since I split my listeners, (ipv4 and ipv6, no wildcard), I haven't had this problem.
>
Had I known that, probably I would have changed my configuration
instead of patching Pound!

> It's something that could be dealt with easily with logparsers/helper scripts.
>
My configuration has Pound sending requests to Apache; on the Apache
machines, mod_extract_forwarded is used to mask the address of the
reverse proxy and to get the client addresses into the Apache vars.

Later on, when I added IPv6 on the Pound machine, I started to
get the v4 MAPPED addresses (those in the form ::ffff:a.b.c.d)
into Apache. [Due to my wildcard listener, as Joe noticed]

ACLs did not match anymore on those addresses. Ouch!

I though, let's duplicate all the whitelisted addresses (it is a
short list, fortunately) in the new form; but Apache is refusing
those addresses as invalid.

Next thing was preventing Pound from generating those, and so this
patch came to light. I didn't send to the list immediately, as I wanted
to test it, and then I forgot about it.

I've used it for some months now, and when I saw the message from
Raj Kishore, I though it was time to send this; maybe it will be
useful to somebody else.

> However, I could see how some might want the ips translated.
>
> I think you'll find this is more simplistic and accomplishes the same result.
>
Didn't have time to test this yet, but yes, it's way cleaner
than mine; thank you!

Best regards,
					Bussi Andrea


> Joe
>
>> -----Original Message-----
>> From: Bussi Andrea [mailto:[email protected]]
>> Sent: Monday, March 25, 2013 11:07 AM
>> To: [email protected]
>> Subject: Re: [Pound Mailing List] Location comparision with backend
>> address & host
>>
>> On 03/18/2013 09:43 AM, Bussi Andrea wrote:
>>> On 03/16/2013 09:30 AM, Raj Kishore1/CHN/TCS wrote:
>>>> Hi Joe,
>>>>
>>>> Thanks for the previous replies.
>>>>
>>>> We have modified the pound 1.9 socket structures to listen IPV4/ipv6
>>>> address.
>>>>
>>>> However when IPV4 request comes,it is mapped to IPV6 address as
>>>> ::ffff:<X.X.X.X> In case of IPV6 client, it is fetching the IPV6
>>>> address only.
>>>>
>>>> Is it the right mechansim to handle ipv4 client in ipv4 mapped ipv6
>>>> address mode in pound?
>>>>
>>>> Do we need to handele explicitly  the IPV4-6 mapped address in dual
>>>> stack?
>>>>
>>> I had the same problem with Pound 2.6; it seems to be an issue in the
>>> inet_ntop function, if I'm not mistaken.
>>>
>>> I don't believe my patch (attached below) is the right way to solve
>>> it; the proper way should be a patched inet_ntop.
>>>
>>> But this solved my troubles; maybe somebody else will find it useful
>>> too.
>>>
>>
>> The patch I previously posted didn't initialize the port variable for
>> IPv4 addresses, so the addr2str function would print an uninitialized
>> stack value as the port number.
>>
>> This is the right patch.
>>
>> Sorry for the mistake,
>> 				Bussi Andrea
>>
>>
>>> Best regards,
>>>                       Bussi Andrea
>>>
>>
>


--
To unsubscribe send an email with subject unsubscribe to [email protected]
Please contact [email protected] for questions.