linux bypass loopback - got to dead end

Hi,

I want to force tcp traffic to go out even that the dest ip
is on the same machine.

I have eth0 with ip 10.56.211.89 and eth1 with 10.56.211.90 (for
example).
I want to open socket from 10.56.211.89 to 10.56.211.90 and bypass the
loopback, I mean that the packet will leave the linux on one interface
and reach the second.

so far couldn't do it.
any idea will be appreciated.

Thanks

Rony

(E-Mail Removed) wrote:
> Hi,
>
> I want to force tcp traffic to go out even that the dest ip
> is on the same machine.
>
> I have eth0 with ip 10.56.211.89 and eth1 with 10.56.211.90 (for
> example).
> I want to open socket from 10.56.211.89 to 10.56.211.90 and bypass the
> loopback, I mean that the packet will leave the linux on one interface
> and reach the second.
>
> so far couldn't do it.
> any idea will be appreciated.

You won't be able to do it without seriously
crippling the IP and ARP code in the kernel.

Even if you succeed in routing to the interface,
you won't get ARP working properly for external
loopback.

Would you please tell why?

In Rubini's book Linux Device Drivers there is
a pseudo-network driver capable of semi-external
loopback, but even that won't use real physical
Ethernet interfaces.

--

Tauno Voipio
tauno voipio (at) iki fi

In comp.os.linux.networking (E-Mail Removed):
> Hi,

> I want to force tcp traffic to go out even that the dest ip
> is on the same machine.

> I have eth0 with ip 10.56.211.89 and eth1 with 10.56.211.90 (for
> example).
> I want to open socket from 10.56.211.89 to 10.56.211.90 and bypass the
> loopback, I mean that the packet will leave the linux on one interface
> and reach the second.

> so far couldn't do it.
> any idea will be appreciated.

No way without modifying the source, the kernel is clever enough
to mention the device/IP is local and won't route a single packet
through the wire as you found out.

Should be easier to put a second (old) box in service.

--
Michael Heiming (X-PGP-Sig > GPG-Key ID: EDD27B94)
mail: echo (E-Mail Removed) | perl -pe 'y/a-z/n-za-m/'
#bofh excuse 144: Too few computrons available.

Tauno Voipio <(E-Mail Removed)> writes:

>(E-Mail Removed) wrote:
>> Hi,
>>
>> I want to force tcp traffic to go out even that the dest ip
>> is on the same machine.

I have no idea what this means. Go out where? The purpose of IP addresses
is to deliver packets to the destination. Not to take some tour of the
world.

>>
>> I have eth0 with ip 10.56.211.89 and eth1 with 10.56.211.90 (for
>> example).
>> I want to open socket from 10.56.211.89 to 10.56.211.90 and bypass the
>> loopback, I mean that the packet will leave the linux on one interface
>> and reach the second.

No idea what this means. Are you not receiving the packets?

>>
>> so far couldn't do it.
>> any idea will be appreciated.

>You won't be able to do it without seriously
>crippling the IP and ARP code in the kernel.

>Even if you succeed in routing to the interface,
>you won't get ARP working properly for external
>loopback.

>Would you please tell why?

>In Rubini's book Linux Device Drivers there is
>a pseudo-network driver capable of semi-external
>loopback, but even that won't use real physical
>Ethernet interfaces.

>--

>Tauno Voipio
>tauno voipio (at) iki fi

Hi,

I have kind of layer 7 traffic generator based on linux. I can (for
example) generate 1000 (concurrency) ftp downloads in wanted rate.
Currenly I need 2 machines to run it. one holds the traffic generator
application and the second run the ftp server. I can't insatll both
application on the same machine bacause then the traffic run on the
loopback and there is no use
for my application.

my goal is to be able to run both servers and application on the same
machine.

Rony

>
>No way without modifying the source, the kernel is clever enough
>to mention the device/IP is local and won't route a single packet
>through the wire as you found out.
>
>Should be easier to put a second (old) box in service.

can someone direct me to information about how can I change the kernel
to do this?

In addtion to "Rubini's book Linux Device Drivers"

Thanks a lot.
--

(E-Mail Removed) wrote:
>>No way without modifying the source, the kernel is clever enough
>>to mention the device/IP is local and won't route a single packet
>>through the wire as you found out.
>>
>>Should be easier to put a second (old) box in service.
>
>
> can someone direct me to information about how can I change the kernel
> to do this?

If you don't know, you - for sure - should not attempt it.
The change will break pretty many places in the kernel.

> In addtion to "Rubini's book Linux Device Drivers"

Read the sources, and understand what is going on
before changing anything.

The change hits both the IP layer and ARP - nothing to
do with the drivers directly.

For traffic tests, get the other host as test target.

--

Tauno Voipio
tauno voipio (at) iki fi