Re: route problem

Le mercredi 18 janvier 2012 à 20:29 -0500, Bill Fink a écrit :

On Wed, 18 Jan 2012, Eric Dumazet wrote:

Actually, if I'm interpreting the "ip route list cache" output
correctly, it appears it has local IP addresses of both 192.168.101.10
and 192.168.101.20.

quoted

local 192.168.101.10 from 192.168.101.101 dev lo  src 192.168.101.10 
    cache <local,src-direct>  iif eth1

quoted

local 192.168.101.20 from 192.168.101.101 dev lo  src 192.168.101.20 
    cache <local,src-direct>  iif eth1

And since his gateway is 192.168.101.10, an apparently local
IP address, that would probably explain the direct ARPS for
172.16.60.*.

Ah ok, thanks.

Pity that such a convoluted setup must be guessed instead of clearly
explained in the initial mail.

Prashant, this is netdev mailing list, with more than a thousand
subscribers, you should send mails with your detailed setup, this will
save time for everyone.

If not, people will flag you as "yet another lazy student trying to
understand how things work"

[Prashant] So is this the correct behavior? What my original intention
was to capture the packets from the 
interface acting as gateway. So with this behavior I will not be able
to capture the packets as 172.16.60.* are not 
present on the local machine, and it will not get any ARP response.
Consider the gateway on a different machine, in which case ARP request
will go for gateway IP, and local machine will 
get the ARP response and send the packets to gateway.

With the setup you gave, you said : 172.16.60.0/24 addresses are
directly reachable on eth1 device, with source address 192.168.101.20

Why should the kernel send an ARP request for 192.168.101.20, since its
one of its own IP address ?

For localy generated packets, why this machine would act as a 'gateway'
at all ? A gateway should forward packets, and first step would be to
_receive_ a packet, not sending one !