LARTC - May 2001 - Strange behaviour of 'ip' with gateways inside non-main tables

Hi, Folks!


	This post deals with ip:
		
~# ip -V
ip utility, iproute2-ss001007


	I want to do this: The box (A) I try to configure has three
network interfaces (eth1, eth2 and eth3) connected through a switch to
a router (let''s say 192.168.100.20). Depending on the source ip
(192.168.100.11 or 192.168.100.12 or 192.168.100.13) I want to send to
the router the packets through one and only one interface, for
instance:

     packets from 192.168.100.11 go through eth1
     packets from 192.168.100.12 go through eth2
     packets from 192.168.100.13 go through eth3

     
<parenthesis> 

     Before attempting it, let''s do a similar thing but sending the
packets to a directly connected host (for example, B). This can easily
be accomplished with three rules and three tables, one rule and one
table for every source address. I have been using this for a long time
and it works fine:

-------                                      ---------
      |eth1-192.168.100.11   ---------       |       |
      |----------------------|       |       |       |
BOX A |                      |       |       | BOX B |
      |eth2-192.168.100.12   |       |       |       |
      |----------------------|switch |-------| 192.168.100.2
      |                      |       |       |       |
      |eth3-192.168.100.13   |       |       |       |
      |----------------------|       |       |       |
-------                      ---------       ---------


     # ip rule add from 192.168.100.11 lookup 11
     # ip rule add from 192.168.100.12 lookup 12
     # ip rule add from 192.168.100.13 lookup 13

     # ip ro a 192.168.100.2 dev eth1 t 11
     # ip ro a 192.168.100.2 dev eth2 t 12
     # ip ro a 192.168.100.2 dev eth3 t 13  

</parenthesis>
  

   The problem arises when I insert a router between my box A and my
   box B, then I have to use ''via''...


------                                      ----------
     |eth1-192.168.100.11   ---------       |        |
     |----------------------|       |       |        |
BOX A|                      |       |       | ROUTER |
     |eth2-192.168.100.12   |       |       |        |
     |----------------------|switch |-------|192.168.100.20
     |                      |       |       |        |
     |eth3-192.168.100.13   |       |       |        |       -------
     |----------------------|       |       |        |-------|     |
------                      ---------       ----------       |     |
                                                             |BOX B|
                                                             |     |
                                                             |     |
                                                             -------
                                                       192.168.100.2

      I try:

     # ip ro a 192.168.100.20 dev eth1 t 11
     # ip ro a 192.168.100.20 dev eth2 t 12
     # ip ro a 192.168.100.20 dev eth3 t 13  
    
    No problem, but...

    # ip ro a 192.168.100.2 via 192.168.100.20 dev eth1 table 11
    RTNETLINK answers: Network is unreachable

    It seemed ip requires a route to 192.168.100.20 _____in the
main table_____. So, before doing this, I try to deceive ''ip'',
making
a fake route that I will delete soon:

    # ip ro a 192.168.100.20 dev eth1
    then,
    # ip ro a 192.168.100.2 via 192.168.100.20 dev eth1 t 11
    works fine.

    And so on with the other two interfaces:

    # ip ro change 192.168.100.20 dev eth2
    # ip ro a 192.168.100.2 via 192.168.100.20 dev eth2 t 12

    # ip ro change 192.168.100.20 dev eth3
    # ip ro a 192.168.100.2 via 192.168.100.20 dev eth2 t 13

    Finally, I delete the fake route from the main table,
    # ip ro del 192.168.100.20


    and, fortunately, get the desired state:

~# ip ru s
0:      from all lookup local
32763:  from 192.168.100.13 lookup 13
32764:  from 192.168.100.12 lookup 12
32765:  from 192.168.100.11 lookup 11
32766:  from all lookup main
32767:  from all lookup default

~# ip r s t 11
192.168.100.20 dev eth1  scope link
default via 192.168.100.20 dev eth1

~# ip r s t 12
192.168.100.20 dev eth2  scope link
default via 192.168.100.20 dev eth2

~# ip r s t 13
192.168.100.20 dev eth3  scope link
default via 192.168.100.20 dev eth3
~# ip r s

~# ip r get 192.168.100.20
RTNETLINK answers: Network is unreachable

~# ip r get 192.168.100.20 from 192.168.100.11
192.168.100.20 from 192.168.100.11 dev eth1
    cache  mtu 1500

~# ip r get 192.168.100.20 from 192.168.100.12
192.168.100.20 from 192.168.100.12 dev eth2
    cache  mtu 1500

~# ip r get 192.168.100.20 from 192.168.100.13
192.168.100.20 from 192.168.100.13 dev eth3
    cache  mtu 1500


    So, it works, but I can only configure my routing tables deceiving
''ip''
with the false routes in the main table, that finally I can delete without
any trouble. 


    Conclussion: It works but the way it is configured is pretty ugly.


    I have tried doing ''ip r flush cache'' between one
''route add'' and
the following ''route add'', also I have tried adding a 
____''from
192.168.100.x''___  and a  ___''t 1x''___  qualifier to
the ''ip route add
.... via ...'' commands; none of these solved the problem.




   *** Q: I wonder if this is a bug of ''ip'' (it seems to check
routes
to desired gateways looking up the main table only and ignoring rules
and source qualifiers) or if I am doing something wrong.


   Thanks in advance for any feedback.
   Read you soon!