[OLSR-users] Question about address assignment

Andreas Tønnesen (spam-protected)
Tue May 4 05:33:29 CEST 2004


Are you sure it won't work if you add a default route on all interfaces 
running OLSR? Didnt the routing code just check for a valid route
to the sender address?
But then again - you have probably tested that :)

But this whole thing sounds like a horrible hack... One should at least 
put a maximum metric on these routes.

As I am currently in the process of writing my thesis(ahh... the 
documentation - always the documentation) and preparing for a exam - 
sadly I don't hve to much time for hacking.
Anybody else got any ideas?

(please keep us updated on what solution you go for)

- Andreas

(spam-protected) wrote:
> Andreas,
> I found that by adding a default route pointing to the my interface I could
> get OLSR to work.
> On Friday I started looking at the AODV code from Uppsala University.  I
> recall
> that you mentioned AODV uses for its protocol messages.  I
> found
> this statement in the AODV implementation README file:
> -----
> On the ad hoc nodes it is also necessary to add a default route in the
> kernel routing table, pointing to the ad hoc interface. For example, if
> the wireless ad hoc interface is eth1:
>>route add default dev eth1
> Otherwise, it will not be possible to communicate with destinations
> outside the ad hoc prefix.
> -----
> Apparently they have experienced the same problem as me.
> My problem, however, is more complex.  My application requires that I run
> the
> MANET protocol on multiple interfaces.  The trick of adding a default
> route will only work for one interface.  The only solution I can see is to
> hack the kernel to allow Martian packets containing OLSR (or AODV) protocol
> packets.
> Any better ideas?
> Thanks,
> Richard Herron
> L-3 Communications
> -----Original Message-----
> From: Andreas Tønnesen [mailto:(spam-protected)]
> Sent: Thursday, April 29, 2004 10:22 AM
> To: (spam-protected)
> Cc: (spam-protected)
> Subject: Re: [OLSR-users] Question about address assignment
> Hi Richard,
> you are ever so right!
> On all my tests I have had routes to the networks trough other
> interfaces. That must have been why the packets have been sent
> up the layers. Sorry about that :)
> I'll look more into it tomorrow.
> - Andreas
> (spam-protected) wrote:
>>I have been looking into the Linux kernel source in
>>kernel/net/ipv4/route.c and I think I know what is happening.
>>Recall that I have two Linux systems with IP addresses from
>>different subnets connected via a switch.  I configured olsrd
>>to use as its broadcast address.
>>         Linux        Eth          Linux
>>        -------      switch       -------
>>        |     |      -----        |     |
>>        |  X  |______|   |________|  Y  |
>>        |     |      |   |        |     |
>>        -------      -----        -------
>>The code in route.c, the function ip_route_input_slow() checks
>>the destination address for 0xffffffff and jumps to the brd_input
>>label.  There it calls fib_validate_source() which returns an
>>error and the code jumps to the martian_source label.  The packet
>>is treated as a Martian packet and dropped because it did not come
>>from a source address with a known route.
>>Has anyone encountered this before?  Does anyone have suggestions
>>for dealing with this besides hacking the kernel?
>>Richard Herron
>>L-3 Communications
>>-----Original Message-----
>>From: Andreas Tønnesen [mailto:(spam-protected)]
>>Sent: Thursday, April 22, 2004 9:47 PM
>>To: (spam-protected)
>>Cc: (spam-protected)
>>Subject: Re: [OLSR-users] Question about address assignment
>>The UDP packets to should be passed up the stack as far 
>>as I know... You are sure there's no netfilter rule stopping the 
>>traffic(iptables -F)?
>>I will check this out when I get back to the lab!
>>- Andreas
>>(spam-protected) wrote:
>>>I have modified my OLSR config file to set IP4BROAD to 
>>> and I have attempted to run my systems with IP
>>>addresses from different subnets, but I am still having trouble.
>>>I have attached traces from both systems showing the OLSRD debug
>>>and some tcpdump printouts.  The tcpdump shows that the Hello
>>>messages are being send and seen at the Ethernet interfaces
>>>of both machines, but the messages don't seem to be getting to
>>>the OLSR protocol.
>>>For example:
>>>09:49:15.022709 >  udp 20 (DF) [tos
>>>09:49:21.000156 >  udp 20 (DF) [tos
>>>Could this be because the Linux network stack is filtering out
>>>messages that do not originate from the same subnet as the
>>>interface that receives them?
>>>I would appreciate any ideas you have on this.
>>>Thank you,
>>>Richard Herron
>>>L-3 Communications
>>>-----Original Message-----
>>>From: Andreas Tønnesen [mailto:(spam-protected)]
>>>Sent: Wednesday, April 14, 2004 12:56 AM
>>>To: (spam-protected)
>>>Cc: (spam-protected)
>>>Subject: Re: [OLSR-users] Question about address assignment
>>>You are not the first to ask this question ;-)
>>>OLSR uses the broadcastaddresses that interfaces are pre-configured 
>>>with. So for two nodes to "hear" eachother they need to use the same 
>>>broadcast address which usually means that they must be on the same 
>>>subnet. This is both good and bad. Good in the sense that it allows for 
>>>subnet design within the manet - bad in the way that it is not very 
>>>"ad-hocish". By "ad-hocish" I mean the idea of nodes roaming freely 
>>>entering and exitting MANETS dynamically.
>>>AODV specifies in the RFC that ALL control traffic should be broadcasted 
>>>on When using this broadcast address ALL nodes will 
>>>pick up the traffic. OLSR on the other hand does not specify this. So I 
>>>figured it was best leave both options to the user. As a result you can 
>>>set the IP4BROAD option in the configfile to either auto, which will 
>>>cause OLSRd to use the broadcastaddress fetched from the interface(s), 
>>>or you can set it to which causes all controltraffic to 
>>>be broadcasted on, meaning that the addresses of your 
>>>nodes does not have to reside within the same subnet.
>>>I hope this answered your question
>>>Andreas T
>>>(spam-protected) wrote:
>>>>The OLSR code seems to be well written and solid.  I have a question
>>>>about assignment of IP addresses and how addresses are handled by
>>>>the UniK OLSR daemon.
>>>>I am running olsrd on Linux machines using an Ethernet interface
>>>>because I currently have no wireless devices.
>>>>Section 11.1 of RFC 3626 says:
>>>>The nodes in the MANET network SHOULD be assigned addresses within a
>>>>defined address sequence, i.e., the nodes in the MANET SHOULD be
>>>>addressable through a network address and a netmask. 
>>>>I understand that this is done to enable us to advertise routes
>>>>from the OLSR MANET into other routing domains, but that it is not
>>>>a requirement.  
>>>>I also understand that generally in a MANET there are no requirements
>>>>as to how addresses are assigned and there may be nodes from different
>>>>"IP subnets" in the same MANET.
>>>>I have the following setup:
>>>>       Linux        Eth          Linux
>>>>      -------      switch       -------
>>>>      |     |      -----        |     |
>>>>      |  X  |______|   |________|  Y  |
>>>>      |     |      |   |        |     |
>>>>      -------      -----        -------
>>>>When workstations X and Y are on the same IP subnet (, 
>>>>OLSR works fine.  When they are on differing subnets (X =
>>>>Y = OLSR doesn't work at all.
>>>>Is this because of the way olsrd is implemented or is it due to the
>>>>way routing works on Linux, or is my understanding incorrect?
>>>>Thanks for any help you can give.
>>>>Richard Herron
>>>>L-3 Communications
>>>>olsr-users mailing list

Andreas Tønnesen((spam-protected))
UniK University Graduation Center
University of Oslo

More information about the Olsr-users mailing list