[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference turris::digital_unix

Title:DIGITAL UNIX(FORMERLY KNOWN AS DEC OSF/1)
Notice:Welcome to the Digital UNIX Conference
Moderator:SMURF::DENHAM
Created:Thu Mar 16 1995
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:10068
Total number of notes:35879

9831.0. "Wrong IP source address" by COPCLU::PALMANN () Thu May 15 1997 10:29

We are runnig UNIX 3.2C, and belive to have found a bug in the IP behavior.

The config:

  UNIX 3.2 C  ( with 2 addresses on same interfase )        
  10.8.41.12          
  194.117.116.105     
        |             
 /---------------------------------/
                    |
               194.117.116.104
               Wanrouter_1
                    |
              (x25 connection)
                    |
               The target node
               147.204.2.5   


The UNIX machine use to only have address 10.8.41.12
And was able to access the 'target node'.
Now the have to use an new address, to access the 'The target node' and 
has added 194.117.116.105, as an alias address on the UNIX machine.
The Unix machine has a static route to 147.204.2.5, and can ping the
'The target node' 147.204.2.5, via Wanrouter_1.

Problem:
When doing a ftp or a telnet connection against the 'The target node', the 
UNIX machine uses 10.8.41.12 as its source IP address.
( this is NOT the case doing a ping ? )
Doiing a telnet, the UNIX machine sends a packet straight to Wanrouter_1 
    (194.117.116.104),
with an IP dest at 147.204.2.5, and a source address as 10.8.41.12.

When this packet returnes to Wanrouter_1, ( who does no know net 10 ) the 
packet is lost.

In simple: 
From the unix node, and
ping'ing.............147.204.2.5, the source address is 194.117.116.104.
ftp or telnet'ing....147.204.2.5, the source address is 10.8.41.12

Why ??

I belive that there is an error in a IP node, that sends straight to a router
in a different IP net, then the one it configured in it self.

Bug, feature, surgestions to a work arround ?

 			Jan


T.RTitleUserPersonal
Name
DateLines
9831.1swap the ip addressesSMURF::DUSTINThu May 15 1997 12:0319
    The outgoing connection simply uses the first IP address
    configured on that network interface (which is called the
    primary IP address).
    
    The alias addresses are not used, and no attempt is made to
    match the destination subnet with one of these aliases to
    avoid routing lookup errors on the return path.
    
    To workaround this problem, you can swap your IP addresses,
    so the alias is the primary and the primary (10.x.x.x) is
    an alias.  Then outgoing connections will always use the
    194.x.x.x address.
    
    Of course, if anyone is relying on seeing 10.x.x.x address,
    then that will suffer the same problems you've seen with using
    the alias.
    
    John
    
9831.2COPCLU::PALMANNFri May 16 1997 03:4822
Hi John,

Thanks for your answer.
I already surgersted the 'swap addresses' to the customer, but they where
affraid that the same problem would appear on the internel net.

I must say, that I still see the behavior of the UNIX machine as illegal
to how a IP node should behave.

>    The outgoing connection simply uses the first IP address
>    configured on that network interface (which is called the
>    primary IP address).

I think that two nodes on the same IP net, should talk to each other using 
this common IP net.

Anyway,
I will try to se if we can implement a 2 interfase solution.
This should work..... Don't you think ?

			Jan 

9831.3not a IP bugSMURF::AJAYFri May 16 1997 16:0517
    When you use alias for an Interface, the system can use these address
    interchangabally. When two aliases are in different (sub)networks,
    interface route is created for both the (sub)network on the UNIX
    m/c. In your example you will see a route to network 10 and network 194
    (depending on your netmask). When doing that and running routing daemon
    on UNIX m/c, it will advertise route for both network 10 and network
    194 so the normal (non-x.25) router would pick-up and will know both
    network 10 and network 194 exist on the same wire.
    
    However, i'm not sure about Wanrouter, but you need to configure in
    Wanrouter that network 10 and network 194 are on the same wire (that
    generally done uisng IGP protocol RIP via routed/gated).
    
    It's not a IP bug, but artifact of X.25 (NBMA networks including ATM)'s
    -- lack of IP routing support.
    
    --Ajay Kachrani
9831.4COPCLU::PALMANNTue May 20 1997 06:4024
So a nonrouting-node with 2 addresses on a LAN.

  10.8.41.12
  194.117.116.105

Who is to talk via an other node on the same LAN

  194.117.116.104
  
Can use the source address 10.8.41.12, and send the packet to 194.117.116.104 ?

Well, if it is no bug, it's a 'thin' implementation. ( I think )

In my opinion, if a node in 'IP-net YYY' 
talks to a node in ,IP-net YYY', 
it should use source address 'IP-net YYY'.

			Jan




 
9831.5it works in the same subnetSMURF::DUSTINTue May 20 1997 10:5112
       > Well, if it is no bug, it's a 'thin' implementation. ( I think )
    
    IP aliases are only officially supported within the same subnet,
    which is why using another subnet results in subtle inconsistencies
    which appear broken, depending on your point of view.
    
    If we officially supported aliases in different subnets, this is
    one of the things that would probably be changed, ie. find the best
    match for the outgoing connections.
    
    John
    
9831.6COPCLU::PALMANNWed May 21 1997 08:063
    OK, Thank you both for the help.
    
    			Jan