Hi all,
I'm experiencing by far the strangest problem I've ever encountered in
Linux.. I'm kind of left with no where to turn for ideas on what might be
happening, I thought maybe there is a slight change it is an actual bug.
First, the system.. A single process PIII-450 (katmai).
Linux 2.2.19 i686
The machine has 4 Network interfaces. a Dual intel etherexpress pro.. an
onboard 3c905, and a pci netgear card. (all 10/100)
The onboard, and dual intel work fine.. using the correct
drivers.
eth0: 3Com 3c905B Cyclone 100baseTx at 0xcc00
eth1: Intel PCI EtherExpress Pro100 82557
eth2: Intel PCI EtherExpress Pro100 82557
eth3: Lite-On 82c168 PNIC rev 32 at 0xc800,
The netgear is using the tulip driver.. (eth3)
Here is the strange part.. The machine has two default routes set..
0.0.0.0 4.10.10.1 0.0.0.0 UG 0 0 0 eth3
0.0.0.0 64.10.10.1 0.0.0.0 UG 0 0 0 eth1
This problem is random.. Usually every 30 minutes to 4 hours.. The route
on eth3 just stops working.. You can still ping 4.10.10.1.. but you can't
get to this machine via that interface, and it can't route through it.
if you issue a:
route delete default gw 4.10.10.1 ; route add default gw 4.10.10.1
It all of a sudden works again.. The route table is identical to how it
was prior. Arp tables are the same before and after.. No errors in any
logs, or on screen.. Just random.
I'm at a loss.. Any thoughts?
---
Erik Parker
---
In article <[email protected]> you wrote:
> I'm at a loss.. Any thoughts?
Just dont use two default routes.
Greetings
Bernd
After doing the same research I've been doing the past week or so.. I'm
coming back to the conclusion that you "can't" have two default routes.
Which doesn't explain why it works at all.
We need this machine to answer on two ip addresses on different networks
to the internet. One that comes in one router, on one t1.. and one that
comes in on another router, on a different t1. And it needs to respond
back out of the same interface.
But it does just that.. I can send mail to mx.domain.com and
mx3.domain.com and it hits whichever interface.. I can send traffic out
both of them. One just dies at random.
I don't even really need two DEFAULTS.. I just need the machine to be seen
on the internet, by two totally different IP's on seperate routes.
Is it a bug with the nic card, or is this impossible? If it's impossible,
how do people multihome linux boxes, without using bgp?
On Wed, 25 Jul 2001, Bernd Eckenfels wrote:
> In article <[email protected]> you wrote:
> > I'm at a loss.. Any thoughts?
>
> Just dont use two default routes.
>
> Greetings
> Bernd
---
Erik Parker
---
In article <[email protected]> you wrote:
> After doing the same research I've been doing the past week or so.. I'm
> coming back to the conclusion that you "can't" have two default routes.
> Which doesn't explain why it works at all.
You can have two default routes. In one case you have different metrics, then
only the one with the smaller metric is used. This is good for situations
where the better route may get dropped (link down).
Yu can have also 2 default routes to the same destination, it is called equal
cost multi route and a kernel option, used for load balancing.
any other configurations are great problem.
It is related to the route cache, why it did worked.
route -C
Greetings
Bernd
Hi,
I have following situation:
Server on 2.4.7 ( with grs pach applied )
Client unit - bootable from floppy ( 2.4.1 kernel ) and it use NFS root from
server.
The problem is:
(...)
Mounting rootfs read only ( nfs root ) succesfull
Freeing unused kernel memory 198kb
NFS server 192.186.1.45 not responding, still trying.
On server side log contains a message that /usr/src/install was mounted
succesfuly by client machine.
No other messages about errors.
And normal mounting of this NFS export from any unit on the network ( mount
-t nfs 192.168.1.45:/usr/src/install /mnt ) works fine.
The problem apeear only when client unit started from floppy try to mount NFS
root.
The same problem is with 2.4.7-pre6
I haven't tested it on pre4 and pre5.
On 2.4.7-pre3 and all previous kernels it works fine - client boot from NFS
root as it should.
All test was made on the same machines with the same system configurations -
only kernels were diffrent.
/etc/export :
/usr/src/install *(rw,no_root_squash)
Slawomir Pol