LinHES Forums
http://forum.linhes.org/

Persistent and mysterious network connectivity failure
http://forum.linhes.org/viewtopic.php?f=2&t=19448
Page 1 of 1

Author:  acronce [ Mon Jan 19, 2009 11:23 am ]
Post subject:  Persistent and mysterious network connectivity failure

Hi all,

I recently upgraded and migrated a backend-only system from R5E50 running on an ancient Athlon to F5.5 on a new MSI Wind Nettop barebones. After the migration, I ran the box on my desk for about a week. The system seemed stable, so I moved it to our server closet.

After the move, I found that I couldn't talk to the backend any more. I did the usual dance of swapping cables, switches, etc. No dice regardless of the equipment I tried, even though there was always a link light on both ends.

I moved the box back to my desk so that I could more conveniently hook up a monitor. I hooked the machine to the same switch and cable that worked before. At boot, there was a parade DHCP attempts, ultimately ending in a "No DHCPOFFERS received" error.

I couldn't ping anything on the network. So I tried giving the box a fixed IP address using netcardconfig. Still couldn't ping. I power cycled my router/firewall and switch, and rebooted the backend. Still nothing. I verified that the BIOS thought that the onboard ethernet was enabled.

After poking around on this site, I came to the conclusion that my onboard Ehernet had gone bad. Seemed strange because I had a link light, but I couldn't think of any other reason why it would suddenly fail just after a physical move. So I RMA'd the box.

Now I have the new box, but unfortunately I'm seeing the exact same problem. Since the problem is following the software installation, I'm suspecting some kind of odd software problem.

After poking around on the box, I found that the /etc/udev/rules.d/z25_persistent-net.rules file had an entry for the Ethernet MAC address of the previous machine. So I moved the file to a backup and rebooted so that the file would be recreated. That removed the old entry.

I then threw away /etc/sysconfig/network-scripts/ifcfg-eth1 file that I found. I have only one onboard Ethernet, so I suspect that this was created by the system at boot after I brought up the new unit.

I then edited the /etc/network/interfaces file to remove any references to eth1. I also ran netcardconfig afterwards and checked the file to make sure that it looked right.

But I still have no network connectivity. If I give myself a fixed IP address I can ping that address (as well as "localhost"), but nothing else.

When I call ifconfig I get back something that looks like what I've seen from other posters on this forum. That is, there's an entry for eth0 with a MAC address that matches my machine's MAC address. The only oddity is that the "RX packets" line shows a ton of "dropped" packets.

I should mention is that the onboard LAN is gigabit, but I'm using it on a 100baseT LAN. Suspecting a link level problem, like failed auto negotiation, I took a look with ethtool. Oddly enough, ethtool says that eth0's supported ports are "FIBER". I would have thought it would be "TP" for twisted pair. Further, it thinks that eth0 is only capable of 1000baseT/full duplex. That's clearly not going to work on my 100baseT switches. Lastly, it says there's no link detected, which I guess makes sense given the fact that the other settings are screwed up.

I tried changing the settings with the following command, but I got back an "Operation not supported" error:

Code:
ethtool -s eth0 port tp speed 100 autoneg 0ff

For grins I resetting things by disabling the LAN in the BIOS and rebooting. I verified that there was no eth0, then I renabled the LAN and rebooted. Same result. I have no network connectivity, tons of "dropped" packets from ifconfig, and the output from ethtool looks just as wrong.

I'm really out of ideas here. Thanks in advance for any suggestions.

Author:  jzigmyth [ Mon Jan 19, 2009 1:15 pm ]
Post subject: 

You could try booting various flavors of live CDs to see if they can identify and run your network chip correctly. That would rule in or out some weird legacy mis-configuration from the migration.

It's possible that your brand new onboard network chip is not supported by linux yet and is being mis-identified by the OS and therefore not working properly.

Author:  acronce [ Mon Jan 19, 2009 6:20 pm ]
Post subject: 

I think I've got a solution. Based on lots of googling, it looks like it's fairly common for the wrong driver to be loaded for the Realtek LAN RTL8111C hardware on my motherboard. Apparently it gets miss-assigned to r8169, when it should be r8168. I could tell that this was the case for my machine via a "ethtool -i eth0" command.

Most of the work arounds that I found included building the r8168 driver. But according to this command, the r8168 driver was included with my KnoppMyth install:

Code:
acronce# lsmod | grep r8168
r8168                  32656  0

So all I had to do was blacklist the r8169 driver like so (as root):

Code:
cd /etc/modprobe.d/
touch blacklist-network
nano blacklist-network

Note that while in "nano", I just added the following line to the blacklist-network file:

Code:
blacklist r8169

Once the above was in place, I performed these commands:

Code:
update-initramfs -u
reboot

After rebooting, I verified that the r8168 driver was in place via the following command:

Code:
acronce# ethtool -i eth0
driver: r8168
version: 8.006.00-NAPI
firmware-version:
bus-info: 0000:01:00.0

The LAN hardware seems to work now. Hopefully this will remain the case. I'm still kind of gun-shy because my original migration worked too until I physically moved the machine.

Author:  cantator [ Tue Feb 10, 2009 9:46 pm ]
Post subject: 

Sweetness! I was bothered by this pesky little annoyance. Thanks for your post and diligent research!

Author:  acronce [ Wed Feb 11, 2009 10:34 am ]
Post subject: 

cantator wrote:
Sweetness! I was bothered by this pesky little annoyance. Thanks for your post and diligent research!


You're welcome. Hope it works for you.

BTW, you might be able to skip the "update-initramfs" step. I think that that might only be necessary if you have to actually build the driver, which as I indicated should not be needed for the F5.5 distribution. I only did it because it was part of the steps I found for other distributions.

Good luck!

Author:  marc.aronson [ Sun Feb 15, 2009 10:35 pm ]
Post subject: 

Thanks for posting this -- solved the problem with the networking issue I was having with the MSI wind barebones I am working on.

Marc

Author:  evilmex [ Mon Mar 16, 2009 7:55 pm ]
Post subject: 

THANK YOU THANK YOU THANK YOU

I just got an ECS GF8200 and was running into this problem. (the lan port worked during install) after rebooting for the first time I could not get a dhcp response. Now I can continue with my upgrade after following your directions.

Author:  tjc [ Sat Mar 26, 2011 2:26 pm ]
Post subject: 

I just ran into this with 6.04 on the new box, and tried to do the same thing. Note that this solution no longer works in R6.

FWIW - This can be simplified to a single step, as root:
Code:
echo "blacklist r8169" >/etc/modprobe.d/blacklist-network

acronce wrote:
So all I had to do was blacklist the r8169 driver like so (as root):

Code:
cd /etc/modprobe.d/
touch blacklist-network
nano blacklist-network

Note that while in "nano", I just added the following line to the blacklist-network file:

Code:
blacklist r8169

Page 1 of 1 All times are UTC - 6 hours
Powered by phpBB® Forum Software © phpBB Group
http://www.phpbb.com/