Cannot set Mellanox card to eth0 - does not show up in Interface List

[Kyle Boddy]

There are many posts saying to move the MAC address of an eth3 or similar to eth0 to get the DNS settings and features of eth0. For example, I have my Mellanox 10G-SFP card connected via DAC and working over 10G link, pulling an IP with no issue in unRAID. It's marked as eth4, and when it is connected while eth0 is not (1GBE), DNS issues occur, because unRAID does not allow eth4 to get the DNS server from the DHCP server.

 

No problem. Re-assign eth4's MAC to eth0 in the Interface List, I've read 20 posts about that subject. No one I've found has this issue, however: The Mellanox card simply does not exist in the Interface List despite being in the Network Settings.

 

Take a look at the attached image - it shows Interface eth0, eth1, eth2, and eth3, which are all onboard NICs and all 1GBE. eth4 is nowhere to be found despite being directly above that section, and working, to boot.

 

At the moment I either have to set the DNS server by manually editing the /etc/resolv.conf file every boot by adding nameservers 172.16.0.1 to the file, or keep eth0 connected with 1GBE along with eth4 connected via 10G-SFP, which causes Fix Common Issues plugin to yell about how two NICs should not be on the same subnet (which is true).

 

Furthermore, on booting with just eth4 plugged in, the unRAID console says that IPv4 address is not set, even though it definitely is by DHCP (another artifact of not having eth0 up, I'm sure).

 

Any resolution to get this to show up?

 

EDIT: The card is indeed a Ethernet controller: Mellanox Technologies MT27500 Family [ConnectX-3], but the mode is set properly to ethernet mode (eth), not infiniband mode (IB) as evidenced by the following command's output:

 

root@Tower:~# cat /sys/bus/pci/devices/0000\:07\:00.0/mlx4_port1
eth

 

tower-diagnostics-20220711-2134.zip

Edited July 12, 2022 by Kyle Boddy
diagnostics attached

[Kyle Boddy]

I suppose this topic is relevant:
 

However, multiple people still seem to have a related issue. Are there fixes in the RC branch available? Any idea what the known regression is? I cannot downgrade as I only started on the recent production version of unRAID, so this is a relatively important issue for me. Thanks.

 

[JorgeB]

There were some issues with Mellanox NICs but AFAIK all known issues were fixed on v6.10.3, for example this one, NIC is being detected, so not sure why it's not listed in the interface rules, maybe @bonienlcan't take a look to see if anything jumps out.

[Kyle Boddy]

Got it thanks. I submitted hardware information via the webGUI to unRAID if that helps.

[Kyle Boddy]

Wondering if there was an update here, thanks. My fallback plan will be to run 4x1GBE bonded, but would really prefer to use the 10G-SFP Mellanox card I have.

[JorgeB]

Just in case there's a regression issue with the number of NICs added to interface rules try disabling one or more of the onboard NICs to see if it makes any difference.

[Kyle Boddy]

I'm not sure I can on this machine - it's a Supermicro X8DT3-LN4F motherboard, and I looked at doing this a few days ago and saw no options to do that. I pulled the motherboard manual from the website and disabling onboard LAN ports is not featured in it.

 

https://www.supermicro.com/manuals/motherboard/5500/MNL-1062.pdf

[JorgeB]

You can always bind one or more to vfio-pci, that way they won't be seen by Unraid.

[Kyle Boddy]

I'm going to try the vfio-pci method next week when I have some downtime I can schedule, but wanted to bump this one more time to see if anyone had any ideas before I do so. Thanks.

[Kyle Boddy]

I successfully was able to bind eth2-4 to vfio-pci which removed them from unRAID. However, the Mellanox card still does not show up in the Interface Rules list, as shown by the images attached.

 

I have also attached anonymized Diagnostics. Can someone please look into this or advise me to file a bug ticket? I can try binding the last 1G ethernet port to vfio-pci and rebooting and seeing what happens, but I doubt that would fix the problem.

 

ifconfig on the command line clearly shows eth1 = Mellanox card, and when it isn't bridged to br1, it gets the same IP (static IP set from the router of 172.16.0.124).

Thanks for reviewing. This seems like a pretty serious bug.

 

@JorgeB @bonienl

tower-diagnostics-20220731-1837.zip

[Kyle Boddy]

Just in case anyone asks, no, there are no other MAC addresses in the pulldown menu under Interface eth0 in Interface Rules:

 

[JorgeB]

6 hours ago, Kyle Boddy said:

However, the Mellanox card still does not show up in the Interface Rules list, as shown by the images attached.

I expected as much, but it was worth trying.

 

6 hours ago, Kyle Boddy said:

This seems like a pretty serious bug.

It's not a general bug, I and many other users have the same Mellanox NIC and it's detected without any issues, the question is why it's not working for you, maybe @bonienlwill have and idea.

[Kyle Boddy]

Got it. Yeah, it's a weird one. Not sure how else to diagnose or try to fix on my end. I recently patched the BIOS, updating it to the latest version (2019), but had no effect.

Edited August 1, 2022 by Kyle Boddy

[wintervaler]

Not sure if you ended up resolving this, it's certainly been a few months, but I thought that on the offchance folks like me find this while searching for a solution, this post helped me:

 

Basically, had to nuke the existing network config files and then reboot. My Mellanox NIC then showed up in the "Interface Rules" list.

 

If you've made changes to the behavior of your network settings or the other NICs, you'll need to redo those with this fix. But in my case that wasn't super difficult.