[LU-11949] Able to define duplicate nids via lnetctl

• Status: Open
• Project: Lustre
• Affects Version/s: Lustre 2.12.0
• Type: Bug
• Priority: Minor
• Reporter: Chris Horn
• Assignee: WC Triage
• Resolution: Unresolved
• Severity: 3

Description

LNet is initially configured via the ip2nets module parameter.

nid00127:~ # cat /etc/modprobe.d/cray-lnet.conf
options lnet ip2nets="gni99 10.129.*.*; gni4 10.129.*.*"
nid00127:~ # lctl list_nids
127@gni99
127@gni4

I'm able to add a duplicate 127@gni4 address with lnetctl:

nid00127:~ # lnetctl net add --ip2net "gni4(ipogif0) 10.129.*.*"
nid00127:~ # lctl list_nids
127@gni99
127@gni4
127@gni4

Can/should we throw an error in this case?

If I add them both with lnetctl then I get an error on the second one:

nid00127:~ # lnetctl net del --net gni4 --if ipogif0
nid00127:~ # lctl list_nids
127@gni99
127@gni4
nid00127:~ # lnetctl net del --net gni4 --if ipogif0
nid00127:~ # lctl list_nids
127@gni99
nid00127:~ # lnetctl net add --ip2net "gni4(ipogif0) 10.129.*.*"
nid00127:~ # lctl list_nids
127@gni99
127@gni4
nid00127:~ # lnetctl net add --ip2net "gni4(ipogif0) 10.129.*.*"
add:
    - net:
          errno: -22
          descr: "cannot add network: Invalid argument"
nid00127:~ #

Comments

[Amir Shehata (Inactive)]

at first glance when you run the command through lnetctl you're specifying the interface name, while in the module parameter case you're not. I suspect this causes the check in the code not to evaluate the second add as a duplicate.

To confirm, you can you make the following change in the module parameters:

 options lnet ip2nets="gni99 10.129.*.*; gni4(ipogif0) 10.129.*.*"

I think this might be clearer if you use:

lnetctl net show 

to dump the configured networks instead of list_nids