[LU-16908] lnet module ip2nets parameter not acting as documented

• Type: Bug
• Resolution: Unresolved
• Priority: Minor
• Labels: lnet patch
• Severity: 3
• Assignee: WC Triage
• Reporter: Josh Samuelson

Description

The ip2nets parameter, when matching a bound IP address on an interface of the node, may activate a different LND interface with the wrong IP address.

This may be related to LUs:
https://jira.whamcloud.com/browse/LU-7563
https://jira.whamcloud.com/browse/LU-11859

The Note/example on page 69 of https://build.whamcloud.com/job/lustre-manual/lastSuccessfulBuild/artifact/lustre_manual.pdf doesn't match what actually happens.

 

If an interface is explicitly specified as well as a pattern, the interface matched using the IP pattern will be sanitized against the explicitly-defined interface.

For example, tcp(eth0) 192.168.*.3 and there exists in the system eth0 == 192.158.19.3 and eth1 == 192.168.3.3, then the configuration will fail, because the pattern contradicts the interface specified.

A clear warning will be displayed if inconsistent configuration is encountered.

 

Example showing the issue with stock lnet.ko, working through the above steps:

[[e-mail] ~]# ip netns add test
[[e-mail] ~]# ip -n test link add eth0 type dummy
[[e-mail] ~]# ip -n test link add eth1 type dummy
[[e-mail] ~]# ip -n test addr add 127.0.0.1/8 brd + dev lo
[[e-mail] ~]# ip -n test addr add 192.158.19.3/24 brd + dev eth0
[[e-mail] ~]# ip -n test addr add 192.168.3.3/24 brd + dev eth1
[[e-mail] ~]# ip -n test link set up dev lo
[[e-mail] ~]# ip -n test link set up dev eth0
[[e-mail] ~]# ip -n test link set up dev eth1

[[e-mail] ~]# ip netns exec test bash

[[e-mail] ~]# ip -4 addr list
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
    inet 127.0.0.1/8 brd 127.255.255.255 scope host lo
       valid_lft forever preferred_lft forever
2: eth0: <BROADCAST,NOARP,UP,LOWER_UP> mtu 1500 qdisc noqueue state UNKNOWN group default qlen 1000
    inet 192.158.19.3/24 brd 192.158.19.255 scope global eth0
       valid_lft forever preferred_lft forever
3: eth1: <BROADCAST,NOARP,UP,LOWER_UP> mtu 1500 qdisc noqueue state UNKNOWN group default qlen 1000
    inet 192.168.3.3/24 brd 192.168.3.255 scope global eth1
       valid_lft forever preferred_lft forever

[[e-mail] ~]# modinfo -F version lnet
0.7.0

[[e-mail] ~]# modinfo -F version lustre
2.15.2

[[e-mail] ~]# cat /etc/modprobe.d/lnet.conf
options lnet "ip2nets=tcp(eth0) 192.168.*.3"

[[e-mail] ~]# modprobe lnet

[[e-mail] ~]# cat /sys/module/lnet/parameters/ip2nets
tcp(eth0) 192.168.*.3

[[e-mail] ~]# lnetctl net show
show:
    - net:
          errno: -100
          descr: "cannot get networks: Network is down"

[[e-mail] ~]# lnetctl lnet configure --all

[[e-mail] ~]# lnetctl net show
net:
    - net type: lo
      local NI(s):
        - nid: 0@lo
          status: up
    - net type: tcp
      local NI(s):
        - nid: 192.158.19.3@tcp
          status: up
          interfaces:
              0: eth0

[[e-mail] ~]# dmesg
LNet: HW NUMA nodes: 2, HW CPU cores: 56, npartitions: 2
alg: No test for adler32 (adler32-zlib)
Key type ._llcrypt registered
Key type .llcrypt registered
LNet: Added LNI 192.158.19.3@tcp [8/256/0/180]
LNet: Accept secure, port 988

 

Same steps with patched lnet.ko (without IP stack setup steps):

[[e-mail] ~]# ip netns exec test bash

[[e-mail] ~]# ip -4 addr list
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
    inet 127.0.0.1/8 brd 127.255.255.255 scope host lo
       valid_lft forever preferred_lft forever
2: eth0: <BROADCAST,NOARP,UP,LOWER_UP> mtu 1500 qdisc noqueue state UNKNOWN group default qlen 1000
    inet 192.158.19.3/24 brd 192.158.19.255 scope global eth0
       valid_lft forever preferred_lft forever
3: eth1: <BROADCAST,NOARP,UP,LOWER_UP> mtu 1500 qdisc noqueue state UNKNOWN group default qlen 1000
    inet 192.168.3.3/24 brd 192.168.3.255 scope global eth1
       valid_lft forever preferred_lft forever

[[e-mail] ~]# cat /etc/modprobe.d/lnet.conf
options lnet "ip2nets=tcp(eth0) 192.168.*.3"

[[e-mail] ~]# modprobe lnet

[[e-mail] ~]# cat /sys/module/lnet/parameters/ip2nets
tcp(eth0) 192.168.*.3

[[e-mail] ~]# lnetctl net show
show:
    - net:
          errno: -100
          descr: "cannot get networks: Network is down"

[[e-mail] ~]# lnetctl lnet configure --all
configure:
    - lnet:
          errno: -22
          descr: "LNet configure error: Invalid argument"

[[e-mail] ~]# dmesg
LNet: HW NUMA nodes: 2, HW CPU cores: 56, npartitions: 2
alg: No test for adler32 (adler32-zlib)
Key type ._llcrypt registered
Key type .llcrypt registered
LNetError: 11a-a: ip2nets does not match any local IP interfaces
LNetError: 47753:0:(config.c:574:lnet_parse_networks()) networks string is undefined

[[e-mail] ~]# modprobe -r lnet ; dmesg -C

[[e-mail] ~]# modprobe lnet "ip2nets=tcp(eth0,eth10,eth1) 192.168.*.3"

[[e-mail] ~]# cat /sys/module/lnet/parameters/ip2nets
tcp(eth0,eth10,eth1) 192.168.*.3

[[e-mail] ~]# lnetctl net show
show:
    - net:
          errno: -100
          descr: "cannot get networks: Network is down"

[[e-mail] ~]# lnetctl lnet configure --all

[[e-mail] ~]# lnetctl net show
net:
    - net type: lo
      local NI(s):
        - nid: 0@lo
          status: up
    - net type: tcp
      local NI(s):
        - nid: 192.168.3.3@tcp
          status: up
          interfaces:
              0: eth1

[[e-mail] ~]# dmesg
LNet: HW NUMA nodes: 2, HW CPU cores: 56, npartitions: 2
alg: No test for adler32 (adler32-zlib)
Key type ._llcrypt registered
Key type .llcrypt registered
LNet: ip2nets matched tcp(eth1) for 192.168.*.3
LNet: Added LNI 192.168.3.3@tcp [8/256/0/180]
LNet: Accept secure, port 988

 

I wrote the patch against the 2.15.2 tag in hopes of making the code follow my interpretation of the documentation for the module parameter. Hopefully the area where I added the check is the correct spot.