Re: No networking due to "net/mlx5e: Add support for devlink-port in non-representors mode"

[Qian Cai]

> On Mar 25, 2020, at 8:17 PM, Saeed Mahameed <saeedm@xxxxxxxxxxxx> wrote:
> 
> On Wed, 2020-03-25 at 16:25 -0400, Qian Cai wrote:
>>> On Mar 25, 2020, at 3:33 PM, Moshe Shemesh <moshe@xxxxxxxxxxxx>
>>> wrote:
>>> 
>>> 
>>> On 3/25/2020 6:01 AM, Qian Cai wrote:
>>>> Reverted the linux-next commit c6acd629ee (ânet/mlx5e: Add
>>>> support for devlink-port in non-representors modeâ)
>>>> and its dependencies,
>>>> 
>>>> 162add8cbae4 (ânet/mlx5e: Use devlink virtual flavour for VF
>>>> devlink portâ)
>>>> 31e87b39ba9d (ânet/mlx5e: Fix devlink port register sequenceâ)
>>>> 
>>>> on the top of next-20200324 allowed NICs to obtain an IPv4
>>>> address from DHCP again.
>>> 
>>> These patches should not interfere DHCP.
>>> 
>>> You might have dependencies on interface name which was changed by
>>> this patch, please check.
>> 
>> Yes,
>> 
>> Before,
>> [  238.225149][ T2021] mlx5_core 0000:0b:00.1 enp11s0f1: renamed from
>> eth1
>> [  238.511324][ T2035] mlx5_core 0000:0b:00.0 enp11s0f0: renamed from
>> eth0
>> 
>> Now,
>> [  234.448420][ T2013] mlx5_core 0000:0b:00.1 enp11s0f1np1: renamed
>> from eth1
>> [  234.664236][ T2042] mlx5_core 0000:0b:00.0 enp11s0f0np0: renamed
>> from eth0
>> 
> 
> it is not a good idea to use the interface name as a unique identifier,
> this is not resilient for kernel updates or configuration updates, e.g
> installing an extra card
> 
> Just use the HW mac address as a unique identifier in the network 
> scripts:
> 
> from [1]:
> HWADDR=MAC-address
> where MAC-address is the hardware address of the Ethernet device in the
> form AA:BB:CC:DD:EE:FF. This directive must be used in machines
> containing more than one NIC to ensure that the interfaces are assigned
> the correct device names regardless of the configured load order for
> each NIC's module. 
> 
> [1] 
> https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/6/html/deployment_guide/s1-networkscripts-interfaces

Is it really necessary to cause a device naming change?

The installer/NetworkManager here (annaconda) will create those files automatically.
They donât usually use hard-coded HWADDR. Ironically, it is RHEL-8.1.

# cat /etc/sysconfig/network-scripts/ifcfg-enp11s0f0
# Generated by dracut initrd
NAME="enp11s0f0"
DEVICE="enp11s0f0"
ONBOOT=yes
NETBOOT=yes
UUID="be2237f3-e40b-4404-be6c-bcb484dbebf6"
IPV6INIT=yes
BOOTPROTO=dhcp
TYPE=Ethernet

# cat /etc/sysconfig/network-scripts/ifcfg-enp11s0f1
TYPE=Ethernet
PROXY_METHOD=none
BROWSER_ONLY=no
BOOTPROTO=dhcp
DEFROUTE=yes
IPV4_FAILURE_FATAL=no
IPV6INIT=yes
IPV6_AUTOCONF=yes
IPV6_DEFROUTE=yes
IPV6_FAILURE_FATAL=no
IPV6_ADDR_GEN_MODE=stable-privacy
NAME=enp11s0f1
UUID=400773fc-5436-4a85-9cee-a8e18cee9230
DEVICE=enp11s0f1
ONBOOT=no

Imagining a kernel update from enterprise distros contain those patches which
would break all those existing setup. It is trivial for me to fix myself, but it could
be a different story for those customers with thousand of machines need to
change.