OpenBMC DHCP Management

[Ed Tanous]

On Wed, Feb 2, 2022 at 11:07 AM Johnathan Mantey
<johnathanx.mantey at intel.com> wrote:
>
> OpenBMC DHCPv4 and DHCPv6 management
>
> I wish to solicit input concerning DHCPv6 handling in OpenBMC.
>
> Background:
>
> Some time ago I submitted and merged changes that allowed IPMI and
> Redfish to enable IPv6 Routing Advertisement. This permitted Stateful
> IPv6 address assignment to occur.
>
> Fast forward to today and the Intel QA group is identifying
> misbehavior in how DHCPv6 is controlled, and state reported. Most of
> the issues appear in the IPMI stack. I am also certain the Redfish
> support is lacking. The phoshpor-network layer does not properly
> control the systemd-network DHCPv6 configuration in a way that allows
> Stateless DHCPv6.
>
> Issues:
>
> OpenBMC uses the systemd-networkd networking stack. The
> phosphor-network code uses D-Bus to get configuration changes from
> user space. It updates the systemd.network configuration file to
> change the state of the network interface. Within the systemd.network
> configuaration file is an entry named "DHCP". This entry, at first
> glance, appears to control IPv4 and IPv6 DHCP. It is also the primary
> configuration entry for reporting DHCP state back to IPMI/Redfish
> about the current DHCP configuration in use by systemd-networkd.
>
> The problem is the "DHCP=" entry does not accurately reflect DHCPv6
> state. Even with "DHCP=v6" assigned it is possible DHCPv6 is not
> active. This entry can really only accurately describe the DHCPv4
> state.
>
> DHCPv6 state is accurately described by the combination of
> "IPv6AcceptRA" and the associated "DHCPv6Client" entries. The
> IPv6AcceptRA control is exposed via D-Bus, DHCPv6Client is not. The
> IPv6 state is reflected in these combinations:
> IPv6AcceptRA=true DHCPv6Client=true  : Stateful IPv6
> IPv6AcceptRA=true DHCPv6Client=false : Stateless IPv6
> IPv6AcceptRA=false                   : DHCPv6 disabled
>
> OpenBMC does not have complete control over these states, nor does it
> accurately report the state of DHCPv6. A first attempt to accurately
> report DHPCv6 state was made in this commit:
> https://gerrit.openbmc-project.xyz/c/openbmc/phosphor-networkd/+/48040
>
> The commit was not approved, which I agree is the correct
> response. The changes bandaged over a larger issue.
>
> Proposal:
> I have begun coding a change to address the issues described. My
> concern is the changes are going to impact four separate repositories,
> specifically:
>
> phosphor-dbus-interfaces
> phoshor-network
> bmcweb
> phosphor-ipmi-host
>
> As such I want to get some feedback on my plan before expending too
> much effort.
>
> My current plan:
> 1. Modify phosphor-dbus-interfaces, changing the DHCPConf entry in
>     EthernetInterface. I propose adding two more DHCPConf
>     enumearations:
>     name: both             # DHCPv4 enabled, DHCPv6 Stateful enabled
>     + name: v4v6stateless  # DHCPv4 enabled, DHCPv6 Stateless enabled
>     name: v6               # DHCPv4 disabled, DHCPv6 Stateful enabled
>     + name: v6stateless    # DHCPv4 disabled, DHCPv6 Stateless enabled
>     name: v4               # DHCPv4 enabled, DHCPv6 disabled
>     name: none             # DHCPv4 and v6 disabled
>
>     The goal is to permit phosphor-network to take the correct actions
>     needed to update the systemd.network file.
> 2. Deprecate the IPv6AcceptRA boolean. This, in my opinion, leaked too
>     much information about how systemd-networkd managed IPv6 dynamic
>     addressing.
> 3. Modify phosphor-network to accept the new DHCPConf entries, and
>     update the systemd.network configuration file based on the higher
>     level mode.
> 4. Update BMCWeb to use the new D-Bus configuration values

I don't understand enough about the above dbus/dhcp changes to have an
opinion on the interface changes, but for bmcweb so long as we
maintain the same Redfish-client API behavior (which I'm assuming it
will), I have no problem with bmcweb being updated to the new APIs
above.  Looking forward to seeing the patches.


> 5. Update phosphor-ipmi-host to use the new D-Bus configuration values
>
> Conclusion:
> The changes suggested abstract the reporting and the modification of
> DHCPv6, and eliminate implementation specific details from the IPMI
> and Redfish clients. They also fill in a gap with Stateless IPv6
> assignment.
>
> --
> Johnathan Mantey
> Senior Software Engineer
> *azad te**chnology partners*
> Contributing to Technology Innovation since 1992
> Phone: (503) 712-6764
> Email: johnathanx.mantey at intel.com <mailto:johnathanx.mantey at intel.com>
>