Re: [PATCH net] team: prevent ipv6 link local address on port devices

[Jiri Pirko <jiri@resnulli.us>]

Thu, Dec 08, 2022 at 06:07:17PM CET, lucien.xin@gmail.com wrote:
>On Thu, Dec 8, 2022 at 6:19 AM Jiri Pirko <jiri@resnulli.us> wrote:
>>
>> Thu, Dec 08, 2022 at 12:35:48AM CET, lucien.xin@gmail.com wrote:
>> >On Wed, Dec 7, 2022 at 8:31 AM Jiri Pirko <jiri@resnulli.us> wrote:
>> >>
>> >> Tue, Dec 06, 2022 at 10:52:33PM CET, lucien.xin@gmail.com wrote:
>> >> >On Tue, Dec 6, 2022 at 8:32 AM Xin Long <lucien.xin@gmail.com> wrote:
>> >> >>
>> >> >> On Tue, Dec 6, 2022 at 3:05 AM Jiri Pirko <jiri@resnulli.us> wrote:
>> >> >> >
>> >> >> > Mon, Dec 05, 2022 at 06:46:05PM CET, lucien.xin@gmail.com wrote:
>> >> >> > >The similar fix from commit c2edacf80e15 ("bonding / ipv6: no addrconf
>> >> >> > >for slaves separately from master") is also needed in Team. Otherwise,
>> >> >> > >DAD and RS packets to be sent from the slaves in turn can confuse the
>> >> >> > >switches and cause them to incorrectly update their forwarding tables
>> >> >> > >as Liang noticed in the test with activebackup mode.
>> >> >> > >
>> >> >> > >Note that the patch also sets IFF_MASTER flag for Team dev accordingly
>> >> >> > >while IFF_SLAVE flag is set for port devs. Although IFF_MASTER flag is
>> >> >> > >not really used in Team, it's good to show in 'ip link':
>> >> >> > >
>> >> >> > >  eth1: <BROADCAST,MULTICAST,SLAVE,UP,LOWER_UP>
>> >> >> > >  team0: <BROADCAST,MULTICAST,MASTER,UP,LOWER_UP>
>> >> >> > >
>> >> >> > >Fixes: 3d249d4ca7d0 ("net: introduce ethernet teaming device")
>> >> >> > >Reported-by: LiLiang <liali@redhat.com>
>> >> >> > >Signed-off-by: Xin Long <lucien.xin@gmail.com>
>> >> >> >
>> >> >> > Nack. Please don't do this. IFF_MASTER and IFF_SLAVE are historical
>> >> >> > flags used by bonding and eql. Should not be used for other devices.
>> >> >> I see. I was wondering why it was not used in Team at the beginning. :)
>> >> >>
>> >> >> >
>> >> >> > addrconf_addr_gen() should not check IFF_SLAVE. It should use:
>> >> >> > netif_is_lag_port() and netif_is_failover_slave() helpers.
>> >> >Hi Jiri,
>> >> >
>> >> >Sorry, it seems not to work with this.
>> >> >
>> >> >As addrconf_addr_gen() is also called in NETDEV_UP event where
>> >> >IFF_TEAM_PORT and IFF_BONDING haven't yet been set before
>> >> >dev_open() when adding the port.
>> >> >
>> >> >If we move IFF_TEAM_PORT setting ahead of dev_open(), it will revert
>> >> >the fix in:
>> >> >
>> >> >commit d7d3c05135f37d8fdf73f9966d27155cada36e56
>> >> >Author: Jiri Pirko <jiri@resnulli.us>
>> >> >Date:   Mon Aug 25 21:38:27 2014 +0200
>> >> >
>> >> >    team: set IFF_TEAM_PORT priv_flag after rx_handler is registered
>> >> >
>> >> >Can we keep IFF_SLAVE here only for no ipv6 addrconf?
>> >>
>> >> So, shouldn't it be rather a new flag specifically for this purpose?
>> >Maybe IFF_NO_ADDRCONF in dev->priv_flags?
>>
>> Sounds fine to me.
>BTW, IFF_LIVE_RENAME_OK flag was just deleted in net-next.git by:
>
>commit bd039b5ea2a91ea707ee8539df26456bd5be80af
>Author: Andy Ren <andy.ren@getcruise.com>
>Date:   Mon Nov 7 09:42:42 2022 -0800
>
>    net/core: Allow live renaming when an interface is up
>
>do you think it is okay to use that vacance and define:
>
>IFF_NO_ADDRCONF = BIT_ULL(30)
>
>in netdev_priv_flags ?

It's a private define, no UAPI, I don't see why not. Let's make the
backporter live a bit harder :)

>
>Thanks.
>
>>
>>
>> >
>> >I will give it a try.
>> >
>> >Thanks.