Re: [PATCH net] ipv6: don't auto-add link-local address to lag ports

[Jarod Wilson]

On Thu, Mar 19, 2020 at 1:06 PM Eric Dumazet <eric.dumazet@xxxxxxxxx> wrote:
>
> On 3/19/20 9:42 AM, Jarod Wilson wrote:
>
> > Interesting. We'll keep digging over here, but that's definitely not
> > working for this particular use case with OVS for whatever reason.
>
> I did a quick test and confirmed that my bonding slaves do not have link-local addresses,
> without anything done to prevent them to appear.
>
> You might add a selftest, if you ever find what is the trigger :)

Okay, have a basic reproducer, courtesy of Marcelo:

# ip link add name bond0 type bond
# ip link set dev ens2f0np0 master bond0
# ip link set dev ens2f1np2 master bond0
# ip link set dev bond0 up
# ip a s
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN
group default qlen 1000
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever
    inet6 ::1/128 scope host
       valid_lft forever preferred_lft forever
2: ens2f0np0: <BROADCAST,MULTICAST,SLAVE,UP,LOWER_UP> mtu 1500 qdisc
mq master bond0 state UP group default qlen 1000
    link/ether 00:0f:53:2f:ea:40 brd ff:ff:ff:ff:ff:ff
5: ens2f1np2: <NO-CARRIER,BROADCAST,MULTICAST,SLAVE,UP> mtu 1500 qdisc
mq master bond0 state DOWN group default qlen 1000
    link/ether 00:0f:53:2f:ea:40 brd ff:ff:ff:ff:ff:ff
11: bond0: <BROADCAST,MULTICAST,MASTER,UP,LOWER_UP> mtu 1500 qdisc
noqueue state UP group default qlen 1000
    link/ether 00:0f:53:2f:ea:40 brd ff:ff:ff:ff:ff:ff
    inet6 fe80::20f:53ff:fe2f:ea40/64 scope link
       valid_lft forever preferred_lft forever

(above trimmed to relevant entries, obviously)

# sysctl net.ipv6.conf.ens2f0np0.addr_gen_mode=0
net.ipv6.conf.ens2f0np0.addr_gen_mode = 0
# sysctl net.ipv6.conf.ens2f1np2.addr_gen_mode=0
net.ipv6.conf.ens2f1np2.addr_gen_mode = 0

# ip a l ens2f0np0
2: ens2f0np0: <BROADCAST,MULTICAST,SLAVE,UP,LOWER_UP> mtu 1500 qdisc
mq master bond0 state UP group default qlen 1000
    link/ether 00:0f:53:2f:ea:40 brd ff:ff:ff:ff:ff:ff
    inet6 fe80::20f:53ff:fe2f:ea40/64 scope link tentative
       valid_lft forever preferred_lft forever
# ip a l ens2f1np2
5: ens2f1np2: <NO-CARRIER,BROADCAST,MULTICAST,SLAVE,UP> mtu 1500 qdisc
mq master bond0 state DOWN group default qlen 1000
    link/ether 00:0f:53:2f:ea:40 brd ff:ff:ff:ff:ff:ff
    inet6 fe80::20f:53ff:fe2f:ea40/64 scope link tentative
       valid_lft forever preferred_lft forever

Looks like addrconf_sysctl_addr_gen_mode() bypasses the original "is
this a slave interface?" check, and results in an address getting
added, while w/the proposed patch added, no address gets added.

Looking back through git history again, I see a bunch of 'Fixes:
d35a00b8e33d ("net/ipv6: allow sysctl to change link-local address
generation mode")' patches, and I guess that's where this issue was
also introduced.

-- 
Jarod Wilson
jarod@xxxxxxxxxx