Messages in this thread | | | From | Jarod Wilson <> | Date | Thu, 19 Mar 2020 17:53:03 -0400 | Subject | Re: [PATCH net] ipv6: don't auto-add link-local address to lag ports |
| |
On Thu, Mar 19, 2020 at 3:52 PM Jay Vosburgh <jay.vosburgh@canonical.com> wrote: > > Jarod Wilson <jarod@redhat.com> wrote: > > >On Thu, Mar 19, 2020 at 1:06 PM Eric Dumazet <eric.dumazet@gmail.com> 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. > > I wonder if this also breaks for the netvsc usage of IFF_SLAVE > to suppress ipv6 addrconf? Adding the hyperv maintainers to Cc. > > In any event, it looks like addrconf_sysctl_addr_gen_mode() > calls addrconf_dev_config() directly, which bypasses the IFF_SLAVE check > in addrconf_notify() that would gate other callers.
Yeah, that's what I was thinking as well.
> From my reading, your patch appears to cover a superset of cases > as compared to the existing IFF_SLAVE test from c2edacf80e15.
I wasn't aware of additional devices that would want to prevent these addresses, could certainly alter the patch to reject anything with IFF_SLAVE too for consistency.
> >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. > > Can the problem be induced via ip link set ... addrgenmode ? > That functionality predates the sysctl interface, looks like it was > introduced with
Doesn't look like it, no. No change after either trying addrgenmode none or random.
-- Jarod Wilson jarod@redhat.com
| |