netdev - [DISCUSS] Bond arp monitor not works with veth due to flag NETIF_F

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 18 Nov 2021 11:11:31 +0800
From:   Hangbin Liu <liuhangbin@...il.com>
To:     netdev@...r.kernel.org
Cc:     Jay Vosburgh <jay.vosburgh@...onical.com>,
        Veaceslav Falico <vfalico@...il.com>,
        Andy Gospodarek <andy@...yhouse.net>,
        Jarod Wilson <jarod@...hat.com>,
        Jakub Kicinski <kuba@...nel.org>,
        Jiri Pirko <jiri@...nulli.us>, davem@...emloft.net,
        Denis Kirjanov <dkirjanov@...e.de>,
        David Ahern <dsahern@...il.com>,
        Eric Dumazet <edumazet@...gle.com>
Subject: [DISCUSS] Bond arp monitor not works with veth due to flag
 NETIF_F_LLTX.

Hi,

When I test bond arp monitor with veth interface, the bond link flaps rapidly.
After checking, in bond_ab_arp_inspect():

                trans_start = dev_trans_start(slave->dev);
                if (bond_is_active_slave(slave) &&
                    (!bond_time_in_interval(bond, trans_start, bond->params.missed_max) ||
                     !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) {
                        bond_propose_link_state(slave, BOND_LINK_DOWN);
                        commit++;
                }

it checks both slave's trans_start and last_rx. While veth interface's
trans_start never get updated due to flag "NETIF_F_LLTX". As when NETIF_F_LLTX
set, in netdev_start_xmit() -> txq_trans_update() the txq->trans_start
never get updated because txq->xmit_lock_owner is always -1.

If we remove the flag NETIF_F_LLTX, the HARD_TX_LOCK() will acquire the
spin_lock and update txq->xmit_lock_owner. I expected there may have some
performance drop. But I tested with xdp_redirect_map and pktgen by forwarding
a 10G NIC's traffic to veth interface and didn't see much performance drop. e.g.
With xdpgeneric mode, with the flag, it's 2.18M pps, after removing the flag,
it's 2.11M pps. Not sure if I missed anything.

So what do you think? Should we remove this flag on veth to fix the issue?
Some user may want to use bonding active-backup arp monitor mode on netns.

Thanks
Hangbin