netdev - Regression introduced by "r8169: simplify rtl_set_mac

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]

Message-ID: <20180702194812.GG3111@calimero.vinschen.de>
Date:   Mon, 2 Jul 2018 21:48:12 +0200
From:   Corinna Vinschen <vinschen@...hat.com>
To:     netdev@...r.kernel.org, Heiner Kallweit <hkallweit1@...il.com>
Subject: Regression introduced by "r8169: simplify rtl_set_mac_address"

Hi,

the patch 1f7aa2bc268e, "r8169: simplify rtl_set_mac_address",
introduced a regression found by trying to team a r8169 NIC.

Try the following (assuming the r8169 NIC is eth0):

$ nmcli con add type team con-name team0 ifname nm-team config \
  '{"runner": {"name": "lacp"}, "link_watch": {"name": "ethtool"}}' \
  ipv4.method disable ipv6.method ignore
$ nmcli con add type ethernet ifname eth0 con-name team0.0 master nm-team
$ nmcli con up team0.0
$ teamdctl nm-team port present eth0
command call failed (No such device)

Bisecting turned up commit 1f7aa2bc268e, "r8169: simplify
rtl_set_mac_address" as the culprit.  Reverting this patch fixes
the issue and the teamdctl call succeeds.

The reason is apparently the usage of eth_mac_addr here.  eth_mac_addr
calls eth_prepare_mac_addr_change which checks for IFF_LIVE_ADDR_CHANGE.
Debugging shows this flag not being set on r8169, thus
eth_prepare_mac_addr_change returns -EBUSY (no idea why userspace claims
"No such device", rather than "Device or resource busy", but that's not
the point here).

Note that other devices like igb, don't call eth_mac_addr either, but
rather call memcpy by themselves to copy the new MAC, just as the
original r8169 code did, too.  Consequentially this problem is not
present on igb.

I suggest to revert this change in the first place, but I wonder if
we're not just missing to set IFF_LIVE_ADDR_CHANGE in a lot of drivers.

Thanks,
Corinna