On Mon, 2014-04-21 at 22:56 +0200, Tom Gundersen wrote: > On Mon, Apr 21, 2014 at 9:33 PM, Tom Gundersen <teg@xxxxxxx> wrote: > > Hi, > > > > If IFLA_ADDRESS is set when creating a bridge over netlink, e.g. > > > > # ip link add test-bridge address b6:83:a2:b3:2f:0e type bride > > > > the bridge id is not set: > > > > # brctl show > > bridge name bridge id STP enabled interfaces > > bridge1 8000.000000000000 no > > My example was incomplete. To see the bug one should of course first > add an interface to the bridge: > > # brctl addif bridge1 eth0 > > > and observe that the bridge id remains unset: > > # brctl show > bridge name bridge id STP enabled interfaces > bridge1 8000.000000000000 no eth0 > > > > If bridge1 had been created without specifying the mac address the > result is as expected: > > # brctl show > bridge name bridge id STP enabled interfaces > bridge1 8000.28d244547cdb no eth0 > Hi. This seems to be a bug. What's worse is that local fdb entry won't be created and we will not be able to receive frames on the bridge device... rtnl_create_link() doesn't call ndo_set_mac_address but populates dev->dev_addr by itself. I'm not sure if it is safe to call ndo_set_mac_address in this function (maybe affects other drivers). I think this can be handled in bridge codes (br_link_ops->newlink). I'm submitting a patch to fix this. Thanks, Toshiaki Makita
