From mboxrd@z Thu Jan 1 00:00:00 1970 From: Vlad Yasevich Subject: Re: [PATCH net-next v2 RESEND 1/4] macvlan: support mac address changes when fwd_priv is enable Date: Thu, 12 Jun 2014 10:24:55 -0400 Message-ID: <5399B837.3020000@redhat.com> References: <1402127143-6456-1-git-send-email-dingtianhong@huawei.com> <1402127143-6456-2-git-send-email-dingtianhong@huawei.com> <5395E617.5080307@redhat.com> <53990628.9070202@huawei.com> Reply-To: vyasevic@redhat.com Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Cc: netdev@vger.kernel.org, John Fastabend To: Ding Tianhong , kaber@trash.net, davem@davemloft.net, edumazet@google.com, makita.toshiaki@lab.ntt.co.jp Return-path: Received: from mx1.redhat.com ([209.132.183.28]:23866 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755910AbaFLOZU (ORCPT ); Thu, 12 Jun 2014 10:25:20 -0400 In-Reply-To: <53990628.9070202@huawei.com> Sender: netdev-owner@vger.kernel.org List-ID: On 06/11/2014 09:45 PM, Ding Tianhong wrote: > On 2014/6/10 0:51, Vlad Yasevich wrote: >> [.. CC John Fastabend ] >> >> On 06/07/2014 03:45 AM, Ding Tianhong wrote: >>> If lowerdev supports L2 forwarding offload, the macvlan's hw address >>> will be set to the rar of the lowerdev and no need to set uc list, >>> and when the macvlan's hw address changes, the macvlan should remove >>> the old fwd and rebuild a new fwd for the macvlan. >>> >>> Signed-off-by: Ding Tianhong >>> --- >>> drivers/net/macvlan.c | 21 +++++++++++++++++++++ >>> 1 file changed, 21 insertions(+) >>> >>> diff --git a/drivers/net/macvlan.c b/drivers/net/macvlan.c >>> index 453d55a..67485ab 100644 >>> --- a/drivers/net/macvlan.c >>> +++ b/drivers/net/macvlan.c >>> @@ -523,6 +523,27 @@ static int macvlan_sync_address(struct net_device *dev, unsigned char *addr) >>> if (macvlan_addr_busy(vlan->port, addr)) >>> return -EBUSY; >>> >>> + if (vlan->fwd_priv) { >>> + lowerdev->netdev_ops->ndo_dfwd_del_station(lowerdev, >>> + vlan->fwd_priv); >>> + vlan->fwd_priv = NULL; >>> + ether_addr_copy(dev->perm_addr, dev->dev_addr); >>> + ether_addr_copy(dev->dev_addr, addr); >>> + vlan->fwd_priv = >>> + lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, >>> + dev); >>> + /* If we get a NULL pointer back, or if we get an error >>> + * then we should restore the old mac address and fwd. >>> + */ >>> + if (IS_ERR_OR_NULL(vlan->fwd_priv)) { >>> + ether_addr_copy(dev->dev_addr, dev->perm_addr); >>> + vlan->fwd_priv = >>> + lowerdev->netdev_ops->ndo_dfwd_add_station(lowerdev, >>> + dev); >>> + return -EINVAL; >>> + } >>> + return 0; >>> + } >> >> Calling del_stations add_station causes all sorts of VMDQ/ring >> operations to happen... Not sure if we can do that while we have a live >> macvlan/macvtap that is capable of transmitting data. >> >> Wouldn't it be sufficient to have a call to call to update with mac >> filter with the appropriate VMDQ. >> >> John, I defer to you here. The above looks really heavy-weight and >> I am not sure if its correct. >> >> Thanks >> -vlad > > I have used two 82599 nic to test this problem, I modify the ixgbe drivers and set the L2 forwarding offload default, > without this patch, when I change the macvlan mac address, the kernel oops, and if I apply this patch, the ixgbe could > work well with the maclvan, so I don't think this solution has problem. Have you tried changing the address while at the same time transmitting data through the macvlan? -vlad > > Thanks > Ding > >> >>> if (!vlan->port->passthru) { >>> err = dev_uc_add(lowerdev, addr); >>> if (err) >>> >> >> >> . >> > >