On Thu, 28 Jul 2005 09:56:03 -0700 Ben Greear <greearb@xxxxxxxxxxxxxxx> wrote: > > ii) This patch will also break any code that reads raw sockets on vlan > > devices and expects to see vlan tags in the frames it reads. Basically > > this patch makes eth0.N look *exactly* like any other ethernet device to > > the networking stack and all other users of the device. > > I haven't looked at the patch, but I don't like this change. If nothing else, > it is a performance hit to reorder the header. > > Why do you need to remove the REORDER flag? (If you need it always enabled on > your box, then just make sure it is always set..) The problem is that if Q in Q is to be supported then the code cannot use your mechanism in vlan_dev_hard_start_xmit() for deciding whether reordering is needed. This routine decides reordering is not needed if the frame to transmit already has an 802.1q tag. Clearly this will not work on the lowest vlan device in a chain, e.g. eth0.1 in eth0.1.2. I thought that since REORDERing was on by default - and I couldn't think of any situations in which you would need to turn it off - the simplest thing to do would be to remove the ability to disable it. However, clearly it *is* being used and so another way must be found. I'll have another look at it.... Alex
