Re: kern/53562: bridge(4) breaks segmentation / TX checksum offloading

[Rin Okuyama <rokuyama%rk.phys.keio.ac.jp@localhost>]

On 2018/09/14 20:17, Masanobu SAITOH wrote:
> On 2018/09/12 21:33, Rin Okuyama wrote:
...
> > This patch disable TX offload when the interface is added to bridge:
> >
> > http://www.netbsd.org/~rin/disable_tx_offload_when_bridged_20180912.patch
>
> I don't know if it's good to doing in L3...
>
> It seems that both FreeBSD and DragonFly clear all bridge member's outgoing
> offload flags when adding a member to a bridge. I also don't know if
> this is good solution or not...

If I understand correctly, since TX offload is for L3 (and L4), we
inevitably need something to do in L3...

I implemented software TX/RX offload engines for ethernet:

  http://www.netbsd.org/~rin/ether_offload_20180916.patch

Using these routines, I wrote a patch where TX offloading is canceled in
bridge_output() when the destination interface is different from the
source interface:

  http://www.netbsd.org/~rin/bridge_tx_offload_20180916.patch

With this patch, the guest successfully communicates with the host even
if any TX offloading is enabled on the host side.

Also, I added TX/RX offloading support to shmif(4):

  http://www.netbsd.org/~rin/shmif_offload_20180916.patch

Since I'm not familiar with rump, this patch has not been tested
systematically yet. But it seems to work to some extent...

Thanks,
rin