This series adds infrastructure and support for using the filtering
support built into recent Linux kernels. The tap device currently
provides a small array for exact MAC address matches, plus a hash
for inexact multicast addresses. By programing this filter, we can
reduce the number of extraneous packets qemu needs to handle.
I've restricted the configuration where this can be used to only
the case of having two devices on a qemu vlan. In this case, we
have a clear consumer/produce model for the backend NIC to request
filtering from the TAP source. To handle hotplug, I've added
notifiers that allow vlan clients to be notified when someone gets
added or removed so they can return to a non-filtered (of self-
filtered) mode.
This series also includes code for both virtio-net and e1000 to
make use of the filtering. The tap devices provides a new
rxfilter=off option to disable exporting this new feature to
the backends. The primary reason for this is that the current
Linux tun driver has a bug that if the exact match array
overflows, there's a possiblity of dropping unicast packets.
There are currently 8 exact matches, so the guest needs to be
trying to filter more than 8 to hit this bug. I've already
submitted a patch for 2.6.29 and stable to fix this.
Please comment, thanks,
Alex
---
Alex Williamson (4):
qemu:e1000: Add support for qemu_vlan_rxfilter
qemu:virtio-net: Add support for qemu_vlan_rxfilter
qemu:net: Add TAP support for RX filtering on Linux
qemu:net: Add infrastructure for setting an RX filter through the vlan
hw/e1000.c | 60 +++++++++++++++++++++++++++++++++
hw/virtio-net.c | 56 ++++++++++++++++++++++++++++++-
net.c | 99 +++++++++++++++++++++++++++++++++++++++++++++++++++----
net.h | 14 ++++++++
qemu-doc.texi | 6 ++-
vl.c | 3 +-
6 files changed, 226 insertions(+), 12 deletions(-)
--
Alex Williamson
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
