diff options
Diffstat (limited to 'results/classifier/accel-gemma3:12b/vmm/2745')
| -rw-r--r-- | results/classifier/accel-gemma3:12b/vmm/2745 | 27 |
1 files changed, 27 insertions, 0 deletions
diff --git a/results/classifier/accel-gemma3:12b/vmm/2745 b/results/classifier/accel-gemma3:12b/vmm/2745 new file mode 100644 index 000000000..7b48d5b86 --- /dev/null +++ b/results/classifier/accel-gemma3:12b/vmm/2745 @@ -0,0 +1,27 @@ + +Qemu should send RARP for vhostuser regardless of whether virtio supports GUEST_ANNOUNCE +Description of problem: +When virtio reports to qemu that GUEST_ANNOUNCE is supported, qemu will not send RARP. The assumption, I think, is that guest kernel will do whatever is needed to announce the guest. The problem with this assumption is that 1) kernel won't send RARPs; 2) for IPv4 and IPv6 it will send GARPs and NAs; 3) for interfaces with no IP addresses, I think it won't send anything at all. + +RARPs are useful because they allow to notify regardless of IP configuration. I've [asked](https://issues.redhat.com/browse/RHEL-71919]) RHEL kernel folks to consider issuing RARPs from the guest kernel, but it won't happen overnight, and regardless, it's not a complete solution since we cannot expect all guests running a patched kernel with such feature. + +RARP packets are also often expected by underlying network components. For example, OVN controller has a special "activation-strategy=rarp" configuration that makes OVN wait for a RARP from destination chassis on live migration, and only then unblock traffic for the port. Since RARP is not issed by Qemu nor virtio-net, the OVN port is never unblocked (until its configuration is reset by CMS). + +I think what should be done from Qemu side is to send RARP for vhostuser ports regardless of whether virtio supports GUEST_ANNOUNCE. I **think** this can be achieved by removing this code: + +``` + /* If guest supports GUEST_ANNOUNCE do nothing */ + if (virtio_has_feature(dev->acked_features, VIRTIO_NET_F_GUEST_ANNOUNCE)) { + return 0; + } +``` +Steps to reproduce: +1. Start a VM with vhostuser* port and fresh virtio guest driver. +2. Live migrate it. +3. Observe that RARP is not sent from the migrated port. GARP (or NA for IPv6) is sent instead. +Additional information: +Some external bugs that may be relevant: + +- RHEL kernel request to send RARPs from virtio: https://issues.redhat.com/browse/RHEL-71919 (won't fix the issue for older unpatched kernels) +- Request for OVN to handle GARPs and NAs: https://issues.redhat.com/browse/FDP-1042 (won't solve for unaddressed ports!) +- An attempt to work around the issue in OpenStack Neutron OVN env by disabling activation strategy: https://issues.redhat.com/browse/OSPRH-12571 (not a great long term solution) |