openvswitch / ovs-issues

Issue tracker repo for Open vSwitch
10 stars 3 forks source link

[Window] BSOD happens because OVSext driver #279

Closed wenyingd closed 9 months ago

wenyingd commented 1 year ago

Hi,

We recently hit BSOD many times on a Windows host running OVS, and the stop code is "KMODE EXCEPTION NOT HANDLED" with failure in "OVSExt.sys". image (2)

aserdean commented 1 year ago

Hey,

To help triage the bug faster, can you please share the minidump or kernel dump?

I can send you instructions on how to enable them if your system doesn’t generate them by default.

wenyingd commented 1 year ago

Having some troubleshooting, we find that the BSoD looks related with deletion of the Windows route. An example is like this,

There are three threads running on the Windows host: 1) T0 maintains a long TCP connection with destination 10.96.0.1:443, 2) T1 works to add routes destination=10.96.0.1, nexthop=169.254.0.253 and destination=169.254.0.253/32, nexthop=192.168.3.1, 192.168.3.1 is configured on a local network adapter, 3) T2 works to delete the above two routes. T2 and T3 continue working in a interval

OVS has OpenFlow entries to perform DNAT from 10.96.0.1 to dst IP 1.1.1.2.

We observed that several acknowledge packets ( the packets are workload packets not used in handshake stage ) were not sent out from the windows host when the BSoD happened, and the listed route entries were not existing at that time.