Between 2017-01-05 and about 2017-01-10 EST we had 17 recorded crashes on the neutron node.
After patching as described above, we have had no crashes. Not being particularly familiar with kernel internals, or the details of the skb data structure, I was not able to confirm explicitly the cause by sending the packets that trigger the crash. I am trusting that the choice to drop the packets with the patch doesn't have any significant side effects for the openvswitch module. A clearer explanation with earlier patch(es) may be found here: https://patchwork.ozlabs.org/patch/559944/ https://patchwork.ozlabs.org/patch/712373/ It also may be of note to others that gso and gro offloading to the NIC are turned off on all of our interfaces. I am not sure how timely the application of the corresponding patch in the upstream stable kernel will be, so here is the patch I used. attached is the patch for my test build, using the sources in comment #3 ** Attachment added: "patch_1655683.txt" https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1655683/+attachment/4803442/+files/patch_1655683.txt -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1655683 Title: kernel BUG at skbuff.h:1486 Insufficient linear data in skb __skb_pull.part.7+0x4/0x6 [openvswitch] To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1655683/+subscriptions -- ubuntu-bugs mailing list ubuntu-bugs@lists.ubuntu.com https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
