6.14.0-1012-azure kernel has broken IPv6 fragmentation (upgrade to 6.14.0-1014-azure is required) #13190
Description
Description
The recent image upgrade brought in the 6.14.0-1012-azure kernel that contains the incorrectly backported kernel changes for IPv6 stack that make all IPv6 sockets have DONTFRAG bit set by default breaking UDP communication for packets larger than MTU. The incorrect beckport was done first in upstream stable kernel. See the report here: https://lore.kernel.org/stable/aElivdUXqd1OqgMY@karahi.gladserv.com/
And ubuntu kernel picked this incorrect backport in one of 6.14.0 kernels.
One of the solutions is to backport the ipv6: replace ipcm6_init calls with ipcm6_init_sk commit as well, which was done in 6.14.0-1014-azure .
Without this change we always get Ncat: Message too long. trying to send data larger than MTU. This breaks system test jobs for OVN project, and will potentially break other users as well.
Please, upgrade the image to 6.14.0-1014-azure.
Platforms affected
- Azure DevOps
- GitHub Actions - Standard Runners
- GitHub Actions - Larger Runners
Runner images affected
- Ubuntu 22.04
- Ubuntu 24.04
- macOS 13
- macOS 13 Arm64
- macOS 14
- macOS 14 Arm64
- macOS 15
- macOS 15 Arm64
- macOS 26 Arm64
- Windows Server 2019
- Windows Server 2022
- Windows Server 2025
Image version and build link
Version: 20251014.76.1
https://github.com/ovn-org/ovn/actions/runs/18622640352/job/53147265121#step:11:4571
Is it regression?
Yes. Last good build: 20250929.60.1 . Link: https://github.com/ovn-org/ovn/actions/runs/18556526459/job/52895345033#step:1:15
Expected behavior
IPv6 fragmentation supposed to work properly, i.e. kernel should fragment large UDPv6 datagrams if the DOTFRAG flag is not set.
Actual behavior
Kernel refuses to fragment large UDPv6 datagrams returning EMSGSIZE when they are larger than MTU.
Repro steps
See the reproducer program here: https://lore.kernel.org/stable/aElivdUXqd1OqgMY@karahi.gladserv.com/
Or start a netcat server and try to send a large file over it.