Reproduced on a clean Ubuntu 24.04.5 install with Canonical's own Ceph, using the reproducer attached to this bug.
Setup: MicroCeph snap, Ceph 19.2.3 squid (c92aebb279828e9c3c1f5d24613efca272649e62), single node with 3 OSDs, HEALTH_OK. A CephFS volume is mounted with the kernel client. Results: - 6.8.0-147-generic (#147-Ubuntu SMP PREEMPT_DYNAMIC Thu Sep 24 11:33:16 UTC 2026), the current Noble GA kernel: 5/9 FAIL. Every hole that touches or crosses the 4 MiB object boundary zeroes the preceding 16 KiB (16384/16384 bytes read as 0x00, expected 0xFF). The 4 cases that do not cross the boundary pass. - 7.0.0-42-generic (#42~24.04.1-Ubuntu, linux-generic-hwe-24.04): 9/9 PASS. So the fix is present in newer kernels but has not been backported to the Noble 6.8 GA kernel, and the Noble task is still New. This is silent data corruption on CephFS for any workload that punches holes (for example, VM disk images with discard enabled). Please consider a backport to Noble linux 6.8; the patch is attached to this bug. Canonical support case 00454344 is open for this issue, and a sosreport from the reproduction cluster has been uploaded to it. -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/2144592 Title: Punching hole through CephFS hosted file causes corruption when crossing 4MB RADOS object boundary To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2144592/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
