On Wed, Feb 11, 2026 at 5:29 PM Fabien Sirjean via ceph-users <[email protected]> wrote: > > Hello, > > We are encountering a reproducible issue with the CephFS kernel client > on Ubuntu 24.04 and would appreciate your feedback. > > Context: > We are preparing an updated Ubuntu 24.04 image for our scientific users. > With kernel |6.17.0-14-generic|, the CephFS mount succeeds and the > filesystem is visible via |df -h|, but any attempt to access the > directory contents (e.g. |ls|) immediately triggers a kernel crash. > > From userland, the process is simply killed: > > |# ls /net/serdon/illdata Killed| > > The mount appears healthy and the filesystem is correctly reported: > > |# mount | grep ceph 192.168.170.61,192.168.170.69,192.168.170.70:/ on > /net/serdon/illdata type ceph (rw,relatime,name=visa,acl) # df -h | grep > serdon 192.168.170.61,192.168.170.69,192.168.170.70:/ 4.1P 2.0P 2.1P 49% > /net/serdon/illdata| > > However, any access to the mount point triggers a kernel NULL pointer > dereference in the Ceph module: > > |BUG: kernel NULL pointer dereference, address: 0000000000000000 RIP: > strcmp+0x10/0x40 Call Trace: ceph_mds_auth_match ceph_mds_check_access > ceph_open| > > Environment: > > |Linux 6.17.0-14-generic #14~24.04.1-Ubuntu CephFS mounted via kernel > client (mount -t ceph)| > > The issue is fully reproducible: any access to the mount point causes > the same kernel crash. > > Full stack trace: > https://pastebin.com/ZgQHpN0q <https://pastebin.com/ZgQHpN0q> > > Questions: > - Is this a known issue with the CephFS kernel client on Linux 6.17?
Hi Fabien, Yes, it has just been fixed: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/fs/ceph?id=7987cce375ac8ce98e170a77aa2399f2cf6eb99f > - Does this look like an upstream CephFS bug, or something specific to > the Ubuntu 6.17 kernel? It's an upstream bug; the fix should be trivial to backport to any Ubuntu kernel that is affected. Thanks, Ilya _______________________________________________ ceph-users mailing list -- [email protected] To unsubscribe send an email to [email protected]
