18 Sep
2025
18 Sep
'25
4:43 a.m.
On Thu, Sep 18, 2025 at 12:51 AM Dave Chinner david@fromorbit.com wrote:
- wait for Josef to finish his inode refcount rework patchset that gets rid of this whole "writeback doesn't hold an inode reference" problem that is the root cause of this the deadlock.
No, it is necessary to have a minimal fix that is eligible for stable backports.
Of course, my patch is a kludge; this problem is much larger and a general, global solution should be preferred. But my patch is minimal, easy to understand, doesn't add overhead and piggybacks on an existing Ceph feature (per-inode work) that is considered mature and stable. It can be backported easily to 6.12 and 6.6 (with minor conflicts due to renamed netfs functions in adjacent lines).