On Mon 2022-04-18 18:52 +0800, Xiubo Li wrote: > Hi Aaron, Hi Xiubo, > Thanks very much for you testing. No problem! > BTW, did you test this by using Livepatch or something else ? I mostly followed your suggestion here [1] by modifying/or patching the kernel to increase the race window so that unsafe_request_wait() may more reliably see a newly registered request with an unprepared session pointer i.e. 'req->r_session == NULL'. Indeed, with this patch we simply skip such a request while traversing the Ceph inode's unsafe directory list in the context of unsafe_request_wait(). [1]: https://tracker.ceph.com/issues/55329 Kind regards, -- Aaron Tomlin
