On 2018/10/09 22:26, Michal Hocko wrote: > On Tue 09-10-18 22:14:24, Tetsuo Handa wrote: >> On 2018/10/09 21:58, Michal Hocko wrote: >>> On Tue 09-10-18 21:52:12, Tetsuo Handa wrote: >>>> On 2018/10/09 20:10, Michal Hocko wrote: >>>>> On Tue 09-10-18 19:00:44, Tetsuo Handa wrote: >>>>>>> 2) add OOM_SCORE_ADJ_MIN and do not kill tasks sharing mm and do not >>>>>>> reap the mm in the rare case of the race. >>>>>> >>>>>> That is no problem. The mistake we made in 4.6 was that we updated oom_score_adj >>>>>> to -1000 (and allowed unprivileged users to OOM-lockup the system). >>>>> >>>>> I do not follow. >>>>> >>>> >>>> http://tomoyo.osdn.jp/cgi-bin/lxr/source/mm/oom_kill.c?v=linux-4.6.7#L493 >>> >>> Ahh, so you are not referring to the current upstream code. Do you see >>> any specific problem with the current one (well, except for the possible >>> race which I have tried to evaluate). >>> >> >> Yes. "task_will_free_mem(current) in out_of_memory() returns false due to MMF_OOM_SKIP >> being already set" is a problem for clone(CLONE_VM without CLONE_THREAD/CLONE_SIGHAND) >> with the current code. > > a) I fail to see how that is related to your previous post and b) could > you be more specific. Is there any other scenario from the two described > in my earlier email? > I do not follow. Just reverting commit 44a70adec910d692 and commit 97fd49c2355ffded is sufficient for closing the copy_process() versus __set_oom_adj() race. We went too far towards complete "struct mm_struct" based OOM handling. But stepping back to "struct signal_struct" based OOM handling solves Yong-Taek's for_each_process() latency problem and your copy_process() versus __set_oom_adj() race problem and my task_will_free_mem(current) race problem.
