On Sat, Dec 2, 2023 at 4:50 AM Serge E. Hallyn <serge@xxxxxxxxxx> wrote: > > On Fri, Dec 01, 2023 at 09:46:32AM +0000, Yafang Shao wrote: > > Considering that MPOL_F_NUMA_BALANCING or mbind(2) using either > > MPOL_MF_MOVE or MPOL_MF_MOVE_ALL are capable of memory movement, it's > > essential to include security_task_movememory() to cover this > > functionality as well. It was identified during a code review. > > Hm - this doesn't have any bad side effects for you when using selinux? > The selinux_task_movememory() hook checks for PROCESS__SETSCHED privs. > The two existing security_task_movememory() calls are in cases where we > expect the caller to be affecting another task identified by pid, so > that makes sense. Is an MPOL_MV_MOVE to move your own pages actually > analogous to that? > > Much like the concern you mentioned in your intro about requiring > CAP_SYS_NICE and thereby expanding its use, it seems that here you > will be regressing some mbind users unless the granting of PROCESS__SETSCHED > is widened. Ah, it appears that this change might lead to regression. I overlooked its association with the PROCESS__SETSCHED privilege. I'll exclude this patch from the upcoming version. Thanks for your review. -- Regards Yafang
