On 12/22, Sukadev Bhattiprolu wrote: > > Oleg Nesterov [oleg@xxxxxxxxxx] wrote: > | On 12/22, Oleg Nesterov wrote: > | > > | > On 12/20, Sukadev Bhattiprolu wrote: > | > > > | > > + * TODO: > | > > + * Making SI_ASYNCIO a kernel signal could make this less hacky. > | > > + */ > | > > +#ifdef CONFIG_PID_NS > | > > +static inline int siginfo_from_user(siginfo_t *info) > | > > +{ > | > > + if (!is_si_special(info) && SI_FROMUSER(info) && > | > > | > OK, if we can trust SI_FROMUSER(), then it is better, i agree. > | > | Aaah, forgot to mention... > | > | But could you explain how are you going to fix another problem, > | .si_pid mangling? This was another reason for (yes, ugly, agreed) > | SIG_FROM_USER in .si_signo. > > Good point. > > I was going through the ->si_pid assignments to try and fix them at > source (like the mqueue patch I sent last week). OK. > The two cases that don't fit the model are sys_kill() and sys_tkill(). > For that I was hoping we could use siginfo_from_user() again. i.e > > if (siginfo_from_user()) > masquerade_si_pid() > > in the default: case of send_signal(). To be safe, masquerade_si_pid() > could do it only iff si_code is either SI_USER or SI_TKILL. > > IOW, with some tweaks, I am trying to see if we can use siginfo_from_user() > in place of the SIG_FROM_USER. sys_rt_sigqueueinfo(). But, perhaps we can just ignore the problems with sigqueueinfo() (and document them). The only thing we must preserve is: we should not change *info when from_parent_ns == F, but this happens "automatically". And, the kernel just can not know what "info" means when it is sent by sigqueueinfo() anyway. So, perhaps we can just do if (!same_ns) masquerade_si_pid() ? Oleg. _______________________________________________ Containers mailing list Containers@xxxxxxxxxxxxxxxxxxxxxxxxxx https://lists.linux-foundation.org/mailman/listinfo/containers