On Sun, Sep 10, 2006 at 11:12:49AM -0700, Pete Zaitcev wrote: > On Sat, 09 Sep 2006 22:42:10 -0600, ebiederm at xmission.com (Eric W. Biederman) wrote: > > > The problem by remember a user space process by it's pid it is > > possible that the process will exit, pid wrap around will occur and a > > different process will appear in it's place. > > ... which is completely all right in this case. We used to have an > implementation which tried to hold onto the task_struct and that sucked. > It is only possible for the task to disappear without notifying devio > under very special conditions only, which involve forking with parent > exiting. In other words, even a buggy application won't trigger this > without deliberately trying. And when it happens, uid checks make sure > that other users are not affected. > > > Holding a reference > > to a struct pid avoid that problem, and paves the way > > for implementing a pid namespace. > > That may be useful. > > The patch itself seems straightforward if we can trust your struct > pid thingies. If OpenVZ people approve, I don't mind. perfectly fine from my side best, Herbert > -- Pete > _______________________________________________ > Containers mailing list > Containers at lists.osdl.org > https://lists.osdl.org/mailman/listinfo/containers
