On Thu, Nov 21, 2024 at 07:33:54AM -0700, Tycho Andersen wrote: > On Wed, Nov 20, 2024 at 10:06:27AM +0100, Christian Brauner wrote: > > On Tue, Nov 05, 2024 at 11:10:24AM +0800, Yun Zhou wrote: > > > It is necessary to have a different pid_max in different containers. > > > For example, multiple containers are running on a host, one of which > > > is Android, and its 32 bit bionic libc only accepts pid <= 65535. So > > > it requires the global pid_max <= 65535. This will cause configuration > > > conflicts with other containers and also limit the maximum number of > > > tasks for the entire system. > > > > > > Signed-off-by: Yun Zhou <yun.zhou@xxxxxxxxxxxxx> > > > --- > > > > Fwiw, I've done a patch like this years ago and then Alex revived it in > > [1] including selftests! There's downsides to consider: > > > > [1]: https://lore.kernel.org/lkml/20240222160915.315255-1-aleksandr.mikhalitsyn@xxxxxxxxxxxxx > > Thanks, looks like this patch has the same oddity. > > For me it's enough to just walk up the tree when changing pid_max. It > seems unlikely that applications will create a sub pidns and then lower > the max in their own pid_max. Famous last words and all that. I think we should revive Alex series. Not just because it has selftests but afaict the implementation is a bit more robust.
