Re: [gentoo-portage-dev] [PATCH v2 3/3] Introduce a tiny init replacement for inside pid namespace

[Michał Górny]

On Sun, 2018-11-18 at 09:21 +0000, M. J. Everitt wrote:
> On 18/11/18 09:18, Michał Górny wrote:
> > On Sun, 2018-11-18 at 09:04 +0000, M. J. Everitt wrote:
> > > On 18/11/18 08:53, Michał Górny wrote:
> > > > Signed-off-by: Michał Górny <mgorny@g.o>
> > > > ---
> > > >  bin/pid-ns-init        | 25 +++++++++++++++++++++++++
> > > >  lib/portage/process.py | 11 ++++++-----
> > > >  2 files changed, 31 insertions(+), 5 deletions(-)
> > > >  create mode 100644 bin/pid-ns-init
> > > >
> > > > diff --git a/bin/pid-ns-init b/bin/pid-ns-init
> > > > new file mode 100644
> > > > index 000000000..90660571a
> > > > --- /dev/null
> > > > +++ b/bin/pid-ns-init
> > > > @@ -0,0 +1,25 @@
> > > > +#!/usr/bin/env python
> > > > +# Copyright 2018 Gentoo Authors
> > > > +# Distributed under the terms of the GNU General Public License v2
> > > > +
> > > > +import os
> > > > +import sys
> > > > +
> > > > +
> > > > +def main(argv):
> > > > +	if len(argv) < 2:
> > > > +		return 'Usage: {} <main-child-pid>'.format(argv[0])
> > > > +	main_child_pid = int(argv[1])
> > > > +
> > > > +	# wait for child processes
> > > > +	while True:
> > > > +		pid, status = os.wait()
> > > > +		if pid == main_child_pid:
> > > > +			return os.WEXITSTATUS(status)
> > > > +
> > > > +	# this should never be reached
> > > > +	return 127
> > > > +
> > > > +
> > > > +if __name__ == '__main__':
> > > > +	sys.exit(main(sys.argv))
> > > > diff --git a/lib/portage/process.py b/lib/portage/process.py
> > > > index dee126c3c..75ec299f0 100644
> > > > --- a/lib/portage/process.py
> > > > +++ b/lib/portage/process.py
> > > > @@ -544,13 +544,14 @@ def _exec(binary, mycommand, opt_name, fd_pipes, env, gid, groups, uid, umask,
> > > >  					else:
> > > >  						if unshare_pid:
> > > >  							# pid namespace requires us to become init
> > > > -							# TODO: do init-ty stuff
> > > > -							# therefore, fork() ASAP
> > > >  							fork_ret = os.fork()
> > > >  							if fork_ret != 0:
> > > > -								pid, status = os.waitpid(fork_ret, 0)
> > > > -								assert pid == fork_ret
> > > > -								os._exit(status)
> > > > +								os.execv(portage._python_interpreter, [
> > > > +									portage._python_interpreter,
> > > > +									os.path.join(portage._bin_path,
> > > > +										'pid-ns-init'),
> > > > +									'%s' % fork_ret,
> > > > +									])
> > > >  						if unshare_mount:
> > > >  							# mark the whole filesystem as slave to avoid
> > > >  							# mounts escaping the namespace
> > >
> > > Why in python?! Surely a small C app would be significantly more efficient ..
> > >
> >
> > Surely adding a new build system for C apps would be entirely justified
> > by the necessity of premature optimization of program that spends ~100%
> > of its time in wait().
> >
>
> Forgive my ignorance then .. what purpose does a process have which simply
> wait()s ?! .. no, on second thoughts, don't answer that question, I'm
> unlikely to like the answer anyway ...
>

Well, if you find it useless, then I would suggest you save some space
by doing:

  rm /sbin/init

-- 
Best regards,
Michał Górny