[PATCHv3 2/2] /proc/PID/status: show all sets of pid according to ns

Chen, Hanxiao chenhanxiao at cn.fujitsu.com
Tue Sep 30 10:37:52 UTC 2014


Hi,

> -----Original Message-----
> From: Serge E. Hallyn [mailto:serge at hallyn.com]
> Sent: Monday, September 29, 2014 10:00 PM
> Subject: Re: [PATCHv3 2/2] /proc/PID/status: show all sets of pid according to
> ns
[snip]
> > >
> > > This patch adds four fields: NStgid, NSpid, NSpgid and NSsid:
> > > a) In init_pid_ns, nothing changed;
> > >
> > > b) In one pidns, will tell the pid inside containers:
> > > NStgid:	21776   5       1
> > > NSpid:  21776   5       1
> > > NSpgid:	21776   5       1
> > > NSsid:  21729   1       0
> > > ** Process id is 21776 in level 0, 5 in level 1, 1 in level 2.
> > >
> > > c) If pidns is nested, it depends on which pidns are you in.
> > > NStgid: 5       1
> > > NSpid:  5       1
> > > NSpgid: 5       1
> > > NSsid:  1       0
> > > ** Views from level 1
> > >
> >
> > This patch is simple, useful and safe.
> > But currently there is not any feedbacks.
> >
> > Any comments or ideas?
> 
> Thanks, Chen.  The code looks fine.  My concern is that you are
> exposing information which cannot be checkpointed and restarted.
> In particular, if I'm inside a nested container, so I'm in pidns
> level 3, then my own NSpid info, when I read it, will show the
> pids at parent namespaces.  If I'm restarted at the third pidns
> level, only the one pid can be restored.

If you're in level 3, read your own proc, only level 3's NSpid info
will be shown. No parent namesapces info could be seen.

Only if not providing a procfs mount point for the new container,
and without a proper pivot_root,
we could see some NSpid info of parent ns.

If each new container got their own procfs mount point,
only its and its child's NSpid info could be seen.

> 
> Now it may be fair to say "this is proc, and proc and sys show
> host info which is not containerized and cannot be checkpointed
> and restarted, deal with it."  But I'm not sure.
> 
> There are two ways you could deal with this.  One would be to
> show the nspids only to the level of the reader of the file - but
> I don't think you need to do that.  I think you're better off
> simply showing the pids up to the level of the struct pid for
> the mounter of the procfs.  So if I'm inside container c2 which
> is inside container c1, my own /proc will only show pids which
> are valid in c2 (and any child namespaces), while the /proc
> mounted in c1 will show pids valid in c1 and c2 (and any children),
> but not those in the init_pid_ns.  It's then just up to the
> container administrators to make sure that c2 cannot see c1's
> /proc to confuse itself and confuddle checkpoint-restart

IIUC, this patch already deal with this scenario:

+	for (g = ns->level; g <= pid->level; g++)
+		seq_printf(m, "\t%d ",
+			task_tgid_nr_ns(p, pid->numbers[g].ns));

With this patch, it did like
a) in init_pid_ns, check /proc/21776/status
NStgid:	21776   5       1

b) in c1, check /proc/5/status:
NStgid:	5       1

c) in c2, check /proc/1/status:
NStgid:	1

Thanks,
- Chen


More information about the Containers mailing list