RE: [PATCH 1/2v6] procfs: show hierarchy of pid namespace

[Chen, Hanxiao]

> -----Original Message-----
> From: Richard Weinberger [mailto:richard@xxxxxx]
> Sent: Wednesday, November 05, 2014 8:52 PM
> To: Serge E. Hallyn
> Cc: Chen, Hanxiao/陈 晗霄; Eric W. Biederman; Serge Hallyn; Oleg Nesterov;
> containers@xxxxxxxxxxxxxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; Mateusz
> Guzik; David Howells
> Subject: Re: [PATCH 1/2v6] procfs: show hierarchy of pid namespace
> 
> Am 05.11.2014 um 13:41 schrieb Serge E. Hallyn:
> > Quoting Richard Weinberger (richard@xxxxxx):
> >> Am 05.11.2014 um 11:41 schrieb Chen Hanxiao:
> >>> We lack of pid hierarchy information, and this will lead to:
> >>> a) we don't know pids' relationship, who is whose child:
> >>>    /proc/PID/ns/pid only tell us whether two pids live in different ns
> >>> b) bring trouble to nested lxc container check/restore/migration
> >>> c) bring trouble to pid translation between containers;
> >>>
> >>> This patch will show the hierarchy of pid namespace
> >>> by pidns_hierarchy like:
> >>>
> >>> [root@localhost ~]#cat /proc/pidns_hierarchy
> >>> 18060 18102 1534
> >>> 18060 18102 1600
> >>> 1550
> >>
> >> Hmm, what about printing the pid hierarchy in the same way as
> /proc/self/mountinfo
> >> does with mount namespaces?
> >> Your current approach is not bad but we should really try to be consistent
> with existing
> >> sources of information.
> >
> > Good point.  How would you structure it to make it look mor elike mountinfo?
> > Adding the pidns inode number (in place of a mount sequence number) might be
> > useful, but it sounds like you have a more concrete idea?
> 
> Just list <init_PID> <parent_of_init_PID>. This way we have exactly one
> information record per line and always exactly two columns to parse.
> 
> e.g.
> [root@localhost ~]#cat /proc/pidns_hierarchy
> 1550 1
> 18060 1
> 18102 18060
> 1534 18102
> 1600 18102
> 
But this style lacks of *level* information:
Ex:
1->18060->18102->1600->1700
If we want to check the 1700's level in pid ns
Style 1:
18060 18102 1600 1700

Style 2:
18060 1
18102 18060
1600 18102
1700 1600

If we had a little more containers,
Style 2 would not be clear enough.
1 line vs $(PID level) line

If there were no more related information to show,
I think style 1 looks better.

Thanks,
- Chen

> >> This function allocates memory per PID. If we have lots of PIDs, how does this
> scale?
> >> I'd go so far and say this can be a DoS'able issue if the pidns_hierarchy file
> is opened multiple times...
> >
> > It's not per pid, but per init-pid.  For non-reaper pids he bails and continue
> > through the loop a few lines above.  This still may be DOS-able if users don't
> > have kmem restrictions to prevent a ton of pid namespaces, but then the
> > namespaces themselves will take a lot more memory than the representation here.
> 
> Ah, I've overlooked that fact. If it is per init-pid it is not that bad. :-)
> 
> Thanks,
> //richard
N?叉??y??b??千v??藓{.n???{?赙zXФ?塄}?财??j:+v???赙zZ+€?zf＂?????i????ア??璀??撷f?^j谦y??@A?囤?0鹅h??i