[Devel] [PATCH] ve/fs: add per-VE limit of mount points

[Stanislav Kinsburskiу]

9 нояб. 2015 г. 4:31 PM пользователь Vladimir Davydov <vdavydov at virtuozzo.com> написал:
>
> On Mon, Nov 09, 2015 at 04:23:34PM +0100, Stanislav Kinsburskiу wrote: 
> > 
> > 9 нояб. 2015 г. 4:15 PM пользователь Vladimir Davydov <vdavydov at virtuozzo.com> написал: 
> > > 
> > > On Mon, Nov 09, 2015 at 03:51:11PM +0100, Stanislav Kinsburskiy wrote: 
> > > > 
> > > > 
> > > > 09.11.2015 15:13, Andrew Vagin пишет: 
> > > > >On Thu, Nov 05, 2015 at 07:37:31PM +0300, Evgenii Shatokhin wrote: 
> > > > >>05.11.2015 18:45, Konstantin Khorenko пишет: 
> > > > >>>Zenya, 
> > > > >>> 
> > > > >>>1) an idea from Volodya Davydov: we can store the pointer to ve_struct, 
> > > > >>>not the veid, 
> > > > >>>    thus we'll eliminate possible races with veid reuse. 
> > > > >>OK. 
> > > > >>The main question is though, whether we should care about the situations 
> > > > >>where something is mounted from a VE and umounted from VE0. If we should 
> > > > >>not, it is not needed to change struct mount at all. 
> > > > >We can't disable this operation, so I think we should. 
> > > > > 
> > > > 
> > > > Could you elaborate on this, please? 
> > > > Why should we care about such situations? 
> > > 
> > > external bind mounts are mounted from ve0 and can be unmounted from 
> > > inside ve 
> > 
> > Ок. But what problems it could bring? 
>
> Unbalanced mnt_nr 

Why should we care about it?
It's not exposed anywhere. And the whole reason for this counter is to prevent "mount bomb".
If one from ve0 helps container to keep this counter low, then we screwed by default. Otherwise it works even in "unbalanced" state.
I don't think, that "counter is balanced" goal worth all this generic code explosion, which is required to achive it.