[Devel] [PATCH rh7] venetstat: Add ioctl interface via VE cgroup

[Vladimir Davydov]

On Fri, Jun 19, 2015 at 11:23:34AM +0300, Vladimir Davydov wrote:
> On Fri, Jun 19, 2015 at 11:10:14AM +0300, Cyrill Gorcunov wrote:
> > On Fri, Jun 19, 2015 at 10:55:29AM +0300, Vladimir Davydov wrote:
> > > > "venet_acct_ioctl" which would emulate old ioctls.
> > > > 
> > > > The write action into this file should pass
> > > > struct vz_tc_cg_req as an argument so that
> > > > it get unpacked into argument and command
> > > > for ioctl call and execute it.
> > > 
> > > IMHO this looks like trading bad for worse. Why can't we just use legacy
> > > ioctls for a start?
> > 
> > Sigh. It seems I misunderstood the problem itself :/ I thought
> > we're having troubles in calling ioctl but the problem is rathe
> > in ioctl internals code, where veid should be passed inside arguments
> > but we need to get rid of them. Drop this patch please, I'll rework.
> 
> I'd prefer to pass legacy_veid to these ioctls instead of introducing a
> new API just to zap it in the next major release...

OK, here goes what we've agreed upon. We do not introduce any substitute
for legacy ioctls used for obtaining stats, because designing a user API
is not an easy task. For UUID-based we will pass ve.legacy_veid to them.
The legacy_veid won't be auto-generated, as it is now. Instead, it will
be set from userspace on CT start. It should conform to two rules: (1)
uniqueness (all CTs on the same node have different IDs) and (2)
persistence (doesn't change after CT restart on the same node).