Re: [PATCH v2 0/3] virtiofsd: Add options to enable/disable posix acl

[Vivek Goyal <vgoyal@redhat.com>]

On Fri, Feb 19, 2021 at 04:55:06PM +0100, Miklos Szeredi wrote:
> On Fri, Feb 19, 2021 at 3:34 PM Vivek Goyal <vgoyal@redhat.com> wrote:
> >
> > On Fri, Feb 19, 2021 at 11:50:54AM +0000, Luis Henriques wrote:
> > > Vivek Goyal <vgoyal@redhat.com> writes:
> > >
> > > > Hi,
> > > >
> > > > This is V2 of the patches. Changes since v1 are.
> > > >
> > > > - Rebased on top of latest master.
> > > > - Took care of Miklos's comments to block acl xattrs if user
> > > >   explicitly disabled posix acl.
> > > >
> > > > Luis Henriques reported that fstest generic/099 fails with virtiofs.
> > > > Little debugging showed that we don't enable acl support. So this
> > > > patch series provides option to enable/disable posix acl support. By
> > > > default it is disabled.
> > > >
> > > > I have run blogbench and pjdfstests with posix acl enabled and
> > > > things work fine.
> > > >
> > > > Luis, can you please apply these patches, and run virtiofsd with
> > > > "-o posix_acl" and see if it fixes the failure you are seeing. I
> > > > ran the steps you provided manually and it fixes the issue for
> > > > me.
> > >
> > > Sorry for the delay.  I've finally tested these patches and they indeed
> > > fix the problem I reported.  My only question about this fix is why is
> > > this option not enabled by default, since this is the documented behavior
> > > in acl(5) and umask(2)?  In fact, why is this an option at all?
> >
> > You mean why to not enable acl by default?
> >
> > I am concerned about performance drop this can lead to because extra
> > GETXATTR(system.posix_acl_*) messages which will trigger if acls are enabled.
> > And not all users might require these. That's why I preferred to not enable
> > acl by default. Those who need it can enable it explicitly.
> >
> > Another example is xattr support. Due to performance concerns, we don't
> > enable xattrs by default either.
> 
> Actually generic xattr is much worse, since there's no caching for
> them currently, as opposed to posix acls, which are cached both when
> positive and negative.
> 
> If we enable ACL by default in case xattrs are enabled, we should be
> safe, I think.

Hi Miklos,

Ok, this sounds reasonable.  I am running some quick tests and if I don't
notice any serious performance regression, I will respin my patch.

> Having an option to disable acls still makes sense,
> but it's an optional plus.

Agreed. If there are no serious negative performance issues with enabling
ACL, then an option to disable is an optional plus.

May be I will drop this for now and add this when somebody needs an
option to disable ACL.

Thanks
Vivek