Re: [Virtio-fs] [PATCH v2 0/2] virtiofsd: Fix xattr and ACL

[Vivek Goyal <vgoyal@redhat.com>]

On Fri, Jan 31, 2020 at 02:06:51AM +0000, misono.tomohiro@fujitsu.com wrote:
> > On Tue, Jan 28, 2020 at 07:18:17PM +0900, Misono Tomohiro wrote:
> > > Hi,
> > >
> > > This is a second version of xattr fix for virtiofsd.
> > > I included ACL fix (which introduces new option posix_acl) in this
> > > version too as ACL mostly depends on xattr.
> > >
> > > I run xfstests with XFS backend using "-o xattr -o posix_acl" option
> > > and only new failure is generic/375 which checks if sgid bit is
> > > cleared after setfacl. I'll try to investigate it.
> > >
> > > change in v1 -> v2
> > >  - rebased to current dev branch
> > >
> > >  - Always chdir for xattr (1st patch)
> > >    In v1, I keep current implementation for regular file/dir since it
> > >    show better performance in my environment. But I notice opening file
> > >    for xattr causes seek sanity test fails (xfstest generic/285, 436).
> > >
> > >    I'm not sure what is the fundamental problem here but I believe
> > >    performance can be improved by introducing some caching mechanism
> > >    in general.
> > 
> > Hi Misono,
> > 
> > How much is performance degradation due to fchdir(). If it is significant, then I will be inclined to keep original code for dir/file
> > till some other mechanism is introduced to offset the perofrmance loss.
> 
> Please refer this replay: https://www.redhat.com/archives/virtio-fs/2020-January/msg00063.html

As per your email, regression due to fchdir() seems to be in the range of
5% to 10%. It is not trivial, IMO. May be its a good idea to keep original
logic and use fchdir() only when need be.

Thanks
Vivek