Re: [PATCH 066/104] virtiofsd: passthrough_ll: add renameat2 support

[Dr. David Alan Gilbert]

* Daniel P. Berrangé (address@hidden) wrote:
> On Mon, Jan 13, 2020 at 08:06:24PM +0000, Dr. David Alan Gilbert wrote:
> > * Dr. David Alan Gilbert (address@hidden) wrote:
> > > * Daniel P. Berrangé (address@hidden) wrote:
> > > > On Thu, Dec 12, 2019 at 04:38:26PM +0000, Dr. David Alan Gilbert (git) 
> > > > wrote:
> > > > > From: Miklos Szeredi <address@hidden>
> > > > > 
> > > > > No glibc support yet, so use syscall().
> > > > 
> > > > It exists in glibc in my Fedora 31 install.
> > > > 
> > > > Presumably this is related to an older version
> > > > 
> > > > > Signed-off-by: Miklos Szeredi <address@hidden>
> > > > > ---
> > > > >  tools/virtiofsd/passthrough_ll.c | 10 ++++++++++
> > > > >  1 file changed, 10 insertions(+)
> > > > > 
> > > > > diff --git a/tools/virtiofsd/passthrough_ll.c 
> > > > > b/tools/virtiofsd/passthrough_ll.c
> > > > > index 91d3120033..bed2270141 100644
> > > > > --- a/tools/virtiofsd/passthrough_ll.c
> > > > > +++ b/tools/virtiofsd/passthrough_ll.c
> > > > > @@ -1083,7 +1083,17 @@ static void lo_rename(fuse_req_t req, 
> > > > > fuse_ino_t parent, const char *name,
> > > > >      }
> > > > >  
> > > > >      if (flags) {
> > > > > +#ifndef SYS_renameat2
> > > > >          fuse_reply_err(req, EINVAL);
> > > > > +#else
> > > > > +        res = syscall(SYS_renameat2, lo_fd(req, parent), name,
> > > > > +                      lo_fd(req, newparent), newname, flags);
> > > > > +        if (res == -1 && errno == ENOSYS) {
> > > > > +            fuse_reply_err(req, EINVAL);
> > > > > +        } else {
> > > > > +            fuse_reply_err(req, res == -1 ? errno : 0);
> > > > > +        }
> > > > > +#endif
> > > > 
> > > > We should use the formal API if available as first choice
> > > 
> > > OK, done - I've kept the 'ifndef SYS_renameat2' that drops back to an
> > > error for truly ancient cases; although I doubt everything else will
> > > build on something that old.
> > 
> > Hmm, and this breaks on middle age distros;  older distros don't have it
> > at all, new ones have both the syscall and the wrapper; but for the
> > middle age ones they have the syscall but not the wrapper.
> >
> > Dan: What's your preference here; should I add a config fragment to
> > detect the wrapper - it seems overkill rather than just reverting it
> > until it becomes common.
> 
> What specific middle age distro in particular is affected ? My general
> thought would be to /not/ support such distros. Focus on modern distros
> since this is a brand new feature in QEMU, where we should try to
> minimize support for legacy stuff at the start. But depending on the
> distro impacted, the might be a reason to stay with SYS_..

The report came from Ubuntu 18.04 (which Intel uses on CI); that's not
that old, so I think it sohuld be supported.   I don't really see the
justification for insisting on using the wrapper.

Dave

> Regards,
> Daniel
> -- 
> |: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
> |: https://libvirt.org         -o-            https://fstop138.berrange.com :|
> |: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK