Re: [PATCH] [REGRESSION] ovl: Handle ENOSYS when fileattr support is missing in lower/upper fs

[Miklos Szeredi]

On Tue, 7 Mar 2023 at 18:14, Jonathan Katz <jkatz@xxxxxxxxxxxx> wrote:
>
> On Tue, Mar 7, 2023 at 12:38 AM Miklos Szeredi <miklos@xxxxxxxxxx> wrote:
> >
> > On Tue, 7 Mar 2023 at 02:12, Jonathan Katz <jkatz@xxxxxxxxxxxx> wrote:
> > >
> > > Hi all,
> > >
> > > In pursuing this issue, I downloaded the kernel source to see if I
> > > could debug it further.  In so doing, it looks like Christian's patch
> > > was never committed to the main source tree (sorry if my terminology
> > > is wrong).  This is up to and including the 6.3-rc1.  I could also
> > > find no mention of the fix in the log.
> > >
> > > I am trying to manually apply this patch now, but, I am wondering if
> > > there was some reason that it was not applied (e.g. it introduces some
> > > instability?)?
> >
> > It's fixing the bug in the wrong place, i.e. it's checking for an
> > -ENOSYS return from vfs_fileattr_get(), but that return value is not
> > valid at that point.
> >
> > The right way to fix this bug is to prevent -ENOSYS from being
> > returned in the first place.
> >
> > Commit 02c0cab8e734 ("fuse: ioctl: translate ENOSYS") fixes one of
> > those bugs, but of course it's possible that I missed something in
> > that fix.
> >
> > Can you please first verify that an upstream kernel (>v6.0) can also
> > reproduce this issue?
>
> Got ya.  that makes a lot of sense, thank you.
>
> I have confirmed that I continue to get the error with 6.2 .
> quick summary of the lowerdir:
>    server ---- NFS(ro) ---- > client "/nfs"
>    client "/nfs" --- bindfs(uidmap) --- > client "/lower"

Can you please run bindfs in debugging mode (-d) and send the
resulting log after reproducing the issue?

Thanks,
Miklos