Re: [PATCH v2] vfs: prevent copy_file_range to copy across devices

[Amir Goldstein]

On Mon, Feb 15, 2021 at 6:53 PM Trond Myklebust <trondmy@xxxxxxxxxxxxxxx> wrote:
>
> On Mon, 2021-02-15 at 18:34 +0200, Amir Goldstein wrote:
> > On Mon, Feb 15, 2021 at 5:42 PM Luis Henriques <lhenriques@xxxxxxx>
> > wrote:
> > >
> > > Nicolas Boichat reported an issue when trying to use the
> > > copy_file_range
> > > syscall on a tracefs file.  It failed silently because the file
> > > content is
> > > generated on-the-fly (reporting a size of zero) and copy_file_range
> > > needs
> > > to know in advance how much data is present.
> > >
> > > This commit restores the cross-fs restrictions that existed prior
> > > to
> > > 5dae222a5ff0 ("vfs: allow copy_file_range to copy across devices")
> > > and
> > > removes generic_copy_file_range() calls from ceph, cifs, fuse, and
> > > nfs.
> > >
> > > Fixes: 5dae222a5ff0 ("vfs: allow copy_file_range to copy across
> > > devices")
> > > Link:
> > > https://lore.kernel.org/linux-fsdevel/20210212044405.4120619-1-drinkcat@xxxxxxxxxxxx/
> > > Cc: Nicolas Boichat <drinkcat@xxxxxxxxxxxx>
> > > Signed-off-by: Luis Henriques <lhenriques@xxxxxxx>
> >
> > Code looks ok.
> > You may add:
> >
> > Reviewed-by: Amir Goldstein <amir73il@xxxxxxxxx>
> >
> > I agree with Trond that the first paragraph of the commit message
> > could
> > be improved.
> > The purpose of this change is to fix the change of behavior that
> > caused the regression.
> >
> > Before v5.3, behavior was -EXDEV and userspace could fallback to
> > read.
> > After v5.3, behavior is zero size copy.
> >
> > It does not matter so much what makes sense for CFR to do in this
> > case (generic cross-fs copy).  What matters is that nobody asked for
> > this change and that it caused problems.
> >
>
> No. I'm saying that this patch should be NACKed unless there is a real
> explanation for why we give crap about this tracefs corner case and why
> it can't be fixed.
>
> There are plenty of reasons why copy offload across filesystems makes
> sense, and particularly when you're doing NAS. Clone just doesn't cut
> it when it comes to disaster recovery (whereas backup to a different
> storage unit does). If the client has to do the copy, then you're
> effectively doubling the load on the server, and you're adding
> potentially unnecessary network traffic (or at the very least you are
> doubling that traffic).
>

I don't understand the use case you are describing.

Which filesystem types are you talking about for source and target
of copy_file_range()?

To be clear, the original change was done to support NFS/CIFS server-side
copy and those should not be affected by this change.

Thanks,
Amir.