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

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Thu, Feb 18, 2021 at 2:14 PM Luis Henriques <lhenriques@xxxxxxx> wrote:
>
> Luis Henriques <lhenriques@xxxxxxx> writes:
>
> > Amir Goldstein <amir73il@xxxxxxxxx> writes:
> >
> >> On Thu, Feb 18, 2021 at 9:42 AM Christoph Hellwig <hch@xxxxxxxxxxxxx> wrote:
> >>>
> >>> Looks good:
> >>>
> >>> Reviewed-by: Christoph Hellwig <hch@xxxxxx>
> >>>
> >>> This whole idea of cross-device copie has always been a horrible idea,
> >>> and I've been arguing against it since the patches were posted.
> >>
> >> Ok. I'm good with this v2 as well, but need to add the fallback to
> >> do_splice_direct()
> >> in nfsd_copy_file_range(), because this patch breaks it.
> >>
> >> And the commit message of v3 is better in describing the reported issue.
> >
> > Except that, as I said in a previous email, v2 doesn't really fix the
> > issue: all the checks need to be done earlier in generic_copy_file_checks().
> >
> > I'll work on getting v4, based on v2 and but moving the checks and
> > implementing your review suggestions to v3 (plus this nfs change).
>
> There's something else:
>
> The filesystems (nfs, ceph, cifs, fuse) rely on the fallback to
> generic_copy_file_range() if something's wrong.  And this "something's
> wrong" is fs specific.  For example: in ceph it is possible to offload the
> file copy to the OSDs even if the files are in different filesystems as
> long as these filesystems are on the *same* ceph cluster.  If the copy
> being done is across two different clusters, then the copy reverts to
> splice.  This means that the boilerplate code being removed in v2 of this
> patch needs to be restored and replace by:
>
>         ret = __ceph_copy_file_range(src_file, src_off, dst_file, dst_off,
>                                      len, flags);
>
>         if (ret == -EOPNOTSUPP || ret == -EXDEV)
>                 ret = do_splice_direct(src_file, &src_off, dst_file, &dst_off,
>                                        len > MAX_RW_COUNT ? MAX_RW_COUNT : len,
>                                        flags);
>         return ret;
>

Why not leave the filesystem code as is and leave the
generic_copy_file_range() helper? Less churn.

Then nfsd_copy_file_range() can also fallback to generic_copy_file_range().

Thanks,
Amir.



[Index of Archives]     [Linux Filesystem Development]     [Linux USB Development]     [Linux Media Development]     [Video for Linux]     [Linux NILFS]     [Linux Audio Users]     [Yosemite Info]     [Linux SCSI]

  Powered by Linux