Re: [PATCH v1 02/11] VFS permit cross device vfs_copy_file_range

[Olga Kornievskaia <olga.kornievskaia@gmail.com>]

On Fri, Oct 19, 2018 at 12:14 PM Trond Myklebust
<trondmy@hammerspace.com> wrote:
>
> On Fri, 2018-10-19 at 11:29 -0400, Olga Kornievskaia wrote:
> > From: Olga Kornievskaia <kolga@netapp.com>
> >
> > Allow copy_file_range to copy between different superblocks but only
> > of the same file system types. This feature was of interest to CIFS
> > as well as NFS.
> >
> > This feature is needed by NFSv4.2 to perform file copy operation on
> > the same server or file copy between different NFSv4.2 servers.
> >
> > If a file system's fileoperations copy_file_range operation prohibits
> > cross-device copies, fall back to do_splice_direct. This would be
> > needed for the NFS (destination) server side implementation of the
> > file copy and currently for CIFS.
> >
> > Besides NFS, there is only 1 implementor of the copy_file_range FS
> > operation -- CIFS. CIFS assumes incoming file descriptors are both
> > CIFS but it will check if they are coming from different servers and
> > return error code to fall back to do_splice_direct.
> >
> > NFS will allow for copies between different NFS servers.
> >
> > Adding to the vfs.txt documentation to explicitly warn about allowing
> > for different superblocks of the same file type to be passed into the
> > copy_file_range for the future users of copy_file_range method.
> >
> > Signed-off-by: Olga Kornievskaia <kolga@netapp.com>
> > ---
> >  Documentation/filesystems/vfs.txt |  4 +++-
> >  fs/read_write.c                   | 13 ++++++-------
> >  2 files changed, 9 insertions(+), 8 deletions(-)
> >
> > diff --git a/Documentation/filesystems/vfs.txt
> > b/Documentation/filesystems/vfs.txt
> > index a6c6a8a..5e520de 100644
> > --- a/Documentation/filesystems/vfs.txt
> > +++ b/Documentation/filesystems/vfs.txt
> > @@ -958,7 +958,9 @@ otherwise noted.
> >
> >    fallocate: called by the VFS to preallocate blocks or punch a
> > hole.
> >
> > -  copy_file_range: called by the copy_file_range(2) system call.
> > +  copy_file_range: called by copy_file_range(2) system call. This
> > method
> > +                works on two file descriptors that might reside on
> > +                different superblocks of the same type of file
> > system.
> >
> >    clone_file_range: called by the ioctl(2) system call for
> > FICLONERANGE and
> >       FICLONE commands.
> > diff --git a/fs/read_write.c b/fs/read_write.c
> > index c60790f..474e740 100644
> > --- a/fs/read_write.c
> > +++ b/fs/read_write.c
> > @@ -1578,10 +1578,6 @@ ssize_t vfs_copy_file_range(struct file
> > *file_in, loff_t pos_in,
> >           (file_out->f_flags & O_APPEND))
> >               return -EBADF;
> >
> > -     /* this could be relaxed once a method supports cross-fs copies
> > */
> > -     if (inode_in->i_sb != inode_out->i_sb)
> > -             return -EXDEV;
> > -
> >       if (len == 0)
> >               return 0;
> >
> > @@ -1591,7 +1587,8 @@ ssize_t vfs_copy_file_range(struct file
> > *file_in, loff_t pos_in,
> >        * Try cloning first, this is supported by more file systems,
> > and
> >        * more efficient if both clone and copy are supported (e.g.
> > NFS).
> >        */
> > -     if (file_in->f_op->clone_file_range) {
> > +     if (inode_in->i_sb == inode_out->i_sb &&
> > +                     file_in->f_op->clone_file_range) {
> >               ret = file_in->f_op->clone_file_range(file_in, pos_in,
> >                               file_out, pos_out, len);
> >               if (ret == 0) {
> > @@ -1600,10 +1597,12 @@ ssize_t vfs_copy_file_range(struct file
> > *file_in, loff_t pos_in,
> >               }
> >       }
> >
> > -     if (file_out->f_op->copy_file_range) {
> > +     if (file_out->f_op->copy_file_range &&
> > +                     (file_in->f_op->copy_file_range ==
> > +                             file_out->f_op->copy_file_range)) {
> >               ret = file_out->f_op->copy_file_range(file_in, pos_in,
> > file_out,
> >                                                     pos_out, len,
> > flags);
> > -             if (ret != -EOPNOTSUPP)
> > +             if (ret != -EOPNOTSUPP && ret != -EXDEV)
> >                       goto done;
> >       }
> >
>
> Ditto.  This also needs an ACK from the VFS maintainers.
>
> Cc: Al and linux-fsdevel

Yeah I sent VFS as separate patches to the linux-fsdevel and included
other folks (glibc/CIFS?) that were interested in this functionality.
Apologizes for double sent. It was easier to send this as a patch
series to just linux-nfs first.

> --
> Trond Myklebust
> Linux NFS client maintainer, Hammerspace
> trond.myklebust@hammerspace.com
>
>