On Tue, Sep 2, 2025 at 6:55 PM Gao Xiang <hsiangkao@xxxxxxxxxxxxxxxxx> wrote:
>
> On 2025/9/3 05:24, Joanne Koong wrote:
> > On Fri, Aug 29, 2025 at 6:54 PM Gao Xiang <xiang@xxxxxxxxxx> wrote:
> >>
> >> Hi Joanne,
> >>
> >> On Fri, Aug 29, 2025 at 04:56:24PM -0700, Joanne Koong wrote:
> >>> Add a void *private arg for read and readahead which filesystems that
> >>> pass in custom read callbacks can use. Stash this in the existing
> >>> private field in the iomap_iter.
> >>>
> >>> Signed-off-by: Joanne Koong <joannelkoong@xxxxxxxxx>
> >>> ---
> >>> block/fops.c | 4 ++--
> >>> fs/erofs/data.c | 4 ++--
> >>> fs/gfs2/aops.c | 4 ++--
> >>> fs/iomap/buffered-io.c | 8 ++++++--
> >>> fs/xfs/xfs_aops.c | 4 ++--
> >>> fs/zonefs/file.c | 4 ++--
> >>> include/linux/iomap.h | 4 ++--
> >>> 7 files changed, 18 insertions(+), 14 deletions(-)
> >>>
> >>
> >> ...
> >>
> >>> int iomap_read_folio(struct folio *folio, const struct iomap_ops *ops,
> >>> - const struct iomap_read_ops *read_ops)
> >>> + const struct iomap_read_ops *read_ops, void *private)
> >>> {
> >>> struct iomap_iter iter = {
> >>> .inode = folio->mapping->host,
> >>> .pos = folio_pos(folio),
> >>> .len = folio_size(folio),
> >>> + .private = private,
> >>> };
> >>
> >> Will this whole work be landed for v6.18?
> >>
> >> If not, may I ask if this patch can be shifted advance in this
> >> patchset for applying separately (I tried but no luck).
> >>
> >> Because I also need some similar approach for EROFS iomap page
> >> cache sharing feature since EROFS uncompressed I/Os go through
> >> iomap and extra information needs a proper way to pass down to
> >> iomap_{begin,end} with extra pointer `.private` too.
> >
> > Hi Gao,
> >
> > I'm not sure whether this will be landed for v6.18 but I'm happy to
> > shift this patch to the beginning of the patchset for applying
> > separately.
>
> Yeah, thanks. At least this common patch can be potentially applied
> easily (e.g. form a common commit id for both features if really
> needed) since other iomap/FUSE patches are not dependency of our new
> feature and shouldn't be coupled with our development branch later.
>
Hi Gao,
I'll be dropping this patch in v2 since all the iomap read stuff is
going to go through a struct ctx arg instead of through iter->private.
Sorry this won't help your use case, but looking forward to seeing your patches.
Thanks,
Joanne
> Thanks,
> Gao Xiang
>
> >
> > Thanks,
> > Joanne
> >>
> >> Thanks,
> >> Gao Xiang
>
