Re: [PATCH 07/31] fuse: create a per-inode flag for toggling iomap

["Darrick J. Wong" <djwong@kernel.org>]

On Sat, Jan 24, 2026 at 08:54:30AM -0800, Darrick J. Wong wrote:
> On Fri, Jan 23, 2026 at 10:05:32AM -0800, Joanne Koong wrote:
> > On Thu, Jan 22, 2026 at 2:22 PM Darrick J. Wong <djwong@kernel.org> wrote:
> > >
> > > On Wed, Jan 21, 2026 at 05:13:39PM -0800, Joanne Koong wrote:
> > > > On Tue, Oct 28, 2025 at 5:46 PM Darrick J. Wong <djwong@kernel.org> wrote:
> > > > >
> > > > > From: Darrick J. Wong <djwong@kernel.org>
> > > > >
> > > > > Create a per-inode flag to control whether or not this inode actually
> > > > > uses iomap.  This is required for non-regular files because iomap
> > > > > doesn't apply there; and enables fuse filesystems to provide some
> > > > > non-iomap files if desired.
> > > > >
> > > > > Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
> > > >
> > > > The logic in this makes sense to me, left just a few comments below.
> > > >
> > > > Reviewed-by: Joanne Koong <joannelkoong@gmail.com>
> > >
> > > Thanks!
> > >
> > > > > ---
> > > > >  fs/fuse/fuse_i.h          |   17 ++++++++++++++++
> > > > >  include/uapi/linux/fuse.h |    3 +++
> > > > >  fs/fuse/file.c            |    1 +
> > > > >  fs/fuse/file_iomap.c      |   49 +++++++++++++++++++++++++++++++++++++++++++++
> > > > >  fs/fuse/inode.c           |   26 ++++++++++++++++++------
> > > > >  5 files changed, 90 insertions(+), 6 deletions(-)
> > > > >
> > > > > diff --git a/fs/fuse/file.c b/fs/fuse/file.c
> > > > > index f1ef77a0be05bb..42c85c19f3b13b 100644
> > > > > --- a/fs/fuse/file.c
> > > > > +++ b/fs/fuse/file.c
> > > > > +void fuse_iomap_init_reg_inode(struct inode *inode, unsigned attr_flags)
> > > > > +{
> > > > > +       struct fuse_conn *conn = get_fuse_conn(inode);
> > > > > +       struct fuse_inode *fi = get_fuse_inode(inode);
> > > > > +
> > > > > +       ASSERT(S_ISREG(inode->i_mode));
> > > > > +
> > > > > +       if (conn->iomap && (attr_flags & FUSE_ATTR_IOMAP)) {
> > > > > +               set_bit(FUSE_I_EXCLUSIVE, &fi->state);
> > > > > +               fuse_inode_set_iomap(inode);
> > > > > +       }
> > > > > +}
> > > > > +
> > > > > +void fuse_iomap_evict_inode(struct inode *inode)
> > > > > +{
> > > > > +       struct fuse_conn *conn = get_fuse_conn(inode);
> > > > > +       struct fuse_inode *fi = get_fuse_inode(inode);
> > > > > +
> > > > > +       if (fuse_inode_has_iomap(inode))
> > > >
> > > > If I'm understanding this correctly, a fuse inode can't have
> > > > FUSE_I_IOMAP set on it if conn>iomap is not enabled, correct?
> > >
> > > Correct.
> > >
> > > > Maybe it makes sense to just return if (!conn->iomap) at the very
> > > > beginning, to make that more clear?
> > >
> > > <shrug> fuse_inode_has_iomap only checks FUSE_I_IOMAP...
> > >
> > > > > +               fuse_inode_clear_iomap(inode);
> > > > > +       if (conn->iomap && fuse_inode_is_exclusive(inode))
> > > > > +               clear_bit(FUSE_I_EXCLUSIVE, &fi->state);
> > >
> > > ...but I wasn't going to assume that iomap is the only way that
> > > FUSE_I_EXCLUSIVE could get set.
> > >
> > > On the other hand, for non-regular files we set FUSE_I_EXCLUSIVE only if
> > > conn->iomap is nonzero *and* attr->flags contains FUSE_ATTR_IOMAP.  So
> > > this clearing code isn't quite the same as the setting code.
> > >
> > > I wonder if that means we should set FUSE_I_IOMAP for non-regular files?
> > > They don't use iomap itself, but I suppose it would be neat if "iomap
> > > directories" also meant that timestamps and whatnot worked in the same
> > > as they do for regular files.
> > >
> > 
> > That seems like a good idea to me. I think that also makes the mental
> > model (at least for me) simpler.
> 
> I tried that, and generic/476 immediately broke.  I'll get back to that
> next week, but turning it on unconditionally is not trivial
> unfortunately. :/

I've tentatively fixed this by defining a FUSE_ATTR_EXCLUSIVE flag in
the uapi so that the fuse server can tell the kernel which files are
"exclusive" files, and hence which ones should have FUSE_I_EXCLUSIVE
set.

(These are ofc files where the kernel can transmogrify ACLs into i_mode
changes and do ACL inheritance because there is no other principal that
could be writing to the ondisk metadata.)

--D

> --D
> 
> > Thanks,
> > Joanne
>