On Thu, Apr 10, 2025 at 11:01 AM Christoph Hellwig <hch@xxxxxxxxxxxxx> wrote:
> On Mon, Apr 07, 2025 at 08:21:01PM +0200, Andreas Gruenbacher wrote:
> > Use a dummy inode as mapping->host of the address spaces for global as
> > well as per-inode metadata. The global metadata address space is now
> > accessed as gfs2_aspace(sdp) instead of sdp->sd_aspace. The per-inode
> > metadata address spaces are still accessed as
> > gfs2_glock2aspace(GFS2_I(inode)->i_gl).
> >
> > Based on a previous version from Bob Peterson from several years ago.
>
> Please explain why you are doing this, not just what.
Right, I have this description now:
Currently, sdp->sd_aspace and the per-inode metadata address spaces use
sb->s_bdev->bd_mapping->host as their ->host. Folios in those address
spaces will thus appear to be on "bdev" rather than on "gfs2"
filesystems. Those "bdev" filesystems will have the SB_I_CGROUPWB flag
set to indicate cgroup writeback support. In fact, gfs2 doesn't support
cgroup writeback, though.
To fix that, use a "dummy" gfs2 inode as ->host of those address spaces
instead. This will then allow functions like inode_to_wb() to determine
that the folio belongs to a a filesystem without cgroup writeback
support.
Thanks,
Andreas
