Re: [PATCH] xfs: fix unwind handling during xfs_mount_sysfs_init()

[Wilfred Mallawa]

On Tue, 2026-10-27 at 16:50 +0100, Carlos Maiolino wrote:
> On Mon, Oct 26, 2026 at 03:34:55PM +1000, Wilfred Mallawa wrote:
> > From: Wilfred Mallawa <wilfred.mallawa@xxxxxxx>
> > 
> > The goto for failing to setup /error/metadata jumps to the wrong
> > place
> > in xfs_mount_sysfs_init(). Fix it.
> > 
> > Fixes: a9a7b9e8686d ("xfs: add support for lazy direct read bounce
> > buffering")
> > Signed-off-by: Wilfred Mallawa <wilfred.mallawa@xxxxxxx>
> > ---
> >  fs/xfs/xfs_sysfs.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/fs/xfs/xfs_sysfs.c b/fs/xfs/xfs_sysfs.c
> > index e77917ac179d..4a6a8c92c3a1 100644
> > --- a/fs/xfs/xfs_sysfs.c
> > +++ b/fs/xfs/xfs_sysfs.c
> > @@ -906,7 +906,7 @@ xfs_mount_sysfs_init(
> >  				"metadata", &mp-
> > >m_error_meta_kobj,
> >  				xfs_error_meta_init);
> >  	if (error)
> > -		goto out_remove_csum_dir;
> > +		goto out_remove_error_dir;
> 
> Hmm. I think this is wrong. Don't we need to remove the
> fail_at_unmount
> file *before* removing the kobject?
> 
> So, we'd need to actually call sysfs_remove_file() before actually
> deleting the m_error_kobj.
> 
> I am honestly not sure if kobject_del would clean up any attributes
> added there, but I think that's the right thing to do here?!

With the xfs_sysfs_del(m_error_kobj), we end up in sysfs_remove_dir(()
which deletes any subdirs/files. So we *shoudn't* need to explicitly do
a per file remove?


Thanks!