[OpenAFS] kernel 2.6.25 and openafs-1.4.7pre3

William John Murray w.j.murray@rl.ac.uk
Tue, 22 Apr 2008 10:33:09 +0100


  Thanks Simon,
        Well, I think that has put me off...sorry.
                Bill

On Tue, 2008-04-22 at 10:30 +0100, Simon Wilkinson wrote:
> On 22 Apr 2008, at 10:16, William Murray wrote:
> >
> >   Hello all,
> >         I am trying to compile AFS for Fedora 9 pre, with a 2.6.25
> > kernel.
> >    osi_groups.c:54: error: =C3=A2=E2=82=AC=CB=9Cstruct task_struct=C3=
=A2=E2=82=AC=E2=84=A2 has no member =20
> > named
> > =C3=A2=E2=82=AC=CB=9Cparent=C3=A2=E2=82=AC=E2=84=A2
> >   This 'parent' member was only inserted last year, but it seems to =20
> > have
> > gone.
> >    Has anyone already patched for this kernel?
>=20
> Fedora 9pre and OpenAFS don't currently play well together. Here's a =20
> quick summary of where we are:
>=20
> *) The dependency structure of the kernel RPMs has changed - Derrick =20
> has patches for the build system and spec file to fix this
> *) struct task_struct doesn't have a 'parent' element. This is due to =20
> a RedHat only patch which gets rid of ptrace. A configure test for =20
> this is now in CVS
> *) rcu_lock_map is no longer exported, so we have to use a weak =20
> reference to determine key_type_keyring. A fix for this is in CVS
> *) Setting tokens crashes. gcc seems to miscompile the picotl =20
> handling code, possibly due to the lack of prototypes. I have a patch =20
> which prototypes the relevant code, and gets this working again
> *) Occasionally, the root vnode seems to have a corrupt pointer to =20
> the Access structure. As yet, unfixed.
> *) Locks seem to be being left dangling, such that ls /afs will only =20
> run once. Again, I'm still investigating this.
>=20
> Cheers,
>=20
> Simon.
>=20