Re: [RFC] Rearranging struct dentry for cache affinity

[Alexander Viro]

On Tue, 16 Jul 2002, Rusty Russell wrote:

> On Sat, 13 Jul 2002 05:11:46 -0400 (EDT)
> Alexander Viro <[e-mail]> wrote:
> 
> > futex.c is seriously b0rken.
> 
> Really?  Other than changing over to get_sb_psuedo(), what does your patch
> fix?  As the filesystem should never be unmounted, what am I missing?
> >  	filp->f_op = &futex_fops;
> > -	filp->f_dentry = dget(futex_dentry);
> > +	filp->f_vfsmnt = mntget(futex_mnt);
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> > +	filp->f_dentry = dget(futex_mnt->mnt_root);

Uninitialized ->f_vfsmnt == quite a few places in the tree very unhappy.
E.g. any access to /proc/<pid>/fd/<n>.

I'd say that "any user can oops the kernel in a couple of lines" does
qualify for serious breakage...