On Sat, Feb 19, 2022 at 05:58:13AM +0000, Al Viro wrote:
> On Sat, Feb 19, 2022 at 05:53:37AM +0000, Al Viro wrote:
> > On Fri, Feb 18, 2022 at 01:31:13PM -0500, Rik van Riel wrote:
> >
> > > struct super_block;
> > > struct vfsmount;
> > > @@ -73,6 +74,7 @@ struct vfsmount {
> > > struct super_block *mnt_sb; /* pointer to superblock */
> > > int mnt_flags;
> > > struct user_namespace *mnt_userns;
> > > + struct rcu_work free_rwork;
> > > } __randomize_layout;
> >
> > Wait, what? First of all, that has no business being in vfsmount -
> > everything that deeply internal belongs in struct mount, not in
> > its public part. Moreover, there's already mount->mnt_rcu, so what's
> > the point duplicating that?
>
> Argh... You need rcu_work there...
>
> OK, so make that a member of the same union mnt_rcu is. In struct mount,
> please. And I'm not sure I like the idea of shoving that much into
> struct mount, TBH...
We might have a plenty of struct mount instances. Very few of them will
ever be internal, in the first place. Fewer yet - using kern_unmount_rcu().
And it's not small. If anything, I would consider something like
call_rcu(&m->mnt_rcu, callback)
with callback adding struct mount to llist and doing schedule_delayed_work()
on that. With work consisting of doing mntput on everything in it.
I'll get some sleep and put together something along those lines tomorrow
morning.
