[PATCH 06/14] audit: Factor out chunk replacement code

[Paul Moore]

On Wed, Nov 7, 2018 at 4:55 AM Jan Kara <jack at suse.cz> wrote:
> On Tue 06-11-18 08:58:36, Paul Moore wrote:
> > On Thu, Oct 18, 2018 at 3:27 PM Richard Guy Briggs <rgb at redhat.com> wrote:
> > > On 2018-10-17 12:14, Jan Kara wrote:
> > > > Chunk replacement code is very similar for the cases where we grow or
> > > > shrink chunk. Factor the code out into a common helper function.
> > >
> > > Noting just the switch from list_replace_init() to list_splice_init().
> >
> > Yeah, I wasn't expecting to see that, maybe it will make sense as I
> > work through the rest of the patchset.
> >
> > Jan, can you explain the reason behind the change?  I'm a little
> > nervous that this is simply hiding a problem (e.g. the list_empty()
> > check in splice).
>
> The reason is very simple: replace_chunk() gets called from tag_chunk()
> *after* we have possibly done:
>
>         if (!tree->root) {
>                 tree->root = chunk;
>                 list_add(&tree->same_root, &chunk->trees);
>         }
>
> So new->trees is possibly non-empty and we need to preserve its contents.
> That's why we need list_splice() and not plain list_replace().

After revisiting this a couple of time this week, I found my problem -
I had reversed list_splice_init() in my mind and was worried that if
the old->trees list was not empty we would end up not adding
new->trees.  Sorry about the noise.

I've updated the audit/working-fsnotify_fixes with the latest patches,
including the revised 12/14 sent as an attachment, and I'm going to
test it over the weekend.  If the kernel is still standing on Monday
morning I'll merge it into audit/next.

Thanks again.

-- 
paul moore
www.paul-moore.com