Re: [PATCH] fork: avoid inappropriate uprobe access to invalid mm

From: Oleg Nesterov
Date: Tue Dec 10 2024 - 12:24:12 EST


On 12/10, Oleg Nesterov wrote:
>
> I must have missed something, but...
>
> On 12/10, Lorenzo Stoakes wrote:
> >
> > @@ -1746,9 +1741,11 @@ static struct mm_struct *dup_mm(struct task_struct *tsk,
> > if (!mm_init(mm, tsk, mm->user_ns))
> > goto fail_nomem;
> >
> > + uprobe_start_dup_mmap();
> > err = dup_mmap(mm, oldmm);
> > if (err)
> > - goto free_pt;
> > + goto free_pt_end_uprobe;
> > + uprobe_end_dup_mmap();
> >
> > mm->hiwater_rss = get_mm_rss(mm);
> > mm->hiwater_vm = mm->total_vm;
> > @@ -1758,6 +1755,8 @@ static struct mm_struct *dup_mm(struct task_struct *tsk,
> >
> > return mm;
> >
> > +free_pt_end_uprobe:
> > + uprobe_end_dup_mmap();
>
> if dup_mmap() fails and "mm" is incomplete, then with this version dup_mmap_sem
> is dropped before __mmput/exit_mmap/etc. How can this help?

Easy to fix, but what ensures that another mmget(mm) is not possible until
dup_mm() calls mmput() and drops dup_mmap_sem? Sorry, my understanding of mm/
is very limited...

Oleg.