[bpf-next] bpf: Fix trampoline usage in preempt

Though the second half of trampoline page is unused a task could be
preempted in the middle of the first half of trampoline and two
updates to trampoline would change the code from underneath the
preempted task. Hence wait for tasks to voluntarily schedule or go
to userspace.
Add similar wait before freeing the trampoline.

Fixes: fec56f5890d9 ("bpf: Introduce BPF trampoline")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
---
 kernel/bpf/trampoline.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

On Mon, Jan 20, 2020 at 07:22:31PM -0800, Alexei Starovoitov wrote:
> Though the second half of trampoline page is unused a task could be
> preempted in the middle of the first half of trampoline and two
> updates to trampoline would change the code from underneath the
> preempted task. Hence wait for tasks to voluntarily schedule or go
> to userspace.
> Add similar wait before freeing the trampoline.
> 
> Fixes: fec56f5890d9 ("bpf: Introduce BPF trampoline")
> Reported-by: Jann Horn <jannh@google.com>
> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
> ---
>  kernel/bpf/trampoline.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)
> 
> diff --git a/kernel/bpf/trampoline.c b/kernel/bpf/trampoline.c
> index 79a04417050d..7657ede7aee2 100644
> --- a/kernel/bpf/trampoline.c
> +++ b/kernel/bpf/trampoline.c
> @@ -160,6 +160,14 @@ static int bpf_trampoline_update(struct bpf_trampoline *tr)
>  	if (fexit_cnt)
>  		flags = BPF_TRAMP_F_CALL_ORIG | BPF_TRAMP_F_SKIP_FRAME;
>  
> +	/* Though the second half of trampoline page is unused a task could be
> +	 * preempted in the middle of the first half of trampoline and two
> +	 * updates to trampoline would change the code from underneath the
> +	 * preempted task. Hence wait for tasks to voluntarily schedule or go
> +	 * to userspace.
> +	 */
> +	synchronize_rcu_tasks();

So in this case, although the trampoline is not freed, it is reused.
And we need to clear everyone off of the trampoline before reusing it.

If this states the situation correctly:

Acked-by: Paul E. McKenney <paulmck@kernel.org>

>  	err = arch_prepare_bpf_trampoline(new_image, new_image + PAGE_SIZE / 2,
>  					  &tr->func.model, flags,
>  					  fentry, fentry_cnt,
> @@ -251,6 +259,8 @@ void bpf_trampoline_put(struct bpf_trampoline *tr)
>  		goto out;
>  	if (WARN_ON_ONCE(!hlist_empty(&tr->progs_hlist[BPF_TRAMP_FEXIT])))
>  		goto out;
> +	/* wait for tasks to get out of trampoline before freeing it */
> +	synchronize_rcu_tasks();
>  	bpf_jit_free_exec(tr->image);
>  	hlist_del(&tr->hlist);
>  	kfree(tr);
> -- 
> 2.23.0
>

On 1/21/20 4:22 AM, Alexei Starovoitov wrote:
> Though the second half of trampoline page is unused a task could be
> preempted in the middle of the first half of trampoline and two
> updates to trampoline would change the code from underneath the
> preempted task. Hence wait for tasks to voluntarily schedule or go
> to userspace.
> Add similar wait before freeing the trampoline.
> 
> Fixes: fec56f5890d9 ("bpf: Introduce BPF trampoline")
> Reported-by: Jann Horn <jannh@google.com>
> Signed-off-by: Alexei Starovoitov <ast@kernel.org>

Applied, thanks!