Re: [PATCH] kprobe: use preempt_{disable, enable}_notrace in kprobe_busy_{begin, end}

[Google]

Hi,

On Sat, 13 May 2023 17:05:48 +0800
Ze Gao <zegao2021@xxxxxxxxx> wrote:

> Replacing preempt_{disable, enable} with its corresponding notrace
> version in kprobe_busy_{begin, end} because they are called by
> fprobe_kprobe_handler. Such changes would resolve the potential
> stack recursion or overflow problem introduced by preempt_count
> _{add, sub} when these functions themselves accidentally traced
> by fprobe+rethook, and does not break the original semantics.

As I explained below;

https://lore.kernel.org/all/20230517010311.f46db3f78b11cf9d92193527@xxxxxxxxxx/

we don't need this for kprobes because kprobes already prohibits
probing on preempt_count_{add,sub}, and we don't want to miss
tracing the preempt-off period by kprobes itself.

Thank you,

> 
> Signed-off-by: Ze Gao <zegao@xxxxxxxxxxx>
> ---
>  kernel/kprobes.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/kprobes.c b/kernel/kprobes.c
> index 00e177de91cc..94d01b5d0646 100644
> --- a/kernel/kprobes.c
> +++ b/kernel/kprobes.c
> @@ -1242,7 +1242,7 @@ void kprobe_busy_begin(void)
>  {
>  	struct kprobe_ctlblk *kcb;
>  
> -	preempt_disable();
> +	preempt_disable_notrace();
>  	__this_cpu_write(current_kprobe, &kprobe_busy);
>  	kcb = get_kprobe_ctlblk();
>  	kcb->kprobe_status = KPROBE_HIT_ACTIVE;
> @@ -1251,7 +1251,7 @@ void kprobe_busy_begin(void)
>  void kprobe_busy_end(void)
>  {
>  	__this_cpu_write(current_kprobe, NULL);
> -	preempt_enable();
> +	preempt_enable_notrace();
>  }
>  
>  /* Add the new probe to 'ap->list'. */
> -- 
> 2.40.1
> 


-- 
Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>