Re: [PATCH bpf-next] bpf: Disable migrate when kprobe_multi attach to access bpf_prog_active

[Tao Chen <chen.dylane@linux.dev>]

在 2025/8/5 17:07, Jiri Olsa 写道:
> On Mon, Aug 04, 2025 at 10:15:46PM +0800, Tao Chen wrote:
>> 在 2025/8/4 21:02, Jiri Olsa 写道:
>>> On Mon, Aug 04, 2025 at 08:16:15PM +0800, Tao Chen wrote:
>>>> The syscall link_create not protected by bpf_disable_instrumentation,
>>>> accessing percpu data bpf_prog_active should use cpu local_lock when
>>>> kprobe_multi program attach.
>>>>
>>>> Fixes: 0dcac2725406 ("bpf: Add multi kprobe link")
>>>> Signed-off-by: Tao Chen <chen.dylane@linux.dev>
>>>> ---
>>>>    kernel/trace/bpf_trace.c | 4 ++--
>>>>    1 file changed, 2 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
>>>> index 3ae52978cae..f6762552e8e 100644
>>>> --- a/kernel/trace/bpf_trace.c
>>>> +++ b/kernel/trace/bpf_trace.c
>>>> @@ -2728,23 +2728,23 @@ kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link,
>>>>    	struct pt_regs *regs;
>>>>    	int err;
>>>> +	migrate_disable();
>>>>    	if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
>>>
>>> this is called all the way from graph tracer, which disables preemption in
>>> function_graph_enter_regs, so I think we can safely use __this_cpu_inc_return
>>>
>>>
>>>>    		bpf_prog_inc_misses_counter(link->link.prog);
>>>>    		err = 1;
>>>>    		goto out;
>>>>    	}
>>>> -	migrate_disable();
>>>
>>> hum, but now I'm not sure why we disable migration in here then
>>>
>>
>> It seems that there is a cant_migrate() check in bpf_prog_run, so it should
>> be disabled before run.
> 
> yes, but disabled preemption will take care of that
> 

I see, you are right, preempt will pass the check, thanks.

void __cant_migrate(const char *file, int line)
{
         static unsigned long prev_jiffy;

         if (irqs_disabled())
                 return;

         if (is_migration_disabled(current))
                 return;

         if (!IS_ENABLED(CONFIG_PREEMPT_COUNT))
                 return;

         if (preempt_count() > 0)
                 return;
...

> I think we can do change below plus some comment that Yonghong
> is suggesting in the other reply
> 

Yes, i will remove the migrate_disable and add some comment as you
and Yonghong suggested.

> jirka
> 
> 
> ---
> diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
> index 3ae52978cae6..74e8d9543c6d 100644
> --- a/kernel/trace/bpf_trace.c
> +++ b/kernel/trace/bpf_trace.c
> @@ -2734,14 +2734,12 @@ kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link,
>   		goto out;
>   	}
>   
> -	migrate_disable();
>   	rcu_read_lock();
>   	regs = ftrace_partial_regs(fregs, bpf_kprobe_multi_pt_regs_ptr());
>   	old_run_ctx = bpf_set_run_ctx(&run_ctx.session_ctx.run_ctx);
>   	err = bpf_prog_run(link->link.prog, regs);
>   	bpf_reset_run_ctx(old_run_ctx);
>   	rcu_read_unlock();
> -	migrate_enable();
>   
>    out:
>   	__this_cpu_dec(bpf_prog_active);
-- 
Best Regards
Tao Chen