Re: [PATCH bpf-next] bpf: Move out synchronize_rcu_tasks_trace from mutex CS

[Jiri Olsa <olsajiri@gmail.com>]

On Tue, Dec 31, 2024 at 03:35:09AM +0000, Pu Lehui wrote:
> From: Pu Lehui <pulehui@huawei.com>
> 
> Commit ef1b808e3b7c ("bpf: Fix UAF via mismatching bpf_prog/attachment
> RCU flavors") resolved a possible UAF issue in uprobes that attach
> non-sleepable bpf prog by explicitly waiting for a tasks-trace-RCU grace
> period. But, in the current implementation, synchronize_rcu_tasks_trace
> is included within the mutex critical section, which increases the
> length of the critical section and may affect performance. So let's move
> out synchronize_rcu_tasks_trace from mutex CS.
> 
> Signed-off-by: Pu Lehui <pulehui@huawei.com>
> ---
>  kernel/trace/bpf_trace.c | 18 +++++++++++-------
>  1 file changed, 11 insertions(+), 7 deletions(-)
> 
> diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
> index 48db147c6c7d..30ef8a6f5ca2 100644
> --- a/kernel/trace/bpf_trace.c
> +++ b/kernel/trace/bpf_trace.c
> @@ -2245,12 +2245,15 @@ void perf_event_detach_bpf_prog(struct perf_event *event)
>  {
>  	struct bpf_prog_array *old_array;
>  	struct bpf_prog_array *new_array;
> +	struct bpf_prog *prog;
>  	int ret;
>  
>  	mutex_lock(&bpf_event_mutex);
>  
> -	if (!event->prog)
> -		goto unlock;
> +	if (!event->prog) {
> +		mutex_unlock(&bpf_event_mutex);
> +		return;
> +	}
>  
>  	old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
>  	if (!old_array)
> @@ -2265,6 +2268,11 @@ void perf_event_detach_bpf_prog(struct perf_event *event)
>  	}
>  
>  put:
> +	prog = event->prog;
> +	event->prog = NULL;
> +
> +	mutex_unlock(&bpf_event_mutex);
> +
>  	/*
>  	 * It could be that the bpf_prog is not sleepable (and will be freed
>  	 * via normal RCU), but is called from a point that supports sleepable
> @@ -2272,11 +2280,7 @@ void perf_event_detach_bpf_prog(struct perf_event *event)
>  	 */
>  	synchronize_rcu_tasks_trace();
>  
> -	bpf_prog_put(event->prog);
> -	event->prog = NULL;
> -
> -unlock:
> -	mutex_unlock(&bpf_event_mutex);
> +	bpf_prog_put(prog);
>  }
>  
>  int perf_event_query_prog_array(struct perf_event *event, void __user *info)
> -- 
> 2.34.1
> 

would something like below be simpler? (not tested)

jirka


---
diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
index 973104f861e9..a4c0efa3a26e 100644
--- a/kernel/trace/bpf_trace.c
+++ b/kernel/trace/bpf_trace.c
@@ -2246,6 +2246,7 @@ void perf_event_detach_bpf_prog(struct perf_event *event)
 {
 	struct bpf_prog_array *old_array;
 	struct bpf_prog_array *new_array;
+	struct bpf_prog *prog = NULL;
 	int ret;
 
 	mutex_lock(&bpf_event_mutex);
@@ -2266,18 +2267,22 @@ void perf_event_detach_bpf_prog(struct perf_event *event)
 	}
 
 put:
-	/*
-	 * It could be that the bpf_prog is not sleepable (and will be freed
-	 * via normal RCU), but is called from a point that supports sleepable
-	 * programs and uses tasks-trace-RCU.
-	 */
-	synchronize_rcu_tasks_trace();
-
-	bpf_prog_put(event->prog);
+	prog = event->prog;
 	event->prog = NULL;
 
 unlock:
 	mutex_unlock(&bpf_event_mutex);
+
+	if (prog) {
+		/*
+		 * It could be that the bpf_prog is not sleepable (and will be freed
+		 * via normal RCU), but is called from a point that supports sleepable
+		 * programs and uses tasks-trace-RCU.
+		 */
+		synchronize_rcu_tasks_trace();
+
+		bpf_prog_put(prog);
+	}
 }
 
 int perf_event_query_prog_array(struct perf_event *event, void __user *info)