On Sat, Jan 04, 2025 at 01:39:46AM +0000, Pu Lehui wrote:
> From: Pu Lehui <pulehui@xxxxxxxxxx>
>
> 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.
lgtm, adding peter
Reviewed-by: Jiri Olsa <jolsa@xxxxxxxxxx>
jirka
>
> Signed-off-by: Pu Lehui <pulehui@xxxxxxxxxx>
> ---
> v2: Simplify code logic. (Jiri)
>
> kernel/trace/bpf_trace.c | 21 +++++++++++++--------
> 1 file changed, 13 insertions(+), 8 deletions(-)
>
> diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
> index 48db147c6c7d..a90880f475af 100644
> --- a/kernel/trace/bpf_trace.c
> +++ b/kernel/trace/bpf_trace.c
> @@ -2245,6 +2245,7 @@ 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);
> @@ -2265,18 +2266,22 @@ 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)
> --
> 2.34.1
>
>
