bpf: smp_wmb before bpf_ringbuf really commit

To guarantee visibility of writing ringbuffer,it is necessary
to call smp_wmb before ringbuffer really commit.
for instance, when updating the data of buffer in cpu1,
it is not visible for the cpu2 which may be accessing buffer. This may
lead to the consumer accessing a incorrect data. using the smp_wmb
before commmit will guarantee that the consume can access the correct data.

CPU1:
    struct mem_event_t* data = bpf_ringbuf_reserve();
    data->type = MEM_EVENT_KSWAPD_WAKE;
    data->event_data.kswapd_wake.node_id = args->nid;
    bpf_ringbuf_commit(data);

CPU2:
    cons_pos = smp_load_acquire(r->consumer_pos);
    len_ptr = r->data + (cons_pos & r->mask);
    sample = (void *)len_ptr + BPF_RINGBUF_HDR_SZ;
    access to sample

Signed-off-by: zhongjinji <[email protected]>

kernel/bpf/ringbuf.c

@@ -508,6 +508,10 @@ static void bpf_ringbuf_commit(void *sample, u64 flags, bool discard)
 	rec_pos = (void *)hdr - (void *)rb->data;
 	cons_pos = smp_load_acquire(&rb->consumer_pos) & rb->mask;
 
+	/* Make sure the modification of data is visible on other CPU's
+	 * before consume the event
+	 */
+	smp_wmb();
 	if (flags & BPF_RB_FORCE_WAKEUP)
 		irq_work_queue(&rb->work);
 	else if (cons_pos == rec_pos && !(flags & BPF_RB_NO_WAKEUP))