[netperf-dev] [PATCH] Avoid a double htonl in udp_stream results

[Rick Jones]

On 07/21/2011 12:10 PM, David MacMahon wrote:
> In recv_udp_stream(), htonl was called on the bytes_received field
> before calling send_response() which also calls htonl on bytes_received
> (and all other fields).  This double htonl had no effect when netserver
> is running on big endian systems, but on little endian systems it caused
> bytes_received to be passed on the wire in little endian form rather
> than big endian form (i.e. network byte order).
> ---
>   src/nettest_bsd.c |    2 +-
>   1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/src/nettest_bsd.c b/src/nettest_bsd.c
> index 69f6d11..79d4ba4 100644
> --- a/src/nettest_bsd.c
> +++ b/src/nettest_bsd.c
> @@ -7296,7 +7296,7 @@ recv_udp_stream()
>     }
>
>     netperf_response.content.response_type       = UDP_STREAM_RESULTS;
> -  udp_stream_results->bytes_received   = htonl(bytes_received);
> +  udp_stream_results->bytes_received   = bytes_received;
>     udp_stream_results->messages_recvd   = messages_recvd;
>     udp_stream_results->elapsed_time     = elapsed_time;
>     udp_stream_results->cpu_method        = cpu_method;

Do you have some before/after output showing the number wrong and then 
correct? (Yes, I'm being a paranoid pedant :)

The omni migration sort of makes the patch moot, but I'll apply the 
patch regardless.  FWIW, it appears there is a similar hton call in 
recv_tcp_stream:


   tcp_stream_results->bytes_received	= htond(bytes_received);
   tcp_stream_results->elapsed_time	= elapsed_time;
   tcp_stream_results->recv_calls	= receive_calls;

not sure what the send side does upon reciept.

rick jones