Re: [PATCH 1/25]: Fix bug in the calculation of very low sending

[Eddie Kohler <kohler@cs.ucla.edu>]

I'm surprised that this case occurs:

Gerrit Renker wrote:
> [CCID 3]: Fix bug in the calculation of very low sending rates
> 
> This fixes an error in the calculation of t_ipi when X converges towards
> very low sending rates (between 1 and 64 bytes per second).
> 
> Although this case may not sound likely, it can be reproduced by connecting,
> hitting enter (1 byte sent) and waiting for some time, during which the
> nofeedback timer halves the sending rate until finally it reaches the region
> 1..64 bytes/sec. Computing X is handled correctly (tested separately); but by
> dividing X _before_ entering the calculation of t_ipi, X becomes zero as
> a result.  This in turn triggers a BUG condition caught in scaled_div().

According to RFC 4342, the allowed sending rate should never be reduced 
below the TCP initial sending rate of two or four packets per RTT, 
depending on the packet size, as the result of an idle or slow period.

Hitting enter and then waiting is an idle period.

Since X is the allowed transmit rate, it seems Linux implementation is 
not following this at the moment; but perhaps there is code elsewhere 
handling this case?

Eddie


> Fixed by replacing with equivalent statement and explicit typecast for good
> measure.
> 
> Calculation verified and effect of patch tested - reduced never below 1 byte
> per 64 seconds afterwards, i.e. not allowing divide-by-zero.
> 
> Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
> ---
>  net/dccp/ccids/ccid3.c |    4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> --- a/net/dccp/ccids/ccid3.c
> +++ b/net/dccp/ccids/ccid3.c
> @@ -102,8 +102,8 @@ static inline u64 rfc3390_initial_rate(s
>  static inline void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx)
>  {
>  	/* Calculate new t_ipi = s / X_inst (X_inst is in 64 * bytes/second) */
> -	hctx->ccid3hctx_t_ipi = scaled_div(hctx->ccid3hctx_s,
> -					   hctx->ccid3hctx_x >> 6);
> +	hctx->ccid3hctx_t_ipi = scaled_div32(((u64)hctx->ccid3hctx_s) << 6,
> +					     hctx->ccid3hctx_x);
>  
>  	/* Calculate new delta by delta = min(t_ipi / 2, t_gran / 2) */
>  	hctx->ccid3hctx_delta = min_t(u32, hctx->ccid3hctx_t_ipi / 2,
> -
> To unsubscribe from this list: send the line "unsubscribe dccp" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html