Re: 2.6.29-rc5-git3 Leak r=1 3

[David Miller <davem@davemloft.net>]

From: "Ilpo Järvinen" <ilpo.jarvinen@helsinki.fi>
Date: Sat, 21 Feb 2009 13:44:27 +0200 (EET)

> [PATCH] tcp: fix retrans_out leaks in shifting code
> 
> There's conflicting assumptions in shifting, the caller assumes
> that dupsack results in S'ed skbs (or a part of it) for sure but
> never gave a hint to tcp_sacktag_one when dsack is actually in
> use. Thus DSACK retrans_out -= pcount was not taken and the
> counter became out of sync. Remove obstacle from that information
> flow to get DSACKs accounted in tcp_sacktag_one as expected.
> 
> Compile tested only.
> 
> In general the call to tcp_sacktag_one with the later full shift
> from already SACKed skb should not be there, it's just doing
> duplicating work already done earlier so it just wastes cycles and
> also seems to duplicate small amount of code unnecessarily :-/.
> However, moving tcp_sacktag_one call one level up is not what I want
> to do in mainline -rc5, the counting & seqno state manipulation that
> is going on there is just too hairy to get right in a hurry (order
> is very significant there as the state transition is in progress
> making normal invariants are not to be trusted). Luckily
> tcp_sacktag_one is effectively a no-op atm with dup_sack set to 0
> for already SACKed skb.
> 
> Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>

Can you do some testing on this before I toss it into
net-next-2.6? :-)  It looks fine to be, but... so did
Herbert's URG change :)