Re: Socket buffer sizes with autotuning

["Jerry Chu" <hkchu@google.com>]

On Fri, May 16, 2008 at 6:29 PM, David Miller <davem@davemloft.net> wrote:
> From: "Jerry Chu" <hkchu@google.com>
> Date: Fri, 16 May 2008 18:13:20 -0700
>
>> The host inflight accounting gets screwed up. It looks like
>> pskb_expand_head() called by tcp_tso_acked()->tcp_trim_head() messes
>> up the accounting but I don't know how to fix it (still trying to
>> understand this complex piece of code). There could be other reason
>> as well.
>
> This is just like freeing up a normal SKB, so decrementing the
> in_flight value the appropriate number of packets should do the right
> thing.
>
> tcp_tso_acked() calculates this adjustment for you, in packets_acked.
>

The current pskb_expand_head() code sets skb_shinfo(skb)->in_flight
to NULL to leave the in_flight accounting solely to the other clone to
handle. To calculate in_flight more accurately, it seems to require
splitting the gso_segs and in_flight accounting bewteen the clone and
the skb (with new header), right? Plus any race condition to take care...
Hard to think clearly in a Friday afternoon :-(

Jerry