Re: [PATCH net-next-2.6] net: Introduce skb_orphan_try()

[David Miller <davem@davemloft.net>]

From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Sat, 17 Apr 2010 00:18:22 +0200

> Le jeudi 15 avril 2010 à 14:33 -0700, David Miller a écrit :
> 
>> If it's not legal to skb_orphan() here then it would not be legal for
>> the drivers to unconditionally skb_orphan(), which they do.
>> 
>> So either your test is unnecessary, or we have a big existing problem
>> :-)
> 
> I cooked following patch, introducing skb_orphan_try() helper, to
> document all known exceptions.

Looks good, applied, thanks Eric.

That timestamping issue, I bet there is some simple solution hiding
in the bushes for that one?

Hmmm, something like... there is a less transient piece of state
(perhaps in the socket) and the skb has a pointer to that piece of
state.  Then the driver writes into this pointed-to place rather
into some member of the skb itself.

Then all of these problems with referencing the skb metadata across
being passed to ->ndo_start_xmit() could go away I think.

> I have a possible followup for this patch :
> 
> Orphaning skbs earlier could also make dev_kfree_skb_irq() faster.
> Instead of queing skb into completion_queue and triggering
> NET_TX_SOFTIRQ, we would directly free an orphaned skb ?

Sounds great.  But I don't see dev_kfree_skb_irq() as much of a
performance priority, since all sane modern drivers free skbs from
softirq context (via NAPI).