Re: tcp_ack of sent skb's

Alan Cox (
Thu, 10 Jul 1997 08:14:16 +0100 (BST)

> What appears to happen is that the kfree_skb gets called
> which doesn't free the skb as it is locked. However, the
> pointer to the skb is lost (forever), and so it never frees.

The dev_kfree_skb() frees it when the packet finally gets off the
transmit queue. There is a pi/pt problem with those queues and they ought
to get removed one day

> Am I making sense?

Sort of - 2.1 does it a bit differently and somewhat more sanely. The
lock is gone - we just clone the buffer head knowing we know have a fast
memory allocator.