Thread (13 messages) flat view 13 messages, 6 authors, 2012-09-20

Re: [PATCH v3] net-tcp: TCP/IP stack bypass for loopback connections

From: Bruce Curtis <hidden>
Date: 2012-09-19 21:19:31

On Wed, Sep 19, 2012 at 2:03 PM, Eric Dumazet [off-list ref] wrote:
On Wed, 2012-09-19 at 16:34 -0400, David Miller wrote:
quoted
I have an idea on how to handle this.

In drivers/net/loopback.c:loopback_tx(), skip the SKB orphan operation
if there is a friend socket at skb->friend.

When sending such friend SKBs out at connection startup, arrange it
such that the skb->destructor will zap the skb->friend pointer to
NULL.

Also, in skb_orphan*(), if necessary, set skb->friend to NULL.

skb->sk will hold a reference to the socket, and since skb->friend
will be equal, this will make sure a pointer to an unreferenced
socket does not escape.
I now am wondering if we still need skb->friend field.

If skb->sk is not zeroed by a premature skb_orphan(), then

skb->sk->sk_friend gives the friend ?
Good question, in the friends data path it's only used to
differentiate a friend directly Q'd skb so should work, I'll look into
it.
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help