Re: [PATCH V5 2/6 net-next] netdevice.h: Add zero-copy flag in netdevice

On Thu, May 19, 2011 at 12:42:49PM -0700, Shirley Ma wrote:

On Wed, 2011-05-18 at 10:00 -0700, Shirley Ma wrote:

quoted

On Wed, 2011-05-18 at 19:51 +0300, Michael S. Tsirkin wrote:

quoted

quoted

quoted

Yes, I agree.  I think for tcpdump, we really need to copy the

data

quoted

quoted

anyway, to avoid guest changing it in between.  So we do that

and

quoted

then

quoted

quoted

use the copy everywhere, release the old one. Hmm? 

Yes. Old one use zerocopy, new one use copy data.

Thanks
Shirley

No, that's wrong, as they might become different with a
malicious guest. As long as we copied already, lets realease
the data and have everyone use the copy. 

Ok, I will patch pskb_expand_head to test it out. 

I am patching skb_copy, skb_clone, pskb_copy, pskb_expand_head to
convert a zero-copy skb to a copy skb to avoid this kind of issue.

This overhead won't impact macvtap/vhost TX zero-copy normally.

Shirley

OK, that will handle packet socket at least in that it won't crash :)

So the requirements are
- data must be released in a timely fashion (e.g. unlike virtio-net
  tun or bridge)
- no filtering based on data (data is mapped in guest)
- SG support
- HIGHDMA support (on arches where this makes sense)
- on fast path no calls to skb_copy, skb_clone, pskb_copy,
  pskb_expand_head as these are slow

First 2 requirements are a must, all other requirements
are just dependencies to make sure zero copy will be faster
than non zero copy.

Using a new feature bit is probably the simplest approach to
this. macvtap on top of most physical NICs most likely works
correctly so it seems a bit more work than it needs to be,
but it's also the safest one I think ...

-- 
MST