Re: [PATCH] pktgen: Clone skb to avoid corruption of skbs in ndo_start_xmit methods

[Neil Horman <nhorman@tuxdriver.com>]

On Wed, Jul 20, 2011 at 06:08:39PM +0200, Michał Mirosław wrote:
> W dniu 20 lipca 2011 17:40 użytkownik Neil Horman
> <nhorman@tuxdriver.com> napisał:
> > On Wed, Jul 20, 2011 at 05:35:47PM +0200, Michał Mirosław wrote:
> >> 2011/7/20 Neil Horman <nhorman@tuxdriver.com>:
> >> > On Wed, Jul 20, 2011 at 06:24:15AM +0200, Eric Dumazet wrote:
> >> >> Le mardi 19 juillet 2011 à 22:07 -0400, Neil Horman a écrit :
> >> >> > >
> >> >> > I think this is a good idea.  It lets pktgen dynamically make the clone/share
> >> >> > decision dynamically and only impacts performance for those systems.
> >> >> >
> >> >>
> >> >> Just let pktgen refuse to use clone_skb command for these devices.
> >> >>
> >> > copy that, This is by no means final, but what do you think of this?  If its
> >> > agreeable to you, Ben, et al. I can add this to my local tree and start auditing
> >> > all the drivers that may need to have the flag set.
> >> >
> >> > Regards
> >> > Neil
> >> >
> >> >
> >> > diff --git a/include/linux/if.h b/include/linux/if.h
> >> > index 3bc63e6..ae904fe 100644
> >> > --- a/include/linux/if.h
> >> > +++ b/include/linux/if.h
> >> > @@ -76,6 +76,7 @@
> >> >  #define IFF_BRIDGE_PORT        0x4000          /* device used as bridge port */
> >> >  #define IFF_OVS_DATAPATH       0x8000  /* device used as Open vSwitch
> >> >                                         * datapath port */
> >> > +#define IFF_CANT_SHARE_SKB     0x10000 /* Device can't share skbs in tx path */
> >> >
> >> >  #define IF_GET_IFACE   0x0001          /* for querying only */
> >> >  #define IF_GET_PROTO   0x0002
> >> > diff --git a/net/core/pktgen.c b/net/core/pktgen.c
> >> > index f76079c..bf6d88d 100644
> >> > --- a/net/core/pktgen.c
> >> > +++ b/net/core/pktgen.c
> >> > @@ -1071,6 +1071,9 @@ static ssize_t pktgen_if_write(struct file *file,
> >> >                if (len < 0)
> >> >                        return len;
> >> >
> >> > +               if (pkt_dev->priv_flags & IFF_CANT_SHARE_SKB)
> >> > +                       return -EOPNOTSUPP;
> >> > +
> >> >                i += len;
> >> >                pkt_dev->clone_skb = value;
> >> >
> >>
> >> I would prefer that the flag be inclusive, i.e. it should mark drivers
> >> which can use shared skbs. And it might be better to clone the skb in
> >> case the flag is disabled to keep functionality working.
> > Ok, I can agree with that.  But you're ok with the general approach?
> 
> I assumed you wanted to use netdev->features and I noticed priv_flags
> just now. If the flag is supposed to be permanent then its fine by me.
> Actually this makes me wonder if NETIF_F_LLTX and similar should get
> moved to netdev->priv_flags.
> 
Yeah, I expect it will be both permanent and invisible to user space.
Neil

> Best Regards,
> Michał Mirosław
>