Re: [PATCH net] gro_cells: Avoid packet re-ordering for cloned skbs

[Thomas Bogendoerfer <tbogendoerfer@suse.de>]

On Mon, 20 Jan 2025 15:55:26 +0100
Eric Dumazet <edumazet@google.com> wrote:

> On Mon, Jan 20, 2025 at 3:32 PM Thomas Bogendoerfer
> <tbogendoerfer@suse.de> wrote:
> >
> > On Mon, 13 Jan 2025 13:55:18 +0100
> > Eric Dumazet <edumazet@google.com> wrote:
> >  
> > > On Mon, Jan 13, 2025 at 12:28 PM Thomas Bogendoerfer
> > > <tbogendoerfer@suse.de> wrote:  
> > > >
> > > > On Thu, 9 Jan 2025 15:56:24 +0100
> > > > Eric Dumazet <edumazet@google.com> wrote:
> > > >  
> > > > > On Thu, Jan 9, 2025 at 3:27 PM Thomas Bogendoerfer
> > > > > <tbogendoerfer@suse.de> wrote:  
> > > > > >
> > > > > > gro_cells_receive() passes a cloned skb directly up the stack and
> > > > > > could cause re-ordering against segments still in GRO. To avoid
> > > > > > this copy the skb and let GRO do it's work.
> > > > > >
> > > > > > Fixes: c9e6bc644e55 ("net: add gro_cells infrastructure")
> > > > > > Signed-off-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
> > > > > > ---
> > > > > >  net/core/gro_cells.c | 11 ++++++++++-
> > > > > >  1 file changed, 10 insertions(+), 1 deletion(-)
> > > > > >
> > > > > > diff --git a/net/core/gro_cells.c b/net/core/gro_cells.c
> > > > > > index ff8e5b64bf6b..2f8d688f9d82 100644
> > > > > > --- a/net/core/gro_cells.c
> > > > > > +++ b/net/core/gro_cells.c
> > > > > > @@ -20,11 +20,20 @@ int gro_cells_receive(struct gro_cells *gcells, struct sk_buff *skb)
> > > > > >         if (unlikely(!(dev->flags & IFF_UP)))
> > > > > >                 goto drop;
> > > > > >
> > > > > > -       if (!gcells->cells || skb_cloned(skb) || netif_elide_gro(dev)) {
> > > > > > +       if (!gcells->cells || netif_elide_gro(dev)) {
> > > > > > +netif_rx:
> > > > > >                 res = netif_rx(skb);
> > > > > >                 goto unlock;
> > > > > >         }
> > > > > > +       if (skb_cloned(skb)) {
> > > > > > +               struct sk_buff *n;
> > > > > >
> > > > > > +               n = skb_copy(skb, GFP_KERNEL);  
> > > > >
> > > > > I do not think we want this skb_copy(). This is going to fail too often.  
> > > >
> > > > ok
> > > >  
> > > > > Can you remind us why we have this skb_cloned() check here ?  
> > > >
> > > > some fields of the ip/tcp header are going to be changed in the first gro
> > > > segment  
> > >
> > > Presumably we should test skb_header_cloned()
> > >
> > > This means something like skb_cow_head(skb, 0) could be much more
> > > reasonable than skb_copy().  
> >
> > I don't think this will work, because at that point it's skb->data points
> > at the IPv6 header in my test case (traffic between two namespaces connected
> > via ip6 tunnel over ipvlan). Correct header offsets are set after later,
> > when gro_cells napi routine runs.
> >
> > Do you see another option ?  
> 
> Anything not attempting order-5 allocations will work :)
> 
> I would try something like that.
> 
> diff --git a/net/core/gro_cells.c b/net/core/gro_cells.c
> index ff8e5b64bf6b..74416194f148 100644
> --- a/net/core/gro_cells.c
> +++ b/net/core/gro_cells.c
> @@ -4,6 +4,7 @@
>  #include <linux/netdevice.h>
>  #include <net/gro_cells.h>
>  #include <net/hotdata.h>
> +#include <net/gro.h>
> 
>  struct gro_cell {
>         struct sk_buff_head     napi_skbs;
> @@ -20,7 +21,7 @@ int gro_cells_receive(struct gro_cells *gcells,
> struct sk_buff *skb)
>         if (unlikely(!(dev->flags & IFF_UP)))
>                 goto drop;
> 
> -       if (!gcells->cells || skb_cloned(skb) || netif_elide_gro(dev)) {
> +       if (!gcells->cells || netif_elide_gro(dev)) {
>                 res = netif_rx(skb);
>                 goto unlock;
>         }
> @@ -58,7 +59,11 @@ static int gro_cell_poll(struct napi_struct *napi,
> int budget)
>                 skb = __skb_dequeue(&cell->napi_skbs);
>                 if (!skb)
>                         break;
> -               napi_gro_receive(napi, skb);
> +               /* Core GRO stack does not play well with clones. */
> +               if (skb_cloned(skb))
> +                       gro_normal_one(napi, skb, 1);
> +               else
> +                       napi_gro_receive(napi, skb);
>                 work_done++;
>         }

works perfectly, thank you. I've sent a v2 of the fix.

Thomas.

-- 
SUSE Software Solutions Germany GmbH
HRB 36809 (AG Nürnberg)
Geschäftsführer: Ivo Totev, Andrew McDonald, Werner Knoblich