Re: [PATCH bpf-next V5 4/5] bpf: drop MTU check when doing TC-BPF redirect to ingress

[Jesper Dangaard Brouer <brouer@redhat.com>]

On Fri, 30 Oct 2020 13:36:05 -0700
John Fastabend <john.fastabend@gmail.com> wrote:

> Jesper Dangaard Brouer wrote:
> > The use-case for dropping the MTU check when TC-BPF does redirect to
> > ingress, is described by Eyal Birger in email[0]. The summary is the
> > ability to increase packet size (e.g. with IPv6 headers for NAT64) and
> > ingress redirect packet and let normal netstack fragment packet as needed.
> > 
> > [0] https://lore.kernel.org/netdev/CAHsH6Gug-hsLGHQ6N0wtixdOa85LDZ3HNRHVd0opR=19Qo4W4Q@mail.gmail.com/
> > 
> > V4:
> >  - Keep net_device "up" (IFF_UP) check.
> >  - Adjustment to handle bpf_redirect_peer() helper
> > 
> > Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
> > ---
> >  include/linux/netdevice.h |   31 +++++++++++++++++++++++++++++--
> >  net/core/dev.c            |   19 ++-----------------
> >  net/core/filter.c         |   14 +++++++++++---
> >  3 files changed, 42 insertions(+), 22 deletions(-)
> > 
> > diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> > index 964b494b0e8d..bd02ddab8dfe 100644
> > --- a/include/linux/netdevice.h
> > +++ b/include/linux/netdevice.h
> > @@ -3891,11 +3891,38 @@ int dev_forward_skb(struct net_device *dev, struct sk_buff *skb);
> >  bool is_skb_forwardable(const struct net_device *dev,
> >  			const struct sk_buff *skb);
> >  
> > +static __always_inline bool __is_skb_forwardable(const struct net_device *dev,
> > +						 const struct sk_buff *skb,
> > +						 const bool check_mtu)  
> 
> It looks like if check_mtu=false then this is just an interface up check.
> Can we leave is_skb_forwardable logic alone and just change the spots where
> this is called with false to something with a name that describes the check,
> such as is_dev_up(dev). I think it will make this change smaller and the
> code easier to read. Did I miss something?

People should realized that this is constructed such, the compiler will
compile-time remove the actual argument (the const bool check_mtu).
And this propagates also to ____dev_forward_skb() where the call places
are also inlined.

Yes, this (check_mtu=false) is basically an interface up check, but the
only place it is used directly is in the ndo_get_peer_dev() case, and
reading the code I find it more readable that is says
__is_skb_forwardable because this is used as part of a forwarding step,
and is_dev_up() doesn't convey the intent in this use-case.


> > +{
> > +	const u32 vlan_hdr_len = 4; /* VLAN_HLEN */
> > +	unsigned int len;
> > +
> > +	if (!(dev->flags & IFF_UP))
> > +		return false;
> > +
> > +	if (!check_mtu)
> > +		return true;
> > +
> > +	len = dev->mtu + dev->hard_header_len + vlan_hdr_len;
> > +	if (skb->len <= len)
> > +		return true;
> > +
> > +	/* if TSO is enabled, we don't care about the length as the packet
> > +	 * could be forwarded without being segmented before
> > +	 */
> > +	if (skb_is_gso(skb))
> > +		return true;
> > +
> > +	return false;
> > +}
> > +
> >  static __always_inline int ____dev_forward_skb(struct net_device *dev,
> > -					       struct sk_buff *skb)
> > +					       struct sk_buff *skb,
> > +					       const bool check_mtu)
> >  {  
> 
> I guess you will get some duplication here if you have a dev_forward_skb()
> and a dev_forward_skb_nocheck() or something. Take it or leave it. I know
> I've added my share of bool swivel bits like this, but better to avoid
> it if possible IMO.

As I wrote the bool will actually get compile-time removed, so I don't
see that as problematic.  And I avoided replicating the code in more
places.

The problematic part (which you didn't comment) on is this:

On Fri, 30 Oct 2020 17:51:07 +0100
Jesper Dangaard Brouer <brouer@redhat.com> wrote:

> diff --git a/net/core/filter.c b/net/core/filter.c
> index bd4a416bd9ad..71b78b8d443c 100644
> --- a/net/core/filter.c
> +++ b/net/core/filter.c
> @@ -2083,13 +2083,21 @@ static const struct bpf_func_proto bpf_csum_level_proto = {
>  
>  static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
>  {
> -	return dev_forward_skb(dev, skb);
> +	int ret = ____dev_forward_skb(dev, skb, false);
> +
> +	if (likely(!ret)) {
> +		skb->protocol = eth_type_trans(skb, dev);
> +		skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
> +		ret = netif_rx(skb);
> +	}
> +
> +	return ret;
>  }

I'm replicating two lines from dev_forward_skb(), but I couldn't find a
way to avoid this, without causing larger code changes (and slower code).



> Other than style aspects it looks correct to me.
> 
> >  	if (skb_orphan_frags(skb, GFP_ATOMIC) ||
> > -	    unlikely(!is_skb_forwardable(dev, skb))) {
> > +	    unlikely(!__is_skb_forwardable(dev, skb, check_mtu))) {
> >  		atomic_long_inc(&dev->rx_dropped);
> >  		kfree_skb(skb);
> >  		return NET_RX_DROP;
> > diff --git a/net/core/dev.c b/net/core/dev.c
> > index 9499a414d67e..445ccf92c149 100644
> > --- a/net/core/dev.c
> > +++ b/net/core/dev.c
> > @@ -2188,28 +2188,13 @@ static inline void net_timestamp_set(struct sk_buff *skb)
> >    
> 



-- 
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer