Re: [PATCH v4 bpf-next 3/3] veth: allow jumbo frames in xdp mode

["Toke Høiland-Jørgensen" <toke@redhat.com>]

Lorenzo Bianconi <lorenzo.bianconi@redhat.com> writes:

>> Lorenzo Bianconi <lorenzo@kernel.org> writes:
>> 
>> > Allow increasing the MTU over page boundaries on veth devices
>> > if the attached xdp program declares to support xdp fragments.
>> > Enable NETIF_F_ALL_TSO when the device is running in xdp mode.
>> >
>> > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
>> > ---
>> >  drivers/net/veth.c | 28 +++++++++++++++++-----------
>> >  1 file changed, 17 insertions(+), 11 deletions(-)
>> >
>> > diff --git a/drivers/net/veth.c b/drivers/net/veth.c
>> > index 47b21b1d2fd9..c5a2dc2b2e4b 100644
>> > --- a/drivers/net/veth.c
>> > +++ b/drivers/net/veth.c
>> > @@ -293,8 +293,7 @@ static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
>> >  /* return true if the specified skb has chances of GRO aggregation
>> >   * Don't strive for accuracy, but try to avoid GRO overhead in the most
>> >   * common scenarios.
>> > - * When XDP is enabled, all traffic is considered eligible, as the xmit
>> > - * device has TSO off.
>> > + * When XDP is enabled, all traffic is considered eligible.
>> >   * When TSO is enabled on the xmit device, we are likely interested only
>> >   * in UDP aggregation, explicitly check for that if the skb is suspected
>> >   * - the sock_wfree destructor is used by UDP, ICMP and XDP sockets -
>> > @@ -302,11 +301,13 @@ static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb,
>> >   */
>> >  static bool veth_skb_is_eligible_for_gro(const struct net_device *dev,
>> >  					 const struct net_device *rcv,
>> > +					 const struct veth_rq *rq,
>> >  					 const struct sk_buff *skb)
>> >  {
>> > -	return !(dev->features & NETIF_F_ALL_TSO) ||
>> > -		(skb->destructor == sock_wfree &&
>> > -		 rcv->features & (NETIF_F_GRO_FRAGLIST | NETIF_F_GRO_UDP_FWD));
>> > +	return rcu_access_pointer(rq->xdp_prog) ||
>> > +	       !(dev->features & NETIF_F_ALL_TSO) ||
>> > +	       (skb->destructor == sock_wfree &&
>> > +		rcv->features & (NETIF_F_GRO_FRAGLIST | NETIF_F_GRO_UDP_FWD));
>> >  }
>> >  
>> >  static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
>> > @@ -335,7 +336,7 @@ static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev)
>> >  		 * Don't bother with napi/GRO if the skb can't be aggregated
>> >  		 */
>> >  		use_napi = rcu_access_pointer(rq->napi) &&
>> > -			   veth_skb_is_eligible_for_gro(dev, rcv, skb);
>> > +			   veth_skb_is_eligible_for_gro(dev, rcv, rq, skb);
>> >  	}
>> >  
>> >  	skb_tx_timestamp(skb);
>> > @@ -1525,9 +1526,14 @@ static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
>> >  			goto err;
>> >  		}
>> >  
>> > -		max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM -
>> > -			  peer->hard_header_len -
>> > -			  SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
>> > +		max_mtu = SKB_WITH_OVERHEAD(PAGE_SIZE - VETH_XDP_HEADROOM) -
>> > +			  peer->hard_header_len;
>> 
>> Why are we no longer accounting the size of the skb_shared_info if the
>> program doesn't support frags?
>
> doing so we do not allow packets over page boundaries (so non-linear xdp_buff)
> if the attached program does not delclare to support them, right?

Oh, sorry, somehow completely skipped over the addition of the
SKB_WITH_OVERHEAD() - so thought you were removing the sizeof(struct
skb_shared_info) from the calculation...

>> > +		/* Allow increasing the max_mtu if the program supports
>> > +		 * XDP fragments.
>> > +		 */
>> > +		if (prog->aux->xdp_has_frags)
>> > +			max_mtu += PAGE_SIZE * MAX_SKB_FRAGS;
>> > +
>> >  		if (peer->mtu > max_mtu) {
>> >  			NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP");
>> >  			err = -ERANGE;
>> > @@ -1549,7 +1555,7 @@ static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog,
>> >  		}
>> >  
>> >  		if (!old_prog) {
>> > -			peer->hw_features &= ~NETIF_F_GSO_SOFTWARE;
>> > +			peer->hw_features &= ~NETIF_F_GSO_FRAGLIST;
>> 
>> The patch description says we're enabling TSO, but this change enables a
>> couple of other flags as well. Also, it's not quite obvious to me why
>> your change makes this possible? Is it because we can now execute XDP on
>> a full TSO packet at once? Because then this should be coupled to the
>> xdp_has_frags flag of the XDP program? Or will the TSO packet be
>> segmented before it hits the XDP program? But then this change has
>> nothing to do with the rest of your series?
>
> actually tso support is not mandatory for this feature (even if it is probably
> meaningful). I will drop it from v5 and we can take care of it in a susequent
> patch.

OK, SGTM

-Toke