Message-ID: <377ae473-a908-6dc3-c658-ca2b81d364e0@nvidia.com>
Date:   Tue, 15 Mar 2022 17:32:50 -0700
From:   Roopa Prabhu <roopa@...dia.com>
To:     Eyal Birger <eyal.birger@...il.com>
Cc:     davem@...emloft.net, kuba@...nel.org, shmulik.ladkani@...il.com,
        netdev@...r.kernel.org
Subject: Re: [PATCH net-next] net: geneve: support IPv4/IPv6 as inner protocol


On 3/15/22 13:22, Eyal Birger wrote:
> Hi Roopa,
>
> On Tue, Mar 15, 2022 at 9:57 PM Roopa Prabhu <roopa@...dia.com> wrote:
>>
>> On 3/15/22 11:56, Eyal Birger wrote:
>>> This patch adds support for encapsulating IPv4/IPv6 within GENEVE.
>>>
>>> In order use this, a new IFLA_GENEVE_TUN flag needs to be provided at
>>> device creation. This property cannot be changed for the time being.
>>>
>>> In case IP traffic is received on a non-tun device the drop count is
>>> increased.
>>>
>>> Signed-off-by: Eyal Birger <eyal.birger@...il.com>
>>> ---
>>>    drivers/net/geneve.c         | 79 +++++++++++++++++++++++++++---------
>>>    include/uapi/linux/if_link.h |  1 +
>>>    2 files changed, 61 insertions(+), 19 deletions(-)
>>>
>>> diff --git a/drivers/net/geneve.c b/drivers/net/geneve.c
>>> index a895ff756093..37305ec26250 100644
>>> --- a/drivers/net/geneve.c
>>> +++ b/drivers/net/geneve.c
>>> @@ -56,6 +56,7 @@ struct geneve_config {
>>>        bool                    use_udp6_rx_checksums;
>>>        bool                    ttl_inherit;
>>>        enum ifla_geneve_df     df;
>>> +     bool                    tun;
>>>    };
>>>
>>>    /* Pseudo network device */
>>> @@ -251,17 +252,24 @@ static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs,
>>>                }
>>>        }
>>>
>>> -     skb_reset_mac_header(skb);
>>> -     skb->protocol = eth_type_trans(skb, geneve->dev);
>>> -     skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
>>> -
>>>        if (tun_dst)
>>>                skb_dst_set(skb, &tun_dst->dst);
>>>
>>> -     /* Ignore packet loops (and multicast echo) */
>>> -     if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) {
>>> -             geneve->dev->stats.rx_errors++;
>>> -             goto drop;
>>> +     if (gnvh->proto_type == htons(ETH_P_TEB)) {
>>> +             skb_reset_mac_header(skb);
>>> +             skb->protocol = eth_type_trans(skb, geneve->dev);
>>> +             skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
>>> +
>>> +             /* Ignore packet loops (and multicast echo) */
>>> +             if (ether_addr_equal(eth_hdr(skb)->h_source,
>>> +                                  geneve->dev->dev_addr)) {
>>> +                     geneve->dev->stats.rx_errors++;
>>> +                     goto drop;
>>> +             }
>>> +     } else {
>>> +             skb_reset_mac_header(skb);
>>> +             skb->dev = geneve->dev;
>>> +             skb->pkt_type = PACKET_HOST;
>>>        }
>>>
>>>        oiph = skb_network_header(skb);
>>> @@ -345,6 +353,7 @@ static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
>>>        struct genevehdr *geneveh;
>>>        struct geneve_dev *geneve;
>>>        struct geneve_sock *gs;
>>> +     __be16 inner_proto;
>>>        int opts_len;
>>>
>>>        /* Need UDP and Geneve header to be present */
>>> @@ -356,8 +365,13 @@ static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
>>>        if (unlikely(geneveh->ver != GENEVE_VER))
>>>                goto drop;
>>>
>>> -     if (unlikely(geneveh->proto_type != htons(ETH_P_TEB)))
>>> +     inner_proto = geneveh->proto_type;
>>> +
>>> +     if (unlikely((inner_proto != htons(ETH_P_TEB) &&
>>> +                   inner_proto != htons(ETH_P_IP) &&
>>> +                   inner_proto != htons(ETH_P_IPV6)))) {
>>>                goto drop;
>>> +     }
>>>
>>
>> unnecessary braces
> Will fix.
>
>>>        gs = rcu_dereference_sk_user_data(sk);
>>>        if (!gs)
>>> @@ -367,9 +381,13 @@ static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
>>>        if (!geneve)
>>>                goto drop;
>>>
>>> +     if (unlikely((!geneve->cfg.tun && inner_proto != htons(ETH_P_TEB)))) {
>>> +             geneve->dev->stats.rx_dropped++;
>>> +             goto drop;
>>> +     }
>> Does this change current default behavior ?.
>>
>> its unclear to be what the current behavior is for a non ETH_P_TEB packet
> Currently non ETH_P_TEB packets are silently dropped.
> I figured that if the driver supported other ethertypes it would make
> sense to increase the count in such case, to assist in debugging wrong
> configurations.
>
> I can remove this if it's better to keep the current behavior.

yes, agree. counting is better.


>>
>>> +
>>>        opts_len = geneveh->opt_len * 4;
>>> -     if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len,
>>> -                              htons(ETH_P_TEB),
>>> +     if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto,
>>>                                 !net_eq(geneve->net, dev_net(geneve->dev)))) {
>>>                geneve->dev->stats.rx_dropped++;
>>>                goto drop;
>>> @@ -717,7 +735,8 @@ static int geneve_stop(struct net_device *dev)
>>>    }
>>>
>>>    static void geneve_build_header(struct genevehdr *geneveh,
>>> -                             const struct ip_tunnel_info *info)
>>> +                             const struct ip_tunnel_info *info,
>>> +                             __be16 inner_proto)
>>>    {
>>>        geneveh->ver = GENEVE_VER;
>>>        geneveh->opt_len = info->options_len / 4;
>>> @@ -725,7 +744,7 @@ static void geneve_build_header(struct genevehdr *geneveh,
>>>        geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT);
>>>        geneveh->rsvd1 = 0;
>>>        tunnel_id_to_vni(info->key.tun_id, geneveh->vni);
>>> -     geneveh->proto_type = htons(ETH_P_TEB);
>>> +     geneveh->proto_type = inner_proto;
>>>        geneveh->rsvd2 = 0;
>>>
>>>        if (info->key.tun_flags & TUNNEL_GENEVE_OPT)
>>> @@ -734,8 +753,9 @@ static void geneve_build_header(struct genevehdr *geneveh,
>>>
>>>    static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
>>>                            const struct ip_tunnel_info *info,
>>> -                         bool xnet, int ip_hdr_len)
>>> +                         bool xnet, int ip_hdr_len, bool tun)
>>>    {
>>> +     __be16 inner_proto = tun ? skb->protocol : htons(ETH_P_TEB);
>>>        bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
>>>        struct genevehdr *gnvh;
>>>        int min_headroom;
>>> @@ -755,8 +775,8 @@ static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb,
>>>                goto free_dst;
>>>
>>>        gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len);
>>> -     geneve_build_header(gnvh, info);
>>> -     skb_set_inner_protocol(skb, htons(ETH_P_TEB));
>>> +     geneve_build_header(gnvh, info, inner_proto);
>>> +     skb_set_inner_protocol(skb, inner_proto);
>>>        return 0;
>>>
>>>    free_dst:
>>> @@ -959,7 +979,8 @@ static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
>>>                }
>>>        }
>>>
>>> -     err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr));
>>> +     err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr),
>>> +                            geneve->cfg.tun);
>>>        if (unlikely(err))
>>>                return err;
>>>
>>> @@ -1038,7 +1059,8 @@ static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
>>>                        ttl = key->ttl;
>>>                ttl = ttl ? : ip6_dst_hoplimit(dst);
>>>        }
>>> -     err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr));
>>> +     err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr),
>>> +                            geneve->cfg.tun);
>>>        if (unlikely(err))
>>>                return err;
>>>
>>> @@ -1388,6 +1410,14 @@ static int geneve_configure(struct net *net, struct net_device *dev,
>>>        dst_cache_reset(&geneve->cfg.info.dst_cache);
>>>        memcpy(&geneve->cfg, cfg, sizeof(*cfg));
>>>
>>> +     if (geneve->cfg.tun) {
>>> +             dev->header_ops = NULL;
>>> +             dev->type = ARPHRD_NONE;
>>> +             dev->hard_header_len = 0;
>>> +             dev->addr_len = 0;
>>> +             dev->flags = IFF_NOARP;
>>> +     }
>>> +
>>>        err = register_netdevice(dev);
>>>        if (err)
>>>                return err;
>>> @@ -1561,10 +1591,18 @@ static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[],
>>>    #endif
>>>        }
>>>
>>> +     if (data[IFLA_GENEVE_TUN]) {
>>> +             if (changelink) {
>>> +                     attrtype = IFLA_GENEVE_TUN;
>>> +                     goto change_notsup;
>>> +             }
>>> +             cfg->tun = true;
>>> +     }
>>> +
>>>        return 0;
>>>    change_notsup:
>>>        NL_SET_ERR_MSG_ATTR(extack, data[attrtype],
>>> -                         "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported");
>>> +                         "Changing VNI, Port, endpoint IP address family, external, tun, and UDP checksum attributes are not supported");
>>>        return -EOPNOTSUPP;
>>>    }
>>>
>>> @@ -1799,6 +1837,9 @@ static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev)
>>>        if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit))
>>>                goto nla_put_failure;
>>>
>>> +     if (geneve->cfg.tun && nla_put_flag(skb, IFLA_GENEVE_TUN))
>>> +             goto nla_put_failure;
>>> +
>>>        return 0;
>>>
>>>    nla_put_failure:
>>> diff --git a/include/uapi/linux/if_link.h b/include/uapi/linux/if_link.h
>>> index bd24c7dc10a2..198aefa2c513 100644
>>> --- a/include/uapi/linux/if_link.h
>>> +++ b/include/uapi/linux/if_link.h
>>> @@ -842,6 +842,7 @@ enum {
>>>        IFLA_GENEVE_LABEL,
>>>        IFLA_GENEVE_TTL_INHERIT,
>>>        IFLA_GENEVE_DF,
>>> +     IFLA_GENEVE_TUN,
>> geneve is itself called a tunnel, i wonder if a different name for the
>> flag would be more appropriate.
> I tried to find a reference to something similar, so went with something like
> the tun vs. tap distinction. I was also concerned about the possible
> confusion, but it felt clearer than something like L3_INNER_PROTO_MODE.
>
> I'd happily replace it with a better suggestion.

o ok, makes sense. I can't think of anything other than simply 
IFLA_GENEVE_INNER_PROTO

(maybe others have better suggestions)


>
>> what is the use case for this ?. is there a RFC reference ?
> I stumbled upon this configuration when trying to receive packets from an
> AWS "Gateway Load Balancer" which sends IP packets encapsulated in GENEVE [1].
>
> But to my understanding the RFC allows this so it doesn't seem something
> specific to AWS.

that makes me wonder if we need a knob atall and if this should be 
allowed by default.

wonder how other network vendor standard geneve implementations behave 
by default.


>
> Thanks for the review!
>
> Eyal.
>
> [1] https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Faws.amazon.com%2Fblogs%2Fnetworking-and-content-delivery%2Fintegrate-your-custom-logic-or-appliance-with-aws-gateway-load-balancer%2F&amp;data=04%7C01%7Croopa%40nvidia.com%7C223c13c616ef430a487f08da06c19610%7C43083d15727340c1b7db39efd9ccc17a%7C0%7C0%7C637829725767772379%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&amp;sdata=ZQTDrFJ8LLn5SdN6h%2B5NECxwlD9PAGV2KzpVUV%2B1chc%3D&amp;reserved=0

Thanks for the details.

Powered by blists - more mailing lists