Hi Jakub, many thanks for your review. Please see my responses inline: On Tue, 10 Nov 2020 15:12:55 -0800 Jakub Kicinski <kuba@xxxxxxxxxx> wrote: > On Sat, 7 Nov 2020 16:31:38 +0100 Andrea Mayer wrote: > > SRv6 End.DT4 is defined in the SRv6 Network Programming [1]. > > > > The SRv6 End.DT4 is used to implement IPv4 L3VPN use-cases in > > multi-tenants environments. It decapsulates the received packets and it > > performs IPv4 routing lookup in the routing table of the tenant. > > > > The SRv6 End.DT4 Linux implementation leverages a VRF device in order to > > force the routing lookup into the associated routing table. > > How does the behavior of DT4 compare to DT6? > The implementation of SRv6 End.DT4 differs from the the implementation of SRv6 End.DT6 due to the different *route input* lookup functions. For IPv6 is it possible to force the routing lookup specifying a routing table through the ip6_pol_route() function (as it is done in the seg6_lookup_any_nexthop()). Conversely, for the IPv4 we cannot force the lookup into a specific table with the functions that are currently exposed by the kernel. > The implementation looks quite different. > Long story short: A long time ago, we discussed here on the mailing list how best to implement the SRv6 DT4. After some time, we identified with the help of David Ahern the VRF as the key infrastructure on which to build the SRv6 End.DT4. Indeed, the use of VRF allows us not to touch in any way the core components of the kernel (i.e.: the ipv4 routing system) and to exploit an already existing infrastructure. I would say that also the SRv6 End.DT6 should leverage the VRF as we did for SRv6 End.DT4. We can also try to change End.DT6 implementation, if needed. > > To make the End.DT4 work properly, it must be guaranteed that the routing > > table used for routing lookup operations is bound to one and only one > > VRF during the tunnel creation. Such constraint has to be enforced by > > enabling the VRF strict_mode sysctl parameter, i.e: > > $ sysctl -wq net.vrf.strict_mode=1. > > > > At JANOG44, LINE corporation presented their multi-tenant DC architecture > > using SRv6 [2]. In the slides, they reported that the Linux kernel is > > missing the support of SRv6 End.DT4 behavior. > > > > The iproute2 counterpart required for configuring the SRv6 End.DT4 > > behavior is already implemented along with the other supported SRv6 > > behaviors [3]. > > > > [1] https://tools.ietf.org/html/draft-ietf-spring-srv6-network-programming > > [2] https://speakerdeck.com/line_developers/line-data-center-networking-with-srv6 > > [3] https://patchwork.ozlabs.org/patch/799837/ > > > > Signed-off-by: Andrea Mayer <andrea.mayer@xxxxxxxxxxx> > > --- > > net/ipv6/seg6_local.c | 205 ++++++++++++++++++++++++++++++++++++++++++ > > 1 file changed, 205 insertions(+) > > > > diff --git a/net/ipv6/seg6_local.c b/net/ipv6/seg6_local.c > > index 4b0f155d641d..a41074acd43e 100644 > > --- a/net/ipv6/seg6_local.c > > +++ b/net/ipv6/seg6_local.c > > @@ -57,6 +57,14 @@ struct bpf_lwt_prog { > > char *name; > > }; > > > > +struct seg6_end_dt4_info { > > + struct net *net; > > + /* VRF device associated to the routing table used by the SRv6 End.DT4 > > + * behavior for routing IPv4 packets. > > + */ > > + int vrf_ifindex; > > +}; > > + > > struct seg6_local_lwt { > > int action; > > struct ipv6_sr_hdr *srh; > > @@ -66,6 +74,7 @@ struct seg6_local_lwt { > > int iif; > > int oif; > > struct bpf_lwt_prog bpf; > > + struct seg6_end_dt4_info dt4_info; > > > > int headroom; > > struct seg6_action_desc *desc; > > @@ -413,6 +422,194 @@ static int input_action_end_dx4(struct sk_buff *skb, > > return -EINVAL; > > } > > > > +#ifdef CONFIG_NET_L3_MASTER_DEV > > + > > no need for this empty line. > Ok. > > +static struct net *fib6_config_get_net(const struct fib6_config *fib6_cfg) > > +{ > > + const struct nl_info *nli = &fib6_cfg->fc_nlinfo; > > + > > + return nli->nl_net; > > +} > > + > > +static int seg6_end_dt4_build(struct seg6_local_lwt *slwt, const void *cfg, > > + struct netlink_ext_ack *extack) > > +{ > > + struct seg6_end_dt4_info *info = &slwt->dt4_info; > > + int vrf_ifindex; > > + struct net *net; > > + > > + net = fib6_config_get_net(cfg); > > + > > + vrf_ifindex = l3mdev_ifindex_lookup_by_table_id(L3MDEV_TYPE_VRF, net, > > + slwt->table); > > + if (vrf_ifindex < 0) { > > + if (vrf_ifindex == -EPERM) { > > + NL_SET_ERR_MSG(extack, > > + "Strict mode for VRF is disabled"); > > + } else if (vrf_ifindex == -ENODEV) { > > + NL_SET_ERR_MSG(extack, "No such device"); > > That's what -ENODEV already says. > Yes, sorry for this very trivial message. I will improve it in v3. > > + } else { > > + NL_SET_ERR_MSG(extack, "Unknown error"); > > Useless error. > Ok, I will remove it and keep only the pr_debug message in v3. > > + pr_debug("seg6local: SRv6 End.DT4 creation error=%d\n", > > + vrf_ifindex); > > + } > > + > > + return vrf_ifindex; > > + } > > + > > + info->net = net; > > + info->vrf_ifindex = vrf_ifindex; > > + > > + return 0; > > +} > > + > > +/* The SRv6 End.DT4 behavior extracts the inner (IPv4) packet and routes the > > + * IPv4 packet by looking at the configured routing table. > > + * > > + * In the SRv6 End.DT4 use case, we can receive traffic (IPv6+Segment Routing > > + * Header packets) from several interfaces and the IPv6 destination address (DA) > > + * is used for retrieving the specific instance of the End.DT4 behavior that > > + * should process the packets. > > + * > > + * However, the inner IPv4 packet is not really bound to any receiving > > + * interface and thus the End.DT4 sets the VRF (associated with the > > + * corresponding routing table) as the *receiving* interface. > > + * In other words, the End.DT4 processes a packet as if it has been received > > + * directly by the VRF (and not by one of its slave devices, if any). > > + * In this way, the VRF interface is used for routing the IPv4 packet in > > + * according to the routing table configured by the End.DT4 instance. > > + * > > + * This design allows you to get some interesting features like: > > + * 1) the statistics on rx packets; > > + * 2) the possibility to install a packet sniffer on the receiving interface > > + * (the VRF one) for looking at the incoming packets; > > + * 3) the possibility to leverage the netfilter prerouting hook for the inner > > + * IPv4 packet. > > + * > > + * This function returns: > > + * - the sk_buff* when the VRF rcv handler has processed the packet correctly; > > + * - NULL when the skb is consumed by the VRF rcv handler; > > + * - a pointer which encodes a negative error number in case of error. > > + * Note that in this case, the function takes care of freeing the skb. > > + */ > > +static struct sk_buff *end_dt4_vrf_rcv(struct sk_buff *skb, > > + struct net_device *dev) > > +{ > > + /* based on l3mdev_ip_rcv; we are only interested in the master */ > > + if (unlikely(!netif_is_l3_master(dev) && !netif_has_l3_rx_handler(dev))) > > + goto drop; > > + > > + if (unlikely(!dev->l3mdev_ops->l3mdev_l3_rcv)) > > + goto drop; > > + > > + /* the decap packet (IPv4) does not come with any mac header info. > > + * We must unset the mac header to allow the VRF device to rebuild it, > > + * just in case there is a sniffer attached on the device. > > + */ > > + skb_unset_mac_header(skb); > > + > > + skb = dev->l3mdev_ops->l3mdev_l3_rcv(dev, skb, AF_INET); > > + if (!skb) > > + /* the skb buffer was consumed by the handler */ > > + return NULL; > > + > > + /* when a packet is received by a VRF or by one of its slaves, the > > + * master device reference is set into the skb. > > + */ > > + if (unlikely(skb->dev != dev || skb->skb_iif != dev->ifindex)) > > + goto drop; > > + > > + return skb; > > + > > +drop: > > + kfree_skb(skb); > > + return ERR_PTR(-EINVAL); > > +} > > + > > +static struct net_device *end_dt4_get_vrf_rcu(struct sk_buff *skb, > > + struct seg6_end_dt4_info *info) > > +{ > > + int vrf_ifindex = info->vrf_ifindex; > > + struct net *net = info->net; > > + > > + if (unlikely(vrf_ifindex < 0)) > > + goto error; > > + > > + if (unlikely(!net_eq(dev_net(skb->dev), net))) > > + goto error; > > + > > + return dev_get_by_index_rcu(net, vrf_ifindex); > > + > > +error: > > + return NULL; > > +} > > + > > +static int input_action_end_dt4(struct sk_buff *skb, > > + struct seg6_local_lwt *slwt) > > +{ > > + struct net_device *vrf; > > + struct iphdr *iph; > > + int err; > > + > > + if (!decap_and_validate(skb, IPPROTO_IPIP)) > > + goto drop; > > + > > + if (!pskb_may_pull(skb, sizeof(struct iphdr))) > > + goto drop; > > + > > + vrf = end_dt4_get_vrf_rcu(skb, &slwt->dt4_info); > > + if (unlikely(!vrf)) > > + goto drop; > > + > > + skb->protocol = htons(ETH_P_IP); > > + > > + skb_dst_drop(skb); > > + > > + skb_set_transport_header(skb, sizeof(struct iphdr)); > > + > > + skb = end_dt4_vrf_rcv(skb, vrf); > > + if (!skb) > > + /* packet has been processed and consumed by the VRF */ > > + return 0; > > + > > + if (IS_ERR(skb)) { > > + err = PTR_ERR(skb); > > + return err; > > return PTR_ERR(skb) > I will fix it in v3. > > + } > > + > > + iph = ip_hdr(skb); > > + > > + err = ip_route_input(skb, iph->daddr, iph->saddr, 0, skb->dev); > > + if (err) > > + goto drop; > > + > > + return dst_input(skb); > > + > > +drop: > > + kfree_skb(skb); > > + return -EINVAL; > > +} > > + > > +#else > > + > > new line not needed > Ok. > > +static int seg6_end_dt4_build(struct seg6_local_lwt *slwt, const void *cfg, > > + struct netlink_ext_ack *extack) > > +{ > > + NL_SET_ERR_MSG(extack, "Operation is not supported"); > > This extack message probably could be more helpful. As it stands it's > basically > Please, see just right below. > > + > > + return -EOPNOTSUPP; > > +} > > + > > +static int input_action_end_dt4(struct sk_buff *skb, > > + struct seg6_local_lwt *slwt) > > Maybe just ifdef out the part of the action table instead of creating > those stubs? > This is a very interesting point and I like your idea. We can eliminate the two stubs while keeping the "unsupported operation" semantics in this way: static struct seg6_action_desc seg6_action_table[] = { [...] { .action = SEG6_LOCAL_ACTION_END_DT4, .attrs = (1 << SEG6_LOCAL_TABLE), #ifdef CONFIG_NET_L3_MASTER_DEV .input = input_action_end_dt4, .slwt_ops = { .build_state = seg6_end_dt4_build, }, #endif }, [...] } when the CONFIG_NET_L3_MASTER_DEV is not defined, the behavior can not be instantiated because the "input" callback is initialized to NULL. This fact forces the parse_nla_action() to fail returning -EOPNOTSUPP to the user (that is exactly what we want to achieve). Note that surrounding the entire DT4 action table entry with #ifdef/#endif does not allow us to distinguish whether the DT4 was really implemented or it was not supported due to the way in which the CONFIG_NET_L3_MASTER_DEV was set. In both cases, when the user tries to instantiate a new DT4 behavior, the kernel replies back with the -EINVAL error. > > +{ > > + kfree_skb(skb); > > + return -EOPNOTSUPP; > > +} > > + > > +#endif > > + > > static int input_action_end_dt6(struct sk_buff *skb, > > struct seg6_local_lwt *slwt) > > { > > @@ -601,6 +798,14 @@ static struct seg6_action_desc seg6_action_table[] = { > > BTW any idea why the action table is not marked as const? > Frankly speaking, I have no idea. I have been working on the seg6 infrastructure for some time now, and I have never seen a single value changed in seg6_action_table[] after its initialization (neither the necessity to carry out an update operation). > Would you mind sending a patch to fix that? > Yes, I will send a fix for this issue adding the 'const' keyword. > > .attrs = (1 << SEG6_LOCAL_NH4), > > .input = input_action_end_dx4, > > }, > > + { > > + .action = SEG6_LOCAL_ACTION_END_DT4, > > + .attrs = (1 << SEG6_LOCAL_TABLE), > > + .input = input_action_end_dt4, > > + .slwt_ops = { > > + .build_state = seg6_end_dt4_build, > > + }, > > + }, > > { > > .action = SEG6_LOCAL_ACTION_END_DT6, > > .attrs = (1 << SEG6_LOCAL_TABLE), > Thank you, Andrea
