ip6_gre: add ip6 gre and gretap collect_md mode
authorWilliam Tu <u9012063@gmail.com>
Fri, 1 Dec 2017 23:26:08 +0000 (15:26 -0800)
committerDavid S. Miller <davem@davemloft.net>
Mon, 4 Dec 2017 16:04:19 +0000 (11:04 -0500)
Similar to gre, vxlan, geneve, ipip tunnels, allow ip6 gre and gretap
tunnels to operate in collect metadata mode.  bpf_skb_[gs]et_tunnel_key()
helpers can make use of it right away.  OVS can use it as well in the
future.

Signed-off-by: William Tu <u9012063@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv6/ip6_gre.c
net/ipv6/ip6_tunnel.c

index 76379f0..1510ce9 100644 (file)
@@ -56,6 +56,7 @@
 #include <net/ip6_tunnel.h>
 #include <net/gre.h>
 #include <net/erspan.h>
+#include <net/dst_metadata.h>
 
 
 static bool log_ecn_error = true;
@@ -69,6 +70,7 @@ static unsigned int ip6gre_net_id __read_mostly;
 struct ip6gre_net {
        struct ip6_tnl __rcu *tunnels[4][IP6_GRE_HASH_SIZE];
 
+       struct ip6_tnl __rcu *collect_md_tun;
        struct net_device *fb_tunnel_dev;
 };
 
@@ -229,6 +231,10 @@ static struct ip6_tnl *ip6gre_tunnel_lookup(struct net_device *dev,
        if (cand)
                return cand;
 
+       t = rcu_dereference(ign->collect_md_tun);
+       if (t && t->dev->flags & IFF_UP)
+               return t;
+
        dev = ign->fb_tunnel_dev;
        if (dev->flags & IFF_UP)
                return netdev_priv(dev);
@@ -264,6 +270,9 @@ static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t)
 {
        struct ip6_tnl __rcu **tp = ip6gre_bucket(ign, t);
 
+       if (t->parms.collect_md)
+               rcu_assign_pointer(ign->collect_md_tun, t);
+
        rcu_assign_pointer(t->next, rtnl_dereference(*tp));
        rcu_assign_pointer(*tp, t);
 }
@@ -273,6 +282,9 @@ static void ip6gre_tunnel_unlink(struct ip6gre_net *ign, struct ip6_tnl *t)
        struct ip6_tnl __rcu **tp;
        struct ip6_tnl *iter;
 
+       if (t->parms.collect_md)
+               rcu_assign_pointer(ign->collect_md_tun, NULL);
+
        for (tp = ip6gre_bucket(ign, t);
             (iter = rtnl_dereference(*tp)) != NULL;
             tp = &iter->next) {
@@ -463,7 +475,22 @@ static int ip6gre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi)
                                      &ipv6h->saddr, &ipv6h->daddr, tpi->key,
                                      tpi->proto);
        if (tunnel) {
-               ip6_tnl_rcv(tunnel, skb, tpi, NULL, log_ecn_error);
+               if (tunnel->parms.collect_md) {
+                       struct metadata_dst *tun_dst;
+                       __be64 tun_id;
+                       __be16 flags;
+
+                       flags = tpi->flags;
+                       tun_id = key32_to_tunnel_id(tpi->key);
+
+                       tun_dst = ipv6_tun_rx_dst(skb, flags, tun_id, 0);
+                       if (!tun_dst)
+                               return PACKET_REJECT;
+
+                       ip6_tnl_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
+               } else {
+                       ip6_tnl_rcv(tunnel, skb, tpi, NULL, log_ecn_error);
+               }
 
                return PACKET_RCVD;
        }
@@ -633,8 +660,38 @@ static netdev_tx_t __gre6_xmit(struct sk_buff *skb,
 
        /* Push GRE header. */
        protocol = (dev->type == ARPHRD_ETHER) ? htons(ETH_P_TEB) : proto;
-       gre_build_header(skb, tunnel->tun_hlen, tunnel->parms.o_flags,
-                        protocol, tunnel->parms.o_key, htonl(tunnel->o_seqno));
+
+       if (tunnel->parms.collect_md) {
+               struct ip_tunnel_info *tun_info;
+               const struct ip_tunnel_key *key;
+               __be16 flags;
+
+               tun_info = skb_tunnel_info(skb);
+               if (unlikely(!tun_info ||
+                            !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
+                            ip_tunnel_info_af(tun_info) != AF_INET6))
+                       return -EINVAL;
+
+               key = &tun_info->key;
+               memset(fl6, 0, sizeof(*fl6));
+               fl6->flowi6_proto = IPPROTO_GRE;
+               fl6->daddr = key->u.ipv6.dst;
+               fl6->flowlabel = key->label;
+               fl6->flowi6_uid = sock_net_uid(dev_net(dev), NULL);
+
+               dsfield = key->tos;
+               flags = key->tun_flags & (TUNNEL_CSUM | TUNNEL_KEY);
+               tunnel->tun_hlen = gre_calc_hlen(flags);
+
+               gre_build_header(skb, tunnel->tun_hlen,
+                                flags, protocol,
+                                tunnel_id_to_key32(tun_info->key.tun_id), 0);
+
+       } else {
+               gre_build_header(skb, tunnel->tun_hlen, tunnel->parms.o_flags,
+                                protocol, tunnel->parms.o_key,
+                                htonl(tunnel->o_seqno));
+       }
 
        return ip6_tnl_xmit(skb, dev, dsfield, fl6, encap_limit, pmtu,
                            NEXTHDR_GRE);
@@ -645,13 +702,15 @@ static inline int ip6gre_xmit_ipv4(struct sk_buff *skb, struct net_device *dev)
        struct ip6_tnl *t = netdev_priv(dev);
        int encap_limit = -1;
        struct flowi6 fl6;
-       __u8 dsfield;
+       __u8 dsfield = 0;
        __u32 mtu;
        int err;
 
        memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 
-       prepare_ip6gre_xmit_ipv4(skb, dev, &fl6, &dsfield, &encap_limit);
+       if (!t->parms.collect_md)
+               prepare_ip6gre_xmit_ipv4(skb, dev, &fl6,
+                                        &dsfield, &encap_limit);
 
        err = gre_handle_offloads(skb, !!(t->parms.o_flags & TUNNEL_CSUM));
        if (err)
@@ -676,14 +735,15 @@ static inline int ip6gre_xmit_ipv6(struct sk_buff *skb, struct net_device *dev)
        struct ipv6hdr *ipv6h = ipv6_hdr(skb);
        int encap_limit = -1;
        struct flowi6 fl6;
-       __u8 dsfield;
+       __u8 dsfield = 0;
        __u32 mtu;
        int err;
 
        if (ipv6_addr_equal(&t->parms.raddr, &ipv6h->saddr))
                return -1;
 
-       if (prepare_ip6gre_xmit_ipv6(skb, dev, &fl6, &dsfield, &encap_limit))
+       if (!t->parms.collect_md &&
+           prepare_ip6gre_xmit_ipv6(skb, dev, &fl6, &dsfield, &encap_limit))
                return -1;
 
        if (gre_handle_offloads(skb, !!(t->parms.o_flags & TUNNEL_CSUM)))
@@ -731,7 +791,8 @@ static int ip6gre_xmit_other(struct sk_buff *skb, struct net_device *dev)
        if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
                encap_limit = t->parms.encap_limit;
 
-       memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
+       if (!t->parms.collect_md)
+               memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
 
        err = gre_handle_offloads(skb, !!(t->parms.o_flags & TUNNEL_CSUM));
        if (err)
@@ -1201,6 +1262,11 @@ static int ip6gre_tunnel_init_common(struct net_device *dev)
        if (!(tunnel->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
                dev->mtu -= 8;
 
+       if (tunnel->parms.collect_md) {
+               dev->features |= NETIF_F_NETNS_LOCAL;
+               netif_keep_dst(dev);
+       }
+
        return 0;
 }
 
@@ -1215,6 +1281,9 @@ static int ip6gre_tunnel_init(struct net_device *dev)
 
        tunnel = netdev_priv(dev);
 
+       if (tunnel->parms.collect_md)
+               return 0;
+
        memcpy(dev->dev_addr, &tunnel->parms.laddr, sizeof(struct in6_addr));
        memcpy(dev->broadcast, &tunnel->parms.raddr, sizeof(struct in6_addr));
 
@@ -1464,6 +1533,9 @@ static void ip6gre_netlink_parms(struct nlattr *data[],
 
        if (data[IFLA_GRE_ERSPAN_INDEX])
                parms->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
+
+       if (data[IFLA_GRE_COLLECT_METADATA])
+               parms->collect_md = true;
 }
 
 static int ip6gre_tap_init(struct net_device *dev)
@@ -1622,8 +1694,13 @@ static int ip6gre_newlink(struct net *src_net, struct net_device *dev,
 
        ip6gre_netlink_parms(data, &nt->parms);
 
-       if (ip6gre_tunnel_find(net, &nt->parms, dev->type))
-               return -EEXIST;
+       if (nt->parms.collect_md) {
+               if (rtnl_dereference(ign->collect_md_tun))
+                       return -EEXIST;
+       } else {
+               if (ip6gre_tunnel_find(net, &nt->parms, dev->type))
+                       return -EEXIST;
+       }
 
        if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
                eth_hw_addr_random(dev);
@@ -1742,6 +1819,8 @@ static size_t ip6gre_get_size(const struct net_device *dev)
                nla_total_size(2) +
                /* IFLA_GRE_ENCAP_DPORT */
                nla_total_size(2) +
+               /* IFLA_GRE_COLLECT_METADATA */
+               nla_total_size(0) +
                /* IFLA_GRE_FWMARK */
                nla_total_size(4) +
                /* IFLA_GRE_ERSPAN_INDEX */
@@ -1781,6 +1860,11 @@ static int ip6gre_fill_info(struct sk_buff *skb, const struct net_device *dev)
                        t->encap.flags))
                goto nla_put_failure;
 
+       if (p->collect_md) {
+               if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
+                       goto nla_put_failure;
+       }
+
        return 0;
 
 nla_put_failure:
@@ -1803,6 +1887,7 @@ static const struct nla_policy ip6gre_policy[IFLA_GRE_MAX + 1] = {
        [IFLA_GRE_ENCAP_FLAGS]  = { .type = NLA_U16 },
        [IFLA_GRE_ENCAP_SPORT]  = { .type = NLA_U16 },
        [IFLA_GRE_ENCAP_DPORT]  = { .type = NLA_U16 },
+       [IFLA_GRE_COLLECT_METADATA] = { .type = NLA_FLAG },
        [IFLA_GRE_FWMARK]       = { .type = NLA_U32 },
        [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 },
 };
index 3d3092a..6a3b1a5 100644 (file)
@@ -861,7 +861,7 @@ int ip6_tnl_rcv(struct ip6_tnl *t, struct sk_buff *skb,
                struct metadata_dst *tun_dst,
                bool log_ecn_err)
 {
-       return __ip6_tnl_rcv(t, skb, tpi, NULL, ip6ip6_dscp_ecn_decapsulate,
+       return __ip6_tnl_rcv(t, skb, tpi, tun_dst, ip6ip6_dscp_ecn_decapsulate,
                             log_ecn_err);
 }
 EXPORT_SYMBOL(ip6_tnl_rcv);
@@ -979,6 +979,9 @@ int ip6_tnl_xmit_ctl(struct ip6_tnl *t,
        int ret = 0;
        struct net *net = t->net;
 
+       if (t->parms.collect_md)
+               return 1;
+
        if ((p->flags & IP6_TNL_F_CAP_XMIT) ||
            ((p->flags & IP6_TNL_F_CAP_PER_PACKET) &&
             (ip6_tnl_get_cap(t, laddr, raddr) & IP6_TNL_F_CAP_XMIT))) {