datapath: define compat ip_tunnel_get_link_net()
[cascardo/ovs.git] / datapath / linux / compat / include / net / ip_tunnels.h
index a786aa9..7083372 100644 (file)
@@ -1,6 +1,14 @@
 #ifndef __NET_IP_TUNNELS_WRAPPER_H
 #define __NET_IP_TUNNELS_WRAPPER_H 1
 
+#include <linux/version.h>
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(3,10,0)
+#include_next <net/ip_tunnels.h>
+#endif
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3,18,0)
+
 #include <linux/if_tunnel.h>
 #include <linux/netdevice.h>
 #include <linux/skbuff.h>
 #include <net/ip.h>
 #include <net/rtnetlink.h>
 
+struct sk_buff *ovs_iptunnel_handle_offloads(struct sk_buff *skb,
+                                            bool csum_help, int gso_type_mask,
+                                            void (*fix_segment)(struct sk_buff *));
+
+#define iptunnel_xmit rpl_iptunnel_xmit
+int rpl_iptunnel_xmit(struct sock *sk, struct rtable *rt, struct sk_buff *skb,
+                     __be32 src, __be32 dst, __u8 proto, __u8 tos, __u8 ttl,
+                     __be16 df, bool xnet);
+
+#define iptunnel_pull_header rpl_iptunnel_pull_header
+int rpl_iptunnel_pull_header(struct sk_buff *skb, int hdr_len, __be16 inner_proto);
+
+#else
+
+#define ovs_iptunnel_handle_offloads(skb, csum_help, gso_type_mask, fix_segment) \
+       iptunnel_handle_offloads(skb, csum_help, gso_type_mask)
+
+#endif /* 3.18 */
+
+#ifndef TUNNEL_CSUM
 #define TUNNEL_CSUM    __cpu_to_be16(0x01)
 #define TUNNEL_ROUTING __cpu_to_be16(0x02)
 #define TUNNEL_KEY     __cpu_to_be16(0x04)
@@ -19,7 +47,6 @@
 #define TUNNEL_REC     __cpu_to_be16(0x20)
 #define TUNNEL_VERSION __cpu_to_be16(0x40)
 #define TUNNEL_NO_KEY  __cpu_to_be16(0x80)
-#define TUNNEL_DONT_FRAGMENT   __cpu_to_be16(0x0100)
 
 struct tnl_ptk_info {
        __be16 flags;
@@ -30,11 +57,255 @@ struct tnl_ptk_info {
 
 #define PACKET_RCVD    0
 #define PACKET_REJECT  1
+#endif
+
+#ifndef TUNNEL_DONT_FRAGMENT
+#define TUNNEL_DONT_FRAGMENT   __cpu_to_be16(0x0100)
+#endif
+
+#ifndef TUNNEL_OAM
+#define TUNNEL_OAM     __cpu_to_be16(0x0200)
+#define TUNNEL_CRIT_OPT        __cpu_to_be16(0x0400)
+#endif
+
+#ifndef TUNNEL_GENEVE_OPT
+#define TUNNEL_GENEVE_OPT      __cpu_to_be16(0x0800)
+#endif
+
+#ifndef TUNNEL_VXLAN_OPT
+#define TUNNEL_VXLAN_OPT       __cpu_to_be16(0x1000)
+#endif
+
+/* Older kernels defined TUNNEL_OPTIONS_PRESENT to GENEVE only */
+#undef TUNNEL_OPTIONS_PRESENT
+#define TUNNEL_OPTIONS_PRESENT (TUNNEL_GENEVE_OPT | TUNNEL_VXLAN_OPT)
+
+#define skb_is_encapsulated ovs_skb_is_encapsulated
+bool ovs_skb_is_encapsulated(struct sk_buff *skb);
+
+#ifndef HAVE_METADATA_DST
+/* Used to memset ip_tunnel padding. */
+#define IP_TUNNEL_KEY_SIZE     offsetofend(struct ip_tunnel_key, tp_dst)
+
+/* Used to memset ipv4 address padding. */
+#define IP_TUNNEL_KEY_IPV4_PAD offsetofend(struct ip_tunnel_key, u.ipv4.dst)
+#define IP_TUNNEL_KEY_IPV4_PAD_LEN                             \
+       (FIELD_SIZEOF(struct ip_tunnel_key, u) -                \
+        FIELD_SIZEOF(struct ip_tunnel_key, u.ipv4))
+
+struct ip_tunnel_key {
+       __be64                  tun_id;
+       union {
+               struct {
+                       __be32  src;
+                       __be32  dst;
+               } ipv4;
+               struct {
+                       struct in6_addr src;
+                       struct in6_addr dst;
+               } ipv6;
+       } u;
+       __be16                  tun_flags;
+       u8                      tos;            /* TOS for IPv4, TC for IPv6 */
+       u8                      ttl;            /* TTL for IPv4, HL for IPv6 */
+       __be16                  tp_src;
+       __be16                  tp_dst;
+};
+
+/* Flags for ip_tunnel_info mode. */
+#define IP_TUNNEL_INFO_TX      0x01    /* represents tx tunnel parameters */
+#define IP_TUNNEL_INFO_IPV6    0x02    /* key contains IPv6 addresses */
+
+struct ip_tunnel_info {
+       struct ip_tunnel_key    key;
+       u8                      options_len;
+       u8                      mode;
+};
+
+static inline unsigned short ip_tunnel_info_af(const struct ip_tunnel_info *tun_info)
+{
+       return tun_info->mode & IP_TUNNEL_INFO_IPV6 ? AF_INET6 : AF_INET;
+}
+
+static inline void *ip_tunnel_info_opts(struct ip_tunnel_info *info)
+{
+       return info + 1;
+}
+
+static inline void ip_tunnel_info_opts_get(void *to,
+                                          const struct ip_tunnel_info *info)
+{
+       memcpy(to, info + 1, info->options_len);
+}
+
+static inline void ip_tunnel_info_opts_set(struct ip_tunnel_info *info,
+                                          const void *from, int len)
+{
+       memcpy(ip_tunnel_info_opts(info), from, len);
+       info->options_len = len;
+}
+
+static inline void ip_tunnel_key_init(struct ip_tunnel_key *key,
+                                     __be32 saddr, __be32 daddr,
+                                     u8 tos, u8 ttl,
+                                     __be16 tp_src, __be16 tp_dst,
+                                     __be64 tun_id, __be16 tun_flags)
+{
+       key->tun_id = tun_id;
+       key->u.ipv4.src = saddr;
+       key->u.ipv4.dst = daddr;
+       memset((unsigned char *)key + IP_TUNNEL_KEY_IPV4_PAD,
+              0, IP_TUNNEL_KEY_IPV4_PAD_LEN);
+       key->tos = tos;
+       key->ttl = ttl;
+       key->tun_flags = tun_flags;
+
+       /* For the tunnel types on the top of IPsec, the tp_src and tp_dst of
+        * the upper tunnel are used.
+        * E.g: GRE over IPSEC, the tp_src and tp_port are zero.
+        */
+       key->tp_src = tp_src;
+       key->tp_dst = tp_dst;
+
+       /* Clear struct padding. */
+       if (sizeof(*key) != IP_TUNNEL_KEY_SIZE)
+               memset((unsigned char *)key + IP_TUNNEL_KEY_SIZE,
+                      0, sizeof(*key) - IP_TUNNEL_KEY_SIZE);
+}
+
+#define ip_tunnel_collect_metadata() true
+
+
+#define ip_tunnel rpl_ip_tunnel
+
+struct ip_tunnel {
+       struct net_device       *dev;
+       struct net              *net;   /* netns for packet i/o */
+
+       int             err_count;      /* Number of arrived ICMP errors */
+       unsigned long   err_time;       /* Time when the last ICMP error
+                                        * arrived
+                                        */
+
+       /* These four fields used only by GRE */
+       u32             i_seqno;        /* The last seen seqno  */
+       u32             o_seqno;        /* The last output seqno */
+       int             tun_hlen;       /* Precalculated header length */
+       int             mlink;
+
+       struct ip_tunnel_parm parms;
+
+       int             encap_hlen;     /* Encap header length (FOU,GUE) */
+       int             hlen;           /* tun_hlen + encap_hlen */
+
+       int             ip_tnl_net_id;
+       bool            collect_md;
+};
+
+#define ip_tunnel_net rpl_ip_tunnel_net
+struct ip_tunnel_net {
+       struct ip_tunnel __rcu *collect_md_tun;
+       struct rtnl_link_ops *rtnl_ops;
+};
+
+
+#ifndef HAVE_PCPU_SW_NETSTATS
+#define ip_tunnel_get_stats64 rpl_ip_tunnel_get_stats64
+struct rtnl_link_stats64 *rpl_ip_tunnel_get_stats64(struct net_device *dev,
+                                                   struct rtnl_link_stats64 *tot);
+#endif
+
+#define ip_tunnel_get_dsfield rpl_ip_tunnel_get_dsfield
+static inline u8 ip_tunnel_get_dsfield(const struct iphdr *iph,
+               const struct sk_buff *skb)
+{
+       if (skb->protocol == htons(ETH_P_IP))
+               return iph->tos;
+       else if (skb->protocol == htons(ETH_P_IPV6))
+               return ipv6_get_dsfield((const struct ipv6hdr *)iph);
+       else
+               return 0;
+}
+
+#define ip_tunnel_ecn_encap rpl_ip_tunnel_ecn_encap
+static inline u8 ip_tunnel_ecn_encap(u8 tos, const struct iphdr *iph,
+               const struct sk_buff *skb)
+{
+       u8 inner = ip_tunnel_get_dsfield(iph, skb);
+
+       return INET_ECN_encapsulate(tos, inner);
+}
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
+#define iptunnel_xmit_stats(err, stats, dummy)         \
+do {                                                   \
+       if (err > 0) {                                  \
+               (stats)->tx_bytes += err;               \
+               (stats)->tx_packets++;                  \
+       } else if (err < 0) {                           \
+               (stats)->tx_errors++;                   \
+               (stats)->tx_aborted_errors++;           \
+       } else {                                        \
+               (stats)->tx_dropped++;                  \
+       }                                               \
+} while (0)
+
+#else
+#define iptunnel_xmit_stats rpl_iptunnel_xmit_stats
+static inline void iptunnel_xmit_stats(int err,
+               struct net_device_stats *err_stats,
+               struct pcpu_sw_netstats __percpu *stats)
+{
+       if (err > 0) {
+               struct pcpu_sw_netstats *tstats = this_cpu_ptr(stats);
+
+               u64_stats_update_begin(&tstats->syncp);
+               tstats->tx_bytes += err;
+               tstats->tx_packets++;
+               u64_stats_update_end(&tstats->syncp);
+       } else if (err < 0) {
+               err_stats->tx_errors++;
+               err_stats->tx_aborted_errors++;
+       } else {
+               err_stats->tx_dropped++;
+       }
+}
+#endif
+
+#define ip_tunnel_init rpl_ip_tunnel_init
+int rpl_ip_tunnel_init(struct net_device *dev);
+
+#define ip_tunnel_uninit rpl_ip_tunnel_uninit
+void rpl_ip_tunnel_uninit(struct net_device *dev);
+
+#define ip_tunnel_change_mtu rpl_ip_tunnel_change_mtu
+int rpl_ip_tunnel_change_mtu(struct net_device *dev, int new_mtu);
+
+#define ip_tunnel_newlink rpl_ip_tunnel_newlink
+int rpl_ip_tunnel_newlink(struct net_device *dev, struct nlattr *tb[],
+                         struct ip_tunnel_parm *p);
+
+#define ip_tunnel_dellink rpl_ip_tunnel_dellink
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,39)
+void rpl_ip_tunnel_dellink(struct net_device *dev, struct list_head *head);
+#else
+void rpl_ip_tunnel_dellink(struct net_device *dev);
+#endif
+
+#define ip_tunnel_init_net rpl_ip_tunnel_init_net
+int rpl_ip_tunnel_init_net(struct net *net, int ip_tnl_net_id,
+                          struct rtnl_link_ops *ops, char *devname);
+
+#define ip_tunnel_delete_net rpl_ip_tunnel_delete_net
+void rpl_ip_tunnel_delete_net(struct ip_tunnel_net *itn, struct rtnl_link_ops *ops);
+
+#define ip_tunnel_setup rpl_ip_tunnel_setup
+void rpl_ip_tunnel_setup(struct net_device *dev, int net_id);
 
-int iptunnel_xmit(struct rtable *rt,
-                 struct sk_buff *skb,
-                 __be32 src, __be32 dst, __u8 proto,
-                 __u8 tos, __u8 ttl, __be16 df);
+#define ip_tunnel_get_iflink rpl_ip_tunnel_get_iflink
+int rpl_ip_tunnel_get_iflink(const struct net_device *dev);
 
-int iptunnel_pull_header(struct sk_buff *skb, int hdr_len, __be16 inner_proto);
+#define ip_tunnel_get_link_net rpl_ip_tunnel_get_link_net
+struct net *rpl_ip_tunnel_get_link_net(const struct net_device *dev);
+#endif /* HAVE_METADATA_DST */
 #endif /* __NET_IP_TUNNELS_H */