include/net/ip_tunnels.h

   1 #ifndef __NET_IP_TUNNELS_H
   2 #define __NET_IP_TUNNELS_H 1
   3
   4 #include <linux/if_tunnel.h>
   5 #include <linux/netdevice.h>
   6 #include <linux/skbuff.h>
   7 #include <linux/socket.h>
   8 #include <linux/types.h>
   9 #include <linux/u64_stats_sync.h>
  10 #include <net/dsfield.h>
  11 #include <net/gro_cells.h>
  12 #include <net/inet_ecn.h>
  13 #include <net/netns/generic.h>
  14 #include <net/rtnetlink.h>
  15 #include <net/lwtunnel.h>
  16
  17 #if IS_ENABLED(CONFIG_IPV6)
  18 #include <net/ipv6.h>
  19 #include <net/ip6_fib.h>
  20 #include <net/ip6_route.h>
  21 #endif
  22
  23 /* Keep error state on tunnel for 30 sec */
  24 #define IPTUNNEL_ERR_TIMEO      (30*HZ)
  25
  26 /* Used to memset ip_tunnel padding. */
  27 #define IP_TUNNEL_KEY_SIZE      offsetofend(struct ip_tunnel_key, tp_dst)
  28
  29 /* Used to memset ipv4 address padding. */
  30 #define IP_TUNNEL_KEY_IPV4_PAD  offsetofend(struct ip_tunnel_key, u.ipv4.dst)
  31 #define IP_TUNNEL_KEY_IPV4_PAD_LEN                              \
  32         (FIELD_SIZEOF(struct ip_tunnel_key, u) -                \
  33          FIELD_SIZEOF(struct ip_tunnel_key, u.ipv4))
  34
  35 struct ip_tunnel_key {
  36         __be64                  tun_id;
  37         union {
  38                 struct {
  39                         __be32  src;
  40                         __be32  dst;
  41                 } ipv4;
  42                 struct {
  43                         struct in6_addr src;
  44                         struct in6_addr dst;
  45                 } ipv6;
  46         } u;
  47         __be16                  tun_flags;
  48         u8                      tos;            /* TOS for IPv4, TC for IPv6 */
  49         u8                      ttl;            /* TTL for IPv4, HL for IPv6 */
  50         __be16                  tp_src;
  51         __be16                  tp_dst;
  52 };
  53
  54 /* Flags for ip_tunnel_info mode. */
  55 #define IP_TUNNEL_INFO_TX       0x01    /* represents tx tunnel parameters */
  56 #define IP_TUNNEL_INFO_IPV6     0x02    /* key contains IPv6 addresses */
  57
  58 struct ip_tunnel_info {
  59         struct ip_tunnel_key    key;
  60         u8                      options_len;
  61         u8                      mode;
  62 };
  63
  64 /* 6rd prefix/relay information */
  65 #ifdef CONFIG_IPV6_SIT_6RD
  66 struct ip_tunnel_6rd_parm {
  67         struct in6_addr         prefix;
  68         __be32                  relay_prefix;
  69         u16                     prefixlen;
  70         u16                     relay_prefixlen;
  71 };
  72 #endif
  73
  74 struct ip_tunnel_encap {
  75         u16                     type;
  76         u16                     flags;
  77         __be16                  sport;
  78         __be16                  dport;
  79 };
  80
  81 struct ip_tunnel_prl_entry {
  82         struct ip_tunnel_prl_entry __rcu *next;
  83         __be32                          addr;
  84         u16                             flags;
  85         struct rcu_head                 rcu_head;
  86 };
  87
  88 struct ip_tunnel_dst {
  89         struct dst_entry __rcu          *dst;
  90         __be32                           saddr;
  91 };
  92
  93 struct metadata_dst;
  94
  95 struct ip_tunnel {
  96         struct ip_tunnel __rcu  *next;
  97         struct hlist_node hash_node;
  98         struct net_device       *dev;
  99         struct net              *net;   /* netns for packet i/o */
 100
 101         int             err_count;      /* Number of arrived ICMP errors */
 102         unsigned long   err_time;       /* Time when the last ICMP error
 103                                          * arrived */
 104
 105         /* These four fields used only by GRE */
 106         u32             i_seqno;        /* The last seen seqno  */
 107         u32             o_seqno;        /* The last output seqno */
 108         int             tun_hlen;       /* Precalculated header length */
 109         int             mlink;
 110
 111         struct ip_tunnel_dst __percpu *dst_cache;
 112
 113         struct ip_tunnel_parm parms;
 114
 115         int             encap_hlen;     /* Encap header length (FOU,GUE) */
 116         struct ip_tunnel_encap encap;
 117
 118         int             hlen;           /* tun_hlen + encap_hlen */
 119
 120         /* for SIT */
 121 #ifdef CONFIG_IPV6_SIT_6RD
 122         struct ip_tunnel_6rd_parm ip6rd;
 123 #endif
 124         struct ip_tunnel_prl_entry __rcu *prl;  /* potential router list */
 125         unsigned int            prl_count;      /* # of entries in PRL */
 126         int                     ip_tnl_net_id;
 127         struct gro_cells        gro_cells;
 128         bool                    collect_md;
 129 };
 130
 131 #define TUNNEL_CSUM             __cpu_to_be16(0x01)
 132 #define TUNNEL_ROUTING          __cpu_to_be16(0x02)
 133 #define TUNNEL_KEY              __cpu_to_be16(0x04)
 134 #define TUNNEL_SEQ              __cpu_to_be16(0x08)
 135 #define TUNNEL_STRICT           __cpu_to_be16(0x10)
 136 #define TUNNEL_REC              __cpu_to_be16(0x20)
 137 #define TUNNEL_VERSION          __cpu_to_be16(0x40)
 138 #define TUNNEL_NO_KEY           __cpu_to_be16(0x80)
 139 #define TUNNEL_DONT_FRAGMENT    __cpu_to_be16(0x0100)
 140 #define TUNNEL_OAM              __cpu_to_be16(0x0200)
 141 #define TUNNEL_CRIT_OPT         __cpu_to_be16(0x0400)
 142 #define TUNNEL_GENEVE_OPT       __cpu_to_be16(0x0800)
 143 #define TUNNEL_VXLAN_OPT        __cpu_to_be16(0x1000)
 144
 145 #define TUNNEL_OPTIONS_PRESENT  (TUNNEL_GENEVE_OPT | TUNNEL_VXLAN_OPT)
 146
 147 struct tnl_ptk_info {
 148         __be16 flags;
 149         __be16 proto;
 150         __be32 key;
 151         __be32 seq;
 152 };
 153
 154 #define PACKET_RCVD     0
 155 #define PACKET_REJECT   1
 156
 157 #define IP_TNL_HASH_BITS   7
 158 #define IP_TNL_HASH_SIZE   (1 << IP_TNL_HASH_BITS)
 159
 160 struct ip_tunnel_net {
 161         struct net_device *fb_tunnel_dev;
 162         struct hlist_head tunnels[IP_TNL_HASH_SIZE];
 163         struct ip_tunnel __rcu *collect_md_tun;
 164 };
 165
 166 struct ip_tunnel_encap_ops {
 167         size_t (*encap_hlen)(struct ip_tunnel_encap *e);
 168         int (*build_header)(struct sk_buff *skb, struct ip_tunnel_encap *e,
 169                             u8 *protocol, struct flowi4 *fl4);
 170 };
 171
 172 #define MAX_IPTUN_ENCAP_OPS 8
 173
 174 extern const struct ip_tunnel_encap_ops __rcu *
 175                 iptun_encaps[MAX_IPTUN_ENCAP_OPS];
 176
 177 int ip_tunnel_encap_add_ops(const struct ip_tunnel_encap_ops *op,
 178                             unsigned int num);
 179 int ip_tunnel_encap_del_ops(const struct ip_tunnel_encap_ops *op,
 180                             unsigned int num);
 181
 182 static inline void ip_tunnel_key_init(struct ip_tunnel_key *key,
 183                                       __be32 saddr, __be32 daddr,
 184                                       u8 tos, u8 ttl,
 185                                       __be16 tp_src, __be16 tp_dst,
 186                                       __be64 tun_id, __be16 tun_flags)
 187 {
 188         key->tun_id = tun_id;
 189         key->u.ipv4.src = saddr;
 190         key->u.ipv4.dst = daddr;
 191         memset((unsigned char *)key + IP_TUNNEL_KEY_IPV4_PAD,
 192                0, IP_TUNNEL_KEY_IPV4_PAD_LEN);
 193         key->tos = tos;
 194         key->ttl = ttl;
 195         key->tun_flags = tun_flags;
 196
 197         /* For the tunnel types on the top of IPsec, the tp_src and tp_dst of
 198          * the upper tunnel are used.
 199          * E.g: GRE over IPSEC, the tp_src and tp_port are zero.
 200          */
 201         key->tp_src = tp_src;
 202         key->tp_dst = tp_dst;
 203
 204         /* Clear struct padding. */
 205         if (sizeof(*key) != IP_TUNNEL_KEY_SIZE)
 206                 memset((unsigned char *)key + IP_TUNNEL_KEY_SIZE,
 207                        0, sizeof(*key) - IP_TUNNEL_KEY_SIZE);
 208 }
 209
 210 static inline unsigned short ip_tunnel_info_af(const struct ip_tunnel_info
 211                                                *tun_info)
 212 {
 213         return tun_info->mode & IP_TUNNEL_INFO_IPV6 ? AF_INET6 : AF_INET;
 214 }
 215
 216 #ifdef CONFIG_INET
 217
 218 int ip_tunnel_init(struct net_device *dev);
 219 void ip_tunnel_uninit(struct net_device *dev);
 220 void  ip_tunnel_dellink(struct net_device *dev, struct list_head *head);
 221 struct net *ip_tunnel_get_link_net(const struct net_device *dev);
 222 int ip_tunnel_get_iflink(const struct net_device *dev);
 223 int ip_tunnel_init_net(struct net *net, int ip_tnl_net_id,
 224                        struct rtnl_link_ops *ops, char *devname);
 225
 226 void ip_tunnel_delete_net(struct ip_tunnel_net *itn, struct rtnl_link_ops *ops);
 227
 228 void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev,
 229                     const struct iphdr *tnl_params, const u8 protocol);
 230 int ip_tunnel_ioctl(struct net_device *dev, struct ip_tunnel_parm *p, int cmd);
 231 int ip_tunnel_encap(struct sk_buff *skb, struct ip_tunnel *t,
 232                     u8 *protocol, struct flowi4 *fl4);
 233 int ip_tunnel_change_mtu(struct net_device *dev, int new_mtu);
 234
 235 struct rtnl_link_stats64 *ip_tunnel_get_stats64(struct net_device *dev,
 236                                                 struct rtnl_link_stats64 *tot);
 237 struct ip_tunnel *ip_tunnel_lookup(struct ip_tunnel_net *itn,
 238                                    int link, __be16 flags,
 239                                    __be32 remote, __be32 local,
 240                                    __be32 key);
 241
 242 int ip_tunnel_rcv(struct ip_tunnel *tunnel, struct sk_buff *skb,
 243                   const struct tnl_ptk_info *tpi, struct metadata_dst *tun_dst,
 244                   bool log_ecn_error);
 245 int ip_tunnel_changelink(struct net_device *dev, struct nlattr *tb[],
 246                          struct ip_tunnel_parm *p);
 247 int ip_tunnel_newlink(struct net_device *dev, struct nlattr *tb[],
 248                       struct ip_tunnel_parm *p);
 249 void ip_tunnel_setup(struct net_device *dev, int net_id);
 250 void ip_tunnel_dst_reset_all(struct ip_tunnel *t);
 251 int ip_tunnel_encap_setup(struct ip_tunnel *t,
 252                           struct ip_tunnel_encap *ipencap);
 253
 254 /* Extract dsfield from inner protocol */
 255 static inline u8 ip_tunnel_get_dsfield(const struct iphdr *iph,
 256                                        const struct sk_buff *skb)
 257 {
 258         if (skb->protocol == htons(ETH_P_IP))
 259                 return iph->tos;
 260         else if (skb->protocol == htons(ETH_P_IPV6))
 261                 return ipv6_get_dsfield((const struct ipv6hdr *)iph);
 262         else
 263                 return 0;
 264 }
 265
 266 /* Propogate ECN bits out */
 267 static inline u8 ip_tunnel_ecn_encap(u8 tos, const struct iphdr *iph,
 268                                      const struct sk_buff *skb)
 269 {
 270         u8 inner = ip_tunnel_get_dsfield(iph, skb);
 271
 272         return INET_ECN_encapsulate(tos, inner);
 273 }
 274
 275 int iptunnel_pull_header(struct sk_buff *skb, int hdr_len, __be16 inner_proto);
 276 int iptunnel_xmit(struct sock *sk, struct rtable *rt, struct sk_buff *skb,
 277                   __be32 src, __be32 dst, u8 proto,
 278                   u8 tos, u8 ttl, __be16 df, bool xnet);
 279 struct metadata_dst *iptunnel_metadata_reply(struct metadata_dst *md,
 280                                              gfp_t flags);
 281
 282 struct sk_buff *iptunnel_handle_offloads(struct sk_buff *skb, bool gre_csum,
 283                                          int gso_type_mask);
 284
 285 static inline void iptunnel_xmit_stats(int err,
 286                                        struct net_device_stats *err_stats,
 287                                        struct pcpu_sw_netstats __percpu *stats)
 288 {
 289         if (err > 0) {
 290                 struct pcpu_sw_netstats *tstats = this_cpu_ptr(stats);
 291
 292                 u64_stats_update_begin(&tstats->syncp);
 293                 tstats->tx_bytes += err;
 294                 tstats->tx_packets++;
 295                 u64_stats_update_end(&tstats->syncp);
 296         } else if (err < 0) {
 297                 err_stats->tx_errors++;
 298                 err_stats->tx_aborted_errors++;
 299         } else {
 300                 err_stats->tx_dropped++;
 301         }
 302 }
 303
 304 static inline void *ip_tunnel_info_opts(struct ip_tunnel_info *info)
 305 {
 306         return info + 1;
 307 }
 308
 309 static inline void ip_tunnel_info_opts_get(void *to,
 310                                            const struct ip_tunnel_info *info)
 311 {
 312         memcpy(to, info + 1, info->options_len);
 313 }
 314
 315 static inline void ip_tunnel_info_opts_set(struct ip_tunnel_info *info,
 316                                            const void *from, int len)
 317 {
 318         memcpy(ip_tunnel_info_opts(info), from, len);
 319         info->options_len = len;
 320 }
 321
 322 static inline struct ip_tunnel_info *lwt_tun_info(struct lwtunnel_state *lwtstate)
 323 {
 324         return (struct ip_tunnel_info *)lwtstate->data;
 325 }
 326
 327 extern struct static_key ip_tunnel_metadata_cnt;
 328
 329 /* Returns > 0 if metadata should be collected */
 330 static inline int ip_tunnel_collect_metadata(void)
 331 {
 332         return static_key_false(&ip_tunnel_metadata_cnt);
 333 }
 334
 335 void __init ip_tunnel_core_init(void);
 336
 337 void ip_tunnel_need_metadata(void);
 338 void ip_tunnel_unneed_metadata(void);
 339
 340 #else /* CONFIG_INET */
 341
 342 static inline struct ip_tunnel_info *lwt_tun_info(struct lwtunnel_state *lwtstate)
 343 {
 344         return NULL;
 345 }
 346
 347 static inline void ip_tunnel_need_metadata(void)
 348 {
 349 }
 350
 351 static inline void ip_tunnel_unneed_metadata(void)
 352 {
 353 }
 354
 355 #endif /* CONFIG_INET */
 356
 357 #endif /* __NET_IP_TUNNELS_H */