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