Commit | Line | Data |
---|---|---|
c5441932 PS |
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> | |
7f9562a1 | 7 | #include <linux/socket.h> |
c5441932 PS |
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> | |
56328486 | 13 | #include <net/netns/generic.h> |
c5441932 | 14 | #include <net/rtnetlink.h> |
3093fbe7 | 15 | #include <net/lwtunnel.h> |
c5441932 PS |
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 | ||
1d8fff90 | 26 | /* Used to memset ip_tunnel padding. */ |
376534a3 | 27 | #define IP_TUNNEL_KEY_SIZE offsetofend(struct ip_tunnel_key, tp_dst) |
1d8fff90 | 28 | |
c1ea5d67 JB |
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 | ||
1d8fff90 TG |
35 | struct ip_tunnel_key { |
36 | __be64 tun_id; | |
c1ea5d67 JB |
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; | |
1d8fff90 | 47 | __be16 tun_flags; |
7c383fb2 JB |
48 | u8 tos; /* TOS for IPv4, TC for IPv6 */ |
49 | u8 ttl; /* TTL for IPv4, HL for IPv6 */ | |
1d8fff90 TG |
50 | __be16 tp_src; |
51 | __be16 tp_dst; | |
ac1cf399 | 52 | }; |
1d8fff90 | 53 | |
46fa062a JB |
54 | /* Flags for ip_tunnel_info mode. */ |
55 | #define IP_TUNNEL_INFO_TX 0x01 /* represents tx tunnel parameters */ | |
7f9562a1 | 56 | #define IP_TUNNEL_INFO_IPV6 0x02 /* key contains IPv6 addresses */ |
ee122c79 | 57 | |
1d8fff90 TG |
58 | struct ip_tunnel_info { |
59 | struct ip_tunnel_key key; | |
1d8fff90 | 60 | u8 options_len; |
ee122c79 | 61 | u8 mode; |
1d8fff90 TG |
62 | }; |
63 | ||
c5441932 PS |
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 | ||
56328486 | 74 | struct ip_tunnel_encap { |
6b8847c5 JB |
75 | u16 type; |
76 | u16 flags; | |
56328486 TH |
77 | __be16 sport; |
78 | __be16 dport; | |
79 | }; | |
80 | ||
c5441932 PS |
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 | ||
9a4aa9af TH |
88 | struct ip_tunnel_dst { |
89 | struct dst_entry __rcu *dst; | |
95cb5745 | 90 | __be32 saddr; |
9a4aa9af TH |
91 | }; |
92 | ||
2e15ea39 PS |
93 | struct metadata_dst; |
94 | ||
c5441932 PS |
95 | struct ip_tunnel { |
96 | struct ip_tunnel __rcu *next; | |
97 | struct hlist_node hash_node; | |
98 | struct net_device *dev; | |
5e6700b3 | 99 | struct net *net; /* netns for packet i/o */ |
c5441932 PS |
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 */ | |
6b8847c5 JB |
106 | u32 i_seqno; /* The last seen seqno */ |
107 | u32 o_seqno; /* The last output seqno */ | |
56328486 | 108 | int tun_hlen; /* Precalculated header length */ |
c5441932 PS |
109 | int mlink; |
110 | ||
9a4aa9af | 111 | struct ip_tunnel_dst __percpu *dst_cache; |
7d442fab | 112 | |
c5441932 PS |
113 | struct ip_tunnel_parm parms; |
114 | ||
56328486 TH |
115 | int encap_hlen; /* Encap header length (FOU,GUE) */ |
116 | struct ip_tunnel_encap encap; | |
117 | ||
118 | int hlen; /* tun_hlen + encap_hlen */ | |
119 | ||
c5441932 PS |
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; | |
2e15ea39 | 128 | bool collect_md; |
c5441932 PS |
129 | }; |
130 | ||
f5796684 JG |
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) | |
9a628224 | 139 | #define TUNNEL_DONT_FRAGMENT __cpu_to_be16(0x0100) |
f5796684 JG |
140 | #define TUNNEL_OAM __cpu_to_be16(0x0200) |
141 | #define TUNNEL_CRIT_OPT __cpu_to_be16(0x0400) | |
1dd144cf TG |
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) | |
c5441932 PS |
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 | ||
6261d983 | 157 | #define IP_TNL_HASH_BITS 7 |
c5441932 PS |
158 | #define IP_TNL_HASH_SIZE (1 << IP_TNL_HASH_BITS) |
159 | ||
160 | struct ip_tunnel_net { | |
c5441932 | 161 | struct net_device *fb_tunnel_dev; |
6261d983 | 162 | struct hlist_head tunnels[IP_TNL_HASH_SIZE]; |
2e15ea39 | 163 | struct ip_tunnel __rcu *collect_md_tun; |
c5441932 PS |
164 | }; |
165 | ||
a8c5f90f TH |
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 | ||
4c222798 PS |
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) | |
1d8fff90 | 187 | { |
4c222798 PS |
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, | |
c1ea5d67 | 192 | 0, IP_TUNNEL_KEY_IPV4_PAD_LEN); |
4c222798 PS |
193 | key->tos = tos; |
194 | key->ttl = ttl; | |
195 | key->tun_flags = tun_flags; | |
1d8fff90 TG |
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 | */ | |
4c222798 PS |
201 | key->tp_src = tp_src; |
202 | key->tp_dst = tp_dst; | |
1d8fff90 TG |
203 | |
204 | /* Clear struct padding. */ | |
4c222798 PS |
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); | |
1d8fff90 TG |
208 | } |
209 | ||
7f9562a1 JB |
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 | ||
5243b6ac JG |
216 | #ifdef CONFIG_INET |
217 | ||
c5441932 PS |
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); | |
1728d4fa | 221 | struct net *ip_tunnel_get_link_net(const struct net_device *dev); |
1e99584b | 222 | int ip_tunnel_get_iflink(const struct net_device *dev); |
d3b6f614 ED |
223 | int ip_tunnel_init_net(struct net *net, int ip_tnl_net_id, |
224 | struct rtnl_link_ops *ops, char *devname); | |
c5441932 | 225 | |
6c742e71 | 226 | void ip_tunnel_delete_net(struct ip_tunnel_net *itn, struct rtnl_link_ops *ops); |
c5441932 PS |
227 | |
228 | void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev, | |
bf3d6a8f | 229 | const struct iphdr *tnl_params, const u8 protocol); |
c5441932 | 230 | int ip_tunnel_ioctl(struct net_device *dev, struct ip_tunnel_parm *p, int cmd); |
56328486 TH |
231 | int ip_tunnel_encap(struct sk_buff *skb, struct ip_tunnel *t, |
232 | u8 *protocol, struct flowi4 *fl4); | |
c5441932 PS |
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, | |
2e15ea39 PS |
243 | const struct tnl_ptk_info *tpi, struct metadata_dst *tun_dst, |
244 | bool log_ecn_error); | |
c5441932 PS |
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); | |
cf71d2bc | 250 | void ip_tunnel_dst_reset_all(struct ip_tunnel *t); |
56328486 TH |
251 | int ip_tunnel_encap_setup(struct ip_tunnel *t, |
252 | struct ip_tunnel_encap *ipencap); | |
c5441932 PS |
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 | ||
3d7b46cd | 275 | int iptunnel_pull_header(struct sk_buff *skb, int hdr_len, __be16 inner_proto); |
039f5062 PS |
276 | void 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); | |
63d008a4 JB |
279 | struct metadata_dst *iptunnel_metadata_reply(struct metadata_dst *md, |
280 | gfp_t flags); | |
0e6fbc5b | 281 | |
6fa79666 | 282 | struct sk_buff *iptunnel_handle_offloads(struct sk_buff *skb, int gso_type_mask); |
2d26f0a3 | 283 | |
039f5062 | 284 | static inline void iptunnel_xmit_stats(struct net_device *dev, int pkt_len) |
c5441932 | 285 | { |
039f5062 PS |
286 | if (pkt_len > 0) { |
287 | struct pcpu_sw_netstats *tstats = get_cpu_ptr(dev->tstats); | |
c5441932 | 288 | |
c5441932 | 289 | u64_stats_update_begin(&tstats->syncp); |
039f5062 | 290 | tstats->tx_bytes += pkt_len; |
c5441932 PS |
291 | tstats->tx_packets++; |
292 | u64_stats_update_end(&tstats->syncp); | |
b4fe85f9 | 293 | put_cpu_ptr(tstats); |
c5441932 | 294 | } else { |
039f5062 PS |
295 | struct net_device_stats *err_stats = &dev->stats; |
296 | ||
297 | if (pkt_len < 0) { | |
298 | err_stats->tx_errors++; | |
299 | err_stats->tx_aborted_errors++; | |
300 | } else { | |
301 | err_stats->tx_dropped++; | |
302 | } | |
c5441932 PS |
303 | } |
304 | } | |
5243b6ac | 305 | |
4c222798 | 306 | static inline void *ip_tunnel_info_opts(struct ip_tunnel_info *info) |
ee122c79 TG |
307 | { |
308 | return info + 1; | |
309 | } | |
310 | ||
4c222798 PS |
311 | static inline void ip_tunnel_info_opts_get(void *to, |
312 | const struct ip_tunnel_info *info) | |
313 | { | |
314 | memcpy(to, info + 1, info->options_len); | |
315 | } | |
316 | ||
317 | static inline void ip_tunnel_info_opts_set(struct ip_tunnel_info *info, | |
318 | const void *from, int len) | |
319 | { | |
320 | memcpy(ip_tunnel_info_opts(info), from, len); | |
321 | info->options_len = len; | |
322 | } | |
323 | ||
3093fbe7 TG |
324 | static inline struct ip_tunnel_info *lwt_tun_info(struct lwtunnel_state *lwtstate) |
325 | { | |
326 | return (struct ip_tunnel_info *)lwtstate->data; | |
327 | } | |
328 | ||
e7030878 TG |
329 | extern struct static_key ip_tunnel_metadata_cnt; |
330 | ||
331 | /* Returns > 0 if metadata should be collected */ | |
332 | static inline int ip_tunnel_collect_metadata(void) | |
333 | { | |
334 | return static_key_false(&ip_tunnel_metadata_cnt); | |
335 | } | |
336 | ||
045a0fa0 TG |
337 | void __init ip_tunnel_core_init(void); |
338 | ||
e7030878 TG |
339 | void ip_tunnel_need_metadata(void); |
340 | void ip_tunnel_unneed_metadata(void); | |
341 | ||
05283187 TG |
342 | #else /* CONFIG_INET */ |
343 | ||
344 | static inline struct ip_tunnel_info *lwt_tun_info(struct lwtunnel_state *lwtstate) | |
345 | { | |
346 | return NULL; | |
347 | } | |
348 | ||
349 | static inline void ip_tunnel_need_metadata(void) | |
350 | { | |
351 | } | |
352 | ||
353 | static inline void ip_tunnel_unneed_metadata(void) | |
354 | { | |
355 | } | |
356 | ||
5243b6ac JG |
357 | #endif /* CONFIG_INET */ |
358 | ||
c5441932 | 359 | #endif /* __NET_IP_TUNNELS_H */ |