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> | |
7 | #include <linux/types.h> | |
8 | #include <linux/u64_stats_sync.h> | |
9 | #include <net/dsfield.h> | |
10 | #include <net/gro_cells.h> | |
11 | #include <net/inet_ecn.h> | |
56328486 | 12 | #include <net/netns/generic.h> |
c5441932 | 13 | #include <net/rtnetlink.h> |
3093fbe7 | 14 | #include <net/lwtunnel.h> |
c5441932 PS |
15 | |
16 | #if IS_ENABLED(CONFIG_IPV6) | |
17 | #include <net/ipv6.h> | |
18 | #include <net/ip6_fib.h> | |
19 | #include <net/ip6_route.h> | |
20 | #endif | |
21 | ||
22 | /* Keep error state on tunnel for 30 sec */ | |
23 | #define IPTUNNEL_ERR_TIMEO (30*HZ) | |
24 | ||
1d8fff90 TG |
25 | /* Used to memset ip_tunnel padding. */ |
26 | #define IP_TUNNEL_KEY_SIZE \ | |
27 | (offsetof(struct ip_tunnel_key, tp_dst) + \ | |
28 | FIELD_SIZEOF(struct ip_tunnel_key, tp_dst)) | |
29 | ||
30 | struct ip_tunnel_key { | |
31 | __be64 tun_id; | |
32 | __be32 ipv4_src; | |
33 | __be32 ipv4_dst; | |
34 | __be16 tun_flags; | |
35 | __u8 ipv4_tos; | |
36 | __u8 ipv4_ttl; | |
37 | __be16 tp_src; | |
38 | __be16 tp_dst; | |
39 | } __packed __aligned(4); /* Minimize padding. */ | |
40 | ||
ee122c79 TG |
41 | /* Indicates whether the tunnel info structure represents receive |
42 | * or transmit tunnel parameters. | |
43 | */ | |
44 | enum { | |
45 | IP_TUNNEL_INFO_RX, | |
46 | IP_TUNNEL_INFO_TX, | |
47 | }; | |
48 | ||
1d8fff90 TG |
49 | struct ip_tunnel_info { |
50 | struct ip_tunnel_key key; | |
51 | const void *options; | |
52 | u8 options_len; | |
ee122c79 | 53 | u8 mode; |
1d8fff90 TG |
54 | }; |
55 | ||
c5441932 PS |
56 | /* 6rd prefix/relay information */ |
57 | #ifdef CONFIG_IPV6_SIT_6RD | |
58 | struct ip_tunnel_6rd_parm { | |
59 | struct in6_addr prefix; | |
60 | __be32 relay_prefix; | |
61 | u16 prefixlen; | |
62 | u16 relay_prefixlen; | |
63 | }; | |
64 | #endif | |
65 | ||
56328486 TH |
66 | struct ip_tunnel_encap { |
67 | __u16 type; | |
68 | __u16 flags; | |
69 | __be16 sport; | |
70 | __be16 dport; | |
71 | }; | |
72 | ||
c5441932 PS |
73 | struct ip_tunnel_prl_entry { |
74 | struct ip_tunnel_prl_entry __rcu *next; | |
75 | __be32 addr; | |
76 | u16 flags; | |
77 | struct rcu_head rcu_head; | |
78 | }; | |
79 | ||
9a4aa9af TH |
80 | struct ip_tunnel_dst { |
81 | struct dst_entry __rcu *dst; | |
95cb5745 | 82 | __be32 saddr; |
9a4aa9af TH |
83 | }; |
84 | ||
2e15ea39 PS |
85 | struct metadata_dst; |
86 | ||
c5441932 PS |
87 | struct ip_tunnel { |
88 | struct ip_tunnel __rcu *next; | |
89 | struct hlist_node hash_node; | |
90 | struct net_device *dev; | |
5e6700b3 | 91 | struct net *net; /* netns for packet i/o */ |
c5441932 PS |
92 | |
93 | int err_count; /* Number of arrived ICMP errors */ | |
94 | unsigned long err_time; /* Time when the last ICMP error | |
95 | * arrived */ | |
96 | ||
97 | /* These four fields used only by GRE */ | |
98 | __u32 i_seqno; /* The last seen seqno */ | |
99 | __u32 o_seqno; /* The last output seqno */ | |
56328486 | 100 | int tun_hlen; /* Precalculated header length */ |
c5441932 PS |
101 | int mlink; |
102 | ||
9a4aa9af | 103 | struct ip_tunnel_dst __percpu *dst_cache; |
7d442fab | 104 | |
c5441932 PS |
105 | struct ip_tunnel_parm parms; |
106 | ||
56328486 TH |
107 | int encap_hlen; /* Encap header length (FOU,GUE) */ |
108 | struct ip_tunnel_encap encap; | |
109 | ||
110 | int hlen; /* tun_hlen + encap_hlen */ | |
111 | ||
c5441932 PS |
112 | /* for SIT */ |
113 | #ifdef CONFIG_IPV6_SIT_6RD | |
114 | struct ip_tunnel_6rd_parm ip6rd; | |
115 | #endif | |
116 | struct ip_tunnel_prl_entry __rcu *prl; /* potential router list */ | |
117 | unsigned int prl_count; /* # of entries in PRL */ | |
118 | int ip_tnl_net_id; | |
119 | struct gro_cells gro_cells; | |
2e15ea39 | 120 | bool collect_md; |
c5441932 PS |
121 | }; |
122 | ||
f5796684 JG |
123 | #define TUNNEL_CSUM __cpu_to_be16(0x01) |
124 | #define TUNNEL_ROUTING __cpu_to_be16(0x02) | |
125 | #define TUNNEL_KEY __cpu_to_be16(0x04) | |
126 | #define TUNNEL_SEQ __cpu_to_be16(0x08) | |
127 | #define TUNNEL_STRICT __cpu_to_be16(0x10) | |
128 | #define TUNNEL_REC __cpu_to_be16(0x20) | |
129 | #define TUNNEL_VERSION __cpu_to_be16(0x40) | |
130 | #define TUNNEL_NO_KEY __cpu_to_be16(0x80) | |
9a628224 | 131 | #define TUNNEL_DONT_FRAGMENT __cpu_to_be16(0x0100) |
f5796684 JG |
132 | #define TUNNEL_OAM __cpu_to_be16(0x0200) |
133 | #define TUNNEL_CRIT_OPT __cpu_to_be16(0x0400) | |
1dd144cf TG |
134 | #define TUNNEL_GENEVE_OPT __cpu_to_be16(0x0800) |
135 | #define TUNNEL_VXLAN_OPT __cpu_to_be16(0x1000) | |
136 | ||
137 | #define TUNNEL_OPTIONS_PRESENT (TUNNEL_GENEVE_OPT | TUNNEL_VXLAN_OPT) | |
c5441932 PS |
138 | |
139 | struct tnl_ptk_info { | |
140 | __be16 flags; | |
141 | __be16 proto; | |
142 | __be32 key; | |
143 | __be32 seq; | |
144 | }; | |
145 | ||
146 | #define PACKET_RCVD 0 | |
147 | #define PACKET_REJECT 1 | |
148 | ||
6261d983 | 149 | #define IP_TNL_HASH_BITS 7 |
c5441932 PS |
150 | #define IP_TNL_HASH_SIZE (1 << IP_TNL_HASH_BITS) |
151 | ||
152 | struct ip_tunnel_net { | |
c5441932 | 153 | struct net_device *fb_tunnel_dev; |
6261d983 | 154 | struct hlist_head tunnels[IP_TNL_HASH_SIZE]; |
2e15ea39 | 155 | struct ip_tunnel __rcu *collect_md_tun; |
c5441932 PS |
156 | }; |
157 | ||
a8c5f90f TH |
158 | struct ip_tunnel_encap_ops { |
159 | size_t (*encap_hlen)(struct ip_tunnel_encap *e); | |
160 | int (*build_header)(struct sk_buff *skb, struct ip_tunnel_encap *e, | |
161 | u8 *protocol, struct flowi4 *fl4); | |
162 | }; | |
163 | ||
164 | #define MAX_IPTUN_ENCAP_OPS 8 | |
165 | ||
166 | extern const struct ip_tunnel_encap_ops __rcu * | |
167 | iptun_encaps[MAX_IPTUN_ENCAP_OPS]; | |
168 | ||
169 | int ip_tunnel_encap_add_ops(const struct ip_tunnel_encap_ops *op, | |
170 | unsigned int num); | |
171 | int ip_tunnel_encap_del_ops(const struct ip_tunnel_encap_ops *op, | |
172 | unsigned int num); | |
173 | ||
1d8fff90 TG |
174 | static inline void __ip_tunnel_info_init(struct ip_tunnel_info *tun_info, |
175 | __be32 saddr, __be32 daddr, | |
176 | u8 tos, u8 ttl, | |
177 | __be16 tp_src, __be16 tp_dst, | |
178 | __be64 tun_id, __be16 tun_flags, | |
179 | const void *opts, u8 opts_len) | |
180 | { | |
181 | tun_info->key.tun_id = tun_id; | |
182 | tun_info->key.ipv4_src = saddr; | |
183 | tun_info->key.ipv4_dst = daddr; | |
184 | tun_info->key.ipv4_tos = tos; | |
185 | tun_info->key.ipv4_ttl = ttl; | |
186 | tun_info->key.tun_flags = tun_flags; | |
187 | ||
188 | /* For the tunnel types on the top of IPsec, the tp_src and tp_dst of | |
189 | * the upper tunnel are used. | |
190 | * E.g: GRE over IPSEC, the tp_src and tp_port are zero. | |
191 | */ | |
192 | tun_info->key.tp_src = tp_src; | |
193 | tun_info->key.tp_dst = tp_dst; | |
194 | ||
195 | /* Clear struct padding. */ | |
196 | if (sizeof(tun_info->key) != IP_TUNNEL_KEY_SIZE) | |
197 | memset((unsigned char *)&tun_info->key + IP_TUNNEL_KEY_SIZE, | |
198 | 0, sizeof(tun_info->key) - IP_TUNNEL_KEY_SIZE); | |
199 | ||
200 | tun_info->options = opts; | |
201 | tun_info->options_len = opts_len; | |
202 | } | |
203 | ||
204 | static inline void ip_tunnel_info_init(struct ip_tunnel_info *tun_info, | |
205 | const struct iphdr *iph, | |
206 | __be16 tp_src, __be16 tp_dst, | |
207 | __be64 tun_id, __be16 tun_flags, | |
208 | const void *opts, u8 opts_len) | |
209 | { | |
210 | __ip_tunnel_info_init(tun_info, iph->saddr, iph->daddr, | |
211 | iph->tos, iph->ttl, tp_src, tp_dst, | |
212 | tun_id, tun_flags, opts, opts_len); | |
213 | } | |
214 | ||
5243b6ac JG |
215 | #ifdef CONFIG_INET |
216 | ||
c5441932 PS |
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); | |
1728d4fa | 220 | struct net *ip_tunnel_get_link_net(const struct net_device *dev); |
1e99584b | 221 | int ip_tunnel_get_iflink(const struct net_device *dev); |
d3b6f614 ED |
222 | int ip_tunnel_init_net(struct net *net, int ip_tnl_net_id, |
223 | struct rtnl_link_ops *ops, char *devname); | |
c5441932 | 224 | |
6c742e71 | 225 | void ip_tunnel_delete_net(struct ip_tunnel_net *itn, struct rtnl_link_ops *ops); |
c5441932 PS |
226 | |
227 | void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev, | |
bf3d6a8f | 228 | const struct iphdr *tnl_params, const u8 protocol); |
c5441932 | 229 | int ip_tunnel_ioctl(struct net_device *dev, struct ip_tunnel_parm *p, int cmd); |
56328486 TH |
230 | int ip_tunnel_encap(struct sk_buff *skb, struct ip_tunnel *t, |
231 | u8 *protocol, struct flowi4 *fl4); | |
c5441932 PS |
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, | |
2e15ea39 PS |
242 | const struct tnl_ptk_info *tpi, struct metadata_dst *tun_dst, |
243 | bool log_ecn_error); | |
c5441932 PS |
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); | |
cf71d2bc | 249 | void ip_tunnel_dst_reset_all(struct ip_tunnel *t); |
56328486 TH |
250 | int ip_tunnel_encap_setup(struct ip_tunnel *t, |
251 | struct ip_tunnel_encap *ipencap); | |
c5441932 PS |
252 | |
253 | /* Extract dsfield from inner protocol */ | |
254 | static inline u8 ip_tunnel_get_dsfield(const struct iphdr *iph, | |
255 | const struct sk_buff *skb) | |
256 | { | |
257 | if (skb->protocol == htons(ETH_P_IP)) | |
258 | return iph->tos; | |
259 | else if (skb->protocol == htons(ETH_P_IPV6)) | |
260 | return ipv6_get_dsfield((const struct ipv6hdr *)iph); | |
261 | else | |
262 | return 0; | |
263 | } | |
264 | ||
265 | /* Propogate ECN bits out */ | |
266 | static inline u8 ip_tunnel_ecn_encap(u8 tos, const struct iphdr *iph, | |
267 | const struct sk_buff *skb) | |
268 | { | |
269 | u8 inner = ip_tunnel_get_dsfield(iph, skb); | |
270 | ||
271 | return INET_ECN_encapsulate(tos, inner); | |
272 | } | |
273 | ||
3d7b46cd | 274 | int iptunnel_pull_header(struct sk_buff *skb, int hdr_len, __be16 inner_proto); |
aad88724 | 275 | int iptunnel_xmit(struct sock *sk, struct rtable *rt, struct sk_buff *skb, |
0e6fbc5b | 276 | __be32 src, __be32 dst, __u8 proto, |
963a88b3 | 277 | __u8 tos, __u8 ttl, __be16 df, bool xnet); |
0e6fbc5b | 278 | |
2d26f0a3 ED |
279 | struct sk_buff *iptunnel_handle_offloads(struct sk_buff *skb, bool gre_csum, |
280 | int gso_type_mask); | |
281 | ||
0e6fbc5b PS |
282 | static inline void iptunnel_xmit_stats(int err, |
283 | struct net_device_stats *err_stats, | |
8f84985f | 284 | struct pcpu_sw_netstats __percpu *stats) |
c5441932 | 285 | { |
0e6fbc5b | 286 | if (err > 0) { |
8f84985f | 287 | struct pcpu_sw_netstats *tstats = this_cpu_ptr(stats); |
c5441932 | 288 | |
c5441932 | 289 | u64_stats_update_begin(&tstats->syncp); |
0e6fbc5b | 290 | tstats->tx_bytes += err; |
c5441932 PS |
291 | tstats->tx_packets++; |
292 | u64_stats_update_end(&tstats->syncp); | |
0e6fbc5b PS |
293 | } else if (err < 0) { |
294 | err_stats->tx_errors++; | |
295 | err_stats->tx_aborted_errors++; | |
c5441932 | 296 | } else { |
0e6fbc5b | 297 | err_stats->tx_dropped++; |
c5441932 PS |
298 | } |
299 | } | |
5243b6ac | 300 | |
ee122c79 TG |
301 | static inline void *ip_tunnel_info_opts(struct ip_tunnel_info *info, size_t n) |
302 | { | |
303 | return info + 1; | |
304 | } | |
305 | ||
3093fbe7 TG |
306 | static inline struct ip_tunnel_info *lwt_tun_info(struct lwtunnel_state *lwtstate) |
307 | { | |
308 | return (struct ip_tunnel_info *)lwtstate->data; | |
309 | } | |
310 | ||
e7030878 TG |
311 | extern struct static_key ip_tunnel_metadata_cnt; |
312 | ||
313 | /* Returns > 0 if metadata should be collected */ | |
314 | static inline int ip_tunnel_collect_metadata(void) | |
315 | { | |
316 | return static_key_false(&ip_tunnel_metadata_cnt); | |
317 | } | |
318 | ||
045a0fa0 TG |
319 | void __init ip_tunnel_core_init(void); |
320 | ||
e7030878 TG |
321 | void ip_tunnel_need_metadata(void); |
322 | void ip_tunnel_unneed_metadata(void); | |
323 | ||
05283187 TG |
324 | #else /* CONFIG_INET */ |
325 | ||
326 | static inline struct ip_tunnel_info *lwt_tun_info(struct lwtunnel_state *lwtstate) | |
327 | { | |
328 | return NULL; | |
329 | } | |
330 | ||
331 | static inline void ip_tunnel_need_metadata(void) | |
332 | { | |
333 | } | |
334 | ||
335 | static inline void ip_tunnel_unneed_metadata(void) | |
336 | { | |
337 | } | |
338 | ||
5243b6ac JG |
339 | #endif /* CONFIG_INET */ |
340 | ||
c5441932 | 341 | #endif /* __NET_IP_TUNNELS_H */ |