2 #define __NET_VXLAN_H 1
5 #include <linux/ipv6.h>
6 #include <linux/if_vlan.h>
7 #include <linux/skbuff.h>
8 #include <linux/netdevice.h>
10 #include <net/dst_metadata.h>
12 /* VXLAN protocol (RFC 7348) header:
13 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
14 * |R|R|R|R|I|R|R|R| Reserved |
15 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
16 * | VXLAN Network Identifier (VNI) | Reserved |
17 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
19 * I = VXLAN Network Identifier (VNI) present.
26 /* VXLAN header flags. */
27 #define VXLAN_HF_VNI cpu_to_be32(BIT(27))
29 #define VXLAN_N_VID (1u << 24)
30 #define VXLAN_VID_MASK (VXLAN_N_VID - 1)
31 #define VXLAN_VNI_MASK cpu_to_be32(VXLAN_VID_MASK << 8)
32 #define VXLAN_HLEN (sizeof(struct udphdr) + sizeof(struct vxlanhdr))
34 #define VNI_HASH_BITS 10
35 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS)
36 #define FDB_HASH_BITS 8
37 #define FDB_HASH_SIZE (1<<FDB_HASH_BITS)
39 /* Remote checksum offload for VXLAN (VXLAN_F_REMCSUM_[RT]X):
40 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
41 * |R|R|R|R|I|R|R|R|R|R|C| Reserved |
42 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
43 * | VXLAN Network Identifier (VNI) |O| Csum start |
44 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
46 * C = Remote checksum offload bit. When set indicates that the
47 * remote checksum offload data is present.
49 * O = Offset bit. Indicates the checksum offset relative to
52 * Csum start = Checksum start divided by two.
54 * http://tools.ietf.org/html/draft-herbert-vxlan-rco
57 /* VXLAN-RCO header flags. */
58 #define VXLAN_HF_RCO cpu_to_be32(BIT(21))
60 /* Remote checksum offload header option */
61 #define VXLAN_RCO_MASK cpu_to_be32(0x7f) /* Last byte of vni field */
62 #define VXLAN_RCO_UDP cpu_to_be32(0x80) /* Indicate UDP RCO (TCP when not set *) */
63 #define VXLAN_RCO_SHIFT 1 /* Left shift of start */
64 #define VXLAN_RCO_SHIFT_MASK ((1 << VXLAN_RCO_SHIFT) - 1)
65 #define VXLAN_MAX_REMCSUM_START (0x7f << VXLAN_RCO_SHIFT)
68 * VXLAN Group Based Policy Extension (VXLAN_F_GBP):
69 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
70 * |G|R|R|R|I|R|R|R|R|D|R|R|A|R|R|R| Group Policy ID |
71 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
72 * | VXLAN Network Identifier (VNI) | Reserved |
73 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
75 * G = Group Policy ID present.
77 * D = Don't Learn bit. When set, this bit indicates that the egress
78 * VTEP MUST NOT learn the source address of the encapsulated frame.
80 * A = Indicates that the group policy has already been applied to
81 * this packet. Policies MUST NOT be applied by devices when the
84 * https://tools.ietf.org/html/draft-smith-vxlan-group-policy
88 #ifdef __LITTLE_ENDIAN_BITFIELD
94 #elif defined(__BIG_ENDIAN_BITFIELD)
101 #error "Please fix <asm/byteorder.h>"
107 /* VXLAN-GBP header flags. */
108 #define VXLAN_HF_GBP cpu_to_be32(BIT(31))
110 #define VXLAN_GBP_USED_BITS (VXLAN_HF_GBP | cpu_to_be32(0xFFFFFF))
114 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
115 * |R|R|R|R|R|R|R|R|R|D|R|R|A|R|R|R| Group Policy ID |
116 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
118 #define VXLAN_GBP_DONT_LEARN (BIT(6) << 16)
119 #define VXLAN_GBP_POLICY_APPLIED (BIT(3) << 16)
120 #define VXLAN_GBP_ID_MASK (0xFFFF)
123 * VXLAN Generic Protocol Extension (VXLAN_F_GPE):
124 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
125 * |R|R|Ver|I|P|R|O| Reserved |Next Protocol |
126 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
127 * | VXLAN Network Identifier (VNI) | Reserved |
128 * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
130 * Ver = Version. Indicates VXLAN GPE protocol version.
132 * P = Next Protocol Bit. The P bit is set to indicate that the
133 * Next Protocol field is present.
135 * O = OAM Flag Bit. The O bit is set to indicate that the packet
138 * Next Protocol = This 8 bit field indicates the protocol header
139 * immediately following the VXLAN GPE header.
141 * https://tools.ietf.org/html/draft-ietf-nvo3-vxlan-gpe-01
144 struct vxlanhdr_gpe {
145 #if defined(__LITTLE_ENDIAN_BITFIELD)
152 #elif defined(__BIG_ENDIAN_BITFIELD)
153 u8 reserved_flags2:2,
166 /* VXLAN-GPE header flags. */
167 #define VXLAN_HF_VER cpu_to_be32(BIT(29) | BIT(28))
168 #define VXLAN_HF_NP cpu_to_be32(BIT(26))
169 #define VXLAN_HF_OAM cpu_to_be32(BIT(24))
171 #define VXLAN_GPE_USED_BITS (VXLAN_HF_VER | VXLAN_HF_NP | VXLAN_HF_OAM | \
174 /* VXLAN-GPE header Next Protocol. */
175 #define VXLAN_GPE_NP_IPV4 0x01
176 #define VXLAN_GPE_NP_IPV6 0x02
177 #define VXLAN_GPE_NP_ETHERNET 0x03
178 #define VXLAN_GPE_NP_NSH 0x04
180 struct vxlan_metadata {
184 /* per UDP socket information */
186 struct hlist_node hlist;
187 struct work_struct del_work;
190 struct hlist_head vni_list[VNI_HASH_SIZE];
192 struct udp_offload udp_offloads;
197 struct sockaddr_in sin;
198 struct sockaddr_in6 sin6;
203 union vxlan_addr remote_ip;
207 struct list_head list;
209 struct dst_cache dst_cache;
212 struct vxlan_config {
213 union vxlan_addr remote_ip;
214 union vxlan_addr saddr;
225 unsigned long age_interval;
226 unsigned int addrmax;
230 /* Pseudo network device */
232 struct hlist_node hlist; /* vni hash table */
233 struct list_head next; /* vxlan's per namespace list */
234 struct vxlan_sock *vn4_sock; /* listening socket for IPv4 */
235 #if IS_ENABLED(CONFIG_IPV6)
236 struct vxlan_sock *vn6_sock; /* listening socket for IPv6 */
238 struct net_device *dev;
239 struct net *net; /* netns for packet i/o */
240 struct vxlan_rdst default_dst; /* default destination */
241 u32 flags; /* VXLAN_F_* in vxlan.h */
243 struct timer_list age_timer;
244 spinlock_t hash_lock;
245 unsigned int addrcnt;
246 struct gro_cells gro_cells;
248 struct vxlan_config cfg;
250 struct hlist_head fdb_head[FDB_HASH_SIZE];
253 #define VXLAN_F_LEARN 0x01
254 #define VXLAN_F_PROXY 0x02
255 #define VXLAN_F_RSC 0x04
256 #define VXLAN_F_L2MISS 0x08
257 #define VXLAN_F_L3MISS 0x10
258 #define VXLAN_F_IPV6 0x20
259 #define VXLAN_F_UDP_ZERO_CSUM_TX 0x40
260 #define VXLAN_F_UDP_ZERO_CSUM6_TX 0x80
261 #define VXLAN_F_UDP_ZERO_CSUM6_RX 0x100
262 #define VXLAN_F_REMCSUM_TX 0x200
263 #define VXLAN_F_REMCSUM_RX 0x400
264 #define VXLAN_F_GBP 0x800
265 #define VXLAN_F_REMCSUM_NOPARTIAL 0x1000
266 #define VXLAN_F_COLLECT_METADATA 0x2000
267 #define VXLAN_F_GPE 0x4000
269 /* Flags that are used in the receive path. These flags must match in
270 * order for a socket to be shareable
272 #define VXLAN_F_RCV_FLAGS (VXLAN_F_GBP | \
274 VXLAN_F_UDP_ZERO_CSUM6_RX | \
275 VXLAN_F_REMCSUM_RX | \
276 VXLAN_F_REMCSUM_NOPARTIAL | \
277 VXLAN_F_COLLECT_METADATA)
279 /* Flags that can be set together with VXLAN_F_GPE. */
280 #define VXLAN_F_ALLOWED_GPE (VXLAN_F_GPE | \
282 VXLAN_F_UDP_ZERO_CSUM_TX | \
283 VXLAN_F_UDP_ZERO_CSUM6_TX | \
284 VXLAN_F_UDP_ZERO_CSUM6_RX | \
285 VXLAN_F_COLLECT_METADATA)
287 struct net_device *vxlan_dev_create(struct net *net, const char *name,
288 u8 name_assign_type, struct vxlan_config *conf);
290 static inline __be16 vxlan_dev_dst_port(struct vxlan_dev *vxlan,
291 unsigned short family)
293 #if IS_ENABLED(CONFIG_IPV6)
294 if (family == AF_INET6)
295 return inet_sk(vxlan->vn6_sock->sock->sk)->inet_sport;
297 return inet_sk(vxlan->vn4_sock->sock->sk)->inet_sport;
300 static inline netdev_features_t vxlan_features_check(struct sk_buff *skb,
301 netdev_features_t features)
305 if (!skb->encapsulation)
308 switch (vlan_get_protocol(skb)) {
309 case htons(ETH_P_IP):
310 l4_hdr = ip_hdr(skb)->protocol;
312 case htons(ETH_P_IPV6):
313 l4_hdr = ipv6_hdr(skb)->nexthdr;
319 if ((l4_hdr == IPPROTO_UDP) &&
320 (skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
321 skb->inner_protocol != htons(ETH_P_TEB) ||
322 (skb_inner_mac_header(skb) - skb_transport_header(skb) !=
323 sizeof(struct udphdr) + sizeof(struct vxlanhdr))))
324 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
329 /* IP header + UDP + VXLAN + Ethernet header */
330 #define VXLAN_HEADROOM (20 + 8 + 8 + 14)
331 /* IPv6 header + UDP + VXLAN + Ethernet header */
332 #define VXLAN6_HEADROOM (40 + 8 + 8 + 14)
334 static inline struct vxlanhdr *vxlan_hdr(struct sk_buff *skb)
336 return (struct vxlanhdr *)(udp_hdr(skb) + 1);
339 static inline __be32 vxlan_vni(__be32 vni_field)
341 #if defined(__BIG_ENDIAN)
342 return (__force __be32)((__force u32)vni_field >> 8);
344 return (__force __be32)((__force u32)(vni_field & VXLAN_VNI_MASK) << 8);
348 static inline __be32 vxlan_vni_field(__be32 vni)
350 #if defined(__BIG_ENDIAN)
351 return (__force __be32)((__force u32)vni << 8);
353 return (__force __be32)((__force u32)vni >> 8);
357 static inline __be32 vxlan_tun_id_to_vni(__be64 tun_id)
359 #if defined(__BIG_ENDIAN)
360 return (__force __be32)tun_id;
362 return (__force __be32)((__force u64)tun_id >> 32);
366 static inline __be64 vxlan_vni_to_tun_id(__be32 vni)
368 #if defined(__BIG_ENDIAN)
369 return (__force __be64)vni;
371 return (__force __be64)((u64)(__force u32)vni << 32);
375 static inline size_t vxlan_rco_start(__be32 vni_field)
377 return be32_to_cpu(vni_field & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
380 static inline size_t vxlan_rco_offset(__be32 vni_field)
382 return (vni_field & VXLAN_RCO_UDP) ?
383 offsetof(struct udphdr, check) :
384 offsetof(struct tcphdr, check);
387 static inline __be32 vxlan_compute_rco(unsigned int start, unsigned int offset)
389 __be32 vni_field = cpu_to_be32(start >> VXLAN_RCO_SHIFT);
391 if (offset == offsetof(struct udphdr, check))
392 vni_field |= VXLAN_RCO_UDP;
396 #if IS_ENABLED(CONFIG_VXLAN)
397 void vxlan_get_rx_port(struct net_device *netdev);
399 static inline void vxlan_get_rx_port(struct net_device *netdev)
404 static inline unsigned short vxlan_get_sk_family(struct vxlan_sock *vs)
406 return vs->sock->sk->sk_family;