include/net/vxlan.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef __NET_VXLAN_H
   3 #define __NET_VXLAN_H 1
   4
   5 #include <linux/if_vlan.h>
   6 #include <net/udp_tunnel.h>
   7 #include <net/dst_metadata.h>
   8 #include <net/udp_tunnel.h>
   9
  10 /* VXLAN protocol (RFC 7348) header:
  11  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  12  * |R|R|R|R|I|R|R|R|               Reserved                        |
  13  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  14  * |                VXLAN Network Identifier (VNI) |   Reserved    |
  15  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  16  *
  17  * I = VXLAN Network Identifier (VNI) present.
  18  */
  19 struct vxlanhdr {
  20         __be32 vx_flags;
  21         __be32 vx_vni;
  22 };
  23
  24 /* VXLAN header flags. */
  25 #define VXLAN_HF_VNI    cpu_to_be32(BIT(27))
  26
  27 #define VXLAN_N_VID     (1u << 24)
  28 #define VXLAN_VID_MASK  (VXLAN_N_VID - 1)
  29 #define VXLAN_VNI_MASK  cpu_to_be32(VXLAN_VID_MASK << 8)
  30 #define VXLAN_HLEN (sizeof(struct udphdr) + sizeof(struct vxlanhdr))
  31
  32 #define VNI_HASH_BITS   10
  33 #define VNI_HASH_SIZE   (1<<VNI_HASH_BITS)
  34 #define FDB_HASH_BITS   8
  35 #define FDB_HASH_SIZE   (1<<FDB_HASH_BITS)
  36
  37 /* Remote checksum offload for VXLAN (VXLAN_F_REMCSUM_[RT]X):
  38  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  39  * |R|R|R|R|I|R|R|R|R|R|C|              Reserved                   |
  40  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  41  * |           VXLAN Network Identifier (VNI)      |O| Csum start  |
  42  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  43  *
  44  * C = Remote checksum offload bit. When set indicates that the
  45  *     remote checksum offload data is present.
  46  *
  47  * O = Offset bit. Indicates the checksum offset relative to
  48  *     checksum start.
  49  *
  50  * Csum start = Checksum start divided by two.
  51  *
  52  * http://tools.ietf.org/html/draft-herbert-vxlan-rco
  53  */
  54
  55 /* VXLAN-RCO header flags. */
  56 #define VXLAN_HF_RCO    cpu_to_be32(BIT(21))
  57
  58 /* Remote checksum offload header option */
  59 #define VXLAN_RCO_MASK  cpu_to_be32(0x7f)  /* Last byte of vni field */
  60 #define VXLAN_RCO_UDP   cpu_to_be32(0x80)  /* Indicate UDP RCO (TCP when not set *) */
  61 #define VXLAN_RCO_SHIFT 1                  /* Left shift of start */
  62 #define VXLAN_RCO_SHIFT_MASK ((1 << VXLAN_RCO_SHIFT) - 1)
  63 #define VXLAN_MAX_REMCSUM_START (0x7f << VXLAN_RCO_SHIFT)
  64
  65 /*
  66  * VXLAN Group Based Policy Extension (VXLAN_F_GBP):
  67  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  68  * |G|R|R|R|I|R|R|R|R|D|R|R|A|R|R|R|        Group Policy ID        |
  69  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  70  * |                VXLAN Network Identifier (VNI) |   Reserved    |
  71  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  72  *
  73  * G = Group Policy ID present.
  74  *
  75  * D = Don't Learn bit. When set, this bit indicates that the egress
  76  *     VTEP MUST NOT learn the source address of the encapsulated frame.
  77  *
  78  * A = Indicates that the group policy has already been applied to
  79  *     this packet. Policies MUST NOT be applied by devices when the
  80  *     A bit is set.
  81  *
  82  * https://tools.ietf.org/html/draft-smith-vxlan-group-policy
  83  */
  84 struct vxlanhdr_gbp {
  85         u8      vx_flags;
  86 #ifdef __LITTLE_ENDIAN_BITFIELD
  87         u8      reserved_flags1:3,
  88                 policy_applied:1,
  89                 reserved_flags2:2,
  90                 dont_learn:1,
  91                 reserved_flags3:1;
  92 #elif defined(__BIG_ENDIAN_BITFIELD)
  93         u8      reserved_flags1:1,
  94                 dont_learn:1,
  95                 reserved_flags2:2,
  96                 policy_applied:1,
  97                 reserved_flags3:3;
  98 #else
  99 #error  "Please fix <asm/byteorder.h>"
 100 #endif
 101         __be16  policy_id;
 102         __be32  vx_vni;
 103 };
 104
 105 /* VXLAN-GBP header flags. */
 106 #define VXLAN_HF_GBP    cpu_to_be32(BIT(31))
 107
 108 #define VXLAN_GBP_USED_BITS (VXLAN_HF_GBP | cpu_to_be32(0xFFFFFF))
 109
 110 /* skb->mark mapping
 111  *
 112  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 113  * |R|R|R|R|R|R|R|R|R|D|R|R|A|R|R|R|        Group Policy ID        |
 114  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 115  */
 116 #define VXLAN_GBP_DONT_LEARN            (BIT(6) << 16)
 117 #define VXLAN_GBP_POLICY_APPLIED        (BIT(3) << 16)
 118 #define VXLAN_GBP_ID_MASK               (0xFFFF)
 119
 120 /*
 121  * VXLAN Generic Protocol Extension (VXLAN_F_GPE):
 122  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 123  * |R|R|Ver|I|P|R|O|       Reserved                |Next Protocol  |
 124  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 125  * |                VXLAN Network Identifier (VNI) |   Reserved    |
 126  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 127  *
 128  * Ver = Version. Indicates VXLAN GPE protocol version.
 129  *
 130  * P = Next Protocol Bit. The P bit is set to indicate that the
 131  *     Next Protocol field is present.
 132  *
 133  * O = OAM Flag Bit. The O bit is set to indicate that the packet
 134  *     is an OAM packet.
 135  *
 136  * Next Protocol = This 8 bit field indicates the protocol header
 137  * immediately following the VXLAN GPE header.
 138  *
 139  * https://tools.ietf.org/html/draft-ietf-nvo3-vxlan-gpe-01
 140  */
 141
 142 struct vxlanhdr_gpe {
 143 #if defined(__LITTLE_ENDIAN_BITFIELD)
 144         u8      oam_flag:1,
 145                 reserved_flags1:1,
 146                 np_applied:1,
 147                 instance_applied:1,
 148                 version:2,
 149                 reserved_flags2:2;
 150 #elif defined(__BIG_ENDIAN_BITFIELD)
 151         u8      reserved_flags2:2,
 152                 version:2,
 153                 instance_applied:1,
 154                 np_applied:1,
 155                 reserved_flags1:1,
 156                 oam_flag:1;
 157 #endif
 158         u8      reserved_flags3;
 159         u8      reserved_flags4;
 160         u8      next_protocol;
 161         __be32  vx_vni;
 162 };
 163
 164 /* VXLAN-GPE header flags. */
 165 #define VXLAN_HF_VER    cpu_to_be32(BIT(29) | BIT(28))
 166 #define VXLAN_HF_NP     cpu_to_be32(BIT(26))
 167 #define VXLAN_HF_OAM    cpu_to_be32(BIT(24))
 168
 169 #define VXLAN_GPE_USED_BITS (VXLAN_HF_VER | VXLAN_HF_NP | VXLAN_HF_OAM | \
 170                              cpu_to_be32(0xff))
 171
 172 struct vxlan_metadata {
 173         u32             gbp;
 174 };
 175
 176 /* per UDP socket information */
 177 struct vxlan_sock {
 178         struct hlist_node hlist;
 179         struct socket    *sock;
 180         struct hlist_head vni_list[VNI_HASH_SIZE];
 181         refcount_t        refcnt;
 182         u32               flags;
 183 };
 184
 185 union vxlan_addr {
 186         struct sockaddr_in sin;
 187         struct sockaddr_in6 sin6;
 188         struct sockaddr sa;
 189 };
 190
 191 struct vxlan_rdst {
 192         union vxlan_addr         remote_ip;
 193         __be16                   remote_port;
 194         __be32                   remote_vni;
 195         u32                      remote_ifindex;
 196         struct list_head         list;
 197         struct rcu_head          rcu;
 198         struct dst_cache         dst_cache;
 199 };
 200
 201 struct vxlan_config {
 202         union vxlan_addr        remote_ip;
 203         union vxlan_addr        saddr;
 204         __be32                  vni;
 205         int                     remote_ifindex;
 206         int                     mtu;
 207         __be16                  dst_port;
 208         u16                     port_min;
 209         u16                     port_max;
 210         u8                      tos;
 211         u8                      ttl;
 212         __be32                  label;
 213         u32                     flags;
 214         unsigned long           age_interval;
 215         unsigned int            addrmax;
 216         bool                    no_share;
 217 };
 218
 219 struct vxlan_dev_node {
 220         struct hlist_node hlist;
 221         struct vxlan_dev *vxlan;
 222 };
 223
 224 /* Pseudo network device */
 225 struct vxlan_dev {
 226         struct vxlan_dev_node hlist4;   /* vni hash table for IPv4 socket */
 227 #if IS_ENABLED(CONFIG_IPV6)
 228         struct vxlan_dev_node hlist6;   /* vni hash table for IPv6 socket */
 229 #endif
 230         struct list_head  next;         /* vxlan's per namespace list */
 231         struct vxlan_sock __rcu *vn4_sock;      /* listening socket for IPv4 */
 232 #if IS_ENABLED(CONFIG_IPV6)
 233         struct vxlan_sock __rcu *vn6_sock;      /* listening socket for IPv6 */
 234 #endif
 235         struct net_device *dev;
 236         struct net        *net;         /* netns for packet i/o */
 237         struct vxlan_rdst default_dst;  /* default destination */
 238
 239         struct timer_list age_timer;
 240         spinlock_t        hash_lock;
 241         unsigned int      addrcnt;
 242         struct gro_cells  gro_cells;
 243
 244         struct vxlan_config     cfg;
 245
 246         struct hlist_head fdb_head[FDB_HASH_SIZE];
 247 };
 248
 249 #define VXLAN_F_LEARN                   0x01
 250 #define VXLAN_F_PROXY                   0x02
 251 #define VXLAN_F_RSC                     0x04
 252 #define VXLAN_F_L2MISS                  0x08
 253 #define VXLAN_F_L3MISS                  0x10
 254 #define VXLAN_F_IPV6                    0x20
 255 #define VXLAN_F_UDP_ZERO_CSUM_TX        0x40
 256 #define VXLAN_F_UDP_ZERO_CSUM6_TX       0x80
 257 #define VXLAN_F_UDP_ZERO_CSUM6_RX       0x100
 258 #define VXLAN_F_REMCSUM_TX              0x200
 259 #define VXLAN_F_REMCSUM_RX              0x400
 260 #define VXLAN_F_GBP                     0x800
 261 #define VXLAN_F_REMCSUM_NOPARTIAL       0x1000
 262 #define VXLAN_F_COLLECT_METADATA        0x2000
 263 #define VXLAN_F_GPE                     0x4000
 264 #define VXLAN_F_IPV6_LINKLOCAL          0x8000
 265 #define VXLAN_F_TTL_INHERIT             0x10000
 266
 267 /* Flags that are used in the receive path. These flags must match in
 268  * order for a socket to be shareable
 269  */
 270 #define VXLAN_F_RCV_FLAGS               (VXLAN_F_GBP |                  \
 271                                          VXLAN_F_GPE |                  \
 272                                          VXLAN_F_UDP_ZERO_CSUM6_RX |    \
 273                                          VXLAN_F_REMCSUM_RX |           \
 274                                          VXLAN_F_REMCSUM_NOPARTIAL |    \
 275                                          VXLAN_F_COLLECT_METADATA)
 276
 277 /* Flags that can be set together with VXLAN_F_GPE. */
 278 #define VXLAN_F_ALLOWED_GPE             (VXLAN_F_GPE |                  \
 279                                          VXLAN_F_IPV6 |                 \
 280                                          VXLAN_F_IPV6_LINKLOCAL |       \
 281                                          VXLAN_F_UDP_ZERO_CSUM_TX |     \
 282                                          VXLAN_F_UDP_ZERO_CSUM6_TX |    \
 283                                          VXLAN_F_UDP_ZERO_CSUM6_RX |    \
 284                                          VXLAN_F_COLLECT_METADATA)
 285
 286 struct net_device *vxlan_dev_create(struct net *net, const char *name,
 287                                     u8 name_assign_type, struct vxlan_config *conf);
 288
 289 static inline netdev_features_t vxlan_features_check(struct sk_buff *skb,
 290                                                      netdev_features_t features)
 291 {
 292         u8 l4_hdr = 0;
 293
 294         if (!skb->encapsulation)
 295                 return features;
 296
 297         switch (vlan_get_protocol(skb)) {
 298         case htons(ETH_P_IP):
 299                 l4_hdr = ip_hdr(skb)->protocol;
 300                 break;
 301         case htons(ETH_P_IPV6):
 302                 l4_hdr = ipv6_hdr(skb)->nexthdr;
 303                 break;
 304         default:
 305                 return features;
 306         }
 307
 308         if ((l4_hdr == IPPROTO_UDP) &&
 309             (skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
 310              skb->inner_protocol != htons(ETH_P_TEB) ||
 311              (skb_inner_mac_header(skb) - skb_transport_header(skb) !=
 312               sizeof(struct udphdr) + sizeof(struct vxlanhdr)) ||
 313              (skb->ip_summed != CHECKSUM_NONE &&
 314               !can_checksum_protocol(features, inner_eth_hdr(skb)->h_proto))))
 315                 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
 316
 317         return features;
 318 }
 319
 320 /* IP header + UDP + VXLAN + Ethernet header */
 321 #define VXLAN_HEADROOM (20 + 8 + 8 + 14)
 322 /* IPv6 header + UDP + VXLAN + Ethernet header */
 323 #define VXLAN6_HEADROOM (40 + 8 + 8 + 14)
 324
 325 static inline struct vxlanhdr *vxlan_hdr(struct sk_buff *skb)
 326 {
 327         return (struct vxlanhdr *)(udp_hdr(skb) + 1);
 328 }
 329
 330 static inline __be32 vxlan_vni(__be32 vni_field)
 331 {
 332 #if defined(__BIG_ENDIAN)
 333         return (__force __be32)((__force u32)vni_field >> 8);
 334 #else
 335         return (__force __be32)((__force u32)(vni_field & VXLAN_VNI_MASK) << 8);
 336 #endif
 337 }
 338
 339 static inline __be32 vxlan_vni_field(__be32 vni)
 340 {
 341 #if defined(__BIG_ENDIAN)
 342         return (__force __be32)((__force u32)vni << 8);
 343 #else
 344         return (__force __be32)((__force u32)vni >> 8);
 345 #endif
 346 }
 347
 348 static inline size_t vxlan_rco_start(__be32 vni_field)
 349 {
 350         return be32_to_cpu(vni_field & VXLAN_RCO_MASK) << VXLAN_RCO_SHIFT;
 351 }
 352
 353 static inline size_t vxlan_rco_offset(__be32 vni_field)
 354 {
 355         return (vni_field & VXLAN_RCO_UDP) ?
 356                 offsetof(struct udphdr, check) :
 357                 offsetof(struct tcphdr, check);
 358 }
 359
 360 static inline __be32 vxlan_compute_rco(unsigned int start, unsigned int offset)
 361 {
 362         __be32 vni_field = cpu_to_be32(start >> VXLAN_RCO_SHIFT);
 363
 364         if (offset == offsetof(struct udphdr, check))
 365                 vni_field |= VXLAN_RCO_UDP;
 366         return vni_field;
 367 }
 368
 369 static inline unsigned short vxlan_get_sk_family(struct vxlan_sock *vs)
 370 {
 371         return vs->sock->sk->sk_family;
 372 }
 373
 374 #endif