1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* Multipath TCP 3 * 4 * Copyright (c) 2017 - 2019, Intel Corporation. 5 */ 6 7 #ifndef __MPTCP_PROTOCOL_H 8 #define __MPTCP_PROTOCOL_H 9 10 #include <linux/random.h> 11 #include <net/tcp.h> 12 #include <net/inet_connection_sock.h> 13 14 #define MPTCP_SUPPORTED_VERSION 1 15 16 /* MPTCP option bits */ 17 #define OPTION_MPTCP_MPC_SYN BIT(0) 18 #define OPTION_MPTCP_MPC_SYNACK BIT(1) 19 #define OPTION_MPTCP_MPC_ACK BIT(2) 20 #define OPTION_MPTCP_MPJ_SYN BIT(3) 21 #define OPTION_MPTCP_MPJ_SYNACK BIT(4) 22 #define OPTION_MPTCP_MPJ_ACK BIT(5) 23 #define OPTION_MPTCP_ADD_ADDR BIT(6) 24 #define OPTION_MPTCP_ADD_ADDR6 BIT(7) 25 #define OPTION_MPTCP_RM_ADDR BIT(8) 26 27 /* MPTCP option subtypes */ 28 #define MPTCPOPT_MP_CAPABLE 0 29 #define MPTCPOPT_MP_JOIN 1 30 #define MPTCPOPT_DSS 2 31 #define MPTCPOPT_ADD_ADDR 3 32 #define MPTCPOPT_RM_ADDR 4 33 #define MPTCPOPT_MP_PRIO 5 34 #define MPTCPOPT_MP_FAIL 6 35 #define MPTCPOPT_MP_FASTCLOSE 7 36 37 /* MPTCP suboption lengths */ 38 #define TCPOLEN_MPTCP_MPC_SYN 4 39 #define TCPOLEN_MPTCP_MPC_SYNACK 12 40 #define TCPOLEN_MPTCP_MPC_ACK 20 41 #define TCPOLEN_MPTCP_MPC_ACK_DATA 22 42 #define TCPOLEN_MPTCP_MPJ_SYN 12 43 #define TCPOLEN_MPTCP_MPJ_SYNACK 16 44 #define TCPOLEN_MPTCP_MPJ_ACK 24 45 #define TCPOLEN_MPTCP_DSS_BASE 4 46 #define TCPOLEN_MPTCP_DSS_ACK32 4 47 #define TCPOLEN_MPTCP_DSS_ACK64 8 48 #define TCPOLEN_MPTCP_DSS_MAP32 10 49 #define TCPOLEN_MPTCP_DSS_MAP64 14 50 #define TCPOLEN_MPTCP_DSS_CHECKSUM 2 51 #define TCPOLEN_MPTCP_ADD_ADDR 16 52 #define TCPOLEN_MPTCP_ADD_ADDR_PORT 18 53 #define TCPOLEN_MPTCP_ADD_ADDR_BASE 8 54 #define TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT 10 55 #define TCPOLEN_MPTCP_ADD_ADDR6 28 56 #define TCPOLEN_MPTCP_ADD_ADDR6_PORT 30 57 #define TCPOLEN_MPTCP_ADD_ADDR6_BASE 20 58 #define TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT 22 59 #define TCPOLEN_MPTCP_PORT_LEN 2 60 #define TCPOLEN_MPTCP_RM_ADDR_BASE 4 61 62 /* MPTCP MP_JOIN flags */ 63 #define MPTCPOPT_BACKUP BIT(0) 64 #define MPTCPOPT_HMAC_LEN 20 65 #define MPTCPOPT_THMAC_LEN 8 66 67 /* MPTCP MP_CAPABLE flags */ 68 #define MPTCP_VERSION_MASK (0x0F) 69 #define MPTCP_CAP_CHECKSUM_REQD BIT(7) 70 #define MPTCP_CAP_EXTENSIBILITY BIT(6) 71 #define MPTCP_CAP_HMAC_SHA256 BIT(0) 72 #define MPTCP_CAP_FLAG_MASK (0x3F) 73 74 /* MPTCP DSS flags */ 75 #define MPTCP_DSS_DATA_FIN BIT(4) 76 #define MPTCP_DSS_DSN64 BIT(3) 77 #define MPTCP_DSS_HAS_MAP BIT(2) 78 #define MPTCP_DSS_ACK64 BIT(1) 79 #define MPTCP_DSS_HAS_ACK BIT(0) 80 #define MPTCP_DSS_FLAG_MASK (0x1F) 81 82 /* MPTCP ADD_ADDR flags */ 83 #define MPTCP_ADDR_ECHO BIT(0) 84 #define MPTCP_ADDR_IPVERSION_4 4 85 #define MPTCP_ADDR_IPVERSION_6 6 86 87 /* MPTCP socket flags */ 88 #define MPTCP_DATA_READY 0 89 #define MPTCP_SEND_SPACE 1 90 #define MPTCP_WORK_RTX 2 91 #define MPTCP_WORK_EOF 3 92 93 struct mptcp_options_received { 94 u64 sndr_key; 95 u64 rcvr_key; 96 u64 data_ack; 97 u64 data_seq; 98 u32 subflow_seq; 99 u16 data_len; 100 u16 mp_capable : 1, 101 mp_join : 1, 102 dss : 1, 103 add_addr : 1, 104 rm_addr : 1, 105 family : 4, 106 echo : 1, 107 backup : 1; 108 u32 token; 109 u32 nonce; 110 u64 thmac; 111 u8 hmac[20]; 112 u8 join_id; 113 u8 use_map:1, 114 dsn64:1, 115 data_fin:1, 116 use_ack:1, 117 ack64:1, 118 mpc_map:1, 119 __unused:2; 120 u8 addr_id; 121 u8 rm_id; 122 union { 123 struct in_addr addr; 124 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 125 struct in6_addr addr6; 126 #endif 127 }; 128 u64 ahmac; 129 u16 port; 130 }; 131 132 static inline __be32 mptcp_option(u8 subopt, u8 len, u8 nib, u8 field) 133 { 134 return htonl((TCPOPT_MPTCP << 24) | (len << 16) | (subopt << 12) | 135 ((nib & 0xF) << 8) | field); 136 } 137 138 struct mptcp_addr_info { 139 sa_family_t family; 140 __be16 port; 141 u8 id; 142 union { 143 struct in_addr addr; 144 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 145 struct in6_addr addr6; 146 #endif 147 }; 148 }; 149 150 enum mptcp_pm_status { 151 MPTCP_PM_ADD_ADDR_RECEIVED, 152 MPTCP_PM_ESTABLISHED, 153 MPTCP_PM_SUBFLOW_ESTABLISHED, 154 }; 155 156 struct mptcp_pm_data { 157 struct mptcp_addr_info local; 158 struct mptcp_addr_info remote; 159 160 spinlock_t lock; /*protects the whole PM data */ 161 162 bool addr_signal; 163 bool server_side; 164 bool work_pending; 165 bool accept_addr; 166 bool accept_subflow; 167 u8 add_addr_signaled; 168 u8 add_addr_accepted; 169 u8 local_addr_used; 170 u8 subflows; 171 u8 add_addr_signal_max; 172 u8 add_addr_accept_max; 173 u8 local_addr_max; 174 u8 subflows_max; 175 u8 status; 176 177 struct work_struct work; 178 }; 179 180 struct mptcp_data_frag { 181 struct list_head list; 182 u64 data_seq; 183 int data_len; 184 int offset; 185 int overhead; 186 struct page *page; 187 }; 188 189 /* MPTCP connection sock */ 190 struct mptcp_sock { 191 /* inet_connection_sock must be the first member */ 192 struct inet_connection_sock sk; 193 u64 local_key; 194 u64 remote_key; 195 u64 write_seq; 196 u64 ack_seq; 197 atomic64_t snd_una; 198 unsigned long timer_ival; 199 u32 token; 200 unsigned long flags; 201 bool can_ack; 202 spinlock_t join_list_lock; 203 struct work_struct work; 204 struct list_head conn_list; 205 struct list_head rtx_queue; 206 struct list_head join_list; 207 struct skb_ext *cached_ext; /* for the next sendmsg */ 208 struct socket *subflow; /* outgoing connect/listener/!mp_capable */ 209 struct sock *first; 210 struct mptcp_pm_data pm; 211 }; 212 213 #define mptcp_for_each_subflow(__msk, __subflow) \ 214 list_for_each_entry(__subflow, &((__msk)->conn_list), node) 215 216 static inline struct mptcp_sock *mptcp_sk(const struct sock *sk) 217 { 218 return (struct mptcp_sock *)sk; 219 } 220 221 static inline struct mptcp_data_frag *mptcp_rtx_tail(const struct sock *sk) 222 { 223 struct mptcp_sock *msk = mptcp_sk(sk); 224 225 if (list_empty(&msk->rtx_queue)) 226 return NULL; 227 228 return list_last_entry(&msk->rtx_queue, struct mptcp_data_frag, list); 229 } 230 231 static inline struct mptcp_data_frag *mptcp_rtx_head(const struct sock *sk) 232 { 233 struct mptcp_sock *msk = mptcp_sk(sk); 234 235 return list_first_entry_or_null(&msk->rtx_queue, struct mptcp_data_frag, list); 236 } 237 238 struct mptcp_subflow_request_sock { 239 struct tcp_request_sock sk; 240 u16 mp_capable : 1, 241 mp_join : 1, 242 backup : 1; 243 u8 local_id; 244 u8 remote_id; 245 u64 local_key; 246 u64 idsn; 247 u32 token; 248 u32 ssn_offset; 249 u64 thmac; 250 u32 local_nonce; 251 u32 remote_nonce; 252 }; 253 254 static inline struct mptcp_subflow_request_sock * 255 mptcp_subflow_rsk(const struct request_sock *rsk) 256 { 257 return (struct mptcp_subflow_request_sock *)rsk; 258 } 259 260 /* MPTCP subflow context */ 261 struct mptcp_subflow_context { 262 struct list_head node;/* conn_list of subflows */ 263 u64 local_key; 264 u64 remote_key; 265 u64 idsn; 266 u64 map_seq; 267 u32 snd_isn; 268 u32 token; 269 u32 rel_write_seq; 270 u32 map_subflow_seq; 271 u32 ssn_offset; 272 u32 map_data_len; 273 u32 request_mptcp : 1, /* send MP_CAPABLE */ 274 request_join : 1, /* send MP_JOIN */ 275 request_bkup : 1, 276 mp_capable : 1, /* remote is MPTCP capable */ 277 mp_join : 1, /* remote is JOINing */ 278 fully_established : 1, /* path validated */ 279 pm_notified : 1, /* PM hook called for established status */ 280 conn_finished : 1, 281 map_valid : 1, 282 mpc_map : 1, 283 backup : 1, 284 data_avail : 1, 285 rx_eof : 1, 286 data_fin_tx_enable : 1, 287 use_64bit_ack : 1, /* Set when we received a 64-bit DSN */ 288 can_ack : 1; /* only after processing the remote a key */ 289 u64 data_fin_tx_seq; 290 u32 remote_nonce; 291 u64 thmac; 292 u32 local_nonce; 293 u32 remote_token; 294 u8 hmac[MPTCPOPT_HMAC_LEN]; 295 u8 local_id; 296 u8 remote_id; 297 298 struct sock *tcp_sock; /* tcp sk backpointer */ 299 struct sock *conn; /* parent mptcp_sock */ 300 const struct inet_connection_sock_af_ops *icsk_af_ops; 301 void (*tcp_data_ready)(struct sock *sk); 302 void (*tcp_state_change)(struct sock *sk); 303 void (*tcp_write_space)(struct sock *sk); 304 305 struct rcu_head rcu; 306 }; 307 308 static inline struct mptcp_subflow_context * 309 mptcp_subflow_ctx(const struct sock *sk) 310 { 311 struct inet_connection_sock *icsk = inet_csk(sk); 312 313 /* Use RCU on icsk_ulp_data only for sock diag code */ 314 return (__force struct mptcp_subflow_context *)icsk->icsk_ulp_data; 315 } 316 317 static inline struct sock * 318 mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow) 319 { 320 return subflow->tcp_sock; 321 } 322 323 static inline u64 324 mptcp_subflow_get_map_offset(const struct mptcp_subflow_context *subflow) 325 { 326 return tcp_sk(mptcp_subflow_tcp_sock(subflow))->copied_seq - 327 subflow->ssn_offset - 328 subflow->map_subflow_seq; 329 } 330 331 static inline u64 332 mptcp_subflow_get_mapped_dsn(const struct mptcp_subflow_context *subflow) 333 { 334 return subflow->map_seq + mptcp_subflow_get_map_offset(subflow); 335 } 336 337 int mptcp_is_enabled(struct net *net); 338 bool mptcp_subflow_data_available(struct sock *sk); 339 void mptcp_subflow_init(void); 340 341 /* called with sk socket lock held */ 342 int __mptcp_subflow_connect(struct sock *sk, int ifindex, 343 const struct mptcp_addr_info *loc, 344 const struct mptcp_addr_info *remote); 345 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock); 346 347 static inline void mptcp_subflow_tcp_fallback(struct sock *sk, 348 struct mptcp_subflow_context *ctx) 349 { 350 sk->sk_data_ready = ctx->tcp_data_ready; 351 sk->sk_state_change = ctx->tcp_state_change; 352 sk->sk_write_space = ctx->tcp_write_space; 353 354 inet_csk(sk)->icsk_af_ops = ctx->icsk_af_ops; 355 } 356 357 extern const struct inet_connection_sock_af_ops ipv4_specific; 358 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 359 extern const struct inet_connection_sock_af_ops ipv6_specific; 360 #endif 361 362 void mptcp_proto_init(void); 363 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 364 int mptcp_proto_v6_init(void); 365 #endif 366 367 struct sock *mptcp_sk_clone(const struct sock *sk, 368 const struct mptcp_options_received *mp_opt, 369 struct request_sock *req); 370 void mptcp_get_options(const struct sk_buff *skb, 371 struct mptcp_options_received *mp_opt); 372 373 void mptcp_finish_connect(struct sock *sk); 374 void mptcp_data_ready(struct sock *sk, struct sock *ssk); 375 bool mptcp_finish_join(struct sock *sk); 376 void mptcp_data_acked(struct sock *sk); 377 void mptcp_subflow_eof(struct sock *sk); 378 379 int mptcp_token_new_request(struct request_sock *req); 380 void mptcp_token_destroy_request(u32 token); 381 int mptcp_token_new_connect(struct sock *sk); 382 int mptcp_token_new_accept(u32 token, struct sock *conn); 383 struct mptcp_sock *mptcp_token_get_sock(u32 token); 384 void mptcp_token_destroy(u32 token); 385 386 void mptcp_crypto_key_sha(u64 key, u32 *token, u64 *idsn); 387 static inline void mptcp_crypto_key_gen_sha(u64 *key, u32 *token, u64 *idsn) 388 { 389 /* we might consider a faster version that computes the key as a 390 * hash of some information available in the MPTCP socket. Use 391 * random data at the moment, as it's probably the safest option 392 * in case multiple sockets are opened in different namespaces at 393 * the same time. 394 */ 395 get_random_bytes(key, sizeof(u64)); 396 mptcp_crypto_key_sha(*key, token, idsn); 397 } 398 399 void mptcp_crypto_hmac_sha(u64 key1, u64 key2, u8 *msg, int len, void *hmac); 400 401 void mptcp_pm_init(void); 402 void mptcp_pm_data_init(struct mptcp_sock *msk); 403 void mptcp_pm_close(struct mptcp_sock *msk); 404 void mptcp_pm_new_connection(struct mptcp_sock *msk, int server_side); 405 void mptcp_pm_fully_established(struct mptcp_sock *msk); 406 bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk); 407 void mptcp_pm_connection_closed(struct mptcp_sock *msk); 408 void mptcp_pm_subflow_established(struct mptcp_sock *msk, 409 struct mptcp_subflow_context *subflow); 410 void mptcp_pm_subflow_closed(struct mptcp_sock *msk, u8 id); 411 void mptcp_pm_add_addr_received(struct mptcp_sock *msk, 412 const struct mptcp_addr_info *addr); 413 414 int mptcp_pm_announce_addr(struct mptcp_sock *msk, 415 const struct mptcp_addr_info *addr); 416 int mptcp_pm_remove_addr(struct mptcp_sock *msk, u8 local_id); 417 int mptcp_pm_remove_subflow(struct mptcp_sock *msk, u8 remote_id); 418 419 static inline bool mptcp_pm_should_signal(struct mptcp_sock *msk) 420 { 421 return READ_ONCE(msk->pm.addr_signal); 422 } 423 424 static inline unsigned int mptcp_add_addr_len(int family) 425 { 426 if (family == AF_INET) 427 return TCPOLEN_MPTCP_ADD_ADDR; 428 return TCPOLEN_MPTCP_ADD_ADDR6; 429 } 430 431 bool mptcp_pm_addr_signal(struct mptcp_sock *msk, unsigned int remaining, 432 struct mptcp_addr_info *saddr); 433 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc); 434 435 void mptcp_pm_nl_init(void); 436 void mptcp_pm_nl_data_init(struct mptcp_sock *msk); 437 void mptcp_pm_nl_fully_established(struct mptcp_sock *msk); 438 void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk); 439 void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk); 440 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc); 441 442 static inline struct mptcp_ext *mptcp_get_ext(struct sk_buff *skb) 443 { 444 return (struct mptcp_ext *)skb_ext_find(skb, SKB_EXT_MPTCP); 445 } 446 447 static inline bool before64(__u64 seq1, __u64 seq2) 448 { 449 return (__s64)(seq1 - seq2) < 0; 450 } 451 452 #define after64(seq2, seq1) before64(seq1, seq2) 453 454 void mptcp_diag_subflow_init(struct tcp_ulp_ops *ops); 455 456 #endif /* __MPTCP_PROTOCOL_H */ 457