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 #define MPTCP_FALLBACK_DONE 4 93 94 struct mptcp_options_received { 95 u64 sndr_key; 96 u64 rcvr_key; 97 u64 data_ack; 98 u64 data_seq; 99 u32 subflow_seq; 100 u16 data_len; 101 u16 mp_capable : 1, 102 mp_join : 1, 103 dss : 1, 104 add_addr : 1, 105 rm_addr : 1, 106 family : 4, 107 echo : 1, 108 backup : 1; 109 u32 token; 110 u32 nonce; 111 u64 thmac; 112 u8 hmac[20]; 113 u8 join_id; 114 u8 use_map:1, 115 dsn64:1, 116 data_fin:1, 117 use_ack:1, 118 ack64:1, 119 mpc_map:1, 120 __unused:2; 121 u8 addr_id; 122 u8 rm_id; 123 union { 124 struct in_addr addr; 125 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 126 struct in6_addr addr6; 127 #endif 128 }; 129 u64 ahmac; 130 u16 port; 131 }; 132 133 static inline __be32 mptcp_option(u8 subopt, u8 len, u8 nib, u8 field) 134 { 135 return htonl((TCPOPT_MPTCP << 24) | (len << 16) | (subopt << 12) | 136 ((nib & 0xF) << 8) | field); 137 } 138 139 struct mptcp_addr_info { 140 sa_family_t family; 141 __be16 port; 142 u8 id; 143 union { 144 struct in_addr addr; 145 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 146 struct in6_addr addr6; 147 #endif 148 }; 149 }; 150 151 enum mptcp_pm_status { 152 MPTCP_PM_ADD_ADDR_RECEIVED, 153 MPTCP_PM_ESTABLISHED, 154 MPTCP_PM_SUBFLOW_ESTABLISHED, 155 }; 156 157 struct mptcp_pm_data { 158 struct mptcp_addr_info local; 159 struct mptcp_addr_info remote; 160 161 spinlock_t lock; /*protects the whole PM data */ 162 163 bool addr_signal; 164 bool server_side; 165 bool work_pending; 166 bool accept_addr; 167 bool accept_subflow; 168 u8 add_addr_signaled; 169 u8 add_addr_accepted; 170 u8 local_addr_used; 171 u8 subflows; 172 u8 add_addr_signal_max; 173 u8 add_addr_accept_max; 174 u8 local_addr_max; 175 u8 subflows_max; 176 u8 status; 177 }; 178 179 struct mptcp_data_frag { 180 struct list_head list; 181 u64 data_seq; 182 int data_len; 183 int offset; 184 int overhead; 185 struct page *page; 186 }; 187 188 /* MPTCP connection sock */ 189 struct mptcp_sock { 190 /* inet_connection_sock must be the first member */ 191 struct inet_connection_sock sk; 192 u64 local_key; 193 u64 remote_key; 194 u64 write_seq; 195 u64 ack_seq; 196 atomic64_t snd_una; 197 unsigned long timer_ival; 198 u32 token; 199 unsigned long flags; 200 bool can_ack; 201 spinlock_t join_list_lock; 202 struct work_struct work; 203 struct list_head conn_list; 204 struct list_head rtx_queue; 205 struct list_head join_list; 206 struct skb_ext *cached_ext; /* for the next sendmsg */ 207 struct socket *subflow; /* outgoing connect/listener/!mp_capable */ 208 struct sock *first; 209 struct mptcp_pm_data pm; 210 struct { 211 u32 space; /* bytes copied in last measurement window */ 212 u32 copied; /* bytes copied in this measurement window */ 213 u64 time; /* start time of measurement window */ 214 u64 rtt_us; /* last maximum rtt of subflows */ 215 } rcvq_space; 216 }; 217 218 #define mptcp_for_each_subflow(__msk, __subflow) \ 219 list_for_each_entry(__subflow, &((__msk)->conn_list), node) 220 221 static inline struct mptcp_sock *mptcp_sk(const struct sock *sk) 222 { 223 return (struct mptcp_sock *)sk; 224 } 225 226 static inline struct mptcp_data_frag *mptcp_rtx_tail(const struct sock *sk) 227 { 228 struct mptcp_sock *msk = mptcp_sk(sk); 229 230 if (list_empty(&msk->rtx_queue)) 231 return NULL; 232 233 return list_last_entry(&msk->rtx_queue, struct mptcp_data_frag, list); 234 } 235 236 static inline struct mptcp_data_frag *mptcp_rtx_head(const struct sock *sk) 237 { 238 struct mptcp_sock *msk = mptcp_sk(sk); 239 240 return list_first_entry_or_null(&msk->rtx_queue, struct mptcp_data_frag, list); 241 } 242 243 struct mptcp_subflow_request_sock { 244 struct tcp_request_sock sk; 245 u16 mp_capable : 1, 246 mp_join : 1, 247 backup : 1; 248 u8 local_id; 249 u8 remote_id; 250 u64 local_key; 251 u64 idsn; 252 u32 token; 253 u32 ssn_offset; 254 u64 thmac; 255 u32 local_nonce; 256 u32 remote_nonce; 257 struct mptcp_sock *msk; 258 struct hlist_nulls_node token_node; 259 }; 260 261 static inline struct mptcp_subflow_request_sock * 262 mptcp_subflow_rsk(const struct request_sock *rsk) 263 { 264 return (struct mptcp_subflow_request_sock *)rsk; 265 } 266 267 /* MPTCP subflow context */ 268 struct mptcp_subflow_context { 269 struct list_head node;/* conn_list of subflows */ 270 u64 local_key; 271 u64 remote_key; 272 u64 idsn; 273 u64 map_seq; 274 u32 snd_isn; 275 u32 token; 276 u32 rel_write_seq; 277 u32 map_subflow_seq; 278 u32 ssn_offset; 279 u32 map_data_len; 280 u32 request_mptcp : 1, /* send MP_CAPABLE */ 281 request_join : 1, /* send MP_JOIN */ 282 request_bkup : 1, 283 mp_capable : 1, /* remote is MPTCP capable */ 284 mp_join : 1, /* remote is JOINing */ 285 fully_established : 1, /* path validated */ 286 pm_notified : 1, /* PM hook called for established status */ 287 conn_finished : 1, 288 map_valid : 1, 289 mpc_map : 1, 290 backup : 1, 291 data_avail : 1, 292 rx_eof : 1, 293 data_fin_tx_enable : 1, 294 use_64bit_ack : 1, /* Set when we received a 64-bit DSN */ 295 can_ack : 1; /* only after processing the remote a key */ 296 u64 data_fin_tx_seq; 297 u32 remote_nonce; 298 u64 thmac; 299 u32 local_nonce; 300 u32 remote_token; 301 u8 hmac[MPTCPOPT_HMAC_LEN]; 302 u8 local_id; 303 u8 remote_id; 304 305 struct sock *tcp_sock; /* tcp sk backpointer */ 306 struct sock *conn; /* parent mptcp_sock */ 307 const struct inet_connection_sock_af_ops *icsk_af_ops; 308 void (*tcp_data_ready)(struct sock *sk); 309 void (*tcp_state_change)(struct sock *sk); 310 void (*tcp_write_space)(struct sock *sk); 311 312 struct rcu_head rcu; 313 }; 314 315 static inline struct mptcp_subflow_context * 316 mptcp_subflow_ctx(const struct sock *sk) 317 { 318 struct inet_connection_sock *icsk = inet_csk(sk); 319 320 /* Use RCU on icsk_ulp_data only for sock diag code */ 321 return (__force struct mptcp_subflow_context *)icsk->icsk_ulp_data; 322 } 323 324 static inline struct sock * 325 mptcp_subflow_tcp_sock(const struct mptcp_subflow_context *subflow) 326 { 327 return subflow->tcp_sock; 328 } 329 330 static inline u64 331 mptcp_subflow_get_map_offset(const struct mptcp_subflow_context *subflow) 332 { 333 return tcp_sk(mptcp_subflow_tcp_sock(subflow))->copied_seq - 334 subflow->ssn_offset - 335 subflow->map_subflow_seq; 336 } 337 338 static inline u64 339 mptcp_subflow_get_mapped_dsn(const struct mptcp_subflow_context *subflow) 340 { 341 return subflow->map_seq + mptcp_subflow_get_map_offset(subflow); 342 } 343 344 int mptcp_is_enabled(struct net *net); 345 bool mptcp_subflow_data_available(struct sock *sk); 346 void __init mptcp_subflow_init(void); 347 348 /* called with sk socket lock held */ 349 int __mptcp_subflow_connect(struct sock *sk, int ifindex, 350 const struct mptcp_addr_info *loc, 351 const struct mptcp_addr_info *remote); 352 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock); 353 354 static inline void mptcp_subflow_tcp_fallback(struct sock *sk, 355 struct mptcp_subflow_context *ctx) 356 { 357 sk->sk_data_ready = ctx->tcp_data_ready; 358 sk->sk_state_change = ctx->tcp_state_change; 359 sk->sk_write_space = ctx->tcp_write_space; 360 361 inet_csk(sk)->icsk_af_ops = ctx->icsk_af_ops; 362 } 363 364 void __init mptcp_proto_init(void); 365 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 366 int __init mptcp_proto_v6_init(void); 367 #endif 368 369 struct sock *mptcp_sk_clone(const struct sock *sk, 370 const struct mptcp_options_received *mp_opt, 371 struct request_sock *req); 372 void mptcp_get_options(const struct sk_buff *skb, 373 struct mptcp_options_received *mp_opt); 374 375 void mptcp_finish_connect(struct sock *sk); 376 void mptcp_rcv_space_init(struct mptcp_sock *msk, const struct sock *ssk); 377 void mptcp_data_ready(struct sock *sk, struct sock *ssk); 378 bool mptcp_finish_join(struct sock *sk); 379 void mptcp_data_acked(struct sock *sk); 380 void mptcp_subflow_eof(struct sock *sk); 381 382 void __init mptcp_token_init(void); 383 static inline void mptcp_token_init_request(struct request_sock *req) 384 { 385 mptcp_subflow_rsk(req)->token_node.pprev = NULL; 386 } 387 388 int mptcp_token_new_request(struct request_sock *req); 389 void mptcp_token_destroy_request(struct request_sock *req); 390 int mptcp_token_new_connect(struct sock *sk); 391 void mptcp_token_accept(struct mptcp_subflow_request_sock *r, 392 struct mptcp_sock *msk); 393 struct mptcp_sock *mptcp_token_get_sock(u32 token); 394 struct mptcp_sock *mptcp_token_iter_next(const struct net *net, long *s_slot, 395 long *s_num); 396 void mptcp_token_destroy(struct mptcp_sock *msk); 397 398 void mptcp_crypto_key_sha(u64 key, u32 *token, u64 *idsn); 399 static inline void mptcp_crypto_key_gen_sha(u64 *key, u32 *token, u64 *idsn) 400 { 401 /* we might consider a faster version that computes the key as a 402 * hash of some information available in the MPTCP socket. Use 403 * random data at the moment, as it's probably the safest option 404 * in case multiple sockets are opened in different namespaces at 405 * the same time. 406 */ 407 get_random_bytes(key, sizeof(u64)); 408 mptcp_crypto_key_sha(*key, token, idsn); 409 } 410 411 void mptcp_crypto_hmac_sha(u64 key1, u64 key2, u8 *msg, int len, void *hmac); 412 413 void __init mptcp_pm_init(void); 414 void mptcp_pm_data_init(struct mptcp_sock *msk); 415 void mptcp_pm_new_connection(struct mptcp_sock *msk, int server_side); 416 void mptcp_pm_fully_established(struct mptcp_sock *msk); 417 bool mptcp_pm_allow_new_subflow(struct mptcp_sock *msk); 418 void mptcp_pm_connection_closed(struct mptcp_sock *msk); 419 void mptcp_pm_subflow_established(struct mptcp_sock *msk, 420 struct mptcp_subflow_context *subflow); 421 void mptcp_pm_subflow_closed(struct mptcp_sock *msk, u8 id); 422 void mptcp_pm_add_addr_received(struct mptcp_sock *msk, 423 const struct mptcp_addr_info *addr); 424 425 int mptcp_pm_announce_addr(struct mptcp_sock *msk, 426 const struct mptcp_addr_info *addr); 427 int mptcp_pm_remove_addr(struct mptcp_sock *msk, u8 local_id); 428 int mptcp_pm_remove_subflow(struct mptcp_sock *msk, u8 remote_id); 429 430 static inline bool mptcp_pm_should_signal(struct mptcp_sock *msk) 431 { 432 return READ_ONCE(msk->pm.addr_signal); 433 } 434 435 static inline unsigned int mptcp_add_addr_len(int family) 436 { 437 if (family == AF_INET) 438 return TCPOLEN_MPTCP_ADD_ADDR; 439 return TCPOLEN_MPTCP_ADD_ADDR6; 440 } 441 442 bool mptcp_pm_addr_signal(struct mptcp_sock *msk, unsigned int remaining, 443 struct mptcp_addr_info *saddr); 444 int mptcp_pm_get_local_id(struct mptcp_sock *msk, struct sock_common *skc); 445 446 void __init mptcp_pm_nl_init(void); 447 void mptcp_pm_nl_data_init(struct mptcp_sock *msk); 448 void mptcp_pm_nl_fully_established(struct mptcp_sock *msk); 449 void mptcp_pm_nl_subflow_established(struct mptcp_sock *msk); 450 void mptcp_pm_nl_add_addr_received(struct mptcp_sock *msk); 451 int mptcp_pm_nl_get_local_id(struct mptcp_sock *msk, struct sock_common *skc); 452 453 static inline struct mptcp_ext *mptcp_get_ext(struct sk_buff *skb) 454 { 455 return (struct mptcp_ext *)skb_ext_find(skb, SKB_EXT_MPTCP); 456 } 457 458 static inline bool before64(__u64 seq1, __u64 seq2) 459 { 460 return (__s64)(seq1 - seq2) < 0; 461 } 462 463 #define after64(seq2, seq1) before64(seq1, seq2) 464 465 void mptcp_diag_subflow_init(struct tcp_ulp_ops *ops); 466 467 static inline bool __mptcp_check_fallback(struct mptcp_sock *msk) 468 { 469 return test_bit(MPTCP_FALLBACK_DONE, &msk->flags); 470 } 471 472 static inline bool mptcp_check_fallback(struct sock *sk) 473 { 474 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 475 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 476 477 return __mptcp_check_fallback(msk); 478 } 479 480 static inline void __mptcp_do_fallback(struct mptcp_sock *msk) 481 { 482 if (test_bit(MPTCP_FALLBACK_DONE, &msk->flags)) { 483 pr_debug("TCP fallback already done (msk=%p)", msk); 484 return; 485 } 486 set_bit(MPTCP_FALLBACK_DONE, &msk->flags); 487 } 488 489 static inline void mptcp_do_fallback(struct sock *sk) 490 { 491 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 492 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 493 494 __mptcp_do_fallback(msk); 495 } 496 497 #define pr_fallback(a) pr_debug("%s:fallback to TCP (msk=%p)", __func__, a) 498 499 static inline bool subflow_simultaneous_connect(struct sock *sk) 500 { 501 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 502 struct sock *parent = subflow->conn; 503 504 return sk->sk_state == TCP_ESTABLISHED && 505 !mptcp_sk(parent)->pm.server_side && 506 !subflow->conn_finished; 507 } 508 509 #endif /* __MPTCP_PROTOCOL_H */ 510