1 // SPDX-License-Identifier: GPL-2.0 2 /* Multipath TCP 3 * 4 * Copyright (c) 2017 - 2019, Intel Corporation. 5 */ 6 7 #define pr_fmt(fmt) "MPTCP: " fmt 8 9 #include <linux/kernel.h> 10 #include <crypto/sha2.h> 11 #include <net/tcp.h> 12 #include <net/mptcp.h> 13 #include "protocol.h" 14 #include "mib.h" 15 16 static bool mptcp_cap_flag_sha256(u8 flags) 17 { 18 return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256; 19 } 20 21 static void mptcp_parse_option(const struct sk_buff *skb, 22 const unsigned char *ptr, int opsize, 23 struct mptcp_options_received *mp_opt) 24 { 25 u8 subtype = *ptr >> 4; 26 int expected_opsize; 27 u8 version; 28 u8 flags; 29 30 switch (subtype) { 31 case MPTCPOPT_MP_CAPABLE: 32 /* strict size checking */ 33 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { 34 if (skb->len > tcp_hdr(skb)->doff << 2) 35 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA; 36 else 37 expected_opsize = TCPOLEN_MPTCP_MPC_ACK; 38 } else { 39 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK) 40 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK; 41 else 42 expected_opsize = TCPOLEN_MPTCP_MPC_SYN; 43 } 44 if (opsize != expected_opsize) 45 break; 46 47 /* try to be gentle vs future versions on the initial syn */ 48 version = *ptr++ & MPTCP_VERSION_MASK; 49 if (opsize != TCPOLEN_MPTCP_MPC_SYN) { 50 if (version != MPTCP_SUPPORTED_VERSION) 51 break; 52 } else if (version < MPTCP_SUPPORTED_VERSION) { 53 break; 54 } 55 56 flags = *ptr++; 57 if (!mptcp_cap_flag_sha256(flags) || 58 (flags & MPTCP_CAP_EXTENSIBILITY)) 59 break; 60 61 /* RFC 6824, Section 3.1: 62 * "For the Checksum Required bit (labeled "A"), if either 63 * host requires the use of checksums, checksums MUST be used. 64 * In other words, the only way for checksums not to be used 65 * is if both hosts in their SYNs set A=0." 66 * 67 * Section 3.3.0: 68 * "If a checksum is not present when its use has been 69 * negotiated, the receiver MUST close the subflow with a RST as 70 * it is considered broken." 71 * 72 * We don't implement DSS checksum - fall back to TCP. 73 */ 74 if (flags & MPTCP_CAP_CHECKSUM_REQD) 75 break; 76 77 mp_opt->mp_capable = 1; 78 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) { 79 mp_opt->sndr_key = get_unaligned_be64(ptr); 80 ptr += 8; 81 } 82 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) { 83 mp_opt->rcvr_key = get_unaligned_be64(ptr); 84 ptr += 8; 85 } 86 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA) { 87 /* Section 3.1.: 88 * "the data parameters in a MP_CAPABLE are semantically 89 * equivalent to those in a DSS option and can be used 90 * interchangeably." 91 */ 92 mp_opt->dss = 1; 93 mp_opt->use_map = 1; 94 mp_opt->mpc_map = 1; 95 mp_opt->data_len = get_unaligned_be16(ptr); 96 ptr += 2; 97 } 98 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d", 99 version, flags, opsize, mp_opt->sndr_key, 100 mp_opt->rcvr_key, mp_opt->data_len); 101 break; 102 103 case MPTCPOPT_MP_JOIN: 104 mp_opt->mp_join = 1; 105 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) { 106 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 107 mp_opt->join_id = *ptr++; 108 mp_opt->token = get_unaligned_be32(ptr); 109 ptr += 4; 110 mp_opt->nonce = get_unaligned_be32(ptr); 111 ptr += 4; 112 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u", 113 mp_opt->backup, mp_opt->join_id, 114 mp_opt->token, mp_opt->nonce); 115 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) { 116 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 117 mp_opt->join_id = *ptr++; 118 mp_opt->thmac = get_unaligned_be64(ptr); 119 ptr += 8; 120 mp_opt->nonce = get_unaligned_be32(ptr); 121 ptr += 4; 122 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u", 123 mp_opt->backup, mp_opt->join_id, 124 mp_opt->thmac, mp_opt->nonce); 125 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) { 126 ptr += 2; 127 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN); 128 pr_debug("MP_JOIN hmac"); 129 } else { 130 pr_warn("MP_JOIN bad option size"); 131 mp_opt->mp_join = 0; 132 } 133 break; 134 135 case MPTCPOPT_DSS: 136 pr_debug("DSS"); 137 ptr++; 138 139 /* we must clear 'mpc_map' be able to detect MP_CAPABLE 140 * map vs DSS map in mptcp_incoming_options(), and reconstruct 141 * map info accordingly 142 */ 143 mp_opt->mpc_map = 0; 144 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK; 145 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0; 146 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0; 147 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0; 148 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0; 149 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK); 150 151 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d", 152 mp_opt->data_fin, mp_opt->dsn64, 153 mp_opt->use_map, mp_opt->ack64, 154 mp_opt->use_ack); 155 156 expected_opsize = TCPOLEN_MPTCP_DSS_BASE; 157 158 if (mp_opt->use_ack) { 159 if (mp_opt->ack64) 160 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64; 161 else 162 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32; 163 } 164 165 if (mp_opt->use_map) { 166 if (mp_opt->dsn64) 167 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64; 168 else 169 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32; 170 } 171 172 /* RFC 6824, Section 3.3: 173 * If a checksum is present, but its use had 174 * not been negotiated in the MP_CAPABLE handshake, 175 * the checksum field MUST be ignored. 176 */ 177 if (opsize != expected_opsize && 178 opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) 179 break; 180 181 mp_opt->dss = 1; 182 183 if (mp_opt->use_ack) { 184 if (mp_opt->ack64) { 185 mp_opt->data_ack = get_unaligned_be64(ptr); 186 ptr += 8; 187 } else { 188 mp_opt->data_ack = get_unaligned_be32(ptr); 189 ptr += 4; 190 } 191 192 pr_debug("data_ack=%llu", mp_opt->data_ack); 193 } 194 195 if (mp_opt->use_map) { 196 if (mp_opt->dsn64) { 197 mp_opt->data_seq = get_unaligned_be64(ptr); 198 ptr += 8; 199 } else { 200 mp_opt->data_seq = get_unaligned_be32(ptr); 201 ptr += 4; 202 } 203 204 mp_opt->subflow_seq = get_unaligned_be32(ptr); 205 ptr += 4; 206 207 mp_opt->data_len = get_unaligned_be16(ptr); 208 ptr += 2; 209 210 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u", 211 mp_opt->data_seq, mp_opt->subflow_seq, 212 mp_opt->data_len); 213 } 214 215 break; 216 217 case MPTCPOPT_ADD_ADDR: 218 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO; 219 if (!mp_opt->echo) { 220 if (opsize == TCPOLEN_MPTCP_ADD_ADDR || 221 opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT) 222 mp_opt->family = MPTCP_ADDR_IPVERSION_4; 223 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 224 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 || 225 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT) 226 mp_opt->family = MPTCP_ADDR_IPVERSION_6; 227 #endif 228 else 229 break; 230 } else { 231 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE || 232 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) 233 mp_opt->family = MPTCP_ADDR_IPVERSION_4; 234 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 235 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE || 236 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) 237 mp_opt->family = MPTCP_ADDR_IPVERSION_6; 238 #endif 239 else 240 break; 241 } 242 243 mp_opt->add_addr = 1; 244 mp_opt->addr_id = *ptr++; 245 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4) { 246 memcpy((u8 *)&mp_opt->addr.s_addr, (u8 *)ptr, 4); 247 ptr += 4; 248 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT || 249 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) { 250 mp_opt->port = get_unaligned_be16(ptr); 251 ptr += 2; 252 } 253 } 254 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 255 else { 256 memcpy(mp_opt->addr6.s6_addr, (u8 *)ptr, 16); 257 ptr += 16; 258 if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT || 259 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) { 260 mp_opt->port = get_unaligned_be16(ptr); 261 ptr += 2; 262 } 263 } 264 #endif 265 if (!mp_opt->echo) { 266 mp_opt->ahmac = get_unaligned_be64(ptr); 267 ptr += 8; 268 } 269 pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d", 270 (mp_opt->family == MPTCP_ADDR_IPVERSION_6) ? "6" : "", 271 mp_opt->addr_id, mp_opt->ahmac, mp_opt->echo, mp_opt->port); 272 break; 273 274 case MPTCPOPT_RM_ADDR: 275 if (opsize != TCPOLEN_MPTCP_RM_ADDR_BASE) 276 break; 277 278 ptr++; 279 280 mp_opt->rm_addr = 1; 281 mp_opt->rm_id = *ptr++; 282 pr_debug("RM_ADDR: id=%d", mp_opt->rm_id); 283 break; 284 285 case MPTCPOPT_MP_PRIO: 286 if (opsize != TCPOLEN_MPTCP_PRIO) 287 break; 288 289 mp_opt->mp_prio = 1; 290 mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP; 291 pr_debug("MP_PRIO: prio=%d", mp_opt->backup); 292 break; 293 294 case MPTCPOPT_MP_FASTCLOSE: 295 if (opsize != TCPOLEN_MPTCP_FASTCLOSE) 296 break; 297 298 ptr += 2; 299 mp_opt->rcvr_key = get_unaligned_be64(ptr); 300 ptr += 8; 301 mp_opt->fastclose = 1; 302 break; 303 304 default: 305 break; 306 } 307 } 308 309 void mptcp_get_options(const struct sk_buff *skb, 310 struct mptcp_options_received *mp_opt) 311 { 312 const struct tcphdr *th = tcp_hdr(skb); 313 const unsigned char *ptr; 314 int length; 315 316 /* initialize option status */ 317 mp_opt->mp_capable = 0; 318 mp_opt->mp_join = 0; 319 mp_opt->add_addr = 0; 320 mp_opt->ahmac = 0; 321 mp_opt->fastclose = 0; 322 mp_opt->port = 0; 323 mp_opt->rm_addr = 0; 324 mp_opt->dss = 0; 325 mp_opt->mp_prio = 0; 326 327 length = (th->doff * 4) - sizeof(struct tcphdr); 328 ptr = (const unsigned char *)(th + 1); 329 330 while (length > 0) { 331 int opcode = *ptr++; 332 int opsize; 333 334 switch (opcode) { 335 case TCPOPT_EOL: 336 return; 337 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */ 338 length--; 339 continue; 340 default: 341 opsize = *ptr++; 342 if (opsize < 2) /* "silly options" */ 343 return; 344 if (opsize > length) 345 return; /* don't parse partial options */ 346 if (opcode == TCPOPT_MPTCP) 347 mptcp_parse_option(skb, ptr, opsize, mp_opt); 348 ptr += opsize - 2; 349 length -= opsize; 350 } 351 } 352 } 353 354 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb, 355 unsigned int *size, struct mptcp_out_options *opts) 356 { 357 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 358 359 /* we will use snd_isn to detect first pkt [re]transmission 360 * in mptcp_established_options_mp() 361 */ 362 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq; 363 if (subflow->request_mptcp) { 364 opts->suboptions = OPTION_MPTCP_MPC_SYN; 365 *size = TCPOLEN_MPTCP_MPC_SYN; 366 return true; 367 } else if (subflow->request_join) { 368 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token, 369 subflow->local_nonce); 370 opts->suboptions = OPTION_MPTCP_MPJ_SYN; 371 opts->join_id = subflow->local_id; 372 opts->token = subflow->remote_token; 373 opts->nonce = subflow->local_nonce; 374 opts->backup = subflow->request_bkup; 375 *size = TCPOLEN_MPTCP_MPJ_SYN; 376 return true; 377 } 378 return false; 379 } 380 381 /* MP_JOIN client subflow must wait for 4th ack before sending any data: 382 * TCP can't schedule delack timer before the subflow is fully established. 383 * MPTCP uses the delack timer to do 3rd ack retransmissions 384 */ 385 static void schedule_3rdack_retransmission(struct sock *sk) 386 { 387 struct inet_connection_sock *icsk = inet_csk(sk); 388 struct tcp_sock *tp = tcp_sk(sk); 389 unsigned long timeout; 390 391 /* reschedule with a timeout above RTT, as we must look only for drop */ 392 if (tp->srtt_us) 393 timeout = tp->srtt_us << 1; 394 else 395 timeout = TCP_TIMEOUT_INIT; 396 397 WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER); 398 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER; 399 icsk->icsk_ack.timeout = timeout; 400 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout); 401 } 402 403 static void clear_3rdack_retransmission(struct sock *sk) 404 { 405 struct inet_connection_sock *icsk = inet_csk(sk); 406 407 sk_stop_timer(sk, &icsk->icsk_delack_timer); 408 icsk->icsk_ack.timeout = 0; 409 icsk->icsk_ack.ato = 0; 410 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER); 411 } 412 413 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb, 414 unsigned int *size, 415 unsigned int remaining, 416 struct mptcp_out_options *opts) 417 { 418 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 419 struct mptcp_ext *mpext; 420 unsigned int data_len; 421 422 /* When skb is not available, we better over-estimate the emitted 423 * options len. A full DSS option (28 bytes) is longer than 424 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so 425 * tell the caller to defer the estimate to 426 * mptcp_established_options_dss(), which will reserve enough space. 427 */ 428 if (!skb) 429 return false; 430 431 /* MPC/MPJ needed only on 3rd ack packet */ 432 if (subflow->fully_established || 433 subflow->snd_isn != TCP_SKB_CB(skb)->seq) 434 return false; 435 436 if (subflow->mp_capable) { 437 mpext = mptcp_get_ext(skb); 438 data_len = mpext ? mpext->data_len : 0; 439 440 /* we will check ext_copy.data_len in mptcp_write_options() to 441 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and 442 * TCPOLEN_MPTCP_MPC_ACK 443 */ 444 opts->ext_copy.data_len = data_len; 445 opts->suboptions = OPTION_MPTCP_MPC_ACK; 446 opts->sndr_key = subflow->local_key; 447 opts->rcvr_key = subflow->remote_key; 448 449 /* Section 3.1. 450 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK 451 * packets that start the first subflow of an MPTCP connection, 452 * as well as the first packet that carries data 453 */ 454 if (data_len > 0) 455 *size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4); 456 else 457 *size = TCPOLEN_MPTCP_MPC_ACK; 458 459 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d", 460 subflow, subflow->local_key, subflow->remote_key, 461 data_len); 462 463 return true; 464 } else if (subflow->mp_join) { 465 opts->suboptions = OPTION_MPTCP_MPJ_ACK; 466 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN); 467 *size = TCPOLEN_MPTCP_MPJ_ACK; 468 pr_debug("subflow=%p", subflow); 469 470 schedule_3rdack_retransmission(sk); 471 return true; 472 } 473 return false; 474 } 475 476 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow, 477 struct sk_buff *skb, struct mptcp_ext *ext) 478 { 479 /* The write_seq value has already been incremented, so the actual 480 * sequence number for the DATA_FIN is one less. 481 */ 482 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1; 483 484 if (!ext->use_map || !skb->len) { 485 /* RFC6824 requires a DSS mapping with specific values 486 * if DATA_FIN is set but no data payload is mapped 487 */ 488 ext->data_fin = 1; 489 ext->use_map = 1; 490 ext->dsn64 = 1; 491 ext->data_seq = data_fin_tx_seq; 492 ext->subflow_seq = 0; 493 ext->data_len = 1; 494 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) { 495 /* If there's an existing DSS mapping and it is the 496 * final mapping, DATA_FIN consumes 1 additional byte of 497 * mapping space. 498 */ 499 ext->data_fin = 1; 500 ext->data_len++; 501 } 502 } 503 504 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb, 505 unsigned int *size, 506 unsigned int remaining, 507 struct mptcp_out_options *opts) 508 { 509 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 510 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 511 unsigned int dss_size = 0; 512 u64 snd_data_fin_enable; 513 struct mptcp_ext *mpext; 514 unsigned int ack_size; 515 bool ret = false; 516 517 mpext = skb ? mptcp_get_ext(skb) : NULL; 518 snd_data_fin_enable = mptcp_data_fin_enabled(msk); 519 520 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) { 521 unsigned int map_size; 522 523 map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64; 524 525 remaining -= map_size; 526 dss_size = map_size; 527 if (mpext) 528 opts->ext_copy = *mpext; 529 530 if (skb && snd_data_fin_enable) 531 mptcp_write_data_fin(subflow, skb, &opts->ext_copy); 532 ret = true; 533 } 534 535 /* passive sockets msk will set the 'can_ack' after accept(), even 536 * if the first subflow may have the already the remote key handy 537 */ 538 opts->ext_copy.use_ack = 0; 539 if (!READ_ONCE(msk->can_ack)) { 540 *size = ALIGN(dss_size, 4); 541 return ret; 542 } 543 544 if (READ_ONCE(msk->use_64bit_ack)) { 545 ack_size = TCPOLEN_MPTCP_DSS_ACK64; 546 opts->ext_copy.data_ack = READ_ONCE(msk->ack_seq); 547 opts->ext_copy.ack64 = 1; 548 } else { 549 ack_size = TCPOLEN_MPTCP_DSS_ACK32; 550 opts->ext_copy.data_ack32 = (uint32_t)READ_ONCE(msk->ack_seq); 551 opts->ext_copy.ack64 = 0; 552 } 553 opts->ext_copy.use_ack = 1; 554 WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk)); 555 556 /* Add kind/length/subtype/flag overhead if mapping is not populated */ 557 if (dss_size == 0) 558 ack_size += TCPOLEN_MPTCP_DSS_BASE; 559 560 dss_size += ack_size; 561 562 *size = ALIGN(dss_size, 4); 563 return true; 564 } 565 566 static u64 add_addr_generate_hmac(u64 key1, u64 key2, u8 addr_id, 567 struct in_addr *addr) 568 { 569 u8 hmac[SHA256_DIGEST_SIZE]; 570 u8 msg[7]; 571 572 msg[0] = addr_id; 573 memcpy(&msg[1], &addr->s_addr, 4); 574 msg[5] = 0; 575 msg[6] = 0; 576 577 mptcp_crypto_hmac_sha(key1, key2, msg, 7, hmac); 578 579 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]); 580 } 581 582 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 583 static u64 add_addr6_generate_hmac(u64 key1, u64 key2, u8 addr_id, 584 struct in6_addr *addr) 585 { 586 u8 hmac[SHA256_DIGEST_SIZE]; 587 u8 msg[19]; 588 589 msg[0] = addr_id; 590 memcpy(&msg[1], &addr->s6_addr, 16); 591 msg[17] = 0; 592 msg[18] = 0; 593 594 mptcp_crypto_hmac_sha(key1, key2, msg, 19, hmac); 595 596 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]); 597 } 598 #endif 599 600 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb, 601 unsigned int *size, 602 unsigned int remaining, 603 struct mptcp_out_options *opts) 604 { 605 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 606 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 607 bool drop_other_suboptions = false; 608 unsigned int opt_size = *size; 609 struct mptcp_addr_info saddr; 610 bool echo; 611 bool port; 612 int len; 613 614 if ((mptcp_pm_should_add_signal_ipv6(msk) || 615 mptcp_pm_should_add_signal_port(msk)) && 616 skb && skb_is_tcp_pure_ack(skb)) { 617 pr_debug("drop other suboptions"); 618 opts->suboptions = 0; 619 opts->ext_copy.use_ack = 0; 620 opts->ext_copy.use_map = 0; 621 remaining += opt_size; 622 drop_other_suboptions = true; 623 } 624 625 if (!mptcp_pm_should_add_signal(msk) || 626 !(mptcp_pm_add_addr_signal(msk, remaining, &saddr, &echo, &port))) 627 return false; 628 629 len = mptcp_add_addr_len(saddr.family, echo, port); 630 if (remaining < len) 631 return false; 632 633 *size = len; 634 if (drop_other_suboptions) 635 *size -= opt_size; 636 opts->addr_id = saddr.id; 637 if (port) 638 opts->port = ntohs(saddr.port); 639 if (saddr.family == AF_INET) { 640 opts->suboptions |= OPTION_MPTCP_ADD_ADDR; 641 opts->addr = saddr.addr; 642 if (!echo) { 643 opts->ahmac = add_addr_generate_hmac(msk->local_key, 644 msk->remote_key, 645 opts->addr_id, 646 &opts->addr); 647 } 648 } 649 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 650 else if (saddr.family == AF_INET6) { 651 opts->suboptions |= OPTION_MPTCP_ADD_ADDR6; 652 opts->addr6 = saddr.addr6; 653 if (!echo) { 654 opts->ahmac = add_addr6_generate_hmac(msk->local_key, 655 msk->remote_key, 656 opts->addr_id, 657 &opts->addr6); 658 } 659 } 660 #endif 661 pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d", 662 opts->addr_id, opts->ahmac, echo, opts->port); 663 664 return true; 665 } 666 667 static bool mptcp_established_options_rm_addr(struct sock *sk, 668 unsigned int *size, 669 unsigned int remaining, 670 struct mptcp_out_options *opts) 671 { 672 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 673 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 674 u8 rm_id; 675 676 if (!mptcp_pm_should_rm_signal(msk) || 677 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_id))) 678 return false; 679 680 if (remaining < TCPOLEN_MPTCP_RM_ADDR_BASE) 681 return false; 682 683 *size = TCPOLEN_MPTCP_RM_ADDR_BASE; 684 opts->suboptions |= OPTION_MPTCP_RM_ADDR; 685 opts->rm_id = rm_id; 686 687 pr_debug("rm_id=%d", opts->rm_id); 688 689 return true; 690 } 691 692 static bool mptcp_established_options_mp_prio(struct sock *sk, 693 unsigned int *size, 694 unsigned int remaining, 695 struct mptcp_out_options *opts) 696 { 697 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 698 699 if (!subflow->send_mp_prio) 700 return false; 701 702 if (remaining < TCPOLEN_MPTCP_PRIO) 703 return false; 704 705 *size = TCPOLEN_MPTCP_PRIO; 706 opts->suboptions |= OPTION_MPTCP_PRIO; 707 opts->backup = subflow->request_bkup; 708 709 pr_debug("prio=%d", opts->backup); 710 711 return true; 712 } 713 714 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb, 715 unsigned int *size, unsigned int remaining, 716 struct mptcp_out_options *opts) 717 { 718 unsigned int opt_size = 0; 719 bool ret = false; 720 721 opts->suboptions = 0; 722 723 if (unlikely(mptcp_check_fallback(sk))) 724 return false; 725 726 /* prevent adding of any MPTCP related options on reset packet 727 * until we support MP_TCPRST/MP_FASTCLOSE 728 */ 729 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) 730 return false; 731 732 if (mptcp_established_options_mp(sk, skb, &opt_size, remaining, opts)) 733 ret = true; 734 else if (mptcp_established_options_dss(sk, skb, &opt_size, remaining, 735 opts)) 736 ret = true; 737 738 /* we reserved enough space for the above options, and exceeding the 739 * TCP option space would be fatal 740 */ 741 if (WARN_ON_ONCE(opt_size > remaining)) 742 return false; 743 744 *size += opt_size; 745 remaining -= opt_size; 746 if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) { 747 *size += opt_size; 748 remaining -= opt_size; 749 ret = true; 750 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) { 751 *size += opt_size; 752 remaining -= opt_size; 753 ret = true; 754 } 755 756 if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) { 757 *size += opt_size; 758 remaining -= opt_size; 759 ret = true; 760 } 761 762 return ret; 763 } 764 765 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size, 766 struct mptcp_out_options *opts) 767 { 768 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 769 770 if (subflow_req->mp_capable) { 771 opts->suboptions = OPTION_MPTCP_MPC_SYNACK; 772 opts->sndr_key = subflow_req->local_key; 773 *size = TCPOLEN_MPTCP_MPC_SYNACK; 774 pr_debug("subflow_req=%p, local_key=%llu", 775 subflow_req, subflow_req->local_key); 776 return true; 777 } else if (subflow_req->mp_join) { 778 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK; 779 opts->backup = subflow_req->backup; 780 opts->join_id = subflow_req->local_id; 781 opts->thmac = subflow_req->thmac; 782 opts->nonce = subflow_req->local_nonce; 783 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u", 784 subflow_req, opts->backup, opts->join_id, 785 opts->thmac, opts->nonce); 786 *size = TCPOLEN_MPTCP_MPJ_SYNACK; 787 return true; 788 } 789 return false; 790 } 791 792 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk, 793 struct mptcp_subflow_context *subflow, 794 struct sk_buff *skb, 795 struct mptcp_options_received *mp_opt) 796 { 797 /* here we can process OoO, in-window pkts, only in-sequence 4th ack 798 * will make the subflow fully established 799 */ 800 if (likely(subflow->fully_established)) { 801 /* on passive sockets, check for 3rd ack retransmission 802 * note that msk is always set by subflow_syn_recv_sock() 803 * for mp_join subflows 804 */ 805 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 && 806 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq && 807 subflow->mp_join && mp_opt->mp_join && 808 READ_ONCE(msk->pm.server_side)) 809 tcp_send_ack(ssk); 810 goto fully_established; 811 } 812 813 /* we must process OoO packets before the first subflow is fully 814 * established. OoO packets are instead a protocol violation 815 * for MP_JOIN subflows as the peer must not send any data 816 * before receiving the forth ack - cfr. RFC 8684 section 3.2. 817 */ 818 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) { 819 if (subflow->mp_join) 820 goto reset; 821 return subflow->mp_capable; 822 } 823 824 if (mp_opt->dss && mp_opt->use_ack) { 825 /* subflows are fully established as soon as we get any 826 * additional ack. 827 */ 828 subflow->fully_established = 1; 829 WRITE_ONCE(msk->fully_established, true); 830 goto fully_established; 831 } 832 833 if (mp_opt->add_addr) { 834 WRITE_ONCE(msk->fully_established, true); 835 return true; 836 } 837 838 /* If the first established packet does not contain MP_CAPABLE + data 839 * then fallback to TCP. Fallback scenarios requires a reset for 840 * MP_JOIN subflows. 841 */ 842 if (!mp_opt->mp_capable) { 843 if (subflow->mp_join) 844 goto reset; 845 subflow->mp_capable = 0; 846 pr_fallback(msk); 847 __mptcp_do_fallback(msk); 848 return false; 849 } 850 851 if (unlikely(!READ_ONCE(msk->pm.server_side))) 852 pr_warn_once("bogus mpc option on established client sk"); 853 mptcp_subflow_fully_established(subflow, mp_opt); 854 855 fully_established: 856 /* if the subflow is not already linked into the conn_list, we can't 857 * notify the PM: this subflow is still on the listener queue 858 * and the PM possibly acquiring the subflow lock could race with 859 * the listener close 860 */ 861 if (likely(subflow->pm_notified) || list_empty(&subflow->node)) 862 return true; 863 864 subflow->pm_notified = 1; 865 if (subflow->mp_join) { 866 clear_3rdack_retransmission(ssk); 867 mptcp_pm_subflow_established(msk, subflow); 868 } else { 869 mptcp_pm_fully_established(msk); 870 } 871 return true; 872 873 reset: 874 mptcp_subflow_reset(ssk); 875 return false; 876 } 877 878 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit) 879 { 880 u32 old_ack32, cur_ack32; 881 882 if (use_64bit) 883 return cur_ack; 884 885 old_ack32 = (u32)old_ack; 886 cur_ack32 = (u32)cur_ack; 887 cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32; 888 if (unlikely(before(cur_ack32, old_ack32))) 889 return cur_ack + (1LL << 32); 890 return cur_ack; 891 } 892 893 static void ack_update_msk(struct mptcp_sock *msk, 894 struct sock *ssk, 895 struct mptcp_options_received *mp_opt) 896 { 897 u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt); 898 struct sock *sk = (struct sock *)msk; 899 u64 old_snd_una; 900 901 mptcp_data_lock(sk); 902 903 /* avoid ack expansion on update conflict, to reduce the risk of 904 * wrongly expanding to a future ack sequence number, which is way 905 * more dangerous than missing an ack 906 */ 907 old_snd_una = msk->snd_una; 908 new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64); 909 910 /* ACK for data not even sent yet? Ignore. */ 911 if (after64(new_snd_una, snd_nxt)) 912 new_snd_una = old_snd_una; 913 914 new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd; 915 916 if (after64(new_wnd_end, msk->wnd_end)) 917 msk->wnd_end = new_wnd_end; 918 919 /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */ 920 if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)) && 921 sk_stream_memory_free(ssk)) 922 __mptcp_check_push(sk, ssk); 923 924 if (after64(new_snd_una, old_snd_una)) { 925 msk->snd_una = new_snd_una; 926 __mptcp_data_acked(sk); 927 } 928 mptcp_data_unlock(sk); 929 } 930 931 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit) 932 { 933 /* Skip if DATA_FIN was already received. 934 * If updating simultaneously with the recvmsg loop, values 935 * should match. If they mismatch, the peer is misbehaving and 936 * we will prefer the most recent information. 937 */ 938 if (READ_ONCE(msk->rcv_data_fin) || !READ_ONCE(msk->first)) 939 return false; 940 941 WRITE_ONCE(msk->rcv_data_fin_seq, 942 expand_ack(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit)); 943 WRITE_ONCE(msk->rcv_data_fin, 1); 944 945 return true; 946 } 947 948 static bool add_addr_hmac_valid(struct mptcp_sock *msk, 949 struct mptcp_options_received *mp_opt) 950 { 951 u64 hmac = 0; 952 953 if (mp_opt->echo) 954 return true; 955 956 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4) 957 hmac = add_addr_generate_hmac(msk->remote_key, 958 msk->local_key, 959 mp_opt->addr_id, &mp_opt->addr); 960 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 961 else 962 hmac = add_addr6_generate_hmac(msk->remote_key, 963 msk->local_key, 964 mp_opt->addr_id, &mp_opt->addr6); 965 #endif 966 967 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n", 968 msk, (unsigned long long)hmac, 969 (unsigned long long)mp_opt->ahmac); 970 971 return hmac == mp_opt->ahmac; 972 } 973 974 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb) 975 { 976 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 977 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 978 struct mptcp_options_received mp_opt; 979 struct mptcp_ext *mpext; 980 981 if (__mptcp_check_fallback(msk)) { 982 /* Keep it simple and unconditionally trigger send data cleanup and 983 * pending queue spooling. We will need to acquire the data lock 984 * for more accurate checks, and once the lock is acquired, such 985 * helpers are cheap. 986 */ 987 mptcp_data_lock(subflow->conn); 988 if (sk_stream_memory_free(sk)) 989 __mptcp_check_push(subflow->conn, sk); 990 __mptcp_data_acked(subflow->conn); 991 mptcp_data_unlock(subflow->conn); 992 return; 993 } 994 995 mptcp_get_options(skb, &mp_opt); 996 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt)) 997 return; 998 999 if (mp_opt.fastclose && 1000 msk->local_key == mp_opt.rcvr_key) { 1001 WRITE_ONCE(msk->rcv_fastclose, true); 1002 mptcp_schedule_work((struct sock *)msk); 1003 } 1004 1005 if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) { 1006 struct mptcp_addr_info addr; 1007 1008 addr.port = htons(mp_opt.port); 1009 addr.id = mp_opt.addr_id; 1010 if (mp_opt.family == MPTCP_ADDR_IPVERSION_4) { 1011 addr.family = AF_INET; 1012 addr.addr = mp_opt.addr; 1013 } 1014 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1015 else if (mp_opt.family == MPTCP_ADDR_IPVERSION_6) { 1016 addr.family = AF_INET6; 1017 addr.addr6 = mp_opt.addr6; 1018 } 1019 #endif 1020 if (!mp_opt.echo) { 1021 mptcp_pm_add_addr_received(msk, &addr); 1022 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR); 1023 } else { 1024 mptcp_pm_del_add_timer(msk, &addr); 1025 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD); 1026 } 1027 mp_opt.add_addr = 0; 1028 } 1029 1030 if (mp_opt.rm_addr) { 1031 mptcp_pm_rm_addr_received(msk, mp_opt.rm_id); 1032 mp_opt.rm_addr = 0; 1033 } 1034 1035 if (mp_opt.mp_prio) { 1036 mptcp_pm_mp_prio_received(sk, mp_opt.backup); 1037 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX); 1038 mp_opt.mp_prio = 0; 1039 } 1040 1041 if (!mp_opt.dss) 1042 return; 1043 1044 /* we can't wait for recvmsg() to update the ack_seq, otherwise 1045 * monodirectional flows will stuck 1046 */ 1047 if (mp_opt.use_ack) 1048 ack_update_msk(msk, sk, &mp_opt); 1049 1050 /* Zero-data-length packets are dropped by the caller and not 1051 * propagated to the MPTCP layer, so the skb extension does not 1052 * need to be allocated or populated. DATA_FIN information, if 1053 * present, needs to be updated here before the skb is freed. 1054 */ 1055 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) { 1056 if (mp_opt.data_fin && mp_opt.data_len == 1 && 1057 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) && 1058 schedule_work(&msk->work)) 1059 sock_hold(subflow->conn); 1060 1061 return; 1062 } 1063 1064 mpext = skb_ext_add(skb, SKB_EXT_MPTCP); 1065 if (!mpext) 1066 return; 1067 1068 memset(mpext, 0, sizeof(*mpext)); 1069 1070 if (mp_opt.use_map) { 1071 if (mp_opt.mpc_map) { 1072 /* this is an MP_CAPABLE carrying MPTCP data 1073 * we know this map the first chunk of data 1074 */ 1075 mptcp_crypto_key_sha(subflow->remote_key, NULL, 1076 &mpext->data_seq); 1077 mpext->data_seq++; 1078 mpext->subflow_seq = 1; 1079 mpext->dsn64 = 1; 1080 mpext->mpc_map = 1; 1081 mpext->data_fin = 0; 1082 } else { 1083 mpext->data_seq = mp_opt.data_seq; 1084 mpext->subflow_seq = mp_opt.subflow_seq; 1085 mpext->dsn64 = mp_opt.dsn64; 1086 mpext->data_fin = mp_opt.data_fin; 1087 } 1088 mpext->data_len = mp_opt.data_len; 1089 mpext->use_map = 1; 1090 } 1091 } 1092 1093 static void mptcp_set_rwin(const struct tcp_sock *tp) 1094 { 1095 const struct sock *ssk = (const struct sock *)tp; 1096 const struct mptcp_subflow_context *subflow; 1097 struct mptcp_sock *msk; 1098 u64 ack_seq; 1099 1100 subflow = mptcp_subflow_ctx(ssk); 1101 msk = mptcp_sk(subflow->conn); 1102 1103 ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd; 1104 1105 if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent))) 1106 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq); 1107 } 1108 1109 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp, 1110 struct mptcp_out_options *opts) 1111 { 1112 if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK | 1113 OPTION_MPTCP_MPC_ACK) & opts->suboptions) { 1114 u8 len; 1115 1116 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) 1117 len = TCPOLEN_MPTCP_MPC_SYN; 1118 else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) 1119 len = TCPOLEN_MPTCP_MPC_SYNACK; 1120 else if (opts->ext_copy.data_len) 1121 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 1122 else 1123 len = TCPOLEN_MPTCP_MPC_ACK; 1124 1125 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len, 1126 MPTCP_SUPPORTED_VERSION, 1127 MPTCP_CAP_HMAC_SHA256); 1128 1129 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) & 1130 opts->suboptions)) 1131 goto mp_capable_done; 1132 1133 put_unaligned_be64(opts->sndr_key, ptr); 1134 ptr += 2; 1135 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions)) 1136 goto mp_capable_done; 1137 1138 put_unaligned_be64(opts->rcvr_key, ptr); 1139 ptr += 2; 1140 if (!opts->ext_copy.data_len) 1141 goto mp_capable_done; 1142 1143 put_unaligned_be32(opts->ext_copy.data_len << 16 | 1144 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1145 ptr += 1; 1146 } 1147 1148 mp_capable_done: 1149 if ((OPTION_MPTCP_ADD_ADDR 1150 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1151 | OPTION_MPTCP_ADD_ADDR6 1152 #endif 1153 ) & opts->suboptions) { 1154 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE; 1155 u8 echo = MPTCP_ADDR_ECHO; 1156 1157 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1158 if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions) 1159 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE; 1160 #endif 1161 1162 if (opts->port) 1163 len += TCPOLEN_MPTCP_PORT_LEN; 1164 1165 if (opts->ahmac) { 1166 len += sizeof(opts->ahmac); 1167 echo = 0; 1168 } 1169 1170 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 1171 len, echo, opts->addr_id); 1172 if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) { 1173 memcpy((u8 *)ptr, (u8 *)&opts->addr.s_addr, 4); 1174 ptr += 1; 1175 } 1176 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1177 else if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions) { 1178 memcpy((u8 *)ptr, opts->addr6.s6_addr, 16); 1179 ptr += 4; 1180 } 1181 #endif 1182 1183 if (!opts->port) { 1184 if (opts->ahmac) { 1185 put_unaligned_be64(opts->ahmac, ptr); 1186 ptr += 2; 1187 } 1188 } else { 1189 if (opts->ahmac) { 1190 u8 *bptr = (u8 *)ptr; 1191 1192 put_unaligned_be16(opts->port, bptr); 1193 bptr += 2; 1194 put_unaligned_be64(opts->ahmac, bptr); 1195 bptr += 8; 1196 put_unaligned_be16(TCPOPT_NOP << 8 | 1197 TCPOPT_NOP, bptr); 1198 1199 ptr += 3; 1200 } else { 1201 put_unaligned_be32(opts->port << 16 | 1202 TCPOPT_NOP << 8 | 1203 TCPOPT_NOP, ptr); 1204 ptr += 1; 1205 } 1206 } 1207 } 1208 1209 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) { 1210 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR, 1211 TCPOLEN_MPTCP_RM_ADDR_BASE, 1212 0, opts->rm_id); 1213 } 1214 1215 if (OPTION_MPTCP_PRIO & opts->suboptions) { 1216 const struct sock *ssk = (const struct sock *)tp; 1217 struct mptcp_subflow_context *subflow; 1218 1219 subflow = mptcp_subflow_ctx(ssk); 1220 subflow->send_mp_prio = 0; 1221 1222 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO, 1223 TCPOLEN_MPTCP_PRIO, 1224 opts->backup, TCPOPT_NOP); 1225 } 1226 1227 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) { 1228 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1229 TCPOLEN_MPTCP_MPJ_SYN, 1230 opts->backup, opts->join_id); 1231 put_unaligned_be32(opts->token, ptr); 1232 ptr += 1; 1233 put_unaligned_be32(opts->nonce, ptr); 1234 ptr += 1; 1235 } 1236 1237 if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) { 1238 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1239 TCPOLEN_MPTCP_MPJ_SYNACK, 1240 opts->backup, opts->join_id); 1241 put_unaligned_be64(opts->thmac, ptr); 1242 ptr += 2; 1243 put_unaligned_be32(opts->nonce, ptr); 1244 ptr += 1; 1245 } 1246 1247 if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) { 1248 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1249 TCPOLEN_MPTCP_MPJ_ACK, 0, 0); 1250 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN); 1251 ptr += 5; 1252 } 1253 1254 if (opts->ext_copy.use_ack || opts->ext_copy.use_map) { 1255 struct mptcp_ext *mpext = &opts->ext_copy; 1256 u8 len = TCPOLEN_MPTCP_DSS_BASE; 1257 u8 flags = 0; 1258 1259 if (mpext->use_ack) { 1260 flags = MPTCP_DSS_HAS_ACK; 1261 if (mpext->ack64) { 1262 len += TCPOLEN_MPTCP_DSS_ACK64; 1263 flags |= MPTCP_DSS_ACK64; 1264 } else { 1265 len += TCPOLEN_MPTCP_DSS_ACK32; 1266 } 1267 } 1268 1269 if (mpext->use_map) { 1270 len += TCPOLEN_MPTCP_DSS_MAP64; 1271 1272 /* Use only 64-bit mapping flags for now, add 1273 * support for optional 32-bit mappings later. 1274 */ 1275 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64; 1276 if (mpext->data_fin) 1277 flags |= MPTCP_DSS_DATA_FIN; 1278 } 1279 1280 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags); 1281 1282 if (mpext->use_ack) { 1283 if (mpext->ack64) { 1284 put_unaligned_be64(mpext->data_ack, ptr); 1285 ptr += 2; 1286 } else { 1287 put_unaligned_be32(mpext->data_ack32, ptr); 1288 ptr += 1; 1289 } 1290 } 1291 1292 if (mpext->use_map) { 1293 put_unaligned_be64(mpext->data_seq, ptr); 1294 ptr += 2; 1295 put_unaligned_be32(mpext->subflow_seq, ptr); 1296 ptr += 1; 1297 put_unaligned_be32(mpext->data_len << 16 | 1298 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1299 } 1300 } 1301 1302 if (tp) 1303 mptcp_set_rwin(tp); 1304 } 1305