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 u8 i; 30 31 switch (subtype) { 32 case MPTCPOPT_MP_CAPABLE: 33 /* strict size checking */ 34 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { 35 if (skb->len > tcp_hdr(skb)->doff << 2) 36 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA; 37 else 38 expected_opsize = TCPOLEN_MPTCP_MPC_ACK; 39 } else { 40 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK) 41 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK; 42 else 43 expected_opsize = TCPOLEN_MPTCP_MPC_SYN; 44 } 45 if (opsize != expected_opsize) 46 break; 47 48 /* try to be gentle vs future versions on the initial syn */ 49 version = *ptr++ & MPTCP_VERSION_MASK; 50 if (opsize != TCPOLEN_MPTCP_MPC_SYN) { 51 if (version != MPTCP_SUPPORTED_VERSION) 52 break; 53 } else if (version < MPTCP_SUPPORTED_VERSION) { 54 break; 55 } 56 57 flags = *ptr++; 58 if (!mptcp_cap_flag_sha256(flags) || 59 (flags & MPTCP_CAP_EXTENSIBILITY)) 60 break; 61 62 /* RFC 6824, Section 3.1: 63 * "For the Checksum Required bit (labeled "A"), if either 64 * host requires the use of checksums, checksums MUST be used. 65 * In other words, the only way for checksums not to be used 66 * is if both hosts in their SYNs set A=0." 67 * 68 * Section 3.3.0: 69 * "If a checksum is not present when its use has been 70 * negotiated, the receiver MUST close the subflow with a RST as 71 * it is considered broken." 72 * 73 * We don't implement DSS checksum - fall back to TCP. 74 */ 75 if (flags & MPTCP_CAP_CHECKSUM_REQD) 76 break; 77 78 mp_opt->mp_capable = 1; 79 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) { 80 mp_opt->sndr_key = get_unaligned_be64(ptr); 81 ptr += 8; 82 } 83 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) { 84 mp_opt->rcvr_key = get_unaligned_be64(ptr); 85 ptr += 8; 86 } 87 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA) { 88 /* Section 3.1.: 89 * "the data parameters in a MP_CAPABLE are semantically 90 * equivalent to those in a DSS option and can be used 91 * interchangeably." 92 */ 93 mp_opt->dss = 1; 94 mp_opt->use_map = 1; 95 mp_opt->mpc_map = 1; 96 mp_opt->data_len = get_unaligned_be16(ptr); 97 ptr += 2; 98 } 99 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d", 100 version, flags, opsize, mp_opt->sndr_key, 101 mp_opt->rcvr_key, mp_opt->data_len); 102 break; 103 104 case MPTCPOPT_MP_JOIN: 105 mp_opt->mp_join = 1; 106 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) { 107 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 108 mp_opt->join_id = *ptr++; 109 mp_opt->token = get_unaligned_be32(ptr); 110 ptr += 4; 111 mp_opt->nonce = get_unaligned_be32(ptr); 112 ptr += 4; 113 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u", 114 mp_opt->backup, mp_opt->join_id, 115 mp_opt->token, mp_opt->nonce); 116 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) { 117 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 118 mp_opt->join_id = *ptr++; 119 mp_opt->thmac = get_unaligned_be64(ptr); 120 ptr += 8; 121 mp_opt->nonce = get_unaligned_be32(ptr); 122 ptr += 4; 123 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u", 124 mp_opt->backup, mp_opt->join_id, 125 mp_opt->thmac, mp_opt->nonce); 126 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) { 127 ptr += 2; 128 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN); 129 pr_debug("MP_JOIN hmac"); 130 } else { 131 pr_warn("MP_JOIN bad option size"); 132 mp_opt->mp_join = 0; 133 } 134 break; 135 136 case MPTCPOPT_DSS: 137 pr_debug("DSS"); 138 ptr++; 139 140 /* we must clear 'mpc_map' be able to detect MP_CAPABLE 141 * map vs DSS map in mptcp_incoming_options(), and reconstruct 142 * map info accordingly 143 */ 144 mp_opt->mpc_map = 0; 145 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK; 146 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0; 147 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0; 148 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0; 149 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0; 150 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK); 151 152 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d", 153 mp_opt->data_fin, mp_opt->dsn64, 154 mp_opt->use_map, mp_opt->ack64, 155 mp_opt->use_ack); 156 157 expected_opsize = TCPOLEN_MPTCP_DSS_BASE; 158 159 if (mp_opt->use_ack) { 160 if (mp_opt->ack64) 161 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64; 162 else 163 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32; 164 } 165 166 if (mp_opt->use_map) { 167 if (mp_opt->dsn64) 168 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64; 169 else 170 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32; 171 } 172 173 /* RFC 6824, Section 3.3: 174 * If a checksum is present, but its use had 175 * not been negotiated in the MP_CAPABLE handshake, 176 * the checksum field MUST be ignored. 177 */ 178 if (opsize != expected_opsize && 179 opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) 180 break; 181 182 mp_opt->dss = 1; 183 184 if (mp_opt->use_ack) { 185 if (mp_opt->ack64) { 186 mp_opt->data_ack = get_unaligned_be64(ptr); 187 ptr += 8; 188 } else { 189 mp_opt->data_ack = get_unaligned_be32(ptr); 190 ptr += 4; 191 } 192 193 pr_debug("data_ack=%llu", mp_opt->data_ack); 194 } 195 196 if (mp_opt->use_map) { 197 if (mp_opt->dsn64) { 198 mp_opt->data_seq = get_unaligned_be64(ptr); 199 ptr += 8; 200 } else { 201 mp_opt->data_seq = get_unaligned_be32(ptr); 202 ptr += 4; 203 } 204 205 mp_opt->subflow_seq = get_unaligned_be32(ptr); 206 ptr += 4; 207 208 mp_opt->data_len = get_unaligned_be16(ptr); 209 ptr += 2; 210 211 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u", 212 mp_opt->data_seq, mp_opt->subflow_seq, 213 mp_opt->data_len); 214 } 215 216 break; 217 218 case MPTCPOPT_ADD_ADDR: 219 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO; 220 if (!mp_opt->echo) { 221 if (opsize == TCPOLEN_MPTCP_ADD_ADDR || 222 opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT) 223 mp_opt->family = MPTCP_ADDR_IPVERSION_4; 224 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 225 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 || 226 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT) 227 mp_opt->family = MPTCP_ADDR_IPVERSION_6; 228 #endif 229 else 230 break; 231 } else { 232 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE || 233 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) 234 mp_opt->family = MPTCP_ADDR_IPVERSION_4; 235 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 236 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE || 237 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) 238 mp_opt->family = MPTCP_ADDR_IPVERSION_6; 239 #endif 240 else 241 break; 242 } 243 244 mp_opt->add_addr = 1; 245 mp_opt->addr_id = *ptr++; 246 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4) { 247 memcpy((u8 *)&mp_opt->addr.s_addr, (u8 *)ptr, 4); 248 ptr += 4; 249 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT || 250 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) { 251 mp_opt->port = get_unaligned_be16(ptr); 252 ptr += 2; 253 } 254 } 255 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 256 else { 257 memcpy(mp_opt->addr6.s6_addr, (u8 *)ptr, 16); 258 ptr += 16; 259 if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT || 260 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) { 261 mp_opt->port = get_unaligned_be16(ptr); 262 ptr += 2; 263 } 264 } 265 #endif 266 if (!mp_opt->echo) { 267 mp_opt->ahmac = get_unaligned_be64(ptr); 268 ptr += 8; 269 } 270 pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d", 271 (mp_opt->family == MPTCP_ADDR_IPVERSION_6) ? "6" : "", 272 mp_opt->addr_id, mp_opt->ahmac, mp_opt->echo, mp_opt->port); 273 break; 274 275 case MPTCPOPT_RM_ADDR: 276 if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 || 277 opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX) 278 break; 279 280 ptr++; 281 282 mp_opt->rm_addr = 1; 283 mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE; 284 for (i = 0; i < mp_opt->rm_list.nr; i++) 285 mp_opt->rm_list.ids[i] = *ptr++; 286 pr_debug("RM_ADDR: rm_list_nr=%d", mp_opt->rm_list.nr); 287 break; 288 289 case MPTCPOPT_MP_PRIO: 290 if (opsize != TCPOLEN_MPTCP_PRIO) 291 break; 292 293 mp_opt->mp_prio = 1; 294 mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP; 295 pr_debug("MP_PRIO: prio=%d", mp_opt->backup); 296 break; 297 298 case MPTCPOPT_MP_FASTCLOSE: 299 if (opsize != TCPOLEN_MPTCP_FASTCLOSE) 300 break; 301 302 ptr += 2; 303 mp_opt->rcvr_key = get_unaligned_be64(ptr); 304 ptr += 8; 305 mp_opt->fastclose = 1; 306 break; 307 308 case MPTCPOPT_RST: 309 if (opsize != TCPOLEN_MPTCP_RST) 310 break; 311 312 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) 313 break; 314 mp_opt->reset = 1; 315 flags = *ptr++; 316 mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT; 317 mp_opt->reset_reason = *ptr; 318 break; 319 320 default: 321 break; 322 } 323 } 324 325 void mptcp_get_options(const struct sk_buff *skb, 326 struct mptcp_options_received *mp_opt) 327 { 328 const struct tcphdr *th = tcp_hdr(skb); 329 const unsigned char *ptr; 330 int length; 331 332 /* initialize option status */ 333 mp_opt->mp_capable = 0; 334 mp_opt->mp_join = 0; 335 mp_opt->add_addr = 0; 336 mp_opt->ahmac = 0; 337 mp_opt->fastclose = 0; 338 mp_opt->port = 0; 339 mp_opt->rm_addr = 0; 340 mp_opt->dss = 0; 341 mp_opt->mp_prio = 0; 342 mp_opt->reset = 0; 343 344 length = (th->doff * 4) - sizeof(struct tcphdr); 345 ptr = (const unsigned char *)(th + 1); 346 347 while (length > 0) { 348 int opcode = *ptr++; 349 int opsize; 350 351 switch (opcode) { 352 case TCPOPT_EOL: 353 return; 354 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */ 355 length--; 356 continue; 357 default: 358 opsize = *ptr++; 359 if (opsize < 2) /* "silly options" */ 360 return; 361 if (opsize > length) 362 return; /* don't parse partial options */ 363 if (opcode == TCPOPT_MPTCP) 364 mptcp_parse_option(skb, ptr, opsize, mp_opt); 365 ptr += opsize - 2; 366 length -= opsize; 367 } 368 } 369 } 370 371 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb, 372 unsigned int *size, struct mptcp_out_options *opts) 373 { 374 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 375 376 /* we will use snd_isn to detect first pkt [re]transmission 377 * in mptcp_established_options_mp() 378 */ 379 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq; 380 if (subflow->request_mptcp) { 381 opts->suboptions = OPTION_MPTCP_MPC_SYN; 382 *size = TCPOLEN_MPTCP_MPC_SYN; 383 return true; 384 } else if (subflow->request_join) { 385 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token, 386 subflow->local_nonce); 387 opts->suboptions = OPTION_MPTCP_MPJ_SYN; 388 opts->join_id = subflow->local_id; 389 opts->token = subflow->remote_token; 390 opts->nonce = subflow->local_nonce; 391 opts->backup = subflow->request_bkup; 392 *size = TCPOLEN_MPTCP_MPJ_SYN; 393 return true; 394 } 395 return false; 396 } 397 398 /* MP_JOIN client subflow must wait for 4th ack before sending any data: 399 * TCP can't schedule delack timer before the subflow is fully established. 400 * MPTCP uses the delack timer to do 3rd ack retransmissions 401 */ 402 static void schedule_3rdack_retransmission(struct sock *sk) 403 { 404 struct inet_connection_sock *icsk = inet_csk(sk); 405 struct tcp_sock *tp = tcp_sk(sk); 406 unsigned long timeout; 407 408 /* reschedule with a timeout above RTT, as we must look only for drop */ 409 if (tp->srtt_us) 410 timeout = tp->srtt_us << 1; 411 else 412 timeout = TCP_TIMEOUT_INIT; 413 414 WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER); 415 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER; 416 icsk->icsk_ack.timeout = timeout; 417 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout); 418 } 419 420 static void clear_3rdack_retransmission(struct sock *sk) 421 { 422 struct inet_connection_sock *icsk = inet_csk(sk); 423 424 sk_stop_timer(sk, &icsk->icsk_delack_timer); 425 icsk->icsk_ack.timeout = 0; 426 icsk->icsk_ack.ato = 0; 427 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER); 428 } 429 430 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb, 431 bool snd_data_fin_enable, 432 unsigned int *size, 433 unsigned int remaining, 434 struct mptcp_out_options *opts) 435 { 436 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 437 struct mptcp_ext *mpext; 438 unsigned int data_len; 439 440 /* When skb is not available, we better over-estimate the emitted 441 * options len. A full DSS option (28 bytes) is longer than 442 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so 443 * tell the caller to defer the estimate to 444 * mptcp_established_options_dss(), which will reserve enough space. 445 */ 446 if (!skb) 447 return false; 448 449 /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */ 450 if (subflow->fully_established || snd_data_fin_enable || 451 subflow->snd_isn != TCP_SKB_CB(skb)->seq || 452 sk->sk_state != TCP_ESTABLISHED) 453 return false; 454 455 if (subflow->mp_capable) { 456 mpext = mptcp_get_ext(skb); 457 data_len = mpext ? mpext->data_len : 0; 458 459 /* we will check ext_copy.data_len in mptcp_write_options() to 460 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and 461 * TCPOLEN_MPTCP_MPC_ACK 462 */ 463 opts->ext_copy.data_len = data_len; 464 opts->suboptions = OPTION_MPTCP_MPC_ACK; 465 opts->sndr_key = subflow->local_key; 466 opts->rcvr_key = subflow->remote_key; 467 468 /* Section 3.1. 469 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK 470 * packets that start the first subflow of an MPTCP connection, 471 * as well as the first packet that carries data 472 */ 473 if (data_len > 0) 474 *size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4); 475 else 476 *size = TCPOLEN_MPTCP_MPC_ACK; 477 478 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d", 479 subflow, subflow->local_key, subflow->remote_key, 480 data_len); 481 482 return true; 483 } else if (subflow->mp_join) { 484 opts->suboptions = OPTION_MPTCP_MPJ_ACK; 485 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN); 486 *size = TCPOLEN_MPTCP_MPJ_ACK; 487 pr_debug("subflow=%p", subflow); 488 489 schedule_3rdack_retransmission(sk); 490 return true; 491 } 492 return false; 493 } 494 495 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow, 496 struct sk_buff *skb, struct mptcp_ext *ext) 497 { 498 /* The write_seq value has already been incremented, so the actual 499 * sequence number for the DATA_FIN is one less. 500 */ 501 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1; 502 503 if (!ext->use_map || !skb->len) { 504 /* RFC6824 requires a DSS mapping with specific values 505 * if DATA_FIN is set but no data payload is mapped 506 */ 507 ext->data_fin = 1; 508 ext->use_map = 1; 509 ext->dsn64 = 1; 510 ext->data_seq = data_fin_tx_seq; 511 ext->subflow_seq = 0; 512 ext->data_len = 1; 513 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) { 514 /* If there's an existing DSS mapping and it is the 515 * final mapping, DATA_FIN consumes 1 additional byte of 516 * mapping space. 517 */ 518 ext->data_fin = 1; 519 ext->data_len++; 520 } 521 } 522 523 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb, 524 bool snd_data_fin_enable, 525 unsigned int *size, 526 unsigned int remaining, 527 struct mptcp_out_options *opts) 528 { 529 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 530 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 531 unsigned int dss_size = 0; 532 struct mptcp_ext *mpext; 533 unsigned int ack_size; 534 bool ret = false; 535 u64 ack_seq; 536 537 mpext = skb ? mptcp_get_ext(skb) : NULL; 538 539 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) { 540 unsigned int map_size; 541 542 map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64; 543 544 remaining -= map_size; 545 dss_size = map_size; 546 if (mpext) 547 opts->ext_copy = *mpext; 548 549 if (skb && snd_data_fin_enable) 550 mptcp_write_data_fin(subflow, skb, &opts->ext_copy); 551 ret = true; 552 } 553 554 /* passive sockets msk will set the 'can_ack' after accept(), even 555 * if the first subflow may have the already the remote key handy 556 */ 557 opts->ext_copy.use_ack = 0; 558 if (!READ_ONCE(msk->can_ack)) { 559 *size = ALIGN(dss_size, 4); 560 return ret; 561 } 562 563 ack_seq = READ_ONCE(msk->ack_seq); 564 if (READ_ONCE(msk->use_64bit_ack)) { 565 ack_size = TCPOLEN_MPTCP_DSS_ACK64; 566 opts->ext_copy.data_ack = ack_seq; 567 opts->ext_copy.ack64 = 1; 568 } else { 569 ack_size = TCPOLEN_MPTCP_DSS_ACK32; 570 opts->ext_copy.data_ack32 = (uint32_t)ack_seq; 571 opts->ext_copy.ack64 = 0; 572 } 573 opts->ext_copy.use_ack = 1; 574 WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk)); 575 576 /* Add kind/length/subtype/flag overhead if mapping is not populated */ 577 if (dss_size == 0) 578 ack_size += TCPOLEN_MPTCP_DSS_BASE; 579 580 dss_size += ack_size; 581 582 *size = ALIGN(dss_size, 4); 583 return true; 584 } 585 586 static u64 add_addr_generate_hmac(u64 key1, u64 key2, u8 addr_id, 587 struct in_addr *addr, u16 port) 588 { 589 u8 hmac[SHA256_DIGEST_SIZE]; 590 u8 msg[7]; 591 592 msg[0] = addr_id; 593 memcpy(&msg[1], &addr->s_addr, 4); 594 msg[5] = port >> 8; 595 msg[6] = port & 0xFF; 596 597 mptcp_crypto_hmac_sha(key1, key2, msg, 7, hmac); 598 599 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]); 600 } 601 602 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 603 static u64 add_addr6_generate_hmac(u64 key1, u64 key2, u8 addr_id, 604 struct in6_addr *addr, u16 port) 605 { 606 u8 hmac[SHA256_DIGEST_SIZE]; 607 u8 msg[19]; 608 609 msg[0] = addr_id; 610 memcpy(&msg[1], &addr->s6_addr, 16); 611 msg[17] = port >> 8; 612 msg[18] = port & 0xFF; 613 614 mptcp_crypto_hmac_sha(key1, key2, msg, 19, hmac); 615 616 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]); 617 } 618 #endif 619 620 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb, 621 unsigned int *size, 622 unsigned int remaining, 623 struct mptcp_out_options *opts) 624 { 625 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 626 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 627 bool drop_other_suboptions = false; 628 unsigned int opt_size = *size; 629 bool echo; 630 bool port; 631 int len; 632 633 if ((mptcp_pm_should_add_signal_ipv6(msk) || 634 mptcp_pm_should_add_signal_port(msk)) && 635 skb && skb_is_tcp_pure_ack(skb)) { 636 pr_debug("drop other suboptions"); 637 opts->suboptions = 0; 638 opts->ext_copy.use_ack = 0; 639 opts->ext_copy.use_map = 0; 640 remaining += opt_size; 641 drop_other_suboptions = true; 642 } 643 644 if (!mptcp_pm_should_add_signal(msk) || 645 !(mptcp_pm_add_addr_signal(msk, remaining, &opts->addr, &echo, &port))) 646 return false; 647 648 len = mptcp_add_addr_len(opts->addr.family, echo, port); 649 if (remaining < len) 650 return false; 651 652 *size = len; 653 if (drop_other_suboptions) 654 *size -= opt_size; 655 if (opts->addr.family == AF_INET) { 656 opts->suboptions |= OPTION_MPTCP_ADD_ADDR; 657 if (!echo) { 658 opts->ahmac = add_addr_generate_hmac(msk->local_key, 659 msk->remote_key, 660 opts->addr.id, 661 &opts->addr.addr, 662 ntohs(opts->addr.port)); 663 } 664 } 665 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 666 else if (opts->addr.family == AF_INET6) { 667 opts->suboptions |= OPTION_MPTCP_ADD_ADDR6; 668 if (!echo) { 669 opts->ahmac = add_addr6_generate_hmac(msk->local_key, 670 msk->remote_key, 671 opts->addr.id, 672 &opts->addr.addr6, 673 ntohs(opts->addr.port)); 674 } 675 } 676 #endif 677 pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d", 678 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port)); 679 680 return true; 681 } 682 683 static bool mptcp_established_options_rm_addr(struct sock *sk, 684 unsigned int *size, 685 unsigned int remaining, 686 struct mptcp_out_options *opts) 687 { 688 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 689 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 690 struct mptcp_rm_list rm_list; 691 int i, len; 692 693 if (!mptcp_pm_should_rm_signal(msk) || 694 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list))) 695 return false; 696 697 len = mptcp_rm_addr_len(&rm_list); 698 if (len < 0) 699 return false; 700 if (remaining < len) 701 return false; 702 703 *size = len; 704 opts->suboptions |= OPTION_MPTCP_RM_ADDR; 705 opts->rm_list = rm_list; 706 707 for (i = 0; i < opts->rm_list.nr; i++) 708 pr_debug("rm_list_ids[%d]=%d", i, opts->rm_list.ids[i]); 709 710 return true; 711 } 712 713 static bool mptcp_established_options_mp_prio(struct sock *sk, 714 unsigned int *size, 715 unsigned int remaining, 716 struct mptcp_out_options *opts) 717 { 718 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 719 720 if (!subflow->send_mp_prio) 721 return false; 722 723 /* account for the trailing 'nop' option */ 724 if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN) 725 return false; 726 727 *size = TCPOLEN_MPTCP_PRIO_ALIGN; 728 opts->suboptions |= OPTION_MPTCP_PRIO; 729 opts->backup = subflow->request_bkup; 730 731 pr_debug("prio=%d", opts->backup); 732 733 return true; 734 } 735 736 static noinline void mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb, 737 unsigned int *size, 738 unsigned int remaining, 739 struct mptcp_out_options *opts) 740 { 741 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 742 743 if (remaining < TCPOLEN_MPTCP_RST) 744 return; 745 746 *size = TCPOLEN_MPTCP_RST; 747 opts->suboptions |= OPTION_MPTCP_RST; 748 opts->reset_transient = subflow->reset_transient; 749 opts->reset_reason = subflow->reset_reason; 750 } 751 752 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb, 753 unsigned int *size, unsigned int remaining, 754 struct mptcp_out_options *opts) 755 { 756 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 757 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 758 unsigned int opt_size = 0; 759 bool snd_data_fin; 760 bool ret = false; 761 762 opts->suboptions = 0; 763 764 if (unlikely(__mptcp_check_fallback(msk))) 765 return false; 766 767 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) { 768 mptcp_established_options_rst(sk, skb, size, remaining, opts); 769 return true; 770 } 771 772 snd_data_fin = mptcp_data_fin_enabled(msk); 773 if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, remaining, opts)) 774 ret = true; 775 else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, remaining, opts)) 776 ret = true; 777 778 /* we reserved enough space for the above options, and exceeding the 779 * TCP option space would be fatal 780 */ 781 if (WARN_ON_ONCE(opt_size > remaining)) 782 return false; 783 784 *size += opt_size; 785 remaining -= opt_size; 786 if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) { 787 *size += opt_size; 788 remaining -= opt_size; 789 ret = true; 790 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) { 791 *size += opt_size; 792 remaining -= opt_size; 793 ret = true; 794 } 795 796 if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) { 797 *size += opt_size; 798 remaining -= opt_size; 799 ret = true; 800 } 801 802 return ret; 803 } 804 805 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size, 806 struct mptcp_out_options *opts) 807 { 808 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 809 810 if (subflow_req->mp_capable) { 811 opts->suboptions = OPTION_MPTCP_MPC_SYNACK; 812 opts->sndr_key = subflow_req->local_key; 813 *size = TCPOLEN_MPTCP_MPC_SYNACK; 814 pr_debug("subflow_req=%p, local_key=%llu", 815 subflow_req, subflow_req->local_key); 816 return true; 817 } else if (subflow_req->mp_join) { 818 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK; 819 opts->backup = subflow_req->backup; 820 opts->join_id = subflow_req->local_id; 821 opts->thmac = subflow_req->thmac; 822 opts->nonce = subflow_req->local_nonce; 823 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u", 824 subflow_req, opts->backup, opts->join_id, 825 opts->thmac, opts->nonce); 826 *size = TCPOLEN_MPTCP_MPJ_SYNACK; 827 return true; 828 } 829 return false; 830 } 831 832 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk, 833 struct mptcp_subflow_context *subflow, 834 struct sk_buff *skb, 835 struct mptcp_options_received *mp_opt) 836 { 837 /* here we can process OoO, in-window pkts, only in-sequence 4th ack 838 * will make the subflow fully established 839 */ 840 if (likely(subflow->fully_established)) { 841 /* on passive sockets, check for 3rd ack retransmission 842 * note that msk is always set by subflow_syn_recv_sock() 843 * for mp_join subflows 844 */ 845 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 && 846 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq && 847 subflow->mp_join && mp_opt->mp_join && 848 READ_ONCE(msk->pm.server_side)) 849 tcp_send_ack(ssk); 850 goto fully_established; 851 } 852 853 /* we must process OoO packets before the first subflow is fully 854 * established. OoO packets are instead a protocol violation 855 * for MP_JOIN subflows as the peer must not send any data 856 * before receiving the forth ack - cfr. RFC 8684 section 3.2. 857 */ 858 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) { 859 if (subflow->mp_join) 860 goto reset; 861 return subflow->mp_capable; 862 } 863 864 if (mp_opt->dss && mp_opt->use_ack) { 865 /* subflows are fully established as soon as we get any 866 * additional ack. 867 */ 868 subflow->fully_established = 1; 869 WRITE_ONCE(msk->fully_established, true); 870 goto fully_established; 871 } 872 873 if (mp_opt->add_addr) { 874 WRITE_ONCE(msk->fully_established, true); 875 return true; 876 } 877 878 /* If the first established packet does not contain MP_CAPABLE + data 879 * then fallback to TCP. Fallback scenarios requires a reset for 880 * MP_JOIN subflows. 881 */ 882 if (!mp_opt->mp_capable) { 883 if (subflow->mp_join) 884 goto reset; 885 subflow->mp_capable = 0; 886 pr_fallback(msk); 887 __mptcp_do_fallback(msk); 888 return false; 889 } 890 891 if (unlikely(!READ_ONCE(msk->pm.server_side))) 892 pr_warn_once("bogus mpc option on established client sk"); 893 mptcp_subflow_fully_established(subflow, mp_opt); 894 895 fully_established: 896 /* if the subflow is not already linked into the conn_list, we can't 897 * notify the PM: this subflow is still on the listener queue 898 * and the PM possibly acquiring the subflow lock could race with 899 * the listener close 900 */ 901 if (likely(subflow->pm_notified) || list_empty(&subflow->node)) 902 return true; 903 904 subflow->pm_notified = 1; 905 if (subflow->mp_join) { 906 clear_3rdack_retransmission(ssk); 907 mptcp_pm_subflow_established(msk); 908 } else { 909 mptcp_pm_fully_established(msk, ssk, GFP_ATOMIC); 910 } 911 return true; 912 913 reset: 914 mptcp_subflow_reset(ssk); 915 return false; 916 } 917 918 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit) 919 { 920 u32 old_ack32, cur_ack32; 921 922 if (use_64bit) 923 return cur_ack; 924 925 old_ack32 = (u32)old_ack; 926 cur_ack32 = (u32)cur_ack; 927 cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32; 928 if (unlikely(before(cur_ack32, old_ack32))) 929 return cur_ack + (1LL << 32); 930 return cur_ack; 931 } 932 933 static void ack_update_msk(struct mptcp_sock *msk, 934 struct sock *ssk, 935 struct mptcp_options_received *mp_opt) 936 { 937 u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt); 938 struct sock *sk = (struct sock *)msk; 939 u64 old_snd_una; 940 941 mptcp_data_lock(sk); 942 943 /* avoid ack expansion on update conflict, to reduce the risk of 944 * wrongly expanding to a future ack sequence number, which is way 945 * more dangerous than missing an ack 946 */ 947 old_snd_una = msk->snd_una; 948 new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64); 949 950 /* ACK for data not even sent yet? Ignore. */ 951 if (after64(new_snd_una, snd_nxt)) 952 new_snd_una = old_snd_una; 953 954 new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd; 955 956 if (after64(new_wnd_end, msk->wnd_end)) 957 msk->wnd_end = new_wnd_end; 958 959 /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */ 960 if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt))) 961 __mptcp_check_push(sk, ssk); 962 963 if (after64(new_snd_una, old_snd_una)) { 964 msk->snd_una = new_snd_una; 965 __mptcp_data_acked(sk); 966 } 967 mptcp_data_unlock(sk); 968 } 969 970 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit) 971 { 972 /* Skip if DATA_FIN was already received. 973 * If updating simultaneously with the recvmsg loop, values 974 * should match. If they mismatch, the peer is misbehaving and 975 * we will prefer the most recent information. 976 */ 977 if (READ_ONCE(msk->rcv_data_fin)) 978 return false; 979 980 WRITE_ONCE(msk->rcv_data_fin_seq, 981 expand_ack(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit)); 982 WRITE_ONCE(msk->rcv_data_fin, 1); 983 984 return true; 985 } 986 987 static bool add_addr_hmac_valid(struct mptcp_sock *msk, 988 struct mptcp_options_received *mp_opt) 989 { 990 u64 hmac = 0; 991 992 if (mp_opt->echo) 993 return true; 994 995 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4) 996 hmac = add_addr_generate_hmac(msk->remote_key, 997 msk->local_key, 998 mp_opt->addr_id, &mp_opt->addr, 999 mp_opt->port); 1000 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1001 else 1002 hmac = add_addr6_generate_hmac(msk->remote_key, 1003 msk->local_key, 1004 mp_opt->addr_id, &mp_opt->addr6, 1005 mp_opt->port); 1006 #endif 1007 1008 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n", 1009 msk, (unsigned long long)hmac, 1010 (unsigned long long)mp_opt->ahmac); 1011 1012 return hmac == mp_opt->ahmac; 1013 } 1014 1015 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb) 1016 { 1017 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 1018 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 1019 struct mptcp_options_received mp_opt; 1020 struct mptcp_ext *mpext; 1021 1022 if (__mptcp_check_fallback(msk)) { 1023 /* Keep it simple and unconditionally trigger send data cleanup and 1024 * pending queue spooling. We will need to acquire the data lock 1025 * for more accurate checks, and once the lock is acquired, such 1026 * helpers are cheap. 1027 */ 1028 mptcp_data_lock(subflow->conn); 1029 if (sk_stream_memory_free(sk)) 1030 __mptcp_check_push(subflow->conn, sk); 1031 __mptcp_data_acked(subflow->conn); 1032 mptcp_data_unlock(subflow->conn); 1033 return; 1034 } 1035 1036 mptcp_get_options(skb, &mp_opt); 1037 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt)) 1038 return; 1039 1040 if (mp_opt.fastclose && 1041 msk->local_key == mp_opt.rcvr_key) { 1042 WRITE_ONCE(msk->rcv_fastclose, true); 1043 mptcp_schedule_work((struct sock *)msk); 1044 } 1045 1046 if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) { 1047 struct mptcp_addr_info addr; 1048 1049 addr.port = htons(mp_opt.port); 1050 addr.id = mp_opt.addr_id; 1051 if (mp_opt.family == MPTCP_ADDR_IPVERSION_4) { 1052 addr.family = AF_INET; 1053 addr.addr = mp_opt.addr; 1054 } 1055 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1056 else if (mp_opt.family == MPTCP_ADDR_IPVERSION_6) { 1057 addr.family = AF_INET6; 1058 addr.addr6 = mp_opt.addr6; 1059 } 1060 #endif 1061 if (!mp_opt.echo) { 1062 mptcp_pm_add_addr_received(msk, &addr); 1063 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR); 1064 } else { 1065 mptcp_pm_add_addr_echoed(msk, &addr); 1066 mptcp_pm_del_add_timer(msk, &addr); 1067 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD); 1068 } 1069 1070 if (mp_opt.port) 1071 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD); 1072 1073 mp_opt.add_addr = 0; 1074 } 1075 1076 if (mp_opt.rm_addr) { 1077 mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list); 1078 mp_opt.rm_addr = 0; 1079 } 1080 1081 if (mp_opt.mp_prio) { 1082 mptcp_pm_mp_prio_received(sk, mp_opt.backup); 1083 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX); 1084 mp_opt.mp_prio = 0; 1085 } 1086 1087 if (mp_opt.reset) { 1088 subflow->reset_seen = 1; 1089 subflow->reset_reason = mp_opt.reset_reason; 1090 subflow->reset_transient = mp_opt.reset_transient; 1091 } 1092 1093 if (!mp_opt.dss) 1094 return; 1095 1096 /* we can't wait for recvmsg() to update the ack_seq, otherwise 1097 * monodirectional flows will stuck 1098 */ 1099 if (mp_opt.use_ack) 1100 ack_update_msk(msk, sk, &mp_opt); 1101 1102 /* Zero-data-length packets are dropped by the caller and not 1103 * propagated to the MPTCP layer, so the skb extension does not 1104 * need to be allocated or populated. DATA_FIN information, if 1105 * present, needs to be updated here before the skb is freed. 1106 */ 1107 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) { 1108 if (mp_opt.data_fin && mp_opt.data_len == 1 && 1109 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) && 1110 schedule_work(&msk->work)) 1111 sock_hold(subflow->conn); 1112 1113 return; 1114 } 1115 1116 mpext = skb_ext_add(skb, SKB_EXT_MPTCP); 1117 if (!mpext) 1118 return; 1119 1120 memset(mpext, 0, sizeof(*mpext)); 1121 1122 if (mp_opt.use_map) { 1123 if (mp_opt.mpc_map) { 1124 /* this is an MP_CAPABLE carrying MPTCP data 1125 * we know this map the first chunk of data 1126 */ 1127 mptcp_crypto_key_sha(subflow->remote_key, NULL, 1128 &mpext->data_seq); 1129 mpext->data_seq++; 1130 mpext->subflow_seq = 1; 1131 mpext->dsn64 = 1; 1132 mpext->mpc_map = 1; 1133 mpext->data_fin = 0; 1134 } else { 1135 mpext->data_seq = mp_opt.data_seq; 1136 mpext->subflow_seq = mp_opt.subflow_seq; 1137 mpext->dsn64 = mp_opt.dsn64; 1138 mpext->data_fin = mp_opt.data_fin; 1139 } 1140 mpext->data_len = mp_opt.data_len; 1141 mpext->use_map = 1; 1142 } 1143 } 1144 1145 static void mptcp_set_rwin(const struct tcp_sock *tp) 1146 { 1147 const struct sock *ssk = (const struct sock *)tp; 1148 const struct mptcp_subflow_context *subflow; 1149 struct mptcp_sock *msk; 1150 u64 ack_seq; 1151 1152 subflow = mptcp_subflow_ctx(ssk); 1153 msk = mptcp_sk(subflow->conn); 1154 1155 ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd; 1156 1157 if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent))) 1158 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq); 1159 } 1160 1161 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp, 1162 struct mptcp_out_options *opts) 1163 { 1164 if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK | 1165 OPTION_MPTCP_MPC_ACK) & opts->suboptions) { 1166 u8 len; 1167 1168 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) 1169 len = TCPOLEN_MPTCP_MPC_SYN; 1170 else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) 1171 len = TCPOLEN_MPTCP_MPC_SYNACK; 1172 else if (opts->ext_copy.data_len) 1173 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 1174 else 1175 len = TCPOLEN_MPTCP_MPC_ACK; 1176 1177 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len, 1178 MPTCP_SUPPORTED_VERSION, 1179 MPTCP_CAP_HMAC_SHA256); 1180 1181 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) & 1182 opts->suboptions)) 1183 goto mp_capable_done; 1184 1185 put_unaligned_be64(opts->sndr_key, ptr); 1186 ptr += 2; 1187 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions)) 1188 goto mp_capable_done; 1189 1190 put_unaligned_be64(opts->rcvr_key, ptr); 1191 ptr += 2; 1192 if (!opts->ext_copy.data_len) 1193 goto mp_capable_done; 1194 1195 put_unaligned_be32(opts->ext_copy.data_len << 16 | 1196 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1197 ptr += 1; 1198 } 1199 1200 mp_capable_done: 1201 if ((OPTION_MPTCP_ADD_ADDR 1202 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1203 | OPTION_MPTCP_ADD_ADDR6 1204 #endif 1205 ) & opts->suboptions) { 1206 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE; 1207 u8 echo = MPTCP_ADDR_ECHO; 1208 1209 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1210 if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions) 1211 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE; 1212 #endif 1213 1214 if (opts->addr.port) 1215 len += TCPOLEN_MPTCP_PORT_LEN; 1216 1217 if (opts->ahmac) { 1218 len += sizeof(opts->ahmac); 1219 echo = 0; 1220 } 1221 1222 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 1223 len, echo, opts->addr.id); 1224 if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) { 1225 memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4); 1226 ptr += 1; 1227 } 1228 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1229 else if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions) { 1230 memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16); 1231 ptr += 4; 1232 } 1233 #endif 1234 1235 if (!opts->addr.port) { 1236 if (opts->ahmac) { 1237 put_unaligned_be64(opts->ahmac, ptr); 1238 ptr += 2; 1239 } 1240 } else { 1241 u16 port = ntohs(opts->addr.port); 1242 1243 if (opts->ahmac) { 1244 u8 *bptr = (u8 *)ptr; 1245 1246 put_unaligned_be16(port, bptr); 1247 bptr += 2; 1248 put_unaligned_be64(opts->ahmac, bptr); 1249 bptr += 8; 1250 put_unaligned_be16(TCPOPT_NOP << 8 | 1251 TCPOPT_NOP, bptr); 1252 1253 ptr += 3; 1254 } else { 1255 put_unaligned_be32(port << 16 | 1256 TCPOPT_NOP << 8 | 1257 TCPOPT_NOP, ptr); 1258 ptr += 1; 1259 } 1260 } 1261 } 1262 1263 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) { 1264 u8 i = 1; 1265 1266 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR, 1267 TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr, 1268 0, opts->rm_list.ids[0]); 1269 1270 while (i < opts->rm_list.nr) { 1271 u8 id1, id2, id3, id4; 1272 1273 id1 = opts->rm_list.ids[i]; 1274 id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP; 1275 id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP; 1276 id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP; 1277 put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr); 1278 ptr += 1; 1279 i += 4; 1280 } 1281 } 1282 1283 if (OPTION_MPTCP_PRIO & opts->suboptions) { 1284 const struct sock *ssk = (const struct sock *)tp; 1285 struct mptcp_subflow_context *subflow; 1286 1287 subflow = mptcp_subflow_ctx(ssk); 1288 subflow->send_mp_prio = 0; 1289 1290 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO, 1291 TCPOLEN_MPTCP_PRIO, 1292 opts->backup, TCPOPT_NOP); 1293 } 1294 1295 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) { 1296 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1297 TCPOLEN_MPTCP_MPJ_SYN, 1298 opts->backup, opts->join_id); 1299 put_unaligned_be32(opts->token, ptr); 1300 ptr += 1; 1301 put_unaligned_be32(opts->nonce, ptr); 1302 ptr += 1; 1303 } 1304 1305 if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) { 1306 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1307 TCPOLEN_MPTCP_MPJ_SYNACK, 1308 opts->backup, opts->join_id); 1309 put_unaligned_be64(opts->thmac, ptr); 1310 ptr += 2; 1311 put_unaligned_be32(opts->nonce, ptr); 1312 ptr += 1; 1313 } 1314 1315 if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) { 1316 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1317 TCPOLEN_MPTCP_MPJ_ACK, 0, 0); 1318 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN); 1319 ptr += 5; 1320 } 1321 1322 if (OPTION_MPTCP_RST & opts->suboptions) 1323 *ptr++ = mptcp_option(MPTCPOPT_RST, 1324 TCPOLEN_MPTCP_RST, 1325 opts->reset_transient, 1326 opts->reset_reason); 1327 1328 if (opts->ext_copy.use_ack || opts->ext_copy.use_map) { 1329 struct mptcp_ext *mpext = &opts->ext_copy; 1330 u8 len = TCPOLEN_MPTCP_DSS_BASE; 1331 u8 flags = 0; 1332 1333 if (mpext->use_ack) { 1334 flags = MPTCP_DSS_HAS_ACK; 1335 if (mpext->ack64) { 1336 len += TCPOLEN_MPTCP_DSS_ACK64; 1337 flags |= MPTCP_DSS_ACK64; 1338 } else { 1339 len += TCPOLEN_MPTCP_DSS_ACK32; 1340 } 1341 } 1342 1343 if (mpext->use_map) { 1344 len += TCPOLEN_MPTCP_DSS_MAP64; 1345 1346 /* Use only 64-bit mapping flags for now, add 1347 * support for optional 32-bit mappings later. 1348 */ 1349 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64; 1350 if (mpext->data_fin) 1351 flags |= MPTCP_DSS_DATA_FIN; 1352 } 1353 1354 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags); 1355 1356 if (mpext->use_ack) { 1357 if (mpext->ack64) { 1358 put_unaligned_be64(mpext->data_ack, ptr); 1359 ptr += 2; 1360 } else { 1361 put_unaligned_be32(mpext->data_ack32, ptr); 1362 ptr += 1; 1363 } 1364 } 1365 1366 if (mpext->use_map) { 1367 put_unaligned_be64(mpext->data_seq, ptr); 1368 ptr += 2; 1369 put_unaligned_be32(mpext->subflow_seq, ptr); 1370 ptr += 1; 1371 put_unaligned_be32(mpext->data_len << 16 | 1372 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1373 } 1374 } 1375 1376 if (tp) 1377 mptcp_set_rwin(tp); 1378 } 1379 1380 __be32 mptcp_get_reset_option(const struct sk_buff *skb) 1381 { 1382 const struct mptcp_ext *ext = mptcp_get_ext(skb); 1383 u8 flags, reason; 1384 1385 if (ext) { 1386 flags = ext->reset_transient; 1387 reason = ext->reset_reason; 1388 1389 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST, 1390 flags, reason); 1391 } 1392 1393 return htonl(0u); 1394 } 1395 EXPORT_SYMBOL_GPL(mptcp_get_reset_option); 1396