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 #include <trace/events/mptcp.h> 17 18 static bool mptcp_cap_flag_sha256(u8 flags) 19 { 20 return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256; 21 } 22 23 static void mptcp_parse_option(const struct sk_buff *skb, 24 const unsigned char *ptr, int opsize, 25 struct mptcp_options_received *mp_opt) 26 { 27 u8 subtype = *ptr >> 4; 28 int expected_opsize; 29 u16 subopt; 30 u8 version; 31 u8 flags; 32 u8 i; 33 34 switch (subtype) { 35 case MPTCPOPT_MP_CAPABLE: 36 /* strict size checking */ 37 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { 38 if (skb->len > tcp_hdr(skb)->doff << 2) 39 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA; 40 else 41 expected_opsize = TCPOLEN_MPTCP_MPC_ACK; 42 subopt = OPTION_MPTCP_MPC_ACK; 43 } else { 44 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK) { 45 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK; 46 subopt = OPTION_MPTCP_MPC_SYNACK; 47 } else { 48 expected_opsize = TCPOLEN_MPTCP_MPC_SYN; 49 subopt = OPTION_MPTCP_MPC_SYN; 50 } 51 } 52 53 /* Cfr RFC 8684 Section 3.3.0: 54 * If a checksum is present but its use had 55 * not been negotiated in the MP_CAPABLE handshake, the receiver MUST 56 * close the subflow with a RST, as it is not behaving as negotiated. 57 * If a checksum is not present when its use has been negotiated, the 58 * receiver MUST close the subflow with a RST, as it is considered 59 * broken 60 * We parse even option with mismatching csum presence, so that 61 * later in subflow_data_ready we can trigger the reset. 62 */ 63 if (opsize != expected_opsize && 64 (expected_opsize != TCPOLEN_MPTCP_MPC_ACK_DATA || 65 opsize != TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM)) 66 break; 67 68 /* try to be gentle vs future versions on the initial syn */ 69 version = *ptr++ & MPTCP_VERSION_MASK; 70 if (opsize != TCPOLEN_MPTCP_MPC_SYN) { 71 if (version != MPTCP_SUPPORTED_VERSION) 72 break; 73 } else if (version < MPTCP_SUPPORTED_VERSION) { 74 break; 75 } 76 77 flags = *ptr++; 78 if (!mptcp_cap_flag_sha256(flags) || 79 (flags & MPTCP_CAP_EXTENSIBILITY)) 80 break; 81 82 /* RFC 6824, Section 3.1: 83 * "For the Checksum Required bit (labeled "A"), if either 84 * host requires the use of checksums, checksums MUST be used. 85 * In other words, the only way for checksums not to be used 86 * is if both hosts in their SYNs set A=0." 87 */ 88 if (flags & MPTCP_CAP_CHECKSUM_REQD) 89 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD; 90 91 mp_opt->deny_join_id0 = !!(flags & MPTCP_CAP_DENY_JOIN_ID0); 92 93 mp_opt->suboptions |= subopt; 94 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) { 95 mp_opt->sndr_key = get_unaligned_be64(ptr); 96 ptr += 8; 97 } 98 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) { 99 mp_opt->rcvr_key = get_unaligned_be64(ptr); 100 ptr += 8; 101 } 102 if (opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA) { 103 /* Section 3.1.: 104 * "the data parameters in a MP_CAPABLE are semantically 105 * equivalent to those in a DSS option and can be used 106 * interchangeably." 107 */ 108 mp_opt->suboptions |= OPTION_MPTCP_DSS; 109 mp_opt->use_map = 1; 110 mp_opt->mpc_map = 1; 111 mp_opt->data_len = get_unaligned_be16(ptr); 112 ptr += 2; 113 } 114 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM) { 115 mp_opt->csum = get_unaligned((__force __sum16 *)ptr); 116 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD; 117 ptr += 2; 118 } 119 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d csum=%u\n", 120 version, flags, opsize, mp_opt->sndr_key, 121 mp_opt->rcvr_key, mp_opt->data_len, mp_opt->csum); 122 break; 123 124 case MPTCPOPT_MP_JOIN: 125 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) { 126 mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYN; 127 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 128 mp_opt->join_id = *ptr++; 129 mp_opt->token = get_unaligned_be32(ptr); 130 ptr += 4; 131 mp_opt->nonce = get_unaligned_be32(ptr); 132 ptr += 4; 133 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u\n", 134 mp_opt->backup, mp_opt->join_id, 135 mp_opt->token, mp_opt->nonce); 136 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) { 137 mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYNACK; 138 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 139 mp_opt->join_id = *ptr++; 140 mp_opt->thmac = get_unaligned_be64(ptr); 141 ptr += 8; 142 mp_opt->nonce = get_unaligned_be32(ptr); 143 ptr += 4; 144 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u\n", 145 mp_opt->backup, mp_opt->join_id, 146 mp_opt->thmac, mp_opt->nonce); 147 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) { 148 mp_opt->suboptions |= OPTION_MPTCP_MPJ_ACK; 149 ptr += 2; 150 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN); 151 pr_debug("MP_JOIN hmac\n"); 152 } 153 break; 154 155 case MPTCPOPT_DSS: 156 pr_debug("DSS\n"); 157 ptr++; 158 159 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK; 160 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0; 161 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0; 162 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0; 163 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0; 164 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK); 165 166 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d\n", 167 mp_opt->data_fin, mp_opt->dsn64, 168 mp_opt->use_map, mp_opt->ack64, 169 mp_opt->use_ack); 170 171 expected_opsize = TCPOLEN_MPTCP_DSS_BASE; 172 173 if (mp_opt->use_ack) { 174 if (mp_opt->ack64) 175 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64; 176 else 177 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32; 178 } 179 180 if (mp_opt->use_map) { 181 if (mp_opt->dsn64) 182 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64; 183 else 184 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32; 185 } 186 187 /* Always parse any csum presence combination, we will enforce 188 * RFC 8684 Section 3.3.0 checks later in subflow_data_ready 189 */ 190 if (opsize != expected_opsize && 191 opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) 192 break; 193 194 mp_opt->suboptions |= OPTION_MPTCP_DSS; 195 if (mp_opt->use_ack) { 196 if (mp_opt->ack64) { 197 mp_opt->data_ack = get_unaligned_be64(ptr); 198 ptr += 8; 199 } else { 200 mp_opt->data_ack = get_unaligned_be32(ptr); 201 ptr += 4; 202 } 203 204 pr_debug("data_ack=%llu\n", mp_opt->data_ack); 205 } 206 207 if (mp_opt->use_map) { 208 if (mp_opt->dsn64) { 209 mp_opt->data_seq = get_unaligned_be64(ptr); 210 ptr += 8; 211 } else { 212 mp_opt->data_seq = get_unaligned_be32(ptr); 213 ptr += 4; 214 } 215 216 mp_opt->subflow_seq = get_unaligned_be32(ptr); 217 ptr += 4; 218 219 mp_opt->data_len = get_unaligned_be16(ptr); 220 ptr += 2; 221 222 if (opsize == expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) { 223 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD; 224 mp_opt->csum = get_unaligned((__force __sum16 *)ptr); 225 ptr += 2; 226 } 227 228 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u\n", 229 mp_opt->data_seq, mp_opt->subflow_seq, 230 mp_opt->data_len, !!(mp_opt->suboptions & OPTION_MPTCP_CSUMREQD), 231 mp_opt->csum); 232 } 233 234 break; 235 236 case MPTCPOPT_ADD_ADDR: 237 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO; 238 if (!mp_opt->echo) { 239 if (opsize == TCPOLEN_MPTCP_ADD_ADDR || 240 opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT) 241 mp_opt->addr.family = AF_INET; 242 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 243 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 || 244 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT) 245 mp_opt->addr.family = AF_INET6; 246 #endif 247 else 248 break; 249 } else { 250 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE || 251 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) 252 mp_opt->addr.family = AF_INET; 253 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 254 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE || 255 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) 256 mp_opt->addr.family = AF_INET6; 257 #endif 258 else 259 break; 260 } 261 262 mp_opt->suboptions |= OPTION_MPTCP_ADD_ADDR; 263 mp_opt->addr.id = *ptr++; 264 mp_opt->addr.port = 0; 265 mp_opt->ahmac = 0; 266 if (mp_opt->addr.family == AF_INET) { 267 memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4); 268 ptr += 4; 269 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT || 270 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) { 271 mp_opt->addr.port = htons(get_unaligned_be16(ptr)); 272 ptr += 2; 273 } 274 } 275 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 276 else { 277 memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16); 278 ptr += 16; 279 if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT || 280 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) { 281 mp_opt->addr.port = htons(get_unaligned_be16(ptr)); 282 ptr += 2; 283 } 284 } 285 #endif 286 if (!mp_opt->echo) { 287 mp_opt->ahmac = get_unaligned_be64(ptr); 288 ptr += 8; 289 } 290 pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d\n", 291 (mp_opt->addr.family == AF_INET6) ? "6" : "", 292 mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port)); 293 break; 294 295 case MPTCPOPT_RM_ADDR: 296 if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 || 297 opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX) 298 break; 299 300 ptr++; 301 302 mp_opt->suboptions |= OPTION_MPTCP_RM_ADDR; 303 mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE; 304 for (i = 0; i < mp_opt->rm_list.nr; i++) 305 mp_opt->rm_list.ids[i] = *ptr++; 306 pr_debug("RM_ADDR: rm_list_nr=%d\n", mp_opt->rm_list.nr); 307 break; 308 309 case MPTCPOPT_MP_PRIO: 310 if (opsize != TCPOLEN_MPTCP_PRIO) 311 break; 312 313 mp_opt->suboptions |= OPTION_MPTCP_PRIO; 314 mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP; 315 pr_debug("MP_PRIO: prio=%d\n", mp_opt->backup); 316 break; 317 318 case MPTCPOPT_MP_FASTCLOSE: 319 if (opsize != TCPOLEN_MPTCP_FASTCLOSE) 320 break; 321 322 ptr += 2; 323 mp_opt->rcvr_key = get_unaligned_be64(ptr); 324 ptr += 8; 325 mp_opt->suboptions |= OPTION_MPTCP_FASTCLOSE; 326 pr_debug("MP_FASTCLOSE: recv_key=%llu\n", mp_opt->rcvr_key); 327 break; 328 329 case MPTCPOPT_RST: 330 if (opsize != TCPOLEN_MPTCP_RST) 331 break; 332 333 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) 334 break; 335 336 mp_opt->suboptions |= OPTION_MPTCP_RST; 337 flags = *ptr++; 338 mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT; 339 mp_opt->reset_reason = *ptr; 340 pr_debug("MP_RST: transient=%u reason=%u\n", 341 mp_opt->reset_transient, mp_opt->reset_reason); 342 break; 343 344 case MPTCPOPT_MP_FAIL: 345 if (opsize != TCPOLEN_MPTCP_FAIL) 346 break; 347 348 ptr += 2; 349 mp_opt->suboptions |= OPTION_MPTCP_FAIL; 350 mp_opt->fail_seq = get_unaligned_be64(ptr); 351 pr_debug("MP_FAIL: data_seq=%llu\n", mp_opt->fail_seq); 352 break; 353 354 default: 355 break; 356 } 357 } 358 359 void mptcp_get_options(const struct sk_buff *skb, 360 struct mptcp_options_received *mp_opt) 361 { 362 const struct tcphdr *th = tcp_hdr(skb); 363 const unsigned char *ptr; 364 int length; 365 366 /* Ensure that casting the whole status to u32 is efficient and safe */ 367 BUILD_BUG_ON(sizeof_field(struct mptcp_options_received, status) != sizeof(u32)); 368 BUILD_BUG_ON(!IS_ALIGNED(offsetof(struct mptcp_options_received, status), 369 sizeof(u32))); 370 *(u32 *)&mp_opt->status = 0; 371 372 length = (th->doff * 4) - sizeof(struct tcphdr); 373 ptr = (const unsigned char *)(th + 1); 374 375 while (length > 0) { 376 int opcode = *ptr++; 377 int opsize; 378 379 switch (opcode) { 380 case TCPOPT_EOL: 381 return; 382 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */ 383 length--; 384 continue; 385 default: 386 if (length < 2) 387 return; 388 opsize = *ptr++; 389 if (opsize < 2) /* "silly options" */ 390 return; 391 if (opsize > length) 392 return; /* don't parse partial options */ 393 if (opcode == TCPOPT_MPTCP) 394 mptcp_parse_option(skb, ptr, opsize, mp_opt); 395 ptr += opsize - 2; 396 length -= opsize; 397 } 398 } 399 } 400 401 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb, 402 unsigned int *size, struct mptcp_out_options *opts) 403 { 404 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 405 406 /* we will use snd_isn to detect first pkt [re]transmission 407 * in mptcp_established_options_mp() 408 */ 409 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq; 410 if (subflow->request_mptcp) { 411 if (unlikely(subflow_simultaneous_connect(sk))) { 412 WARN_ON_ONCE(!mptcp_try_fallback(sk, MPTCP_MIB_SIMULTCONNFALLBACK)); 413 414 /* Ensure mptcp_finish_connect() will not process the 415 * MPC handshake. 416 */ 417 subflow->request_mptcp = 0; 418 return false; 419 } 420 421 opts->suboptions = OPTION_MPTCP_MPC_SYN; 422 opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk)); 423 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk)); 424 *size = TCPOLEN_MPTCP_MPC_SYN; 425 return true; 426 } else if (subflow->request_join) { 427 pr_debug("remote_token=%u, nonce=%u\n", subflow->remote_token, 428 subflow->local_nonce); 429 opts->suboptions = OPTION_MPTCP_MPJ_SYN; 430 opts->join_id = subflow->local_id; 431 opts->token = subflow->remote_token; 432 opts->nonce = subflow->local_nonce; 433 opts->backup = subflow->request_bkup; 434 *size = TCPOLEN_MPTCP_MPJ_SYN; 435 return true; 436 } 437 return false; 438 } 439 440 static void clear_3rdack_retransmission(struct sock *sk) 441 { 442 struct inet_connection_sock *icsk = inet_csk(sk); 443 444 sk_stop_timer(sk, &icsk->icsk_delack_timer); 445 icsk->icsk_ack.ato = 0; 446 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER); 447 } 448 449 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb, 450 bool snd_data_fin_enable, 451 unsigned int *size, 452 struct mptcp_out_options *opts) 453 { 454 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 455 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 456 struct mptcp_ext *mpext; 457 unsigned int data_len; 458 u8 len; 459 460 /* When skb is not available, we better over-estimate the emitted 461 * options len. A full DSS option (28 bytes) is longer than 462 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so 463 * tell the caller to defer the estimate to 464 * mptcp_established_options_dss(), which will reserve enough space. 465 */ 466 if (!skb) 467 return false; 468 469 /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */ 470 if (READ_ONCE(subflow->fully_established) || snd_data_fin_enable || 471 subflow->snd_isn != TCP_SKB_CB(skb)->seq || 472 sk->sk_state != TCP_ESTABLISHED) 473 return false; 474 475 if (subflow->mp_capable) { 476 mpext = mptcp_get_ext(skb); 477 data_len = mpext ? mpext->data_len : 0; 478 479 /* we will check ops->data_len in mptcp_write_options() to 480 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and 481 * TCPOLEN_MPTCP_MPC_ACK 482 */ 483 opts->data_len = data_len; 484 opts->suboptions = OPTION_MPTCP_MPC_ACK; 485 opts->sndr_key = subflow->local_key; 486 opts->rcvr_key = subflow->remote_key; 487 opts->csum_reqd = READ_ONCE(msk->csum_enabled); 488 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk)); 489 490 /* Section 3.1. 491 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK 492 * packets that start the first subflow of an MPTCP connection, 493 * as well as the first packet that carries data 494 */ 495 if (data_len > 0) { 496 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 497 if (opts->csum_reqd) { 498 /* we need to propagate more info to csum the pseudo hdr */ 499 opts->data_seq = mpext->data_seq; 500 opts->subflow_seq = mpext->subflow_seq; 501 opts->csum = mpext->csum; 502 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 503 } 504 *size = ALIGN(len, 4); 505 } else { 506 *size = TCPOLEN_MPTCP_MPC_ACK; 507 } 508 509 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d\n", 510 subflow, subflow->local_key, subflow->remote_key, 511 data_len); 512 513 return true; 514 } else if (subflow->mp_join) { 515 opts->suboptions = OPTION_MPTCP_MPJ_ACK; 516 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN); 517 *size = TCPOLEN_MPTCP_MPJ_ACK; 518 pr_debug("subflow=%p\n", subflow); 519 520 /* we can use the full delegate action helper only from BH context 521 * If we are in process context - sk is flushing the backlog at 522 * socket lock release time - just set the appropriate flag, will 523 * be handled by the release callback 524 */ 525 if (sock_owned_by_user(sk)) 526 set_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status); 527 else 528 mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_ACK); 529 return true; 530 } 531 return false; 532 } 533 534 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow, 535 struct sk_buff *skb, struct mptcp_ext *ext) 536 { 537 /* The write_seq value has already been incremented, so the actual 538 * sequence number for the DATA_FIN is one less. 539 */ 540 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1; 541 542 if (!ext->use_map || !skb->len) { 543 /* RFC6824 requires a DSS mapping with specific values 544 * if DATA_FIN is set but no data payload is mapped 545 */ 546 ext->data_fin = 1; 547 ext->use_map = 1; 548 ext->dsn64 = 1; 549 ext->data_seq = data_fin_tx_seq; 550 ext->subflow_seq = 0; 551 ext->data_len = 1; 552 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) { 553 /* If there's an existing DSS mapping and it is the 554 * final mapping, DATA_FIN consumes 1 additional byte of 555 * mapping space. 556 */ 557 ext->data_fin = 1; 558 ext->data_len++; 559 } 560 } 561 562 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb, 563 bool snd_data_fin_enable, 564 unsigned int *size, 565 struct mptcp_out_options *opts) 566 { 567 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 568 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 569 struct tcp_sock *tp = tcp_sk(sk); 570 unsigned int dss_size = 0; 571 struct mptcp_ext *mpext; 572 unsigned int ack_size; 573 bool ret = false; 574 575 /* Zero `use_ack` and `use_map` flags with one shot. */ 576 BUILD_BUG_ON(sizeof_field(struct mptcp_ext, flags) != sizeof(u16)); 577 BUILD_BUG_ON(!IS_ALIGNED(offsetof(struct mptcp_ext, flags), 578 sizeof(u16))); 579 *(u16 *)&opts->ext_copy.flags = 0; 580 opts->csum_reqd = READ_ONCE(msk->csum_enabled); 581 mpext = skb ? mptcp_get_ext(skb) : NULL; 582 583 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) { 584 unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64; 585 586 if (mpext) { 587 if (opts->csum_reqd) 588 map_size += TCPOLEN_MPTCP_DSS_CHECKSUM; 589 590 opts->ext_copy = *mpext; 591 } 592 593 dss_size = map_size; 594 if (skb && snd_data_fin_enable) 595 mptcp_write_data_fin(subflow, skb, &opts->ext_copy); 596 opts->suboptions = OPTION_MPTCP_DSS; 597 ret = true; 598 } 599 600 /* passive sockets msk will set the 'can_ack' after accept(), even 601 * if the first subflow may have the already the remote key handy 602 */ 603 if (!READ_ONCE(msk->can_ack)) { 604 *size = ALIGN(dss_size, 4); 605 return ret; 606 } 607 608 if (READ_ONCE(msk->use_64bit_ack)) { 609 ack_size = TCPOLEN_MPTCP_DSS_ACK64; 610 opts->ext_copy.ack64 = 1; 611 } else { 612 ack_size = TCPOLEN_MPTCP_DSS_ACK32; 613 opts->ext_copy.ack64 = 0; 614 } 615 opts->ext_copy.use_ack = 1; 616 opts->suboptions = OPTION_MPTCP_DSS; 617 618 /* Add kind/length/subtype/flag overhead if mapping is not populated */ 619 if (dss_size == 0) 620 ack_size += TCPOLEN_MPTCP_DSS_BASE; 621 622 /* The caller is __tcp_transmit_skb(), and will compute the new rcv 623 * wnd soon: ensure that the window can shrink. 624 */ 625 if (skb) 626 tp->rcv_wnd = tp->rcv_nxt - tp->rcv_wup; 627 628 dss_size += ack_size; 629 630 *size = ALIGN(dss_size, 4); 631 return true; 632 } 633 634 static u64 add_addr_generate_hmac(u64 key1, u64 key2, 635 struct mptcp_addr_info *addr) 636 { 637 u16 port = ntohs(addr->port); 638 u8 hmac[SHA256_DIGEST_SIZE]; 639 u8 msg[19]; 640 int i = 0; 641 642 msg[i++] = addr->id; 643 if (addr->family == AF_INET) { 644 memcpy(&msg[i], &addr->addr.s_addr, 4); 645 i += 4; 646 } 647 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 648 else if (addr->family == AF_INET6) { 649 memcpy(&msg[i], &addr->addr6.s6_addr, 16); 650 i += 16; 651 } 652 #endif 653 msg[i++] = port >> 8; 654 msg[i++] = port & 0xFF; 655 656 mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac); 657 658 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]); 659 } 660 661 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb, 662 unsigned int *size, 663 unsigned int remaining, 664 struct mptcp_out_options *opts) 665 { 666 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 667 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 668 unsigned int opt_size = *size; 669 struct mptcp_addr_info addr; 670 bool echo; 671 int len; 672 673 /* add addr will strip the existing options, be sure to avoid breaking 674 * MPC/MPJ handshakes 675 */ 676 if (!mptcp_pm_should_add_signal(msk) || 677 (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) || 678 !skb || !skb_is_tcp_pure_ack(skb) || 679 !mptcp_pm_add_addr_signal(msk, opt_size, remaining, &addr, &echo)) 680 return false; 681 682 remaining += opt_size; 683 684 len = mptcp_add_addr_len(addr.family, echo, !!addr.port); 685 if (remaining < len) 686 return false; 687 688 *size = len; 689 pr_debug("drop other suboptions\n"); 690 opts->suboptions = 0; 691 *size -= opt_size; 692 opts->addr = addr; 693 opts->suboptions |= OPTION_MPTCP_ADD_ADDR; 694 if (!echo) { 695 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDRTX); 696 opts->ahmac = add_addr_generate_hmac(READ_ONCE(msk->local_key), 697 READ_ONCE(msk->remote_key), 698 &opts->addr); 699 } else { 700 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADDTX); 701 opts->ahmac = 0; 702 } 703 pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d\n", 704 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port)); 705 706 return true; 707 } 708 709 static bool mptcp_established_options_rm_addr(struct sock *sk, 710 unsigned int *size, 711 unsigned int remaining, 712 struct mptcp_out_options *opts) 713 { 714 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 715 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 716 struct mptcp_rm_list rm_list; 717 int i, len; 718 719 if (!mptcp_pm_should_rm_signal(msk) || 720 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list))) 721 return false; 722 723 len = mptcp_rm_addr_len(&rm_list); 724 if (len < 0) 725 return false; 726 if (remaining < len) 727 return false; 728 729 *size = len; 730 opts->suboptions |= OPTION_MPTCP_RM_ADDR; 731 opts->rm_list = rm_list; 732 733 for (i = 0; i < opts->rm_list.nr; i++) 734 pr_debug("rm_list_ids[%d]=%d\n", i, opts->rm_list.ids[i]); 735 MPTCP_ADD_STATS(sock_net(sk), MPTCP_MIB_RMADDRTX, opts->rm_list.nr); 736 return true; 737 } 738 739 static bool mptcp_established_options_mp_prio(struct sock *sk, 740 unsigned int *size, 741 unsigned int remaining, 742 struct mptcp_out_options *opts) 743 { 744 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 745 746 /* can't send MP_PRIO with MPC, as they share the same option space: 747 * 'backup'. Also it makes no sense at all 748 */ 749 if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC)) 750 return false; 751 752 /* account for the trailing 'nop' option */ 753 if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN) 754 return false; 755 756 *size = TCPOLEN_MPTCP_PRIO_ALIGN; 757 opts->suboptions |= OPTION_MPTCP_PRIO; 758 opts->backup = subflow->request_bkup; 759 760 pr_debug("prio=%d\n", opts->backup); 761 762 return true; 763 } 764 765 static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb, 766 unsigned int *size, 767 unsigned int remaining, 768 struct mptcp_out_options *opts) 769 { 770 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 771 772 if (remaining < TCPOLEN_MPTCP_RST) 773 return false; 774 775 *size = TCPOLEN_MPTCP_RST; 776 opts->suboptions |= OPTION_MPTCP_RST; 777 opts->reset_transient = subflow->reset_transient; 778 opts->reset_reason = subflow->reset_reason; 779 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTTX); 780 781 return true; 782 } 783 784 static bool mptcp_established_options_fastclose(struct sock *sk, 785 unsigned int *size, 786 unsigned int remaining, 787 struct mptcp_out_options *opts) 788 { 789 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 790 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 791 792 if (likely(!subflow->send_fastclose)) 793 return false; 794 795 if (remaining < TCPOLEN_MPTCP_FASTCLOSE) 796 return false; 797 798 *size = TCPOLEN_MPTCP_FASTCLOSE; 799 opts->suboptions |= OPTION_MPTCP_FASTCLOSE; 800 opts->rcvr_key = READ_ONCE(msk->remote_key); 801 802 pr_debug("FASTCLOSE key=%llu\n", opts->rcvr_key); 803 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSETX); 804 return true; 805 } 806 807 static bool mptcp_established_options_mp_fail(struct sock *sk, 808 unsigned int *size, 809 unsigned int remaining, 810 struct mptcp_out_options *opts) 811 { 812 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 813 814 if (likely(!subflow->send_mp_fail)) 815 return false; 816 817 if (remaining < TCPOLEN_MPTCP_FAIL) 818 return false; 819 820 *size = TCPOLEN_MPTCP_FAIL; 821 opts->suboptions |= OPTION_MPTCP_FAIL; 822 opts->fail_seq = subflow->map_seq; 823 824 pr_debug("MP_FAIL fail_seq=%llu\n", opts->fail_seq); 825 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILTX); 826 827 return true; 828 } 829 830 int mptcp_established_options(struct sock *sk, struct sk_buff *skb, 831 unsigned int remaining, 832 struct mptcp_out_options *opts) 833 { 834 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 835 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 836 unsigned int opt_size = 0; 837 int total_size = 0; 838 bool snd_data_fin; 839 bool ret = false; 840 841 opts->suboptions = 0; 842 843 /* Force later mptcp_write_options(), but do not use any actual 844 * option space. 845 */ 846 if (unlikely(__mptcp_check_fallback(msk) && !mptcp_check_infinite_map(skb))) 847 return 0; 848 849 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) { 850 if (mptcp_established_options_fastclose(sk, &opt_size, remaining, opts) || 851 mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) { 852 total_size += opt_size; 853 remaining -= opt_size; 854 } 855 /* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */ 856 if (mptcp_established_options_rst(sk, skb, &opt_size, remaining, opts)) { 857 total_size += opt_size; 858 remaining -= opt_size; 859 } 860 return total_size; 861 } 862 863 snd_data_fin = mptcp_data_fin_enabled(msk); 864 if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, opts)) 865 ret = true; 866 else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, opts)) { 867 unsigned int mp_fail_size; 868 869 ret = true; 870 if (mptcp_established_options_mp_fail(sk, &mp_fail_size, 871 remaining - opt_size, opts)) { 872 total_size += opt_size + mp_fail_size; 873 remaining -= opt_size - mp_fail_size; 874 return total_size; 875 } 876 } 877 878 /* we reserved enough space for the above options, and exceeding the 879 * TCP option space would be fatal 880 */ 881 if (WARN_ON_ONCE(opt_size > remaining)) 882 return -1; 883 884 total_size += opt_size; 885 remaining -= opt_size; 886 if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) { 887 total_size += opt_size; 888 remaining -= opt_size; 889 ret = true; 890 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) { 891 total_size += opt_size; 892 remaining -= opt_size; 893 ret = true; 894 } 895 896 if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) { 897 total_size += opt_size; 898 remaining -= opt_size; 899 ret = true; 900 } 901 902 return ret ? total_size : -1; 903 } 904 905 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size, 906 struct mptcp_out_options *opts) 907 { 908 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 909 910 if (subflow_req->mp_capable) { 911 opts->suboptions = OPTION_MPTCP_MPC_SYNACK; 912 opts->sndr_key = subflow_req->local_key; 913 opts->csum_reqd = subflow_req->csum_reqd; 914 opts->allow_join_id0 = subflow_req->allow_join_id0; 915 *size = TCPOLEN_MPTCP_MPC_SYNACK; 916 pr_debug("subflow_req=%p, local_key=%llu\n", 917 subflow_req, subflow_req->local_key); 918 return true; 919 } else if (subflow_req->mp_join) { 920 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK; 921 opts->backup = subflow_req->request_bkup; 922 opts->join_id = subflow_req->local_id; 923 opts->thmac = subflow_req->thmac; 924 opts->nonce = subflow_req->local_nonce; 925 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u\n", 926 subflow_req, opts->backup, opts->join_id, 927 opts->thmac, opts->nonce); 928 *size = TCPOLEN_MPTCP_MPJ_SYNACK; 929 return true; 930 } 931 return false; 932 } 933 934 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk, 935 struct mptcp_subflow_context *subflow, 936 struct sk_buff *skb, 937 struct mptcp_options_received *mp_opt) 938 { 939 /* here we can process OoO, in-window pkts, only in-sequence 4th ack 940 * will make the subflow fully established 941 */ 942 if (likely(READ_ONCE(subflow->fully_established))) { 943 /* on passive sockets, check for 3rd ack retransmission 944 * note that msk is always set by subflow_syn_recv_sock() 945 * for mp_join subflows 946 */ 947 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 && 948 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq && 949 subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) && 950 !subflow->request_join) 951 tcp_send_ack(ssk); 952 goto check_notify; 953 } 954 955 /* we must process OoO packets before the first subflow is fully 956 * established. OoO packets are instead a protocol violation 957 * for MP_JOIN subflows as the peer must not send any data 958 * before receiving the forth ack - cfr. RFC 8684 section 3.2. 959 */ 960 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) { 961 if (subflow->mp_join) 962 goto reset; 963 if (subflow->is_mptfo && mp_opt->suboptions & OPTION_MPTCP_MPC_ACK) 964 goto set_fully_established; 965 return subflow->mp_capable; 966 } 967 968 if (subflow->remote_key_valid && 969 (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) || 970 ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) && 971 (!mp_opt->echo || subflow->mp_join)))) { 972 /* subflows are fully established as soon as we get any 973 * additional ack, including ADD_ADDR. 974 */ 975 goto set_fully_established; 976 } 977 978 /* If the first established packet does not contain MP_CAPABLE + data 979 * then fallback to TCP. Fallback scenarios requires a reset for 980 * MP_JOIN subflows. 981 */ 982 if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) { 983 if (subflow->mp_join) 984 goto reset; 985 subflow->mp_capable = 0; 986 if (!mptcp_try_fallback(ssk, MPTCP_MIB_MPCAPABLEDATAFALLBACK)) { 987 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_FALLBACKFAILED); 988 goto reset; 989 } 990 return false; 991 } 992 993 if (unlikely(!READ_ONCE(msk->pm.server_side))) 994 pr_warn_once("bogus mpc option on established client sk"); 995 996 set_fully_established: 997 if (mp_opt->deny_join_id0) 998 WRITE_ONCE(msk->pm.remote_deny_join_id0, true); 999 1000 mptcp_data_lock((struct sock *)msk); 1001 __mptcp_subflow_fully_established(msk, subflow, mp_opt); 1002 mptcp_data_unlock((struct sock *)msk); 1003 1004 check_notify: 1005 /* if the subflow is not already linked into the conn_list, we can't 1006 * notify the PM: this subflow is still on the listener queue 1007 * and the PM possibly acquiring the subflow lock could race with 1008 * the listener close 1009 */ 1010 if (likely(subflow->pm_notified) || list_empty(&subflow->node)) 1011 return true; 1012 1013 subflow->pm_notified = 1; 1014 if (subflow->mp_join) { 1015 clear_3rdack_retransmission(ssk); 1016 mptcp_pm_subflow_established(msk); 1017 } else { 1018 mptcp_pm_fully_established(msk, ssk); 1019 } 1020 return true; 1021 1022 reset: 1023 mptcp_subflow_reset(ssk); 1024 return false; 1025 } 1026 1027 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq) 1028 { 1029 u32 old_seq32, cur_seq32; 1030 1031 old_seq32 = (u32)old_seq; 1032 cur_seq32 = (u32)cur_seq; 1033 cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32; 1034 if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32))) 1035 return cur_seq + (1LL << 32); 1036 1037 /* reverse wrap could happen, too */ 1038 if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32))) 1039 return cur_seq - (1LL << 32); 1040 return cur_seq; 1041 } 1042 1043 static void __mptcp_snd_una_update(struct mptcp_sock *msk, u64 new_snd_una) 1044 { 1045 msk->bytes_acked += new_snd_una - msk->snd_una; 1046 WRITE_ONCE(msk->snd_una, new_snd_una); 1047 } 1048 1049 static void rwin_update(struct mptcp_sock *msk, struct sock *ssk, 1050 struct sk_buff *skb) 1051 { 1052 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 1053 struct tcp_sock *tp = tcp_sk(ssk); 1054 u64 mptcp_rcv_wnd; 1055 1056 /* Avoid touching extra cachelines if TCP is going to accept this 1057 * skb without filling the TCP-level window even with a possibly 1058 * outdated mptcp-level rwin. 1059 */ 1060 if (!skb->len || skb->len < tcp_receive_window(tp)) 1061 return; 1062 1063 mptcp_rcv_wnd = atomic64_read(&msk->rcv_wnd_sent); 1064 if (!after64(mptcp_rcv_wnd, subflow->rcv_wnd_sent)) 1065 return; 1066 1067 /* Some other subflow grew the mptcp-level rwin since rcv_wup, 1068 * resync. 1069 */ 1070 tp->rcv_wnd += mptcp_rcv_wnd - subflow->rcv_wnd_sent; 1071 tcp_update_max_rcv_wnd_seq(tp); 1072 subflow->rcv_wnd_sent = mptcp_rcv_wnd; 1073 } 1074 1075 static void ack_update_msk(struct mptcp_sock *msk, 1076 struct sock *ssk, 1077 struct mptcp_options_received *mp_opt) 1078 { 1079 u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt); 1080 struct sock *sk = (struct sock *)msk; 1081 u64 old_snd_una; 1082 1083 mptcp_data_lock(sk); 1084 1085 /* avoid ack expansion on update conflict, to reduce the risk of 1086 * wrongly expanding to a future ack sequence number, which is way 1087 * more dangerous than missing an ack 1088 */ 1089 old_snd_una = msk->snd_una; 1090 new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64); 1091 1092 /* ACK for data not even sent yet? Ignore.*/ 1093 if (unlikely(after64(new_snd_una, snd_nxt))) 1094 new_snd_una = old_snd_una; 1095 1096 new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd; 1097 1098 if (after64(new_wnd_end, msk->wnd_end)) 1099 WRITE_ONCE(msk->wnd_end, new_wnd_end); 1100 1101 /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */ 1102 if (after64(msk->wnd_end, snd_nxt)) 1103 __mptcp_check_push(sk, ssk); 1104 1105 if (after64(new_snd_una, old_snd_una)) { 1106 __mptcp_snd_una_update(msk, new_snd_una); 1107 __mptcp_data_acked(sk); 1108 } 1109 msk->last_ack_recv = tcp_jiffies32; 1110 mptcp_data_unlock(sk); 1111 1112 trace_ack_update_msk(mp_opt->data_ack, 1113 old_snd_una, new_snd_una, 1114 new_wnd_end, READ_ONCE(msk->wnd_end)); 1115 } 1116 1117 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit) 1118 { 1119 /* Skip if DATA_FIN was already received. 1120 * If updating simultaneously with the recvmsg loop, values 1121 * should match. If they mismatch, the peer is misbehaving and 1122 * we will prefer the most recent information. 1123 */ 1124 if (READ_ONCE(msk->rcv_data_fin)) 1125 return false; 1126 1127 WRITE_ONCE(msk->rcv_data_fin_seq, 1128 mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit)); 1129 WRITE_ONCE(msk->rcv_data_fin, 1); 1130 1131 return true; 1132 } 1133 1134 static bool add_addr_hmac_valid(struct mptcp_sock *msk, 1135 struct mptcp_options_received *mp_opt) 1136 { 1137 u64 hmac = 0; 1138 1139 if (mp_opt->echo) 1140 return true; 1141 1142 hmac = add_addr_generate_hmac(READ_ONCE(msk->remote_key), 1143 READ_ONCE(msk->local_key), 1144 &mp_opt->addr); 1145 1146 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n", 1147 msk, hmac, mp_opt->ahmac); 1148 1149 return hmac == mp_opt->ahmac; 1150 } 1151 1152 /* Return false in case of error (or subflow has been reset), 1153 * else return true. 1154 */ 1155 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb) 1156 { 1157 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 1158 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 1159 struct mptcp_options_received mp_opt; 1160 struct mptcp_ext *mpext; 1161 1162 if (__mptcp_check_fallback(msk)) { 1163 /* Keep it simple and unconditionally trigger send data cleanup and 1164 * pending queue spooling. We will need to acquire the data lock 1165 * for more accurate checks, and once the lock is acquired, such 1166 * helpers are cheap. 1167 */ 1168 mptcp_data_lock(subflow->conn); 1169 if (sk_stream_memory_free(sk)) 1170 __mptcp_check_push(subflow->conn, sk); 1171 1172 /* on fallback we just need to ignore the msk-level snd_una, as 1173 * this is really plain TCP 1174 */ 1175 __mptcp_snd_una_update(msk, READ_ONCE(msk->snd_nxt)); 1176 1177 __mptcp_data_acked(subflow->conn); 1178 mptcp_data_unlock(subflow->conn); 1179 return true; 1180 } 1181 1182 mptcp_get_options(skb, &mp_opt); 1183 1184 /* The subflow can be in close state only if check_fully_established() 1185 * just sent a reset. If so, tell the caller to ignore the current packet. 1186 */ 1187 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt)) 1188 return sk->sk_state != TCP_CLOSE; 1189 1190 if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) { 1191 if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) && 1192 READ_ONCE(msk->local_key) == mp_opt.rcvr_key) { 1193 WRITE_ONCE(msk->rcv_fastclose, true); 1194 mptcp_schedule_work((struct sock *)msk); 1195 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSERX); 1196 } 1197 1198 if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) && 1199 add_addr_hmac_valid(msk, &mp_opt)) { 1200 if (!mp_opt.echo) { 1201 mptcp_pm_add_addr_received(sk, &mp_opt.addr); 1202 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR); 1203 } else { 1204 mptcp_pm_add_addr_echoed(msk, &mp_opt.addr); 1205 mptcp_pm_del_add_timer(msk, &mp_opt.addr, true); 1206 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD); 1207 } 1208 1209 if (mp_opt.addr.port) 1210 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD); 1211 } 1212 1213 if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR) 1214 mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list); 1215 1216 if (mp_opt.suboptions & OPTION_MPTCP_PRIO) { 1217 mptcp_pm_mp_prio_received(sk, mp_opt.backup); 1218 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX); 1219 } 1220 1221 if (mp_opt.suboptions & OPTION_MPTCP_FAIL) { 1222 mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq); 1223 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX); 1224 } 1225 1226 if (mp_opt.suboptions & OPTION_MPTCP_RST) { 1227 subflow->reset_seen = 1; 1228 subflow->reset_reason = mp_opt.reset_reason; 1229 subflow->reset_transient = mp_opt.reset_transient; 1230 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTRX); 1231 } 1232 1233 if (!(mp_opt.suboptions & OPTION_MPTCP_DSS)) 1234 return true; 1235 } 1236 1237 /* we can't wait for recvmsg() to update the ack_seq, otherwise 1238 * monodirectional flows will stuck 1239 */ 1240 if (mp_opt.use_ack) 1241 ack_update_msk(msk, sk, &mp_opt); 1242 rwin_update(msk, sk, skb); 1243 1244 /* Zero-data-length packets are dropped by the caller and not 1245 * propagated to the MPTCP layer, so the skb extension does not 1246 * need to be allocated or populated. DATA_FIN information, if 1247 * present, needs to be updated here before the skb is freed. 1248 */ 1249 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) { 1250 if (mp_opt.data_fin && mp_opt.data_len == 1 && 1251 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64)) 1252 mptcp_schedule_work((struct sock *)msk); 1253 1254 return true; 1255 } 1256 1257 mpext = skb_ext_add(skb, SKB_EXT_MPTCP); 1258 if (!mpext) 1259 return false; 1260 1261 memset(mpext, 0, sizeof(*mpext)); 1262 1263 if (likely(mp_opt.use_map)) { 1264 if (mp_opt.mpc_map) { 1265 /* this is an MP_CAPABLE carrying MPTCP data 1266 * we know this map the first chunk of data 1267 */ 1268 mptcp_crypto_key_sha(subflow->remote_key, NULL, 1269 &mpext->data_seq); 1270 mpext->data_seq++; 1271 mpext->subflow_seq = 1; 1272 mpext->dsn64 = 1; 1273 mpext->mpc_map = 1; 1274 mpext->data_fin = 0; 1275 } else { 1276 mpext->data_seq = mp_opt.data_seq; 1277 mpext->subflow_seq = mp_opt.subflow_seq; 1278 mpext->dsn64 = mp_opt.dsn64; 1279 mpext->data_fin = mp_opt.data_fin; 1280 } 1281 mpext->data_len = mp_opt.data_len; 1282 mpext->use_map = 1; 1283 mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD); 1284 1285 if (mpext->csum_reqd) 1286 mpext->csum = mp_opt.csum; 1287 } 1288 1289 return true; 1290 } 1291 1292 static u64 mptcp_set_rwin(struct mptcp_sock *msk, struct tcp_sock *tp, 1293 struct tcphdr *th, u64 ack_seq) 1294 { 1295 const struct sock *ssk = (const struct sock *)tp; 1296 u64 rcv_wnd_old, rcv_wnd_new; 1297 u32 new_win; 1298 u64 win; 1299 1300 rcv_wnd_new = ack_seq + tp->rcv_wnd; 1301 1302 rcv_wnd_old = atomic64_read(&msk->rcv_wnd_sent); 1303 if (after64(rcv_wnd_new, rcv_wnd_old)) { 1304 u64 rcv_wnd; 1305 1306 for (;;) { 1307 rcv_wnd = atomic64_cmpxchg(&msk->rcv_wnd_sent, rcv_wnd_old, rcv_wnd_new); 1308 1309 if (rcv_wnd == rcv_wnd_old) 1310 break; 1311 1312 rcv_wnd_old = rcv_wnd; 1313 if (before64(rcv_wnd_new, rcv_wnd_old)) { 1314 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICTUPDATE); 1315 goto raise_win; 1316 } 1317 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICT); 1318 } 1319 goto update_wspace; 1320 } 1321 1322 if (rcv_wnd_new != rcv_wnd_old) { 1323 raise_win: 1324 /* The msk-level rcv wnd is after the tcp level one, 1325 * sync the latter. 1326 */ 1327 rcv_wnd_new = rcv_wnd_old; 1328 win = rcv_wnd_old - ack_seq; 1329 new_win = min_t(u64, win, U32_MAX); 1330 tp->rcv_wnd = new_win; 1331 tcp_update_max_rcv_wnd_seq(tp); 1332 1333 /* Make sure we do not exceed the maximum possible 1334 * scaled window. 1335 */ 1336 if (unlikely(th->syn)) 1337 new_win = min(new_win, 65535U) << tp->rx_opt.rcv_wscale; 1338 if (!tp->rx_opt.rcv_wscale && 1339 READ_ONCE(sock_net(ssk)->ipv4.sysctl_tcp_workaround_signed_windows)) 1340 new_win = min(new_win, MAX_TCP_WINDOW); 1341 else 1342 new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale)); 1343 1344 /* RFC1323 scaling applied */ 1345 new_win >>= tp->rx_opt.rcv_wscale; 1346 th->window = htons(new_win); 1347 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDSHARED); 1348 } 1349 1350 update_wspace: 1351 WRITE_ONCE(msk->old_wspace, tp->rcv_wnd); 1352 return rcv_wnd_new; 1353 } 1354 1355 static void mptcp_track_rwin(struct tcp_sock *tp) 1356 { 1357 const struct sock *ssk = (const struct sock *)tp; 1358 struct mptcp_subflow_context *subflow; 1359 struct mptcp_sock *msk; 1360 1361 if (!ssk) 1362 return; 1363 1364 subflow = mptcp_subflow_ctx(ssk); 1365 msk = mptcp_sk(subflow->conn); 1366 WRITE_ONCE(msk->old_wspace, tp->rcv_wnd); 1367 } 1368 1369 __sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum) 1370 { 1371 struct csum_pseudo_header header; 1372 __wsum csum; 1373 1374 /* cfr RFC 8684 3.3.1.: 1375 * the data sequence number used in the pseudo-header is 1376 * always the 64-bit value, irrespective of what length is used in the 1377 * DSS option itself. 1378 */ 1379 header.data_seq = cpu_to_be64(data_seq); 1380 header.subflow_seq = htonl(subflow_seq); 1381 header.data_len = htons(data_len); 1382 header.csum = 0; 1383 1384 csum = csum_partial(&header, sizeof(header), sum); 1385 return csum_fold(csum); 1386 } 1387 1388 static __sum16 mptcp_make_csum(const struct mptcp_ext *mpext) 1389 { 1390 return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len, 1391 ~csum_unfold(mpext->csum)); 1392 } 1393 1394 static void put_len_csum(u16 len, __sum16 csum, void *data) 1395 { 1396 __sum16 *sumptr = data + 2; 1397 __be16 *ptr = data; 1398 1399 put_unaligned_be16(len, ptr); 1400 1401 put_unaligned(csum, sumptr); 1402 } 1403 1404 void mptcp_write_options(struct tcphdr *th, __be32 *ptr, struct tcp_sock *tp, 1405 struct mptcp_out_options *opts) 1406 { 1407 const struct sock *ssk = (const struct sock *)tp; 1408 struct mptcp_subflow_context *subflow; 1409 1410 /* Which options can be used together? 1411 * 1412 * X: mutually exclusive 1413 * O: often used together 1414 * C: can be used together in some cases 1415 * P: could be used together but we prefer not to (optimisations) 1416 * 1417 * Opt: | MPC | MPJ | DSS | ADD | RM | PRIO | FAIL | FC | 1418 * ------|------|------|------|------|------|------|------|------| 1419 * MPC |------|------|------|------|------|------|------|------| 1420 * MPJ | X |------|------|------|------|------|------|------| 1421 * DSS | X | X |------|------|------|------|------|------| 1422 * ADD | X | X | P |------|------|------|------|------| 1423 * RM | C | C | C | P |------|------|------|------| 1424 * PRIO | X | C | C | C | C |------|------|------| 1425 * FAIL | X | X | C | X | X | X |------|------| 1426 * FC | X | X | X | X | X | X | X |------| 1427 * RST | X | X | X | X | X | X | O | O | 1428 * ------|------|------|------|------|------|------|------|------| 1429 * 1430 * The same applies in mptcp_established_options() function. 1431 */ 1432 if (likely(OPTION_MPTCP_DSS & opts->suboptions)) { 1433 struct mptcp_ext *mpext = &opts->ext_copy; 1434 u8 len = TCPOLEN_MPTCP_DSS_BASE; 1435 u8 flags = 0; 1436 1437 if (mpext->use_ack) { 1438 flags = MPTCP_DSS_HAS_ACK; 1439 if (mpext->ack64) { 1440 len += TCPOLEN_MPTCP_DSS_ACK64; 1441 flags |= MPTCP_DSS_ACK64; 1442 } else { 1443 len += TCPOLEN_MPTCP_DSS_ACK32; 1444 } 1445 } 1446 1447 if (mpext->use_map) { 1448 len += TCPOLEN_MPTCP_DSS_MAP64; 1449 1450 /* Use only 64-bit mapping flags for now, add 1451 * support for optional 32-bit mappings later. 1452 */ 1453 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64; 1454 if (mpext->data_fin) 1455 flags |= MPTCP_DSS_DATA_FIN; 1456 1457 if (opts->csum_reqd) 1458 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 1459 } 1460 1461 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags); 1462 1463 if (mpext->use_ack) { 1464 struct mptcp_sock *msk; 1465 u64 ack_seq; 1466 1467 /* DSS option is set only by mptcp_established_options, 1468 * the caller is __tcp_transmit_skb() and ssk is always 1469 * not NULL. 1470 */ 1471 subflow = mptcp_subflow_ctx(ssk); 1472 msk = mptcp_sk(subflow->conn); 1473 ack_seq = READ_ONCE(msk->ack_seq); 1474 if (mpext->ack64) { 1475 put_unaligned_be64(ack_seq, ptr); 1476 ptr += 2; 1477 } else { 1478 put_unaligned_be32(ack_seq, ptr); 1479 ptr += 1; 1480 } 1481 subflow->rcv_wnd_sent = mptcp_set_rwin(msk, tp, th, 1482 ack_seq); 1483 } 1484 1485 if (mpext->use_map) { 1486 put_unaligned_be64(mpext->data_seq, ptr); 1487 ptr += 2; 1488 put_unaligned_be32(mpext->subflow_seq, ptr); 1489 ptr += 1; 1490 if (opts->csum_reqd) { 1491 /* data_len == 0 is reserved for the infinite mapping, 1492 * the checksum will also be set to 0. 1493 */ 1494 put_len_csum(mpext->data_len, 1495 (mpext->data_len ? mptcp_make_csum(mpext) : 0), 1496 ptr); 1497 } else { 1498 put_unaligned_be32(mpext->data_len << 16 | 1499 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1500 } 1501 ptr += 1; 1502 } 1503 1504 /* We might need to add MP_FAIL options in rare cases */ 1505 if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) 1506 goto mp_fail; 1507 } else if (OPTIONS_MPTCP_MPC & opts->suboptions) { 1508 u8 len, flag = MPTCP_CAP_HMAC_SHA256; 1509 1510 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) { 1511 len = TCPOLEN_MPTCP_MPC_SYN; 1512 } else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) { 1513 len = TCPOLEN_MPTCP_MPC_SYNACK; 1514 } else if (opts->data_len) { 1515 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 1516 if (opts->csum_reqd) 1517 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 1518 } else { 1519 len = TCPOLEN_MPTCP_MPC_ACK; 1520 } 1521 1522 if (opts->csum_reqd) 1523 flag |= MPTCP_CAP_CHECKSUM_REQD; 1524 1525 if (!opts->allow_join_id0) 1526 flag |= MPTCP_CAP_DENY_JOIN_ID0; 1527 1528 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len, 1529 MPTCP_SUPPORTED_VERSION, 1530 flag); 1531 1532 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) & 1533 opts->suboptions)) 1534 goto mp_capable_done; 1535 1536 put_unaligned_be64(opts->sndr_key, ptr); 1537 ptr += 2; 1538 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions)) 1539 goto mp_capable_done; 1540 1541 put_unaligned_be64(opts->rcvr_key, ptr); 1542 ptr += 2; 1543 if (!opts->data_len) 1544 goto mp_capable_done; 1545 1546 if (opts->csum_reqd) { 1547 put_len_csum(opts->data_len, 1548 __mptcp_make_csum(opts->data_seq, 1549 opts->subflow_seq, 1550 opts->data_len, 1551 ~csum_unfold(opts->csum)), 1552 ptr); 1553 } else { 1554 put_unaligned_be32(opts->data_len << 16 | 1555 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1556 } 1557 ptr += 1; 1558 1559 /* MPC is additionally mutually exclusive with MP_PRIO */ 1560 goto mp_capable_done; 1561 } else if (OPTIONS_MPTCP_MPJ & opts->suboptions) { 1562 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) { 1563 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1564 TCPOLEN_MPTCP_MPJ_SYN, 1565 opts->backup, opts->join_id); 1566 put_unaligned_be32(opts->token, ptr); 1567 ptr += 1; 1568 put_unaligned_be32(opts->nonce, ptr); 1569 ptr += 1; 1570 } else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) { 1571 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1572 TCPOLEN_MPTCP_MPJ_SYNACK, 1573 opts->backup, opts->join_id); 1574 put_unaligned_be64(opts->thmac, ptr); 1575 ptr += 2; 1576 put_unaligned_be32(opts->nonce, ptr); 1577 ptr += 1; 1578 } else { 1579 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1580 TCPOLEN_MPTCP_MPJ_ACK, 0, 0); 1581 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN); 1582 ptr += 5; 1583 } 1584 } else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) { 1585 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE; 1586 u8 echo = MPTCP_ADDR_ECHO; 1587 1588 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1589 if (opts->addr.family == AF_INET6) 1590 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE; 1591 #endif 1592 1593 if (opts->addr.port) 1594 len += TCPOLEN_MPTCP_PORT_LEN; 1595 1596 if (opts->ahmac) { 1597 len += sizeof(opts->ahmac); 1598 echo = 0; 1599 } 1600 1601 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 1602 len, echo, opts->addr.id); 1603 if (opts->addr.family == AF_INET) { 1604 memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4); 1605 ptr += 1; 1606 } 1607 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1608 else if (opts->addr.family == AF_INET6) { 1609 memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16); 1610 ptr += 4; 1611 } 1612 #endif 1613 1614 if (!opts->addr.port) { 1615 if (opts->ahmac) { 1616 put_unaligned_be64(opts->ahmac, ptr); 1617 ptr += 2; 1618 } 1619 } else { 1620 u16 port = ntohs(opts->addr.port); 1621 1622 if (opts->ahmac) { 1623 u8 *bptr = (u8 *)ptr; 1624 1625 put_unaligned_be16(port, bptr); 1626 bptr += 2; 1627 put_unaligned_be64(opts->ahmac, bptr); 1628 bptr += 8; 1629 put_unaligned_be16(TCPOPT_NOP << 8 | 1630 TCPOPT_NOP, bptr); 1631 1632 ptr += 3; 1633 } else { 1634 put_unaligned_be32(port << 16 | 1635 TCPOPT_NOP << 8 | 1636 TCPOPT_NOP, ptr); 1637 ptr += 1; 1638 } 1639 } 1640 } else if (unlikely(OPTION_MPTCP_FASTCLOSE & opts->suboptions)) { 1641 /* FASTCLOSE is mutually exclusive with others except RST */ 1642 *ptr++ = mptcp_option(MPTCPOPT_MP_FASTCLOSE, 1643 TCPOLEN_MPTCP_FASTCLOSE, 1644 0, 0); 1645 put_unaligned_be64(opts->rcvr_key, ptr); 1646 ptr += 2; 1647 1648 if (OPTION_MPTCP_RST & opts->suboptions) 1649 goto mp_rst; 1650 return; 1651 } else if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) { 1652 mp_fail: 1653 /* MP_FAIL is mutually exclusive with others except RST */ 1654 subflow = mptcp_subflow_ctx(ssk); 1655 subflow->send_mp_fail = 0; 1656 1657 *ptr++ = mptcp_option(MPTCPOPT_MP_FAIL, 1658 TCPOLEN_MPTCP_FAIL, 1659 0, 0); 1660 put_unaligned_be64(opts->fail_seq, ptr); 1661 ptr += 2; 1662 1663 if (OPTION_MPTCP_RST & opts->suboptions) 1664 goto mp_rst; 1665 return; 1666 } else if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) { 1667 mp_rst: 1668 *ptr++ = mptcp_option(MPTCPOPT_RST, 1669 TCPOLEN_MPTCP_RST, 1670 opts->reset_transient, 1671 opts->reset_reason); 1672 return; 1673 } else if (unlikely(!opts->suboptions)) { 1674 /* Fallback to TCP */ 1675 mptcp_track_rwin(tp); 1676 return; 1677 } 1678 1679 if (OPTION_MPTCP_PRIO & opts->suboptions) { 1680 subflow = mptcp_subflow_ctx(ssk); 1681 subflow->send_mp_prio = 0; 1682 1683 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO, 1684 TCPOLEN_MPTCP_PRIO, 1685 opts->backup, TCPOPT_NOP); 1686 1687 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPPRIOTX); 1688 } 1689 1690 mp_capable_done: 1691 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) { 1692 u8 i = 1; 1693 1694 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR, 1695 TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr, 1696 0, opts->rm_list.ids[0]); 1697 1698 while (i < opts->rm_list.nr) { 1699 u8 id1, id2, id3, id4; 1700 1701 id1 = opts->rm_list.ids[i]; 1702 id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP; 1703 id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP; 1704 id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP; 1705 put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr); 1706 ptr += 1; 1707 i += 4; 1708 } 1709 } 1710 } 1711 1712 __be32 mptcp_get_reset_option(const struct sk_buff *skb) 1713 { 1714 const struct mptcp_ext *ext = mptcp_get_ext(skb); 1715 u8 flags, reason; 1716 1717 if (ext) { 1718 flags = ext->reset_transient; 1719 reason = ext->reset_reason; 1720 1721 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST, 1722 flags, reason); 1723 } 1724 1725 return htonl(0u); 1726 } 1727 EXPORT_SYMBOL_GPL(mptcp_get_reset_option); 1728