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 <net/tcp.h> 11 #include <net/mptcp.h> 12 #include "protocol.h" 13 14 static bool mptcp_cap_flag_sha256(u8 flags) 15 { 16 return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256; 17 } 18 19 static void mptcp_parse_option(const struct sk_buff *skb, 20 const unsigned char *ptr, int opsize, 21 struct mptcp_options_received *mp_opt) 22 { 23 u8 subtype = *ptr >> 4; 24 int expected_opsize; 25 u8 version; 26 u8 flags; 27 28 switch (subtype) { 29 case MPTCPOPT_MP_CAPABLE: 30 /* strict size checking */ 31 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { 32 if (skb->len > tcp_hdr(skb)->doff << 2) 33 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA; 34 else 35 expected_opsize = TCPOLEN_MPTCP_MPC_ACK; 36 } else { 37 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK) 38 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK; 39 else 40 expected_opsize = TCPOLEN_MPTCP_MPC_SYN; 41 } 42 if (opsize != expected_opsize) 43 break; 44 45 /* try to be gentle vs future versions on the initial syn */ 46 version = *ptr++ & MPTCP_VERSION_MASK; 47 if (opsize != TCPOLEN_MPTCP_MPC_SYN) { 48 if (version != MPTCP_SUPPORTED_VERSION) 49 break; 50 } else if (version < MPTCP_SUPPORTED_VERSION) { 51 break; 52 } 53 54 flags = *ptr++; 55 if (!mptcp_cap_flag_sha256(flags) || 56 (flags & MPTCP_CAP_EXTENSIBILITY)) 57 break; 58 59 /* RFC 6824, Section 3.1: 60 * "For the Checksum Required bit (labeled "A"), if either 61 * host requires the use of checksums, checksums MUST be used. 62 * In other words, the only way for checksums not to be used 63 * is if both hosts in their SYNs set A=0." 64 * 65 * Section 3.3.0: 66 * "If a checksum is not present when its use has been 67 * negotiated, the receiver MUST close the subflow with a RST as 68 * it is considered broken." 69 * 70 * We don't implement DSS checksum - fall back to TCP. 71 */ 72 if (flags & MPTCP_CAP_CHECKSUM_REQD) 73 break; 74 75 mp_opt->mp_capable = 1; 76 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) { 77 mp_opt->sndr_key = get_unaligned_be64(ptr); 78 ptr += 8; 79 } 80 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) { 81 mp_opt->rcvr_key = get_unaligned_be64(ptr); 82 ptr += 8; 83 } 84 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA) { 85 /* Section 3.1.: 86 * "the data parameters in a MP_CAPABLE are semantically 87 * equivalent to those in a DSS option and can be used 88 * interchangeably." 89 */ 90 mp_opt->dss = 1; 91 mp_opt->use_map = 1; 92 mp_opt->mpc_map = 1; 93 mp_opt->data_len = get_unaligned_be16(ptr); 94 ptr += 2; 95 } 96 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d", 97 version, flags, opsize, mp_opt->sndr_key, 98 mp_opt->rcvr_key, mp_opt->data_len); 99 break; 100 101 case MPTCPOPT_MP_JOIN: 102 mp_opt->mp_join = 1; 103 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) { 104 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 105 mp_opt->join_id = *ptr++; 106 mp_opt->token = get_unaligned_be32(ptr); 107 ptr += 4; 108 mp_opt->nonce = get_unaligned_be32(ptr); 109 ptr += 4; 110 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u", 111 mp_opt->backup, mp_opt->join_id, 112 mp_opt->token, mp_opt->nonce); 113 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) { 114 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 115 mp_opt->join_id = *ptr++; 116 mp_opt->thmac = get_unaligned_be64(ptr); 117 ptr += 8; 118 mp_opt->nonce = get_unaligned_be32(ptr); 119 ptr += 4; 120 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u", 121 mp_opt->backup, mp_opt->join_id, 122 mp_opt->thmac, mp_opt->nonce); 123 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) { 124 ptr += 2; 125 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN); 126 pr_debug("MP_JOIN hmac"); 127 } else { 128 pr_warn("MP_JOIN bad option size"); 129 mp_opt->mp_join = 0; 130 } 131 break; 132 133 case MPTCPOPT_DSS: 134 pr_debug("DSS"); 135 ptr++; 136 137 /* we must clear 'mpc_map' be able to detect MP_CAPABLE 138 * map vs DSS map in mptcp_incoming_options(), and reconstruct 139 * map info accordingly 140 */ 141 mp_opt->mpc_map = 0; 142 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK; 143 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0; 144 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0; 145 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0; 146 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0; 147 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK); 148 149 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d", 150 mp_opt->data_fin, mp_opt->dsn64, 151 mp_opt->use_map, mp_opt->ack64, 152 mp_opt->use_ack); 153 154 expected_opsize = TCPOLEN_MPTCP_DSS_BASE; 155 156 if (mp_opt->use_ack) { 157 if (mp_opt->ack64) 158 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64; 159 else 160 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32; 161 } 162 163 if (mp_opt->use_map) { 164 if (mp_opt->dsn64) 165 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64; 166 else 167 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32; 168 } 169 170 /* RFC 6824, Section 3.3: 171 * If a checksum is present, but its use had 172 * not been negotiated in the MP_CAPABLE handshake, 173 * the checksum field MUST be ignored. 174 */ 175 if (opsize != expected_opsize && 176 opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) 177 break; 178 179 mp_opt->dss = 1; 180 181 if (mp_opt->use_ack) { 182 if (mp_opt->ack64) { 183 mp_opt->data_ack = get_unaligned_be64(ptr); 184 ptr += 8; 185 } else { 186 mp_opt->data_ack = get_unaligned_be32(ptr); 187 ptr += 4; 188 } 189 190 pr_debug("data_ack=%llu", mp_opt->data_ack); 191 } 192 193 if (mp_opt->use_map) { 194 if (mp_opt->dsn64) { 195 mp_opt->data_seq = get_unaligned_be64(ptr); 196 ptr += 8; 197 } else { 198 mp_opt->data_seq = get_unaligned_be32(ptr); 199 ptr += 4; 200 } 201 202 mp_opt->subflow_seq = get_unaligned_be32(ptr); 203 ptr += 4; 204 205 mp_opt->data_len = get_unaligned_be16(ptr); 206 ptr += 2; 207 208 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u", 209 mp_opt->data_seq, mp_opt->subflow_seq, 210 mp_opt->data_len); 211 } 212 213 break; 214 215 case MPTCPOPT_ADD_ADDR: 216 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO; 217 if (!mp_opt->echo) { 218 if (opsize == TCPOLEN_MPTCP_ADD_ADDR || 219 opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT) 220 mp_opt->family = MPTCP_ADDR_IPVERSION_4; 221 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 222 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 || 223 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT) 224 mp_opt->family = MPTCP_ADDR_IPVERSION_6; 225 #endif 226 else 227 break; 228 } else { 229 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE || 230 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) 231 mp_opt->family = MPTCP_ADDR_IPVERSION_4; 232 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 233 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE || 234 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) 235 mp_opt->family = MPTCP_ADDR_IPVERSION_6; 236 #endif 237 else 238 break; 239 } 240 241 mp_opt->add_addr = 1; 242 mp_opt->port = 0; 243 mp_opt->addr_id = *ptr++; 244 pr_debug("ADD_ADDR: id=%d", mp_opt->addr_id); 245 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4) { 246 memcpy((u8 *)&mp_opt->addr.s_addr, (u8 *)ptr, 4); 247 ptr += 4; 248 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT || 249 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) { 250 mp_opt->port = get_unaligned_be16(ptr); 251 ptr += 2; 252 } 253 } 254 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 255 else { 256 memcpy(mp_opt->addr6.s6_addr, (u8 *)ptr, 16); 257 ptr += 16; 258 if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT || 259 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) { 260 mp_opt->port = get_unaligned_be16(ptr); 261 ptr += 2; 262 } 263 } 264 #endif 265 if (!mp_opt->echo) { 266 mp_opt->ahmac = get_unaligned_be64(ptr); 267 ptr += 8; 268 } 269 break; 270 271 case MPTCPOPT_RM_ADDR: 272 if (opsize != TCPOLEN_MPTCP_RM_ADDR_BASE) 273 break; 274 275 mp_opt->rm_addr = 1; 276 mp_opt->rm_id = *ptr++; 277 pr_debug("RM_ADDR: id=%d", mp_opt->rm_id); 278 break; 279 280 default: 281 break; 282 } 283 } 284 285 void mptcp_get_options(const struct sk_buff *skb, 286 struct mptcp_options_received *mp_opt) 287 { 288 const struct tcphdr *th = tcp_hdr(skb); 289 const unsigned char *ptr; 290 int length; 291 292 /* initialize option status */ 293 mp_opt->mp_capable = 0; 294 mp_opt->mp_join = 0; 295 mp_opt->add_addr = 0; 296 mp_opt->rm_addr = 0; 297 mp_opt->dss = 0; 298 299 length = (th->doff * 4) - sizeof(struct tcphdr); 300 ptr = (const unsigned char *)(th + 1); 301 302 while (length > 0) { 303 int opcode = *ptr++; 304 int opsize; 305 306 switch (opcode) { 307 case TCPOPT_EOL: 308 return; 309 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */ 310 length--; 311 continue; 312 default: 313 opsize = *ptr++; 314 if (opsize < 2) /* "silly options" */ 315 return; 316 if (opsize > length) 317 return; /* don't parse partial options */ 318 if (opcode == TCPOPT_MPTCP) 319 mptcp_parse_option(skb, ptr, opsize, mp_opt); 320 ptr += opsize - 2; 321 length -= opsize; 322 } 323 } 324 } 325 326 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb, 327 unsigned int *size, struct mptcp_out_options *opts) 328 { 329 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 330 331 /* we will use snd_isn to detect first pkt [re]transmission 332 * in mptcp_established_options_mp() 333 */ 334 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq; 335 if (subflow->request_mptcp) { 336 pr_debug("local_key=%llu", subflow->local_key); 337 opts->suboptions = OPTION_MPTCP_MPC_SYN; 338 opts->sndr_key = subflow->local_key; 339 *size = TCPOLEN_MPTCP_MPC_SYN; 340 return true; 341 } else if (subflow->request_join) { 342 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token, 343 subflow->local_nonce); 344 opts->suboptions = OPTION_MPTCP_MPJ_SYN; 345 opts->join_id = subflow->local_id; 346 opts->token = subflow->remote_token; 347 opts->nonce = subflow->local_nonce; 348 opts->backup = subflow->request_bkup; 349 *size = TCPOLEN_MPTCP_MPJ_SYN; 350 return true; 351 } 352 return false; 353 } 354 355 /* MP_JOIN client subflow must wait for 4th ack before sending any data: 356 * TCP can't schedule delack timer before the subflow is fully established. 357 * MPTCP uses the delack timer to do 3rd ack retransmissions 358 */ 359 static void schedule_3rdack_retransmission(struct sock *sk) 360 { 361 struct inet_connection_sock *icsk = inet_csk(sk); 362 struct tcp_sock *tp = tcp_sk(sk); 363 unsigned long timeout; 364 365 /* reschedule with a timeout above RTT, as we must look only for drop */ 366 if (tp->srtt_us) 367 timeout = tp->srtt_us << 1; 368 else 369 timeout = TCP_TIMEOUT_INIT; 370 371 WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER); 372 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER; 373 icsk->icsk_ack.timeout = timeout; 374 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout); 375 } 376 377 static void clear_3rdack_retransmission(struct sock *sk) 378 { 379 struct inet_connection_sock *icsk = inet_csk(sk); 380 381 sk_stop_timer(sk, &icsk->icsk_delack_timer); 382 icsk->icsk_ack.timeout = 0; 383 icsk->icsk_ack.ato = 0; 384 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER); 385 } 386 387 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb, 388 unsigned int *size, 389 unsigned int remaining, 390 struct mptcp_out_options *opts) 391 { 392 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 393 struct mptcp_ext *mpext; 394 unsigned int data_len; 395 396 /* When skb is not available, we better over-estimate the emitted 397 * options len. A full DSS option (28 bytes) is longer than 398 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so 399 * tell the caller to defer the estimate to 400 * mptcp_established_options_dss(), which will reserve enough space. 401 */ 402 if (!skb) 403 return false; 404 405 /* MPC/MPJ needed only on 3rd ack packet */ 406 if (subflow->fully_established || 407 subflow->snd_isn != TCP_SKB_CB(skb)->seq) 408 return false; 409 410 if (subflow->mp_capable) { 411 mpext = mptcp_get_ext(skb); 412 data_len = mpext ? mpext->data_len : 0; 413 414 /* we will check ext_copy.data_len in mptcp_write_options() to 415 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and 416 * TCPOLEN_MPTCP_MPC_ACK 417 */ 418 opts->ext_copy.data_len = data_len; 419 opts->suboptions = OPTION_MPTCP_MPC_ACK; 420 opts->sndr_key = subflow->local_key; 421 opts->rcvr_key = subflow->remote_key; 422 423 /* Section 3.1. 424 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK 425 * packets that start the first subflow of an MPTCP connection, 426 * as well as the first packet that carries data 427 */ 428 if (data_len > 0) 429 *size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4); 430 else 431 *size = TCPOLEN_MPTCP_MPC_ACK; 432 433 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d", 434 subflow, subflow->local_key, subflow->remote_key, 435 data_len); 436 437 return true; 438 } else if (subflow->mp_join) { 439 opts->suboptions = OPTION_MPTCP_MPJ_ACK; 440 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN); 441 *size = TCPOLEN_MPTCP_MPJ_ACK; 442 pr_debug("subflow=%p", subflow); 443 444 schedule_3rdack_retransmission(sk); 445 return true; 446 } 447 return false; 448 } 449 450 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow, 451 struct mptcp_ext *ext) 452 { 453 if (!ext->use_map) { 454 /* RFC6824 requires a DSS mapping with specific values 455 * if DATA_FIN is set but no data payload is mapped 456 */ 457 ext->data_fin = 1; 458 ext->use_map = 1; 459 ext->dsn64 = 1; 460 ext->data_seq = subflow->data_fin_tx_seq; 461 ext->subflow_seq = 0; 462 ext->data_len = 1; 463 } else if (ext->data_seq + ext->data_len == subflow->data_fin_tx_seq) { 464 /* If there's an existing DSS mapping and it is the 465 * final mapping, DATA_FIN consumes 1 additional byte of 466 * mapping space. 467 */ 468 ext->data_fin = 1; 469 ext->data_len++; 470 } 471 } 472 473 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb, 474 unsigned int *size, 475 unsigned int remaining, 476 struct mptcp_out_options *opts) 477 { 478 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 479 unsigned int dss_size = 0; 480 struct mptcp_ext *mpext; 481 struct mptcp_sock *msk; 482 unsigned int ack_size; 483 bool ret = false; 484 u8 tcp_fin; 485 486 if (skb) { 487 mpext = mptcp_get_ext(skb); 488 tcp_fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN; 489 } else { 490 mpext = NULL; 491 tcp_fin = 0; 492 } 493 494 if (!skb || (mpext && mpext->use_map) || tcp_fin) { 495 unsigned int map_size; 496 497 map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64; 498 499 remaining -= map_size; 500 dss_size = map_size; 501 if (mpext) 502 opts->ext_copy = *mpext; 503 504 if (skb && tcp_fin && subflow->data_fin_tx_enable) 505 mptcp_write_data_fin(subflow, &opts->ext_copy); 506 ret = true; 507 } 508 509 /* passive sockets msk will set the 'can_ack' after accept(), even 510 * if the first subflow may have the already the remote key handy 511 */ 512 opts->ext_copy.use_ack = 0; 513 msk = mptcp_sk(subflow->conn); 514 if (!READ_ONCE(msk->can_ack)) { 515 *size = ALIGN(dss_size, 4); 516 return ret; 517 } 518 519 if (subflow->use_64bit_ack) { 520 ack_size = TCPOLEN_MPTCP_DSS_ACK64; 521 opts->ext_copy.data_ack = msk->ack_seq; 522 opts->ext_copy.ack64 = 1; 523 } else { 524 ack_size = TCPOLEN_MPTCP_DSS_ACK32; 525 opts->ext_copy.data_ack32 = (uint32_t)(msk->ack_seq); 526 opts->ext_copy.ack64 = 0; 527 } 528 opts->ext_copy.use_ack = 1; 529 530 /* Add kind/length/subtype/flag overhead if mapping is not populated */ 531 if (dss_size == 0) 532 ack_size += TCPOLEN_MPTCP_DSS_BASE; 533 534 dss_size += ack_size; 535 536 *size = ALIGN(dss_size, 4); 537 return true; 538 } 539 540 static u64 add_addr_generate_hmac(u64 key1, u64 key2, u8 addr_id, 541 struct in_addr *addr) 542 { 543 u8 hmac[MPTCP_ADDR_HMAC_LEN]; 544 u8 msg[7]; 545 546 msg[0] = addr_id; 547 memcpy(&msg[1], &addr->s_addr, 4); 548 msg[5] = 0; 549 msg[6] = 0; 550 551 mptcp_crypto_hmac_sha(key1, key2, msg, 7, hmac); 552 553 return get_unaligned_be64(hmac); 554 } 555 556 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 557 static u64 add_addr6_generate_hmac(u64 key1, u64 key2, u8 addr_id, 558 struct in6_addr *addr) 559 { 560 u8 hmac[MPTCP_ADDR_HMAC_LEN]; 561 u8 msg[19]; 562 563 msg[0] = addr_id; 564 memcpy(&msg[1], &addr->s6_addr, 16); 565 msg[17] = 0; 566 msg[18] = 0; 567 568 mptcp_crypto_hmac_sha(key1, key2, msg, 19, hmac); 569 570 return get_unaligned_be64(hmac); 571 } 572 #endif 573 574 static bool mptcp_established_options_addr(struct sock *sk, 575 unsigned int *size, 576 unsigned int remaining, 577 struct mptcp_out_options *opts) 578 { 579 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 580 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 581 struct mptcp_addr_info saddr; 582 int len; 583 584 if (!mptcp_pm_should_signal(msk) || 585 !(mptcp_pm_addr_signal(msk, remaining, &saddr))) 586 return false; 587 588 len = mptcp_add_addr_len(saddr.family); 589 if (remaining < len) 590 return false; 591 592 *size = len; 593 opts->addr_id = saddr.id; 594 if (saddr.family == AF_INET) { 595 opts->suboptions |= OPTION_MPTCP_ADD_ADDR; 596 opts->addr = saddr.addr; 597 opts->ahmac = add_addr_generate_hmac(msk->local_key, 598 msk->remote_key, 599 opts->addr_id, 600 &opts->addr); 601 } 602 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 603 else if (saddr.family == AF_INET6) { 604 opts->suboptions |= OPTION_MPTCP_ADD_ADDR6; 605 opts->addr6 = saddr.addr6; 606 opts->ahmac = add_addr6_generate_hmac(msk->local_key, 607 msk->remote_key, 608 opts->addr_id, 609 &opts->addr6); 610 } 611 #endif 612 pr_debug("addr_id=%d, ahmac=%llu", opts->addr_id, opts->ahmac); 613 614 return true; 615 } 616 617 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb, 618 unsigned int *size, unsigned int remaining, 619 struct mptcp_out_options *opts) 620 { 621 unsigned int opt_size = 0; 622 bool ret = false; 623 624 opts->suboptions = 0; 625 626 if (mptcp_established_options_mp(sk, skb, &opt_size, remaining, opts)) 627 ret = true; 628 else if (mptcp_established_options_dss(sk, skb, &opt_size, remaining, 629 opts)) 630 ret = true; 631 632 /* we reserved enough space for the above options, and exceeding the 633 * TCP option space would be fatal 634 */ 635 if (WARN_ON_ONCE(opt_size > remaining)) 636 return false; 637 638 *size += opt_size; 639 remaining -= opt_size; 640 if (mptcp_established_options_addr(sk, &opt_size, remaining, opts)) { 641 *size += opt_size; 642 remaining -= opt_size; 643 ret = true; 644 } 645 646 return ret; 647 } 648 649 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size, 650 struct mptcp_out_options *opts) 651 { 652 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 653 654 if (subflow_req->mp_capable) { 655 opts->suboptions = OPTION_MPTCP_MPC_SYNACK; 656 opts->sndr_key = subflow_req->local_key; 657 *size = TCPOLEN_MPTCP_MPC_SYNACK; 658 pr_debug("subflow_req=%p, local_key=%llu", 659 subflow_req, subflow_req->local_key); 660 return true; 661 } else if (subflow_req->mp_join) { 662 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK; 663 opts->backup = subflow_req->backup; 664 opts->join_id = subflow_req->local_id; 665 opts->thmac = subflow_req->thmac; 666 opts->nonce = subflow_req->local_nonce; 667 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u", 668 subflow_req, opts->backup, opts->join_id, 669 opts->thmac, opts->nonce); 670 *size = TCPOLEN_MPTCP_MPJ_SYNACK; 671 return true; 672 } 673 return false; 674 } 675 676 static bool check_fully_established(struct mptcp_sock *msk, struct sock *sk, 677 struct mptcp_subflow_context *subflow, 678 struct sk_buff *skb, 679 struct mptcp_options_received *mp_opt) 680 { 681 /* here we can process OoO, in-window pkts, only in-sequence 4th ack 682 * will make the subflow fully established 683 */ 684 if (likely(subflow->fully_established)) { 685 /* on passive sockets, check for 3rd ack retransmission 686 * note that msk is always set by subflow_syn_recv_sock() 687 * for mp_join subflows 688 */ 689 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 && 690 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq && 691 subflow->mp_join && mp_opt->mp_join && 692 READ_ONCE(msk->pm.server_side)) 693 tcp_send_ack(sk); 694 goto fully_established; 695 } 696 697 /* we should process OoO packets before the first subflow is fully 698 * established, but not expected for MP_JOIN subflows 699 */ 700 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) 701 return subflow->mp_capable; 702 703 if (mp_opt->dss && mp_opt->use_ack) { 704 /* subflows are fully established as soon as we get any 705 * additional ack. 706 */ 707 subflow->fully_established = 1; 708 goto fully_established; 709 } 710 711 /* If the first established packet does not contain MP_CAPABLE + data 712 * then fallback to TCP 713 */ 714 if (!mp_opt->mp_capable) { 715 subflow->mp_capable = 0; 716 tcp_sk(sk)->is_mptcp = 0; 717 return false; 718 } 719 720 if (unlikely(!READ_ONCE(msk->pm.server_side))) 721 pr_warn_once("bogus mpc option on established client sk"); 722 subflow->fully_established = 1; 723 subflow->remote_key = mp_opt->sndr_key; 724 subflow->can_ack = 1; 725 726 fully_established: 727 if (likely(subflow->pm_notified)) 728 return true; 729 730 subflow->pm_notified = 1; 731 if (subflow->mp_join) { 732 clear_3rdack_retransmission(sk); 733 mptcp_pm_subflow_established(msk, subflow); 734 } else { 735 mptcp_pm_fully_established(msk); 736 } 737 return true; 738 } 739 740 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit) 741 { 742 u32 old_ack32, cur_ack32; 743 744 if (use_64bit) 745 return cur_ack; 746 747 old_ack32 = (u32)old_ack; 748 cur_ack32 = (u32)cur_ack; 749 cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32; 750 if (unlikely(before(cur_ack32, old_ack32))) 751 return cur_ack + (1LL << 32); 752 return cur_ack; 753 } 754 755 static void update_una(struct mptcp_sock *msk, 756 struct mptcp_options_received *mp_opt) 757 { 758 u64 new_snd_una, snd_una, old_snd_una = atomic64_read(&msk->snd_una); 759 u64 write_seq = READ_ONCE(msk->write_seq); 760 761 /* avoid ack expansion on update conflict, to reduce the risk of 762 * wrongly expanding to a future ack sequence number, which is way 763 * more dangerous than missing an ack 764 */ 765 new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64); 766 767 /* ACK for data not even sent yet? Ignore. */ 768 if (after64(new_snd_una, write_seq)) 769 new_snd_una = old_snd_una; 770 771 while (after64(new_snd_una, old_snd_una)) { 772 snd_una = old_snd_una; 773 old_snd_una = atomic64_cmpxchg(&msk->snd_una, snd_una, 774 new_snd_una); 775 if (old_snd_una == snd_una) { 776 mptcp_data_acked((struct sock *)msk); 777 break; 778 } 779 } 780 } 781 782 static bool add_addr_hmac_valid(struct mptcp_sock *msk, 783 struct mptcp_options_received *mp_opt) 784 { 785 u64 hmac = 0; 786 787 if (mp_opt->echo) 788 return true; 789 790 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4) 791 hmac = add_addr_generate_hmac(msk->remote_key, 792 msk->local_key, 793 mp_opt->addr_id, &mp_opt->addr); 794 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 795 else 796 hmac = add_addr6_generate_hmac(msk->remote_key, 797 msk->local_key, 798 mp_opt->addr_id, &mp_opt->addr6); 799 #endif 800 801 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n", 802 msk, (unsigned long long)hmac, 803 (unsigned long long)mp_opt->ahmac); 804 805 return hmac == mp_opt->ahmac; 806 } 807 808 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb, 809 struct tcp_options_received *opt_rx) 810 { 811 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 812 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 813 struct mptcp_options_received mp_opt; 814 struct mptcp_ext *mpext; 815 816 mptcp_get_options(skb, &mp_opt); 817 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt)) 818 return; 819 820 if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) { 821 struct mptcp_addr_info addr; 822 823 addr.port = htons(mp_opt.port); 824 addr.id = mp_opt.addr_id; 825 if (mp_opt.family == MPTCP_ADDR_IPVERSION_4) { 826 addr.family = AF_INET; 827 addr.addr = mp_opt.addr; 828 } 829 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 830 else if (mp_opt.family == MPTCP_ADDR_IPVERSION_6) { 831 addr.family = AF_INET6; 832 addr.addr6 = mp_opt.addr6; 833 } 834 #endif 835 if (!mp_opt.echo) 836 mptcp_pm_add_addr_received(msk, &addr); 837 mp_opt.add_addr = 0; 838 } 839 840 if (!mp_opt.dss) 841 return; 842 843 /* we can't wait for recvmsg() to update the ack_seq, otherwise 844 * monodirectional flows will stuck 845 */ 846 if (mp_opt.use_ack) 847 update_una(msk, &mp_opt); 848 849 mpext = skb_ext_add(skb, SKB_EXT_MPTCP); 850 if (!mpext) 851 return; 852 853 memset(mpext, 0, sizeof(*mpext)); 854 855 if (mp_opt.use_map) { 856 if (mp_opt.mpc_map) { 857 /* this is an MP_CAPABLE carrying MPTCP data 858 * we know this map the first chunk of data 859 */ 860 mptcp_crypto_key_sha(subflow->remote_key, NULL, 861 &mpext->data_seq); 862 mpext->data_seq++; 863 mpext->subflow_seq = 1; 864 mpext->dsn64 = 1; 865 mpext->mpc_map = 1; 866 mpext->data_fin = 0; 867 } else { 868 mpext->data_seq = mp_opt.data_seq; 869 mpext->subflow_seq = mp_opt.subflow_seq; 870 mpext->dsn64 = mp_opt.dsn64; 871 mpext->data_fin = mp_opt.data_fin; 872 } 873 mpext->data_len = mp_opt.data_len; 874 mpext->use_map = 1; 875 } 876 } 877 878 void mptcp_write_options(__be32 *ptr, struct mptcp_out_options *opts) 879 { 880 if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK | 881 OPTION_MPTCP_MPC_ACK) & opts->suboptions) { 882 u8 len; 883 884 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) 885 len = TCPOLEN_MPTCP_MPC_SYN; 886 else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) 887 len = TCPOLEN_MPTCP_MPC_SYNACK; 888 else if (opts->ext_copy.data_len) 889 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 890 else 891 len = TCPOLEN_MPTCP_MPC_ACK; 892 893 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len, 894 MPTCP_SUPPORTED_VERSION, 895 MPTCP_CAP_HMAC_SHA256); 896 897 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) & 898 opts->suboptions)) 899 goto mp_capable_done; 900 901 put_unaligned_be64(opts->sndr_key, ptr); 902 ptr += 2; 903 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions)) 904 goto mp_capable_done; 905 906 put_unaligned_be64(opts->rcvr_key, ptr); 907 ptr += 2; 908 if (!opts->ext_copy.data_len) 909 goto mp_capable_done; 910 911 put_unaligned_be32(opts->ext_copy.data_len << 16 | 912 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 913 ptr += 1; 914 } 915 916 mp_capable_done: 917 if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) { 918 if (opts->ahmac) 919 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 920 TCPOLEN_MPTCP_ADD_ADDR, 0, 921 opts->addr_id); 922 else 923 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 924 TCPOLEN_MPTCP_ADD_ADDR_BASE, 925 MPTCP_ADDR_ECHO, 926 opts->addr_id); 927 memcpy((u8 *)ptr, (u8 *)&opts->addr.s_addr, 4); 928 ptr += 1; 929 if (opts->ahmac) { 930 put_unaligned_be64(opts->ahmac, ptr); 931 ptr += 2; 932 } 933 } 934 935 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 936 if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions) { 937 if (opts->ahmac) 938 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 939 TCPOLEN_MPTCP_ADD_ADDR6, 0, 940 opts->addr_id); 941 else 942 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 943 TCPOLEN_MPTCP_ADD_ADDR6_BASE, 944 MPTCP_ADDR_ECHO, 945 opts->addr_id); 946 memcpy((u8 *)ptr, opts->addr6.s6_addr, 16); 947 ptr += 4; 948 if (opts->ahmac) { 949 put_unaligned_be64(opts->ahmac, ptr); 950 ptr += 2; 951 } 952 } 953 #endif 954 955 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) { 956 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR, 957 TCPOLEN_MPTCP_RM_ADDR_BASE, 958 0, opts->rm_id); 959 } 960 961 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) { 962 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 963 TCPOLEN_MPTCP_MPJ_SYN, 964 opts->backup, opts->join_id); 965 put_unaligned_be32(opts->token, ptr); 966 ptr += 1; 967 put_unaligned_be32(opts->nonce, ptr); 968 ptr += 1; 969 } 970 971 if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) { 972 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 973 TCPOLEN_MPTCP_MPJ_SYNACK, 974 opts->backup, opts->join_id); 975 put_unaligned_be64(opts->thmac, ptr); 976 ptr += 2; 977 put_unaligned_be32(opts->nonce, ptr); 978 ptr += 1; 979 } 980 981 if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) { 982 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 983 TCPOLEN_MPTCP_MPJ_ACK, 0, 0); 984 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN); 985 ptr += 5; 986 } 987 988 if (opts->ext_copy.use_ack || opts->ext_copy.use_map) { 989 struct mptcp_ext *mpext = &opts->ext_copy; 990 u8 len = TCPOLEN_MPTCP_DSS_BASE; 991 u8 flags = 0; 992 993 if (mpext->use_ack) { 994 flags = MPTCP_DSS_HAS_ACK; 995 if (mpext->ack64) { 996 len += TCPOLEN_MPTCP_DSS_ACK64; 997 flags |= MPTCP_DSS_ACK64; 998 } else { 999 len += TCPOLEN_MPTCP_DSS_ACK32; 1000 } 1001 } 1002 1003 if (mpext->use_map) { 1004 len += TCPOLEN_MPTCP_DSS_MAP64; 1005 1006 /* Use only 64-bit mapping flags for now, add 1007 * support for optional 32-bit mappings later. 1008 */ 1009 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64; 1010 if (mpext->data_fin) 1011 flags |= MPTCP_DSS_DATA_FIN; 1012 } 1013 1014 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags); 1015 1016 if (mpext->use_ack) { 1017 if (mpext->ack64) { 1018 put_unaligned_be64(mpext->data_ack, ptr); 1019 ptr += 2; 1020 } else { 1021 put_unaligned_be32(mpext->data_ack32, ptr); 1022 ptr += 1; 1023 } 1024 } 1025 1026 if (mpext->use_map) { 1027 put_unaligned_be64(mpext->data_seq, ptr); 1028 ptr += 2; 1029 put_unaligned_be32(mpext->subflow_seq, ptr); 1030 ptr += 1; 1031 put_unaligned_be32(mpext->data_len << 16 | 1032 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1033 } 1034 } 1035 } 1036