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/sha.h> 11 #include <net/tcp.h> 12 #include <net/mptcp.h> 13 #include "protocol.h" 14 15 static bool mptcp_cap_flag_sha256(u8 flags) 16 { 17 return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256; 18 } 19 20 static void mptcp_parse_option(const struct sk_buff *skb, 21 const unsigned char *ptr, int opsize, 22 struct mptcp_options_received *mp_opt) 23 { 24 u8 subtype = *ptr >> 4; 25 int expected_opsize; 26 u8 version; 27 u8 flags; 28 29 switch (subtype) { 30 case MPTCPOPT_MP_CAPABLE: 31 /* strict size checking */ 32 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { 33 if (skb->len > tcp_hdr(skb)->doff << 2) 34 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA; 35 else 36 expected_opsize = TCPOLEN_MPTCP_MPC_ACK; 37 } else { 38 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK) 39 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK; 40 else 41 expected_opsize = TCPOLEN_MPTCP_MPC_SYN; 42 } 43 if (opsize != expected_opsize) 44 break; 45 46 /* try to be gentle vs future versions on the initial syn */ 47 version = *ptr++ & MPTCP_VERSION_MASK; 48 if (opsize != TCPOLEN_MPTCP_MPC_SYN) { 49 if (version != MPTCP_SUPPORTED_VERSION) 50 break; 51 } else if (version < MPTCP_SUPPORTED_VERSION) { 52 break; 53 } 54 55 flags = *ptr++; 56 if (!mptcp_cap_flag_sha256(flags) || 57 (flags & MPTCP_CAP_EXTENSIBILITY)) 58 break; 59 60 /* RFC 6824, Section 3.1: 61 * "For the Checksum Required bit (labeled "A"), if either 62 * host requires the use of checksums, checksums MUST be used. 63 * In other words, the only way for checksums not to be used 64 * is if both hosts in their SYNs set A=0." 65 * 66 * Section 3.3.0: 67 * "If a checksum is not present when its use has been 68 * negotiated, the receiver MUST close the subflow with a RST as 69 * it is considered broken." 70 * 71 * We don't implement DSS checksum - fall back to TCP. 72 */ 73 if (flags & MPTCP_CAP_CHECKSUM_REQD) 74 break; 75 76 mp_opt->mp_capable = 1; 77 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) { 78 mp_opt->sndr_key = get_unaligned_be64(ptr); 79 ptr += 8; 80 } 81 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) { 82 mp_opt->rcvr_key = get_unaligned_be64(ptr); 83 ptr += 8; 84 } 85 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA) { 86 /* Section 3.1.: 87 * "the data parameters in a MP_CAPABLE are semantically 88 * equivalent to those in a DSS option and can be used 89 * interchangeably." 90 */ 91 mp_opt->dss = 1; 92 mp_opt->use_map = 1; 93 mp_opt->mpc_map = 1; 94 mp_opt->data_len = get_unaligned_be16(ptr); 95 ptr += 2; 96 } 97 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d", 98 version, flags, opsize, mp_opt->sndr_key, 99 mp_opt->rcvr_key, mp_opt->data_len); 100 break; 101 102 case MPTCPOPT_MP_JOIN: 103 mp_opt->mp_join = 1; 104 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) { 105 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 106 mp_opt->join_id = *ptr++; 107 mp_opt->token = get_unaligned_be32(ptr); 108 ptr += 4; 109 mp_opt->nonce = get_unaligned_be32(ptr); 110 ptr += 4; 111 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u", 112 mp_opt->backup, mp_opt->join_id, 113 mp_opt->token, mp_opt->nonce); 114 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) { 115 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 116 mp_opt->join_id = *ptr++; 117 mp_opt->thmac = get_unaligned_be64(ptr); 118 ptr += 8; 119 mp_opt->nonce = get_unaligned_be32(ptr); 120 ptr += 4; 121 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u", 122 mp_opt->backup, mp_opt->join_id, 123 mp_opt->thmac, mp_opt->nonce); 124 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) { 125 ptr += 2; 126 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN); 127 pr_debug("MP_JOIN hmac"); 128 } else { 129 pr_warn("MP_JOIN bad option size"); 130 mp_opt->mp_join = 0; 131 } 132 break; 133 134 case MPTCPOPT_DSS: 135 pr_debug("DSS"); 136 ptr++; 137 138 /* we must clear 'mpc_map' be able to detect MP_CAPABLE 139 * map vs DSS map in mptcp_incoming_options(), and reconstruct 140 * map info accordingly 141 */ 142 mp_opt->mpc_map = 0; 143 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK; 144 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0; 145 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0; 146 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0; 147 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0; 148 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK); 149 150 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d", 151 mp_opt->data_fin, mp_opt->dsn64, 152 mp_opt->use_map, mp_opt->ack64, 153 mp_opt->use_ack); 154 155 expected_opsize = TCPOLEN_MPTCP_DSS_BASE; 156 157 if (mp_opt->use_ack) { 158 if (mp_opt->ack64) 159 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64; 160 else 161 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32; 162 } 163 164 if (mp_opt->use_map) { 165 if (mp_opt->dsn64) 166 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64; 167 else 168 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32; 169 } 170 171 /* RFC 6824, Section 3.3: 172 * If a checksum is present, but its use had 173 * not been negotiated in the MP_CAPABLE handshake, 174 * the checksum field MUST be ignored. 175 */ 176 if (opsize != expected_opsize && 177 opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) 178 break; 179 180 mp_opt->dss = 1; 181 182 if (mp_opt->use_ack) { 183 if (mp_opt->ack64) { 184 mp_opt->data_ack = get_unaligned_be64(ptr); 185 ptr += 8; 186 } else { 187 mp_opt->data_ack = get_unaligned_be32(ptr); 188 ptr += 4; 189 } 190 191 pr_debug("data_ack=%llu", mp_opt->data_ack); 192 } 193 194 if (mp_opt->use_map) { 195 if (mp_opt->dsn64) { 196 mp_opt->data_seq = get_unaligned_be64(ptr); 197 ptr += 8; 198 } else { 199 mp_opt->data_seq = get_unaligned_be32(ptr); 200 ptr += 4; 201 } 202 203 mp_opt->subflow_seq = get_unaligned_be32(ptr); 204 ptr += 4; 205 206 mp_opt->data_len = get_unaligned_be16(ptr); 207 ptr += 2; 208 209 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u", 210 mp_opt->data_seq, mp_opt->subflow_seq, 211 mp_opt->data_len); 212 } 213 214 break; 215 216 case MPTCPOPT_ADD_ADDR: 217 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO; 218 if (!mp_opt->echo) { 219 if (opsize == TCPOLEN_MPTCP_ADD_ADDR || 220 opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT) 221 mp_opt->family = MPTCP_ADDR_IPVERSION_4; 222 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 223 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 || 224 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT) 225 mp_opt->family = MPTCP_ADDR_IPVERSION_6; 226 #endif 227 else 228 break; 229 } else { 230 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE || 231 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) 232 mp_opt->family = MPTCP_ADDR_IPVERSION_4; 233 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 234 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE || 235 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) 236 mp_opt->family = MPTCP_ADDR_IPVERSION_6; 237 #endif 238 else 239 break; 240 } 241 242 mp_opt->add_addr = 1; 243 mp_opt->port = 0; 244 mp_opt->addr_id = *ptr++; 245 pr_debug("ADD_ADDR: id=%d", mp_opt->addr_id); 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 break; 271 272 case MPTCPOPT_RM_ADDR: 273 if (opsize != TCPOLEN_MPTCP_RM_ADDR_BASE) 274 break; 275 276 ptr++; 277 278 mp_opt->rm_addr = 1; 279 mp_opt->rm_id = *ptr++; 280 pr_debug("RM_ADDR: id=%d", mp_opt->rm_id); 281 break; 282 283 default: 284 break; 285 } 286 } 287 288 void mptcp_get_options(const struct sk_buff *skb, 289 struct mptcp_options_received *mp_opt) 290 { 291 const struct tcphdr *th = tcp_hdr(skb); 292 const unsigned char *ptr; 293 int length; 294 295 /* initialize option status */ 296 mp_opt->mp_capable = 0; 297 mp_opt->mp_join = 0; 298 mp_opt->add_addr = 0; 299 mp_opt->rm_addr = 0; 300 mp_opt->dss = 0; 301 302 length = (th->doff * 4) - sizeof(struct tcphdr); 303 ptr = (const unsigned char *)(th + 1); 304 305 while (length > 0) { 306 int opcode = *ptr++; 307 int opsize; 308 309 switch (opcode) { 310 case TCPOPT_EOL: 311 return; 312 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */ 313 length--; 314 continue; 315 default: 316 opsize = *ptr++; 317 if (opsize < 2) /* "silly options" */ 318 return; 319 if (opsize > length) 320 return; /* don't parse partial options */ 321 if (opcode == TCPOPT_MPTCP) 322 mptcp_parse_option(skb, ptr, opsize, mp_opt); 323 ptr += opsize - 2; 324 length -= opsize; 325 } 326 } 327 } 328 329 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb, 330 unsigned int *size, struct mptcp_out_options *opts) 331 { 332 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 333 334 /* we will use snd_isn to detect first pkt [re]transmission 335 * in mptcp_established_options_mp() 336 */ 337 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq; 338 if (subflow->request_mptcp) { 339 opts->suboptions = OPTION_MPTCP_MPC_SYN; 340 *size = TCPOLEN_MPTCP_MPC_SYN; 341 return true; 342 } else if (subflow->request_join) { 343 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token, 344 subflow->local_nonce); 345 opts->suboptions = OPTION_MPTCP_MPJ_SYN; 346 opts->join_id = subflow->local_id; 347 opts->token = subflow->remote_token; 348 opts->nonce = subflow->local_nonce; 349 opts->backup = subflow->request_bkup; 350 *size = TCPOLEN_MPTCP_MPJ_SYN; 351 return true; 352 } 353 return false; 354 } 355 356 /* MP_JOIN client subflow must wait for 4th ack before sending any data: 357 * TCP can't schedule delack timer before the subflow is fully established. 358 * MPTCP uses the delack timer to do 3rd ack retransmissions 359 */ 360 static void schedule_3rdack_retransmission(struct sock *sk) 361 { 362 struct inet_connection_sock *icsk = inet_csk(sk); 363 struct tcp_sock *tp = tcp_sk(sk); 364 unsigned long timeout; 365 366 /* reschedule with a timeout above RTT, as we must look only for drop */ 367 if (tp->srtt_us) 368 timeout = tp->srtt_us << 1; 369 else 370 timeout = TCP_TIMEOUT_INIT; 371 372 WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER); 373 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER; 374 icsk->icsk_ack.timeout = timeout; 375 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout); 376 } 377 378 static void clear_3rdack_retransmission(struct sock *sk) 379 { 380 struct inet_connection_sock *icsk = inet_csk(sk); 381 382 sk_stop_timer(sk, &icsk->icsk_delack_timer); 383 icsk->icsk_ack.timeout = 0; 384 icsk->icsk_ack.ato = 0; 385 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER); 386 } 387 388 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb, 389 unsigned int *size, 390 unsigned int remaining, 391 struct mptcp_out_options *opts) 392 { 393 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 394 struct mptcp_ext *mpext; 395 unsigned int data_len; 396 397 /* When skb is not available, we better over-estimate the emitted 398 * options len. A full DSS option (28 bytes) is longer than 399 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so 400 * tell the caller to defer the estimate to 401 * mptcp_established_options_dss(), which will reserve enough space. 402 */ 403 if (!skb) 404 return false; 405 406 /* MPC/MPJ needed only on 3rd ack packet */ 407 if (subflow->fully_established || 408 subflow->snd_isn != TCP_SKB_CB(skb)->seq) 409 return false; 410 411 if (subflow->mp_capable) { 412 mpext = mptcp_get_ext(skb); 413 data_len = mpext ? mpext->data_len : 0; 414 415 /* we will check ext_copy.data_len in mptcp_write_options() to 416 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and 417 * TCPOLEN_MPTCP_MPC_ACK 418 */ 419 opts->ext_copy.data_len = data_len; 420 opts->suboptions = OPTION_MPTCP_MPC_ACK; 421 opts->sndr_key = subflow->local_key; 422 opts->rcvr_key = subflow->remote_key; 423 424 /* Section 3.1. 425 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK 426 * packets that start the first subflow of an MPTCP connection, 427 * as well as the first packet that carries data 428 */ 429 if (data_len > 0) 430 *size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4); 431 else 432 *size = TCPOLEN_MPTCP_MPC_ACK; 433 434 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d", 435 subflow, subflow->local_key, subflow->remote_key, 436 data_len); 437 438 return true; 439 } else if (subflow->mp_join) { 440 opts->suboptions = OPTION_MPTCP_MPJ_ACK; 441 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN); 442 *size = TCPOLEN_MPTCP_MPJ_ACK; 443 pr_debug("subflow=%p", subflow); 444 445 schedule_3rdack_retransmission(sk); 446 return true; 447 } 448 return false; 449 } 450 451 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow, 452 struct sk_buff *skb, struct mptcp_ext *ext) 453 { 454 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq); 455 456 if (!ext->use_map || !skb->len) { 457 /* RFC6824 requires a DSS mapping with specific values 458 * if DATA_FIN is set but no data payload is mapped 459 */ 460 ext->data_fin = 1; 461 ext->use_map = 1; 462 ext->dsn64 = 1; 463 /* The write_seq value has already been incremented, so 464 * the actual sequence number for the DATA_FIN is one less. 465 */ 466 ext->data_seq = data_fin_tx_seq - 1; 467 ext->subflow_seq = 0; 468 ext->data_len = 1; 469 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) { 470 /* If there's an existing DSS mapping and it is the 471 * final mapping, DATA_FIN consumes 1 additional byte of 472 * mapping space. 473 */ 474 ext->data_fin = 1; 475 ext->data_len++; 476 } 477 } 478 479 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb, 480 unsigned int *size, 481 unsigned int remaining, 482 struct mptcp_out_options *opts) 483 { 484 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 485 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 486 unsigned int dss_size = 0; 487 u64 snd_data_fin_enable; 488 struct mptcp_ext *mpext; 489 unsigned int ack_size; 490 bool ret = false; 491 492 mpext = skb ? mptcp_get_ext(skb) : NULL; 493 snd_data_fin_enable = READ_ONCE(msk->snd_data_fin_enable); 494 495 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) { 496 unsigned int map_size; 497 498 map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64; 499 500 remaining -= map_size; 501 dss_size = map_size; 502 if (mpext) 503 opts->ext_copy = *mpext; 504 505 if (skb && snd_data_fin_enable) 506 mptcp_write_data_fin(subflow, skb, &opts->ext_copy); 507 ret = true; 508 } 509 510 /* passive sockets msk will set the 'can_ack' after accept(), even 511 * if the first subflow may have the already the remote key handy 512 */ 513 opts->ext_copy.use_ack = 0; 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[SHA256_DIGEST_SIZE]; 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[SHA256_DIGEST_SIZE - sizeof(u64)]); 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[SHA256_DIGEST_SIZE]; 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[SHA256_DIGEST_SIZE - sizeof(u64)]); 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 (unlikely(mptcp_check_fallback(sk))) 627 return false; 628 629 if (mptcp_established_options_mp(sk, skb, &opt_size, remaining, opts)) 630 ret = true; 631 else if (mptcp_established_options_dss(sk, skb, &opt_size, remaining, 632 opts)) 633 ret = true; 634 635 /* we reserved enough space for the above options, and exceeding the 636 * TCP option space would be fatal 637 */ 638 if (WARN_ON_ONCE(opt_size > remaining)) 639 return false; 640 641 *size += opt_size; 642 remaining -= opt_size; 643 if (mptcp_established_options_addr(sk, &opt_size, remaining, opts)) { 644 *size += opt_size; 645 remaining -= opt_size; 646 ret = true; 647 } 648 649 return ret; 650 } 651 652 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size, 653 struct mptcp_out_options *opts) 654 { 655 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 656 657 if (subflow_req->mp_capable) { 658 opts->suboptions = OPTION_MPTCP_MPC_SYNACK; 659 opts->sndr_key = subflow_req->local_key; 660 *size = TCPOLEN_MPTCP_MPC_SYNACK; 661 pr_debug("subflow_req=%p, local_key=%llu", 662 subflow_req, subflow_req->local_key); 663 return true; 664 } else if (subflow_req->mp_join) { 665 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK; 666 opts->backup = subflow_req->backup; 667 opts->join_id = subflow_req->local_id; 668 opts->thmac = subflow_req->thmac; 669 opts->nonce = subflow_req->local_nonce; 670 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u", 671 subflow_req, opts->backup, opts->join_id, 672 opts->thmac, opts->nonce); 673 *size = TCPOLEN_MPTCP_MPJ_SYNACK; 674 return true; 675 } 676 return false; 677 } 678 679 static bool check_fully_established(struct mptcp_sock *msk, struct sock *sk, 680 struct mptcp_subflow_context *subflow, 681 struct sk_buff *skb, 682 struct mptcp_options_received *mp_opt) 683 { 684 /* here we can process OoO, in-window pkts, only in-sequence 4th ack 685 * will make the subflow fully established 686 */ 687 if (likely(subflow->fully_established)) { 688 /* on passive sockets, check for 3rd ack retransmission 689 * note that msk is always set by subflow_syn_recv_sock() 690 * for mp_join subflows 691 */ 692 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 && 693 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq && 694 subflow->mp_join && mp_opt->mp_join && 695 READ_ONCE(msk->pm.server_side)) 696 tcp_send_ack(sk); 697 goto fully_established; 698 } 699 700 /* we should process OoO packets before the first subflow is fully 701 * established, but not expected for MP_JOIN subflows 702 */ 703 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) 704 return subflow->mp_capable; 705 706 if (mp_opt->dss && mp_opt->use_ack) { 707 /* subflows are fully established as soon as we get any 708 * additional ack. 709 */ 710 subflow->fully_established = 1; 711 WRITE_ONCE(msk->fully_established, true); 712 goto fully_established; 713 } 714 715 /* If the first established packet does not contain MP_CAPABLE + data 716 * then fallback to TCP 717 */ 718 if (!mp_opt->mp_capable) { 719 subflow->mp_capable = 0; 720 pr_fallback(msk); 721 __mptcp_do_fallback(msk); 722 return false; 723 } 724 725 if (unlikely(!READ_ONCE(msk->pm.server_side))) 726 pr_warn_once("bogus mpc option on established client sk"); 727 mptcp_subflow_fully_established(subflow, mp_opt); 728 729 fully_established: 730 if (likely(subflow->pm_notified)) 731 return true; 732 733 subflow->pm_notified = 1; 734 if (subflow->mp_join) { 735 clear_3rdack_retransmission(sk); 736 mptcp_pm_subflow_established(msk, subflow); 737 } else { 738 mptcp_pm_fully_established(msk); 739 } 740 return true; 741 } 742 743 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit) 744 { 745 u32 old_ack32, cur_ack32; 746 747 if (use_64bit) 748 return cur_ack; 749 750 old_ack32 = (u32)old_ack; 751 cur_ack32 = (u32)cur_ack; 752 cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32; 753 if (unlikely(before(cur_ack32, old_ack32))) 754 return cur_ack + (1LL << 32); 755 return cur_ack; 756 } 757 758 static void update_una(struct mptcp_sock *msk, 759 struct mptcp_options_received *mp_opt) 760 { 761 u64 new_snd_una, snd_una, old_snd_una = atomic64_read(&msk->snd_una); 762 u64 write_seq = READ_ONCE(msk->write_seq); 763 764 /* avoid ack expansion on update conflict, to reduce the risk of 765 * wrongly expanding to a future ack sequence number, which is way 766 * more dangerous than missing an ack 767 */ 768 new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64); 769 770 /* ACK for data not even sent yet? Ignore. */ 771 if (after64(new_snd_una, write_seq)) 772 new_snd_una = old_snd_una; 773 774 while (after64(new_snd_una, old_snd_una)) { 775 snd_una = old_snd_una; 776 old_snd_una = atomic64_cmpxchg(&msk->snd_una, snd_una, 777 new_snd_una); 778 if (old_snd_una == snd_una) { 779 mptcp_data_acked((struct sock *)msk); 780 break; 781 } 782 } 783 } 784 785 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq) 786 { 787 /* Skip if DATA_FIN was already received. 788 * If updating simultaneously with the recvmsg loop, values 789 * should match. If they mismatch, the peer is misbehaving and 790 * we will prefer the most recent information. 791 */ 792 if (READ_ONCE(msk->rcv_data_fin) || !READ_ONCE(msk->first)) 793 return false; 794 795 WRITE_ONCE(msk->rcv_data_fin_seq, data_fin_seq); 796 WRITE_ONCE(msk->rcv_data_fin, 1); 797 798 return true; 799 } 800 801 static bool add_addr_hmac_valid(struct mptcp_sock *msk, 802 struct mptcp_options_received *mp_opt) 803 { 804 u64 hmac = 0; 805 806 if (mp_opt->echo) 807 return true; 808 809 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4) 810 hmac = add_addr_generate_hmac(msk->remote_key, 811 msk->local_key, 812 mp_opt->addr_id, &mp_opt->addr); 813 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 814 else 815 hmac = add_addr6_generate_hmac(msk->remote_key, 816 msk->local_key, 817 mp_opt->addr_id, &mp_opt->addr6); 818 #endif 819 820 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n", 821 msk, (unsigned long long)hmac, 822 (unsigned long long)mp_opt->ahmac); 823 824 return hmac == mp_opt->ahmac; 825 } 826 827 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb, 828 struct tcp_options_received *opt_rx) 829 { 830 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 831 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 832 struct mptcp_options_received mp_opt; 833 struct mptcp_ext *mpext; 834 835 if (__mptcp_check_fallback(msk)) 836 return; 837 838 mptcp_get_options(skb, &mp_opt); 839 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt)) 840 return; 841 842 if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) { 843 struct mptcp_addr_info addr; 844 845 addr.port = htons(mp_opt.port); 846 addr.id = mp_opt.addr_id; 847 if (mp_opt.family == MPTCP_ADDR_IPVERSION_4) { 848 addr.family = AF_INET; 849 addr.addr = mp_opt.addr; 850 } 851 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 852 else if (mp_opt.family == MPTCP_ADDR_IPVERSION_6) { 853 addr.family = AF_INET6; 854 addr.addr6 = mp_opt.addr6; 855 } 856 #endif 857 if (!mp_opt.echo) 858 mptcp_pm_add_addr_received(msk, &addr); 859 mp_opt.add_addr = 0; 860 } 861 862 if (!mp_opt.dss) 863 return; 864 865 /* we can't wait for recvmsg() to update the ack_seq, otherwise 866 * monodirectional flows will stuck 867 */ 868 if (mp_opt.use_ack) 869 update_una(msk, &mp_opt); 870 871 /* Zero-data-length packets are dropped by the caller and not 872 * propagated to the MPTCP layer, so the skb extension does not 873 * need to be allocated or populated. DATA_FIN information, if 874 * present, needs to be updated here before the skb is freed. 875 */ 876 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) { 877 if (mp_opt.data_fin && mp_opt.data_len == 1 && 878 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq) && 879 schedule_work(&msk->work)) 880 sock_hold(subflow->conn); 881 882 return; 883 } 884 885 mpext = skb_ext_add(skb, SKB_EXT_MPTCP); 886 if (!mpext) 887 return; 888 889 memset(mpext, 0, sizeof(*mpext)); 890 891 if (mp_opt.use_map) { 892 if (mp_opt.mpc_map) { 893 /* this is an MP_CAPABLE carrying MPTCP data 894 * we know this map the first chunk of data 895 */ 896 mptcp_crypto_key_sha(subflow->remote_key, NULL, 897 &mpext->data_seq); 898 mpext->data_seq++; 899 mpext->subflow_seq = 1; 900 mpext->dsn64 = 1; 901 mpext->mpc_map = 1; 902 mpext->data_fin = 0; 903 } else { 904 mpext->data_seq = mp_opt.data_seq; 905 mpext->subflow_seq = mp_opt.subflow_seq; 906 mpext->dsn64 = mp_opt.dsn64; 907 mpext->data_fin = mp_opt.data_fin; 908 } 909 mpext->data_len = mp_opt.data_len; 910 mpext->use_map = 1; 911 } 912 } 913 914 void mptcp_write_options(__be32 *ptr, struct mptcp_out_options *opts) 915 { 916 if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK | 917 OPTION_MPTCP_MPC_ACK) & opts->suboptions) { 918 u8 len; 919 920 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) 921 len = TCPOLEN_MPTCP_MPC_SYN; 922 else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) 923 len = TCPOLEN_MPTCP_MPC_SYNACK; 924 else if (opts->ext_copy.data_len) 925 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 926 else 927 len = TCPOLEN_MPTCP_MPC_ACK; 928 929 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len, 930 MPTCP_SUPPORTED_VERSION, 931 MPTCP_CAP_HMAC_SHA256); 932 933 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) & 934 opts->suboptions)) 935 goto mp_capable_done; 936 937 put_unaligned_be64(opts->sndr_key, ptr); 938 ptr += 2; 939 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions)) 940 goto mp_capable_done; 941 942 put_unaligned_be64(opts->rcvr_key, ptr); 943 ptr += 2; 944 if (!opts->ext_copy.data_len) 945 goto mp_capable_done; 946 947 put_unaligned_be32(opts->ext_copy.data_len << 16 | 948 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 949 ptr += 1; 950 } 951 952 mp_capable_done: 953 if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) { 954 if (opts->ahmac) 955 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 956 TCPOLEN_MPTCP_ADD_ADDR, 0, 957 opts->addr_id); 958 else 959 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 960 TCPOLEN_MPTCP_ADD_ADDR_BASE, 961 MPTCP_ADDR_ECHO, 962 opts->addr_id); 963 memcpy((u8 *)ptr, (u8 *)&opts->addr.s_addr, 4); 964 ptr += 1; 965 if (opts->ahmac) { 966 put_unaligned_be64(opts->ahmac, ptr); 967 ptr += 2; 968 } 969 } 970 971 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 972 if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions) { 973 if (opts->ahmac) 974 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 975 TCPOLEN_MPTCP_ADD_ADDR6, 0, 976 opts->addr_id); 977 else 978 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 979 TCPOLEN_MPTCP_ADD_ADDR6_BASE, 980 MPTCP_ADDR_ECHO, 981 opts->addr_id); 982 memcpy((u8 *)ptr, opts->addr6.s6_addr, 16); 983 ptr += 4; 984 if (opts->ahmac) { 985 put_unaligned_be64(opts->ahmac, ptr); 986 ptr += 2; 987 } 988 } 989 #endif 990 991 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) { 992 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR, 993 TCPOLEN_MPTCP_RM_ADDR_BASE, 994 0, opts->rm_id); 995 } 996 997 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) { 998 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 999 TCPOLEN_MPTCP_MPJ_SYN, 1000 opts->backup, opts->join_id); 1001 put_unaligned_be32(opts->token, ptr); 1002 ptr += 1; 1003 put_unaligned_be32(opts->nonce, ptr); 1004 ptr += 1; 1005 } 1006 1007 if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) { 1008 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1009 TCPOLEN_MPTCP_MPJ_SYNACK, 1010 opts->backup, opts->join_id); 1011 put_unaligned_be64(opts->thmac, ptr); 1012 ptr += 2; 1013 put_unaligned_be32(opts->nonce, ptr); 1014 ptr += 1; 1015 } 1016 1017 if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) { 1018 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1019 TCPOLEN_MPTCP_MPJ_ACK, 0, 0); 1020 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN); 1021 ptr += 5; 1022 } 1023 1024 if (opts->ext_copy.use_ack || opts->ext_copy.use_map) { 1025 struct mptcp_ext *mpext = &opts->ext_copy; 1026 u8 len = TCPOLEN_MPTCP_DSS_BASE; 1027 u8 flags = 0; 1028 1029 if (mpext->use_ack) { 1030 flags = MPTCP_DSS_HAS_ACK; 1031 if (mpext->ack64) { 1032 len += TCPOLEN_MPTCP_DSS_ACK64; 1033 flags |= MPTCP_DSS_ACK64; 1034 } else { 1035 len += TCPOLEN_MPTCP_DSS_ACK32; 1036 } 1037 } 1038 1039 if (mpext->use_map) { 1040 len += TCPOLEN_MPTCP_DSS_MAP64; 1041 1042 /* Use only 64-bit mapping flags for now, add 1043 * support for optional 32-bit mappings later. 1044 */ 1045 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64; 1046 if (mpext->data_fin) 1047 flags |= MPTCP_DSS_DATA_FIN; 1048 } 1049 1050 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags); 1051 1052 if (mpext->use_ack) { 1053 if (mpext->ack64) { 1054 put_unaligned_be64(mpext->data_ack, ptr); 1055 ptr += 2; 1056 } else { 1057 put_unaligned_be32(mpext->data_ack32, ptr); 1058 ptr += 1; 1059 } 1060 } 1061 1062 if (mpext->use_map) { 1063 put_unaligned_be64(mpext->data_seq, ptr); 1064 ptr += 2; 1065 put_unaligned_be32(mpext->subflow_seq, ptr); 1066 ptr += 1; 1067 put_unaligned_be32(mpext->data_len << 16 | 1068 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1069 } 1070 } 1071 } 1072