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 #include "mib.h" 15 16 static bool mptcp_cap_flag_sha256(u8 flags) 17 { 18 return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256; 19 } 20 21 static void mptcp_parse_option(const struct sk_buff *skb, 22 const unsigned char *ptr, int opsize, 23 struct mptcp_options_received *mp_opt) 24 { 25 u8 subtype = *ptr >> 4; 26 int expected_opsize; 27 u8 version; 28 u8 flags; 29 30 switch (subtype) { 31 case MPTCPOPT_MP_CAPABLE: 32 /* strict size checking */ 33 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { 34 if (skb->len > tcp_hdr(skb)->doff << 2) 35 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA; 36 else 37 expected_opsize = TCPOLEN_MPTCP_MPC_ACK; 38 } else { 39 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK) 40 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK; 41 else 42 expected_opsize = TCPOLEN_MPTCP_MPC_SYN; 43 } 44 if (opsize != expected_opsize) 45 break; 46 47 /* try to be gentle vs future versions on the initial syn */ 48 version = *ptr++ & MPTCP_VERSION_MASK; 49 if (opsize != TCPOLEN_MPTCP_MPC_SYN) { 50 if (version != MPTCP_SUPPORTED_VERSION) 51 break; 52 } else if (version < MPTCP_SUPPORTED_VERSION) { 53 break; 54 } 55 56 flags = *ptr++; 57 if (!mptcp_cap_flag_sha256(flags) || 58 (flags & MPTCP_CAP_EXTENSIBILITY)) 59 break; 60 61 /* RFC 6824, Section 3.1: 62 * "For the Checksum Required bit (labeled "A"), if either 63 * host requires the use of checksums, checksums MUST be used. 64 * In other words, the only way for checksums not to be used 65 * is if both hosts in their SYNs set A=0." 66 * 67 * Section 3.3.0: 68 * "If a checksum is not present when its use has been 69 * negotiated, the receiver MUST close the subflow with a RST as 70 * it is considered broken." 71 * 72 * We don't implement DSS checksum - fall back to TCP. 73 */ 74 if (flags & MPTCP_CAP_CHECKSUM_REQD) 75 break; 76 77 mp_opt->mp_capable = 1; 78 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) { 79 mp_opt->sndr_key = get_unaligned_be64(ptr); 80 ptr += 8; 81 } 82 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) { 83 mp_opt->rcvr_key = get_unaligned_be64(ptr); 84 ptr += 8; 85 } 86 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA) { 87 /* Section 3.1.: 88 * "the data parameters in a MP_CAPABLE are semantically 89 * equivalent to those in a DSS option and can be used 90 * interchangeably." 91 */ 92 mp_opt->dss = 1; 93 mp_opt->use_map = 1; 94 mp_opt->mpc_map = 1; 95 mp_opt->data_len = get_unaligned_be16(ptr); 96 ptr += 2; 97 } 98 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d", 99 version, flags, opsize, mp_opt->sndr_key, 100 mp_opt->rcvr_key, mp_opt->data_len); 101 break; 102 103 case MPTCPOPT_MP_JOIN: 104 mp_opt->mp_join = 1; 105 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) { 106 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 107 mp_opt->join_id = *ptr++; 108 mp_opt->token = get_unaligned_be32(ptr); 109 ptr += 4; 110 mp_opt->nonce = get_unaligned_be32(ptr); 111 ptr += 4; 112 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u", 113 mp_opt->backup, mp_opt->join_id, 114 mp_opt->token, mp_opt->nonce); 115 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) { 116 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 117 mp_opt->join_id = *ptr++; 118 mp_opt->thmac = get_unaligned_be64(ptr); 119 ptr += 8; 120 mp_opt->nonce = get_unaligned_be32(ptr); 121 ptr += 4; 122 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u", 123 mp_opt->backup, mp_opt->join_id, 124 mp_opt->thmac, mp_opt->nonce); 125 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) { 126 ptr += 2; 127 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN); 128 pr_debug("MP_JOIN hmac"); 129 } else { 130 pr_warn("MP_JOIN bad option size"); 131 mp_opt->mp_join = 0; 132 } 133 break; 134 135 case MPTCPOPT_DSS: 136 pr_debug("DSS"); 137 ptr++; 138 139 /* we must clear 'mpc_map' be able to detect MP_CAPABLE 140 * map vs DSS map in mptcp_incoming_options(), and reconstruct 141 * map info accordingly 142 */ 143 mp_opt->mpc_map = 0; 144 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK; 145 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0; 146 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0; 147 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0; 148 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0; 149 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK); 150 151 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d", 152 mp_opt->data_fin, mp_opt->dsn64, 153 mp_opt->use_map, mp_opt->ack64, 154 mp_opt->use_ack); 155 156 expected_opsize = TCPOLEN_MPTCP_DSS_BASE; 157 158 if (mp_opt->use_ack) { 159 if (mp_opt->ack64) 160 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64; 161 else 162 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32; 163 } 164 165 if (mp_opt->use_map) { 166 if (mp_opt->dsn64) 167 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64; 168 else 169 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32; 170 } 171 172 /* RFC 6824, Section 3.3: 173 * If a checksum is present, but its use had 174 * not been negotiated in the MP_CAPABLE handshake, 175 * the checksum field MUST be ignored. 176 */ 177 if (opsize != expected_opsize && 178 opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) 179 break; 180 181 mp_opt->dss = 1; 182 183 if (mp_opt->use_ack) { 184 if (mp_opt->ack64) { 185 mp_opt->data_ack = get_unaligned_be64(ptr); 186 ptr += 8; 187 } else { 188 mp_opt->data_ack = get_unaligned_be32(ptr); 189 ptr += 4; 190 } 191 192 pr_debug("data_ack=%llu", mp_opt->data_ack); 193 } 194 195 if (mp_opt->use_map) { 196 if (mp_opt->dsn64) { 197 mp_opt->data_seq = get_unaligned_be64(ptr); 198 ptr += 8; 199 } else { 200 mp_opt->data_seq = get_unaligned_be32(ptr); 201 ptr += 4; 202 } 203 204 mp_opt->subflow_seq = get_unaligned_be32(ptr); 205 ptr += 4; 206 207 mp_opt->data_len = get_unaligned_be16(ptr); 208 ptr += 2; 209 210 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u", 211 mp_opt->data_seq, mp_opt->subflow_seq, 212 mp_opt->data_len); 213 } 214 215 break; 216 217 case MPTCPOPT_ADD_ADDR: 218 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO; 219 if (!mp_opt->echo) { 220 if (opsize == TCPOLEN_MPTCP_ADD_ADDR || 221 opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT) 222 mp_opt->family = MPTCP_ADDR_IPVERSION_4; 223 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 224 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 || 225 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT) 226 mp_opt->family = MPTCP_ADDR_IPVERSION_6; 227 #endif 228 else 229 break; 230 } else { 231 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE || 232 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) 233 mp_opt->family = MPTCP_ADDR_IPVERSION_4; 234 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 235 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE || 236 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) 237 mp_opt->family = MPTCP_ADDR_IPVERSION_6; 238 #endif 239 else 240 break; 241 } 242 243 mp_opt->add_addr = 1; 244 mp_opt->addr_id = *ptr++; 245 pr_debug("ADD_ADDR%s: id=%d, echo=%d", 246 (mp_opt->family == MPTCP_ADDR_IPVERSION_6) ? "6" : "", 247 mp_opt->addr_id, mp_opt->echo); 248 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4) { 249 memcpy((u8 *)&mp_opt->addr.s_addr, (u8 *)ptr, 4); 250 ptr += 4; 251 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT || 252 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) { 253 mp_opt->port = get_unaligned_be16(ptr); 254 ptr += 2; 255 } 256 } 257 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 258 else { 259 memcpy(mp_opt->addr6.s6_addr, (u8 *)ptr, 16); 260 ptr += 16; 261 if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT || 262 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) { 263 mp_opt->port = get_unaligned_be16(ptr); 264 ptr += 2; 265 } 266 } 267 #endif 268 if (!mp_opt->echo) { 269 mp_opt->ahmac = get_unaligned_be64(ptr); 270 ptr += 8; 271 } 272 break; 273 274 case MPTCPOPT_RM_ADDR: 275 if (opsize != TCPOLEN_MPTCP_RM_ADDR_BASE) 276 break; 277 278 ptr++; 279 280 mp_opt->rm_addr = 1; 281 mp_opt->rm_id = *ptr++; 282 pr_debug("RM_ADDR: id=%d", mp_opt->rm_id); 283 break; 284 285 default: 286 break; 287 } 288 } 289 290 void mptcp_get_options(const struct sk_buff *skb, 291 struct mptcp_options_received *mp_opt) 292 { 293 const struct tcphdr *th = tcp_hdr(skb); 294 const unsigned char *ptr; 295 int length; 296 297 /* initialize option status */ 298 mp_opt->mp_capable = 0; 299 mp_opt->mp_join = 0; 300 mp_opt->add_addr = 0; 301 mp_opt->ahmac = 0; 302 mp_opt->port = 0; 303 mp_opt->rm_addr = 0; 304 mp_opt->dss = 0; 305 306 length = (th->doff * 4) - sizeof(struct tcphdr); 307 ptr = (const unsigned char *)(th + 1); 308 309 while (length > 0) { 310 int opcode = *ptr++; 311 int opsize; 312 313 switch (opcode) { 314 case TCPOPT_EOL: 315 return; 316 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */ 317 length--; 318 continue; 319 default: 320 opsize = *ptr++; 321 if (opsize < 2) /* "silly options" */ 322 return; 323 if (opsize > length) 324 return; /* don't parse partial options */ 325 if (opcode == TCPOPT_MPTCP) 326 mptcp_parse_option(skb, ptr, opsize, mp_opt); 327 ptr += opsize - 2; 328 length -= opsize; 329 } 330 } 331 } 332 333 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb, 334 unsigned int *size, struct mptcp_out_options *opts) 335 { 336 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 337 338 /* we will use snd_isn to detect first pkt [re]transmission 339 * in mptcp_established_options_mp() 340 */ 341 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq; 342 if (subflow->request_mptcp) { 343 opts->suboptions = OPTION_MPTCP_MPC_SYN; 344 *size = TCPOLEN_MPTCP_MPC_SYN; 345 return true; 346 } else if (subflow->request_join) { 347 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token, 348 subflow->local_nonce); 349 opts->suboptions = OPTION_MPTCP_MPJ_SYN; 350 opts->join_id = subflow->local_id; 351 opts->token = subflow->remote_token; 352 opts->nonce = subflow->local_nonce; 353 opts->backup = subflow->request_bkup; 354 *size = TCPOLEN_MPTCP_MPJ_SYN; 355 return true; 356 } 357 return false; 358 } 359 360 /* MP_JOIN client subflow must wait for 4th ack before sending any data: 361 * TCP can't schedule delack timer before the subflow is fully established. 362 * MPTCP uses the delack timer to do 3rd ack retransmissions 363 */ 364 static void schedule_3rdack_retransmission(struct sock *sk) 365 { 366 struct inet_connection_sock *icsk = inet_csk(sk); 367 struct tcp_sock *tp = tcp_sk(sk); 368 unsigned long timeout; 369 370 /* reschedule with a timeout above RTT, as we must look only for drop */ 371 if (tp->srtt_us) 372 timeout = tp->srtt_us << 1; 373 else 374 timeout = TCP_TIMEOUT_INIT; 375 376 WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER); 377 icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER; 378 icsk->icsk_ack.timeout = timeout; 379 sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout); 380 } 381 382 static void clear_3rdack_retransmission(struct sock *sk) 383 { 384 struct inet_connection_sock *icsk = inet_csk(sk); 385 386 sk_stop_timer(sk, &icsk->icsk_delack_timer); 387 icsk->icsk_ack.timeout = 0; 388 icsk->icsk_ack.ato = 0; 389 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER); 390 } 391 392 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb, 393 unsigned int *size, 394 unsigned int remaining, 395 struct mptcp_out_options *opts) 396 { 397 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 398 struct mptcp_ext *mpext; 399 unsigned int data_len; 400 401 /* When skb is not available, we better over-estimate the emitted 402 * options len. A full DSS option (28 bytes) is longer than 403 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so 404 * tell the caller to defer the estimate to 405 * mptcp_established_options_dss(), which will reserve enough space. 406 */ 407 if (!skb) 408 return false; 409 410 /* MPC/MPJ needed only on 3rd ack packet */ 411 if (subflow->fully_established || 412 subflow->snd_isn != TCP_SKB_CB(skb)->seq) 413 return false; 414 415 if (subflow->mp_capable) { 416 mpext = mptcp_get_ext(skb); 417 data_len = mpext ? mpext->data_len : 0; 418 419 /* we will check ext_copy.data_len in mptcp_write_options() to 420 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and 421 * TCPOLEN_MPTCP_MPC_ACK 422 */ 423 opts->ext_copy.data_len = data_len; 424 opts->suboptions = OPTION_MPTCP_MPC_ACK; 425 opts->sndr_key = subflow->local_key; 426 opts->rcvr_key = subflow->remote_key; 427 428 /* Section 3.1. 429 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK 430 * packets that start the first subflow of an MPTCP connection, 431 * as well as the first packet that carries data 432 */ 433 if (data_len > 0) 434 *size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4); 435 else 436 *size = TCPOLEN_MPTCP_MPC_ACK; 437 438 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d", 439 subflow, subflow->local_key, subflow->remote_key, 440 data_len); 441 442 return true; 443 } else if (subflow->mp_join) { 444 opts->suboptions = OPTION_MPTCP_MPJ_ACK; 445 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN); 446 *size = TCPOLEN_MPTCP_MPJ_ACK; 447 pr_debug("subflow=%p", subflow); 448 449 schedule_3rdack_retransmission(sk); 450 return true; 451 } 452 return false; 453 } 454 455 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow, 456 struct sk_buff *skb, struct mptcp_ext *ext) 457 { 458 /* The write_seq value has already been incremented, so the actual 459 * sequence number for the DATA_FIN is one less. 460 */ 461 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1; 462 463 if (!ext->use_map || !skb->len) { 464 /* RFC6824 requires a DSS mapping with specific values 465 * if DATA_FIN is set but no data payload is mapped 466 */ 467 ext->data_fin = 1; 468 ext->use_map = 1; 469 ext->dsn64 = 1; 470 ext->data_seq = data_fin_tx_seq; 471 ext->subflow_seq = 0; 472 ext->data_len = 1; 473 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) { 474 /* If there's an existing DSS mapping and it is the 475 * final mapping, DATA_FIN consumes 1 additional byte of 476 * mapping space. 477 */ 478 ext->data_fin = 1; 479 ext->data_len++; 480 } 481 } 482 483 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb, 484 unsigned int *size, 485 unsigned int remaining, 486 struct mptcp_out_options *opts) 487 { 488 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 489 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 490 unsigned int dss_size = 0; 491 u64 snd_data_fin_enable; 492 struct mptcp_ext *mpext; 493 unsigned int ack_size; 494 bool ret = false; 495 496 mpext = skb ? mptcp_get_ext(skb) : NULL; 497 snd_data_fin_enable = mptcp_data_fin_enabled(msk); 498 499 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) { 500 unsigned int map_size; 501 502 map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64; 503 504 remaining -= map_size; 505 dss_size = map_size; 506 if (mpext) 507 opts->ext_copy = *mpext; 508 509 if (skb && snd_data_fin_enable) 510 mptcp_write_data_fin(subflow, skb, &opts->ext_copy); 511 ret = true; 512 } 513 514 /* passive sockets msk will set the 'can_ack' after accept(), even 515 * if the first subflow may have the already the remote key handy 516 */ 517 opts->ext_copy.use_ack = 0; 518 if (!READ_ONCE(msk->can_ack)) { 519 *size = ALIGN(dss_size, 4); 520 return ret; 521 } 522 523 if (READ_ONCE(msk->use_64bit_ack)) { 524 ack_size = TCPOLEN_MPTCP_DSS_ACK64; 525 opts->ext_copy.data_ack = READ_ONCE(msk->ack_seq); 526 opts->ext_copy.ack64 = 1; 527 } else { 528 ack_size = TCPOLEN_MPTCP_DSS_ACK32; 529 opts->ext_copy.data_ack32 = (uint32_t)READ_ONCE(msk->ack_seq); 530 opts->ext_copy.ack64 = 0; 531 } 532 opts->ext_copy.use_ack = 1; 533 WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk)); 534 535 /* Add kind/length/subtype/flag overhead if mapping is not populated */ 536 if (dss_size == 0) 537 ack_size += TCPOLEN_MPTCP_DSS_BASE; 538 539 dss_size += ack_size; 540 541 *size = ALIGN(dss_size, 4); 542 return true; 543 } 544 545 static u64 add_addr_generate_hmac(u64 key1, u64 key2, u8 addr_id, 546 struct in_addr *addr) 547 { 548 u8 hmac[SHA256_DIGEST_SIZE]; 549 u8 msg[7]; 550 551 msg[0] = addr_id; 552 memcpy(&msg[1], &addr->s_addr, 4); 553 msg[5] = 0; 554 msg[6] = 0; 555 556 mptcp_crypto_hmac_sha(key1, key2, msg, 7, hmac); 557 558 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]); 559 } 560 561 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 562 static u64 add_addr6_generate_hmac(u64 key1, u64 key2, u8 addr_id, 563 struct in6_addr *addr) 564 { 565 u8 hmac[SHA256_DIGEST_SIZE]; 566 u8 msg[19]; 567 568 msg[0] = addr_id; 569 memcpy(&msg[1], &addr->s6_addr, 16); 570 msg[17] = 0; 571 msg[18] = 0; 572 573 mptcp_crypto_hmac_sha(key1, key2, msg, 19, hmac); 574 575 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]); 576 } 577 #endif 578 579 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb, 580 unsigned int *size, 581 unsigned int remaining, 582 struct mptcp_out_options *opts) 583 { 584 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 585 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 586 bool drop_other_suboptions = false; 587 unsigned int opt_size = *size; 588 struct mptcp_addr_info saddr; 589 bool echo; 590 int len; 591 592 if (mptcp_pm_should_add_signal_ipv6(msk) && 593 skb && skb_is_tcp_pure_ack(skb)) { 594 pr_debug("drop other suboptions"); 595 opts->suboptions = 0; 596 remaining += opt_size; 597 drop_other_suboptions = true; 598 } 599 600 if (!mptcp_pm_should_add_signal(msk) || 601 !(mptcp_pm_add_addr_signal(msk, remaining, &saddr, &echo))) 602 return false; 603 604 len = mptcp_add_addr_len(saddr.family, echo); 605 if (remaining < len) 606 return false; 607 608 *size = len; 609 if (drop_other_suboptions) 610 *size -= opt_size; 611 opts->addr_id = saddr.id; 612 if (saddr.family == AF_INET) { 613 opts->suboptions |= OPTION_MPTCP_ADD_ADDR; 614 opts->addr = saddr.addr; 615 if (!echo) { 616 opts->ahmac = add_addr_generate_hmac(msk->local_key, 617 msk->remote_key, 618 opts->addr_id, 619 &opts->addr); 620 } 621 } 622 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 623 else if (saddr.family == AF_INET6) { 624 opts->suboptions |= OPTION_MPTCP_ADD_ADDR6; 625 opts->addr6 = saddr.addr6; 626 if (!echo) { 627 opts->ahmac = add_addr6_generate_hmac(msk->local_key, 628 msk->remote_key, 629 opts->addr_id, 630 &opts->addr6); 631 } 632 } 633 #endif 634 pr_debug("addr_id=%d, ahmac=%llu, echo=%d", opts->addr_id, opts->ahmac, echo); 635 636 return true; 637 } 638 639 static bool mptcp_established_options_rm_addr(struct sock *sk, 640 unsigned int *size, 641 unsigned int remaining, 642 struct mptcp_out_options *opts) 643 { 644 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 645 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 646 u8 rm_id; 647 648 if (!mptcp_pm_should_rm_signal(msk) || 649 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_id))) 650 return false; 651 652 if (remaining < TCPOLEN_MPTCP_RM_ADDR_BASE) 653 return false; 654 655 *size = TCPOLEN_MPTCP_RM_ADDR_BASE; 656 opts->suboptions |= OPTION_MPTCP_RM_ADDR; 657 opts->rm_id = rm_id; 658 659 pr_debug("rm_id=%d", opts->rm_id); 660 661 return true; 662 } 663 664 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb, 665 unsigned int *size, unsigned int remaining, 666 struct mptcp_out_options *opts) 667 { 668 unsigned int opt_size = 0; 669 bool ret = false; 670 671 opts->suboptions = 0; 672 673 if (unlikely(mptcp_check_fallback(sk))) 674 return false; 675 676 /* prevent adding of any MPTCP related options on reset packet 677 * until we support MP_TCPRST/MP_FASTCLOSE 678 */ 679 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) 680 return false; 681 682 if (mptcp_established_options_mp(sk, skb, &opt_size, remaining, opts)) 683 ret = true; 684 else if (mptcp_established_options_dss(sk, skb, &opt_size, remaining, 685 opts)) 686 ret = true; 687 688 /* we reserved enough space for the above options, and exceeding the 689 * TCP option space would be fatal 690 */ 691 if (WARN_ON_ONCE(opt_size > remaining)) 692 return false; 693 694 *size += opt_size; 695 remaining -= opt_size; 696 if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) { 697 *size += opt_size; 698 remaining -= opt_size; 699 ret = true; 700 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) { 701 *size += opt_size; 702 remaining -= opt_size; 703 ret = true; 704 } 705 706 return ret; 707 } 708 709 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size, 710 struct mptcp_out_options *opts) 711 { 712 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 713 714 if (subflow_req->mp_capable) { 715 opts->suboptions = OPTION_MPTCP_MPC_SYNACK; 716 opts->sndr_key = subflow_req->local_key; 717 *size = TCPOLEN_MPTCP_MPC_SYNACK; 718 pr_debug("subflow_req=%p, local_key=%llu", 719 subflow_req, subflow_req->local_key); 720 return true; 721 } else if (subflow_req->mp_join) { 722 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK; 723 opts->backup = subflow_req->backup; 724 opts->join_id = subflow_req->local_id; 725 opts->thmac = subflow_req->thmac; 726 opts->nonce = subflow_req->local_nonce; 727 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u", 728 subflow_req, opts->backup, opts->join_id, 729 opts->thmac, opts->nonce); 730 *size = TCPOLEN_MPTCP_MPJ_SYNACK; 731 return true; 732 } 733 return false; 734 } 735 736 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk, 737 struct mptcp_subflow_context *subflow, 738 struct sk_buff *skb, 739 struct mptcp_options_received *mp_opt) 740 { 741 /* here we can process OoO, in-window pkts, only in-sequence 4th ack 742 * will make the subflow fully established 743 */ 744 if (likely(subflow->fully_established)) { 745 /* on passive sockets, check for 3rd ack retransmission 746 * note that msk is always set by subflow_syn_recv_sock() 747 * for mp_join subflows 748 */ 749 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 && 750 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq && 751 subflow->mp_join && mp_opt->mp_join && 752 READ_ONCE(msk->pm.server_side)) 753 tcp_send_ack(ssk); 754 goto fully_established; 755 } 756 757 /* we must process OoO packets before the first subflow is fully 758 * established. OoO packets are instead a protocol violation 759 * for MP_JOIN subflows as the peer must not send any data 760 * before receiving the forth ack - cfr. RFC 8684 section 3.2. 761 */ 762 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) { 763 if (subflow->mp_join) 764 goto reset; 765 return subflow->mp_capable; 766 } 767 768 if (mp_opt->dss && mp_opt->use_ack) { 769 /* subflows are fully established as soon as we get any 770 * additional ack. 771 */ 772 subflow->fully_established = 1; 773 WRITE_ONCE(msk->fully_established, true); 774 goto fully_established; 775 } 776 777 if (mp_opt->add_addr) { 778 WRITE_ONCE(msk->fully_established, true); 779 return true; 780 } 781 782 /* If the first established packet does not contain MP_CAPABLE + data 783 * then fallback to TCP. Fallback scenarios requires a reset for 784 * MP_JOIN subflows. 785 */ 786 if (!mp_opt->mp_capable) { 787 if (subflow->mp_join) 788 goto reset; 789 subflow->mp_capable = 0; 790 pr_fallback(msk); 791 __mptcp_do_fallback(msk); 792 return false; 793 } 794 795 if (unlikely(!READ_ONCE(msk->pm.server_side))) 796 pr_warn_once("bogus mpc option on established client sk"); 797 mptcp_subflow_fully_established(subflow, mp_opt); 798 799 fully_established: 800 if (likely(subflow->pm_notified)) 801 return true; 802 803 subflow->pm_notified = 1; 804 if (subflow->mp_join) { 805 clear_3rdack_retransmission(ssk); 806 mptcp_pm_subflow_established(msk, subflow); 807 } else { 808 mptcp_pm_fully_established(msk); 809 } 810 return true; 811 812 reset: 813 mptcp_subflow_reset(ssk); 814 return false; 815 } 816 817 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit) 818 { 819 u32 old_ack32, cur_ack32; 820 821 if (use_64bit) 822 return cur_ack; 823 824 old_ack32 = (u32)old_ack; 825 cur_ack32 = (u32)cur_ack; 826 cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32; 827 if (unlikely(before(cur_ack32, old_ack32))) 828 return cur_ack + (1LL << 32); 829 return cur_ack; 830 } 831 832 static void ack_update_msk(struct mptcp_sock *msk, 833 const struct sock *ssk, 834 struct mptcp_options_received *mp_opt) 835 { 836 u64 new_snd_una, snd_una, old_snd_una = atomic64_read(&msk->snd_una); 837 u64 new_wnd_end, wnd_end, old_wnd_end = atomic64_read(&msk->wnd_end); 838 u64 snd_nxt = READ_ONCE(msk->snd_nxt); 839 struct sock *sk = (struct sock *)msk; 840 841 /* avoid ack expansion on update conflict, to reduce the risk of 842 * wrongly expanding to a future ack sequence number, which is way 843 * more dangerous than missing an ack 844 */ 845 new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64); 846 847 /* ACK for data not even sent yet? Ignore. */ 848 if (after64(new_snd_una, snd_nxt)) 849 new_snd_una = old_snd_una; 850 851 new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd; 852 853 while (after64(new_wnd_end, old_wnd_end)) { 854 wnd_end = old_wnd_end; 855 old_wnd_end = atomic64_cmpxchg(&msk->wnd_end, wnd_end, 856 new_wnd_end); 857 if (old_wnd_end == wnd_end) { 858 if (mptcp_send_head(sk)) 859 mptcp_schedule_work(sk); 860 break; 861 } 862 } 863 864 while (after64(new_snd_una, old_snd_una)) { 865 snd_una = old_snd_una; 866 old_snd_una = atomic64_cmpxchg(&msk->snd_una, snd_una, 867 new_snd_una); 868 if (old_snd_una == snd_una) { 869 mptcp_data_acked(sk); 870 break; 871 } 872 } 873 } 874 875 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit) 876 { 877 /* Skip if DATA_FIN was already received. 878 * If updating simultaneously with the recvmsg loop, values 879 * should match. If they mismatch, the peer is misbehaving and 880 * we will prefer the most recent information. 881 */ 882 if (READ_ONCE(msk->rcv_data_fin) || !READ_ONCE(msk->first)) 883 return false; 884 885 WRITE_ONCE(msk->rcv_data_fin_seq, 886 expand_ack(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit)); 887 WRITE_ONCE(msk->rcv_data_fin, 1); 888 889 return true; 890 } 891 892 static bool add_addr_hmac_valid(struct mptcp_sock *msk, 893 struct mptcp_options_received *mp_opt) 894 { 895 u64 hmac = 0; 896 897 if (mp_opt->echo) 898 return true; 899 900 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4) 901 hmac = add_addr_generate_hmac(msk->remote_key, 902 msk->local_key, 903 mp_opt->addr_id, &mp_opt->addr); 904 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 905 else 906 hmac = add_addr6_generate_hmac(msk->remote_key, 907 msk->local_key, 908 mp_opt->addr_id, &mp_opt->addr6); 909 #endif 910 911 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n", 912 msk, (unsigned long long)hmac, 913 (unsigned long long)mp_opt->ahmac); 914 915 return hmac == mp_opt->ahmac; 916 } 917 918 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb) 919 { 920 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 921 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 922 struct mptcp_options_received mp_opt; 923 struct mptcp_ext *mpext; 924 925 if (__mptcp_check_fallback(msk)) 926 return; 927 928 mptcp_get_options(skb, &mp_opt); 929 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt)) 930 return; 931 932 if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) { 933 struct mptcp_addr_info addr; 934 935 addr.port = htons(mp_opt.port); 936 addr.id = mp_opt.addr_id; 937 if (mp_opt.family == MPTCP_ADDR_IPVERSION_4) { 938 addr.family = AF_INET; 939 addr.addr = mp_opt.addr; 940 } 941 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 942 else if (mp_opt.family == MPTCP_ADDR_IPVERSION_6) { 943 addr.family = AF_INET6; 944 addr.addr6 = mp_opt.addr6; 945 } 946 #endif 947 if (!mp_opt.echo) { 948 mptcp_pm_add_addr_received(msk, &addr); 949 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR); 950 } else { 951 mptcp_pm_del_add_timer(msk, &addr); 952 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD); 953 } 954 mp_opt.add_addr = 0; 955 } 956 957 if (mp_opt.rm_addr) { 958 mptcp_pm_rm_addr_received(msk, mp_opt.rm_id); 959 mp_opt.rm_addr = 0; 960 } 961 962 if (!mp_opt.dss) 963 return; 964 965 /* we can't wait for recvmsg() to update the ack_seq, otherwise 966 * monodirectional flows will stuck 967 */ 968 if (mp_opt.use_ack) 969 ack_update_msk(msk, sk, &mp_opt); 970 971 /* Zero-data-length packets are dropped by the caller and not 972 * propagated to the MPTCP layer, so the skb extension does not 973 * need to be allocated or populated. DATA_FIN information, if 974 * present, needs to be updated here before the skb is freed. 975 */ 976 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) { 977 if (mp_opt.data_fin && mp_opt.data_len == 1 && 978 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) && 979 schedule_work(&msk->work)) 980 sock_hold(subflow->conn); 981 982 return; 983 } 984 985 mpext = skb_ext_add(skb, SKB_EXT_MPTCP); 986 if (!mpext) 987 return; 988 989 memset(mpext, 0, sizeof(*mpext)); 990 991 if (mp_opt.use_map) { 992 if (mp_opt.mpc_map) { 993 /* this is an MP_CAPABLE carrying MPTCP data 994 * we know this map the first chunk of data 995 */ 996 mptcp_crypto_key_sha(subflow->remote_key, NULL, 997 &mpext->data_seq); 998 mpext->data_seq++; 999 mpext->subflow_seq = 1; 1000 mpext->dsn64 = 1; 1001 mpext->mpc_map = 1; 1002 mpext->data_fin = 0; 1003 } else { 1004 mpext->data_seq = mp_opt.data_seq; 1005 mpext->subflow_seq = mp_opt.subflow_seq; 1006 mpext->dsn64 = mp_opt.dsn64; 1007 mpext->data_fin = mp_opt.data_fin; 1008 } 1009 mpext->data_len = mp_opt.data_len; 1010 mpext->use_map = 1; 1011 } 1012 } 1013 1014 static void mptcp_set_rwin(const struct tcp_sock *tp) 1015 { 1016 const struct sock *ssk = (const struct sock *)tp; 1017 const struct mptcp_subflow_context *subflow; 1018 struct mptcp_sock *msk; 1019 u64 ack_seq; 1020 1021 subflow = mptcp_subflow_ctx(ssk); 1022 msk = mptcp_sk(subflow->conn); 1023 1024 ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd; 1025 1026 if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent))) 1027 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq); 1028 } 1029 1030 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp, 1031 struct mptcp_out_options *opts) 1032 { 1033 if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK | 1034 OPTION_MPTCP_MPC_ACK) & opts->suboptions) { 1035 u8 len; 1036 1037 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) 1038 len = TCPOLEN_MPTCP_MPC_SYN; 1039 else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) 1040 len = TCPOLEN_MPTCP_MPC_SYNACK; 1041 else if (opts->ext_copy.data_len) 1042 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 1043 else 1044 len = TCPOLEN_MPTCP_MPC_ACK; 1045 1046 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len, 1047 MPTCP_SUPPORTED_VERSION, 1048 MPTCP_CAP_HMAC_SHA256); 1049 1050 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) & 1051 opts->suboptions)) 1052 goto mp_capable_done; 1053 1054 put_unaligned_be64(opts->sndr_key, ptr); 1055 ptr += 2; 1056 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions)) 1057 goto mp_capable_done; 1058 1059 put_unaligned_be64(opts->rcvr_key, ptr); 1060 ptr += 2; 1061 if (!opts->ext_copy.data_len) 1062 goto mp_capable_done; 1063 1064 put_unaligned_be32(opts->ext_copy.data_len << 16 | 1065 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1066 ptr += 1; 1067 } 1068 1069 mp_capable_done: 1070 if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) { 1071 if (opts->ahmac) 1072 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 1073 TCPOLEN_MPTCP_ADD_ADDR, 0, 1074 opts->addr_id); 1075 else 1076 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 1077 TCPOLEN_MPTCP_ADD_ADDR_BASE, 1078 MPTCP_ADDR_ECHO, 1079 opts->addr_id); 1080 memcpy((u8 *)ptr, (u8 *)&opts->addr.s_addr, 4); 1081 ptr += 1; 1082 if (opts->ahmac) { 1083 put_unaligned_be64(opts->ahmac, ptr); 1084 ptr += 2; 1085 } 1086 } 1087 1088 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1089 if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions) { 1090 if (opts->ahmac) 1091 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 1092 TCPOLEN_MPTCP_ADD_ADDR6, 0, 1093 opts->addr_id); 1094 else 1095 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 1096 TCPOLEN_MPTCP_ADD_ADDR6_BASE, 1097 MPTCP_ADDR_ECHO, 1098 opts->addr_id); 1099 memcpy((u8 *)ptr, opts->addr6.s6_addr, 16); 1100 ptr += 4; 1101 if (opts->ahmac) { 1102 put_unaligned_be64(opts->ahmac, ptr); 1103 ptr += 2; 1104 } 1105 } 1106 #endif 1107 1108 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) { 1109 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR, 1110 TCPOLEN_MPTCP_RM_ADDR_BASE, 1111 0, opts->rm_id); 1112 } 1113 1114 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) { 1115 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1116 TCPOLEN_MPTCP_MPJ_SYN, 1117 opts->backup, opts->join_id); 1118 put_unaligned_be32(opts->token, ptr); 1119 ptr += 1; 1120 put_unaligned_be32(opts->nonce, ptr); 1121 ptr += 1; 1122 } 1123 1124 if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) { 1125 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1126 TCPOLEN_MPTCP_MPJ_SYNACK, 1127 opts->backup, opts->join_id); 1128 put_unaligned_be64(opts->thmac, ptr); 1129 ptr += 2; 1130 put_unaligned_be32(opts->nonce, ptr); 1131 ptr += 1; 1132 } 1133 1134 if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) { 1135 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1136 TCPOLEN_MPTCP_MPJ_ACK, 0, 0); 1137 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN); 1138 ptr += 5; 1139 } 1140 1141 if (opts->ext_copy.use_ack || opts->ext_copy.use_map) { 1142 struct mptcp_ext *mpext = &opts->ext_copy; 1143 u8 len = TCPOLEN_MPTCP_DSS_BASE; 1144 u8 flags = 0; 1145 1146 if (mpext->use_ack) { 1147 flags = MPTCP_DSS_HAS_ACK; 1148 if (mpext->ack64) { 1149 len += TCPOLEN_MPTCP_DSS_ACK64; 1150 flags |= MPTCP_DSS_ACK64; 1151 } else { 1152 len += TCPOLEN_MPTCP_DSS_ACK32; 1153 } 1154 } 1155 1156 if (mpext->use_map) { 1157 len += TCPOLEN_MPTCP_DSS_MAP64; 1158 1159 /* Use only 64-bit mapping flags for now, add 1160 * support for optional 32-bit mappings later. 1161 */ 1162 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64; 1163 if (mpext->data_fin) 1164 flags |= MPTCP_DSS_DATA_FIN; 1165 } 1166 1167 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags); 1168 1169 if (mpext->use_ack) { 1170 if (mpext->ack64) { 1171 put_unaligned_be64(mpext->data_ack, ptr); 1172 ptr += 2; 1173 } else { 1174 put_unaligned_be32(mpext->data_ack32, ptr); 1175 ptr += 1; 1176 } 1177 } 1178 1179 if (mpext->use_map) { 1180 put_unaligned_be64(mpext->data_seq, ptr); 1181 ptr += 2; 1182 put_unaligned_be32(mpext->subflow_seq, ptr); 1183 ptr += 1; 1184 put_unaligned_be32(mpext->data_len << 16 | 1185 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1186 } 1187 } 1188 1189 if (tp) 1190 mptcp_set_rwin(tp); 1191 } 1192