1 // SPDX-License-Identifier: GPL-2.0 2 /* Multipath TCP 3 * 4 * Copyright (c) 2017 - 2019, Intel Corporation. 5 */ 6 7 #define pr_fmt(fmt) "MPTCP: " fmt 8 9 #include <linux/kernel.h> 10 #include <crypto/sha2.h> 11 #include <net/tcp.h> 12 #include <net/mptcp.h> 13 #include "protocol.h" 14 #include "mib.h" 15 16 #include <trace/events/mptcp.h> 17 18 static bool mptcp_cap_flag_sha256(u8 flags) 19 { 20 return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256; 21 } 22 23 static void mptcp_parse_option(const struct sk_buff *skb, 24 const unsigned char *ptr, int opsize, 25 struct mptcp_options_received *mp_opt) 26 { 27 u8 subtype = *ptr >> 4; 28 int expected_opsize; 29 u16 subopt; 30 u8 version; 31 u8 flags; 32 u8 i; 33 34 switch (subtype) { 35 case MPTCPOPT_MP_CAPABLE: 36 /* strict size checking */ 37 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { 38 if (skb->len > tcp_hdr(skb)->doff << 2) 39 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA; 40 else 41 expected_opsize = TCPOLEN_MPTCP_MPC_ACK; 42 subopt = OPTION_MPTCP_MPC_ACK; 43 } else { 44 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK) { 45 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK; 46 subopt = OPTION_MPTCP_MPC_SYNACK; 47 } else { 48 expected_opsize = TCPOLEN_MPTCP_MPC_SYN; 49 subopt = OPTION_MPTCP_MPC_SYN; 50 } 51 } 52 53 /* Cfr RFC 8684 Section 3.3.0: 54 * If a checksum is present but its use had 55 * not been negotiated in the MP_CAPABLE handshake, the receiver MUST 56 * close the subflow with a RST, as it is not behaving as negotiated. 57 * If a checksum is not present when its use has been negotiated, the 58 * receiver MUST close the subflow with a RST, as it is considered 59 * broken 60 * We parse even option with mismatching csum presence, so that 61 * later in subflow_data_ready we can trigger the reset. 62 */ 63 if (opsize != expected_opsize && 64 (expected_opsize != TCPOLEN_MPTCP_MPC_ACK_DATA || 65 opsize != TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM)) 66 break; 67 68 /* try to be gentle vs future versions on the initial syn */ 69 version = *ptr++ & MPTCP_VERSION_MASK; 70 if (opsize != TCPOLEN_MPTCP_MPC_SYN) { 71 if (version != MPTCP_SUPPORTED_VERSION) 72 break; 73 } else if (version < MPTCP_SUPPORTED_VERSION) { 74 break; 75 } 76 77 flags = *ptr++; 78 if (!mptcp_cap_flag_sha256(flags) || 79 (flags & MPTCP_CAP_EXTENSIBILITY)) 80 break; 81 82 /* RFC 6824, Section 3.1: 83 * "For the Checksum Required bit (labeled "A"), if either 84 * host requires the use of checksums, checksums MUST be used. 85 * In other words, the only way for checksums not to be used 86 * is if both hosts in their SYNs set A=0." 87 */ 88 if (flags & MPTCP_CAP_CHECKSUM_REQD) 89 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD; 90 91 mp_opt->deny_join_id0 = !!(flags & MPTCP_CAP_DENY_JOIN_ID0); 92 93 mp_opt->suboptions |= subopt; 94 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) { 95 mp_opt->sndr_key = get_unaligned_be64(ptr); 96 ptr += 8; 97 } 98 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) { 99 mp_opt->rcvr_key = get_unaligned_be64(ptr); 100 ptr += 8; 101 } 102 if (opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA) { 103 /* Section 3.1.: 104 * "the data parameters in a MP_CAPABLE are semantically 105 * equivalent to those in a DSS option and can be used 106 * interchangeably." 107 */ 108 mp_opt->suboptions |= OPTION_MPTCP_DSS; 109 mp_opt->use_map = 1; 110 mp_opt->mpc_map = 1; 111 mp_opt->data_len = get_unaligned_be16(ptr); 112 ptr += 2; 113 } 114 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM) { 115 mp_opt->csum = get_unaligned((__force __sum16 *)ptr); 116 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD; 117 ptr += 2; 118 } 119 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d csum=%u", 120 version, flags, opsize, mp_opt->sndr_key, 121 mp_opt->rcvr_key, mp_opt->data_len, mp_opt->csum); 122 break; 123 124 case MPTCPOPT_MP_JOIN: 125 mp_opt->suboptions |= OPTIONS_MPTCP_MPJ; 126 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) { 127 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 128 mp_opt->join_id = *ptr++; 129 mp_opt->token = get_unaligned_be32(ptr); 130 ptr += 4; 131 mp_opt->nonce = get_unaligned_be32(ptr); 132 ptr += 4; 133 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u", 134 mp_opt->backup, mp_opt->join_id, 135 mp_opt->token, mp_opt->nonce); 136 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) { 137 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 138 mp_opt->join_id = *ptr++; 139 mp_opt->thmac = get_unaligned_be64(ptr); 140 ptr += 8; 141 mp_opt->nonce = get_unaligned_be32(ptr); 142 ptr += 4; 143 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u", 144 mp_opt->backup, mp_opt->join_id, 145 mp_opt->thmac, mp_opt->nonce); 146 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) { 147 ptr += 2; 148 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN); 149 pr_debug("MP_JOIN hmac"); 150 } else { 151 mp_opt->suboptions &= ~OPTIONS_MPTCP_MPJ; 152 } 153 break; 154 155 case MPTCPOPT_DSS: 156 pr_debug("DSS"); 157 ptr++; 158 159 /* we must clear 'mpc_map' be able to detect MP_CAPABLE 160 * map vs DSS map in mptcp_incoming_options(), and reconstruct 161 * map info accordingly 162 */ 163 mp_opt->mpc_map = 0; 164 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK; 165 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0; 166 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0; 167 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0; 168 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0; 169 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK); 170 171 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d", 172 mp_opt->data_fin, mp_opt->dsn64, 173 mp_opt->use_map, mp_opt->ack64, 174 mp_opt->use_ack); 175 176 expected_opsize = TCPOLEN_MPTCP_DSS_BASE; 177 178 if (mp_opt->use_ack) { 179 if (mp_opt->ack64) 180 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64; 181 else 182 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32; 183 } 184 185 if (mp_opt->use_map) { 186 if (mp_opt->dsn64) 187 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64; 188 else 189 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32; 190 } 191 192 /* Always parse any csum presence combination, we will enforce 193 * RFC 8684 Section 3.3.0 checks later in subflow_data_ready 194 */ 195 if (opsize != expected_opsize && 196 opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) 197 break; 198 199 mp_opt->suboptions |= OPTION_MPTCP_DSS; 200 if (mp_opt->use_ack) { 201 if (mp_opt->ack64) { 202 mp_opt->data_ack = get_unaligned_be64(ptr); 203 ptr += 8; 204 } else { 205 mp_opt->data_ack = get_unaligned_be32(ptr); 206 ptr += 4; 207 } 208 209 pr_debug("data_ack=%llu", mp_opt->data_ack); 210 } 211 212 if (mp_opt->use_map) { 213 if (mp_opt->dsn64) { 214 mp_opt->data_seq = get_unaligned_be64(ptr); 215 ptr += 8; 216 } else { 217 mp_opt->data_seq = get_unaligned_be32(ptr); 218 ptr += 4; 219 } 220 221 mp_opt->subflow_seq = get_unaligned_be32(ptr); 222 ptr += 4; 223 224 mp_opt->data_len = get_unaligned_be16(ptr); 225 ptr += 2; 226 227 if (opsize == expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) { 228 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD; 229 mp_opt->csum = get_unaligned((__force __sum16 *)ptr); 230 ptr += 2; 231 } 232 233 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u", 234 mp_opt->data_seq, mp_opt->subflow_seq, 235 mp_opt->data_len, !!(mp_opt->suboptions & OPTION_MPTCP_CSUMREQD), 236 mp_opt->csum); 237 } 238 239 break; 240 241 case MPTCPOPT_ADD_ADDR: 242 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO; 243 if (!mp_opt->echo) { 244 if (opsize == TCPOLEN_MPTCP_ADD_ADDR || 245 opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT) 246 mp_opt->addr.family = AF_INET; 247 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 248 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 || 249 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT) 250 mp_opt->addr.family = AF_INET6; 251 #endif 252 else 253 break; 254 } else { 255 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE || 256 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) 257 mp_opt->addr.family = AF_INET; 258 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 259 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE || 260 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) 261 mp_opt->addr.family = AF_INET6; 262 #endif 263 else 264 break; 265 } 266 267 mp_opt->suboptions |= OPTION_MPTCP_ADD_ADDR; 268 mp_opt->addr.id = *ptr++; 269 mp_opt->addr.port = 0; 270 mp_opt->ahmac = 0; 271 if (mp_opt->addr.family == AF_INET) { 272 memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4); 273 ptr += 4; 274 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT || 275 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) { 276 mp_opt->addr.port = htons(get_unaligned_be16(ptr)); 277 ptr += 2; 278 } 279 } 280 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 281 else { 282 memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16); 283 ptr += 16; 284 if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT || 285 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) { 286 mp_opt->addr.port = htons(get_unaligned_be16(ptr)); 287 ptr += 2; 288 } 289 } 290 #endif 291 if (!mp_opt->echo) { 292 mp_opt->ahmac = get_unaligned_be64(ptr); 293 ptr += 8; 294 } 295 pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d", 296 (mp_opt->addr.family == AF_INET6) ? "6" : "", 297 mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port)); 298 break; 299 300 case MPTCPOPT_RM_ADDR: 301 if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 || 302 opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX) 303 break; 304 305 ptr++; 306 307 mp_opt->suboptions |= OPTION_MPTCP_RM_ADDR; 308 mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE; 309 for (i = 0; i < mp_opt->rm_list.nr; i++) 310 mp_opt->rm_list.ids[i] = *ptr++; 311 pr_debug("RM_ADDR: rm_list_nr=%d", mp_opt->rm_list.nr); 312 break; 313 314 case MPTCPOPT_MP_PRIO: 315 if (opsize != TCPOLEN_MPTCP_PRIO) 316 break; 317 318 mp_opt->suboptions |= OPTION_MPTCP_PRIO; 319 mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP; 320 pr_debug("MP_PRIO: prio=%d", mp_opt->backup); 321 break; 322 323 case MPTCPOPT_MP_FASTCLOSE: 324 if (opsize != TCPOLEN_MPTCP_FASTCLOSE) 325 break; 326 327 ptr += 2; 328 mp_opt->rcvr_key = get_unaligned_be64(ptr); 329 ptr += 8; 330 mp_opt->suboptions |= OPTION_MPTCP_FASTCLOSE; 331 pr_debug("MP_FASTCLOSE: recv_key=%llu", mp_opt->rcvr_key); 332 break; 333 334 case MPTCPOPT_RST: 335 if (opsize != TCPOLEN_MPTCP_RST) 336 break; 337 338 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) 339 break; 340 341 mp_opt->suboptions |= OPTION_MPTCP_RST; 342 flags = *ptr++; 343 mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT; 344 mp_opt->reset_reason = *ptr; 345 pr_debug("MP_RST: transient=%u reason=%u", 346 mp_opt->reset_transient, mp_opt->reset_reason); 347 break; 348 349 case MPTCPOPT_MP_FAIL: 350 if (opsize != TCPOLEN_MPTCP_FAIL) 351 break; 352 353 ptr += 2; 354 mp_opt->suboptions |= OPTION_MPTCP_FAIL; 355 mp_opt->fail_seq = get_unaligned_be64(ptr); 356 pr_debug("MP_FAIL: data_seq=%llu", mp_opt->fail_seq); 357 break; 358 359 default: 360 break; 361 } 362 } 363 364 void mptcp_get_options(const struct sk_buff *skb, 365 struct mptcp_options_received *mp_opt) 366 { 367 const struct tcphdr *th = tcp_hdr(skb); 368 const unsigned char *ptr; 369 int length; 370 371 /* initialize option status */ 372 mp_opt->suboptions = 0; 373 374 length = (th->doff * 4) - sizeof(struct tcphdr); 375 ptr = (const unsigned char *)(th + 1); 376 377 while (length > 0) { 378 int opcode = *ptr++; 379 int opsize; 380 381 switch (opcode) { 382 case TCPOPT_EOL: 383 return; 384 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */ 385 length--; 386 continue; 387 default: 388 if (length < 2) 389 return; 390 opsize = *ptr++; 391 if (opsize < 2) /* "silly options" */ 392 return; 393 if (opsize > length) 394 return; /* don't parse partial options */ 395 if (opcode == TCPOPT_MPTCP) 396 mptcp_parse_option(skb, ptr, opsize, mp_opt); 397 ptr += opsize - 2; 398 length -= opsize; 399 } 400 } 401 } 402 403 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb, 404 unsigned int *size, struct mptcp_out_options *opts) 405 { 406 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 407 408 /* we will use snd_isn to detect first pkt [re]transmission 409 * in mptcp_established_options_mp() 410 */ 411 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq; 412 if (subflow->request_mptcp) { 413 opts->suboptions = OPTION_MPTCP_MPC_SYN; 414 opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk)); 415 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk)); 416 *size = TCPOLEN_MPTCP_MPC_SYN; 417 return true; 418 } else if (subflow->request_join) { 419 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token, 420 subflow->local_nonce); 421 opts->suboptions = OPTION_MPTCP_MPJ_SYN; 422 opts->join_id = subflow->local_id; 423 opts->token = subflow->remote_token; 424 opts->nonce = subflow->local_nonce; 425 opts->backup = subflow->request_bkup; 426 *size = TCPOLEN_MPTCP_MPJ_SYN; 427 return true; 428 } 429 return false; 430 } 431 432 static void clear_3rdack_retransmission(struct sock *sk) 433 { 434 struct inet_connection_sock *icsk = inet_csk(sk); 435 436 sk_stop_timer(sk, &icsk->icsk_delack_timer); 437 icsk->icsk_ack.timeout = 0; 438 icsk->icsk_ack.ato = 0; 439 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER); 440 } 441 442 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb, 443 bool snd_data_fin_enable, 444 unsigned int *size, 445 unsigned int remaining, 446 struct mptcp_out_options *opts) 447 { 448 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 449 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 450 struct mptcp_ext *mpext; 451 unsigned int data_len; 452 u8 len; 453 454 /* When skb is not available, we better over-estimate the emitted 455 * options len. A full DSS option (28 bytes) is longer than 456 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so 457 * tell the caller to defer the estimate to 458 * mptcp_established_options_dss(), which will reserve enough space. 459 */ 460 if (!skb) 461 return false; 462 463 /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */ 464 if (subflow->fully_established || snd_data_fin_enable || 465 subflow->snd_isn != TCP_SKB_CB(skb)->seq || 466 sk->sk_state != TCP_ESTABLISHED) 467 return false; 468 469 if (subflow->mp_capable) { 470 mpext = mptcp_get_ext(skb); 471 data_len = mpext ? mpext->data_len : 0; 472 473 /* we will check ops->data_len in mptcp_write_options() to 474 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and 475 * TCPOLEN_MPTCP_MPC_ACK 476 */ 477 opts->data_len = data_len; 478 opts->suboptions = OPTION_MPTCP_MPC_ACK; 479 opts->sndr_key = subflow->local_key; 480 opts->rcvr_key = subflow->remote_key; 481 opts->csum_reqd = READ_ONCE(msk->csum_enabled); 482 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk)); 483 484 /* Section 3.1. 485 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK 486 * packets that start the first subflow of an MPTCP connection, 487 * as well as the first packet that carries data 488 */ 489 if (data_len > 0) { 490 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 491 if (opts->csum_reqd) { 492 /* we need to propagate more info to csum the pseudo hdr */ 493 opts->data_seq = mpext->data_seq; 494 opts->subflow_seq = mpext->subflow_seq; 495 opts->csum = mpext->csum; 496 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 497 } 498 *size = ALIGN(len, 4); 499 } else { 500 *size = TCPOLEN_MPTCP_MPC_ACK; 501 } 502 503 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d", 504 subflow, subflow->local_key, subflow->remote_key, 505 data_len); 506 507 return true; 508 } else if (subflow->mp_join) { 509 opts->suboptions = OPTION_MPTCP_MPJ_ACK; 510 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN); 511 *size = TCPOLEN_MPTCP_MPJ_ACK; 512 pr_debug("subflow=%p", subflow); 513 514 /* we can use the full delegate action helper only from BH context 515 * If we are in process context - sk is flushing the backlog at 516 * socket lock release time - just set the appropriate flag, will 517 * be handled by the release callback 518 */ 519 if (sock_owned_by_user(sk)) 520 set_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status); 521 else 522 mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_ACK); 523 return true; 524 } 525 return false; 526 } 527 528 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow, 529 struct sk_buff *skb, struct mptcp_ext *ext) 530 { 531 /* The write_seq value has already been incremented, so the actual 532 * sequence number for the DATA_FIN is one less. 533 */ 534 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1; 535 536 if (!ext->use_map || !skb->len) { 537 /* RFC6824 requires a DSS mapping with specific values 538 * if DATA_FIN is set but no data payload is mapped 539 */ 540 ext->data_fin = 1; 541 ext->use_map = 1; 542 ext->dsn64 = 1; 543 ext->data_seq = data_fin_tx_seq; 544 ext->subflow_seq = 0; 545 ext->data_len = 1; 546 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) { 547 /* If there's an existing DSS mapping and it is the 548 * final mapping, DATA_FIN consumes 1 additional byte of 549 * mapping space. 550 */ 551 ext->data_fin = 1; 552 ext->data_len++; 553 } 554 } 555 556 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb, 557 bool snd_data_fin_enable, 558 unsigned int *size, 559 unsigned int remaining, 560 struct mptcp_out_options *opts) 561 { 562 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 563 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 564 unsigned int dss_size = 0; 565 struct mptcp_ext *mpext; 566 unsigned int ack_size; 567 bool ret = false; 568 u64 ack_seq; 569 570 opts->csum_reqd = READ_ONCE(msk->csum_enabled); 571 mpext = skb ? mptcp_get_ext(skb) : NULL; 572 573 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) { 574 unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64; 575 576 if (mpext) { 577 if (opts->csum_reqd) 578 map_size += TCPOLEN_MPTCP_DSS_CHECKSUM; 579 580 opts->ext_copy = *mpext; 581 } 582 583 remaining -= map_size; 584 dss_size = map_size; 585 if (skb && snd_data_fin_enable) 586 mptcp_write_data_fin(subflow, skb, &opts->ext_copy); 587 opts->suboptions = OPTION_MPTCP_DSS; 588 ret = true; 589 } 590 591 /* passive sockets msk will set the 'can_ack' after accept(), even 592 * if the first subflow may have the already the remote key handy 593 */ 594 opts->ext_copy.use_ack = 0; 595 if (!READ_ONCE(msk->can_ack)) { 596 *size = ALIGN(dss_size, 4); 597 return ret; 598 } 599 600 ack_seq = READ_ONCE(msk->ack_seq); 601 if (READ_ONCE(msk->use_64bit_ack)) { 602 ack_size = TCPOLEN_MPTCP_DSS_ACK64; 603 opts->ext_copy.data_ack = ack_seq; 604 opts->ext_copy.ack64 = 1; 605 } else { 606 ack_size = TCPOLEN_MPTCP_DSS_ACK32; 607 opts->ext_copy.data_ack32 = (uint32_t)ack_seq; 608 opts->ext_copy.ack64 = 0; 609 } 610 opts->ext_copy.use_ack = 1; 611 opts->suboptions = OPTION_MPTCP_DSS; 612 WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk)); 613 614 /* Add kind/length/subtype/flag overhead if mapping is not populated */ 615 if (dss_size == 0) 616 ack_size += TCPOLEN_MPTCP_DSS_BASE; 617 618 dss_size += ack_size; 619 620 *size = ALIGN(dss_size, 4); 621 return true; 622 } 623 624 static u64 add_addr_generate_hmac(u64 key1, u64 key2, 625 struct mptcp_addr_info *addr) 626 { 627 u16 port = ntohs(addr->port); 628 u8 hmac[SHA256_DIGEST_SIZE]; 629 u8 msg[19]; 630 int i = 0; 631 632 msg[i++] = addr->id; 633 if (addr->family == AF_INET) { 634 memcpy(&msg[i], &addr->addr.s_addr, 4); 635 i += 4; 636 } 637 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 638 else if (addr->family == AF_INET6) { 639 memcpy(&msg[i], &addr->addr6.s6_addr, 16); 640 i += 16; 641 } 642 #endif 643 msg[i++] = port >> 8; 644 msg[i++] = port & 0xFF; 645 646 mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac); 647 648 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]); 649 } 650 651 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb, 652 unsigned int *size, 653 unsigned int remaining, 654 struct mptcp_out_options *opts) 655 { 656 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 657 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 658 bool drop_other_suboptions = false; 659 unsigned int opt_size = *size; 660 bool echo; 661 int len; 662 663 /* add addr will strip the existing options, be sure to avoid breaking 664 * MPC/MPJ handshakes 665 */ 666 if (!mptcp_pm_should_add_signal(msk) || 667 (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) || 668 !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, &opts->addr, 669 &echo, &drop_other_suboptions)) 670 return false; 671 672 if (drop_other_suboptions) 673 remaining += opt_size; 674 len = mptcp_add_addr_len(opts->addr.family, echo, !!opts->addr.port); 675 if (remaining < len) 676 return false; 677 678 *size = len; 679 if (drop_other_suboptions) { 680 pr_debug("drop other suboptions"); 681 opts->suboptions = 0; 682 683 /* note that e.g. DSS could have written into the memory 684 * aliased by ahmac, we must reset the field here 685 * to avoid appending the hmac even for ADD_ADDR echo 686 * options 687 */ 688 opts->ahmac = 0; 689 *size -= opt_size; 690 } 691 opts->suboptions |= OPTION_MPTCP_ADD_ADDR; 692 if (!echo) { 693 opts->ahmac = add_addr_generate_hmac(msk->local_key, 694 msk->remote_key, 695 &opts->addr); 696 } 697 pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d", 698 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port)); 699 700 return true; 701 } 702 703 static bool mptcp_established_options_rm_addr(struct sock *sk, 704 unsigned int *size, 705 unsigned int remaining, 706 struct mptcp_out_options *opts) 707 { 708 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 709 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 710 struct mptcp_rm_list rm_list; 711 int i, len; 712 713 if (!mptcp_pm_should_rm_signal(msk) || 714 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list))) 715 return false; 716 717 len = mptcp_rm_addr_len(&rm_list); 718 if (len < 0) 719 return false; 720 if (remaining < len) 721 return false; 722 723 *size = len; 724 opts->suboptions |= OPTION_MPTCP_RM_ADDR; 725 opts->rm_list = rm_list; 726 727 for (i = 0; i < opts->rm_list.nr; i++) 728 pr_debug("rm_list_ids[%d]=%d", i, opts->rm_list.ids[i]); 729 730 return true; 731 } 732 733 static bool mptcp_established_options_mp_prio(struct sock *sk, 734 unsigned int *size, 735 unsigned int remaining, 736 struct mptcp_out_options *opts) 737 { 738 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 739 740 /* can't send MP_PRIO with MPC, as they share the same option space: 741 * 'backup'. Also it makes no sense at all 742 */ 743 if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC)) 744 return false; 745 746 /* account for the trailing 'nop' option */ 747 if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN) 748 return false; 749 750 *size = TCPOLEN_MPTCP_PRIO_ALIGN; 751 opts->suboptions |= OPTION_MPTCP_PRIO; 752 opts->backup = subflow->request_bkup; 753 754 pr_debug("prio=%d", opts->backup); 755 756 return true; 757 } 758 759 static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb, 760 unsigned int *size, 761 unsigned int remaining, 762 struct mptcp_out_options *opts) 763 { 764 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 765 766 if (remaining < TCPOLEN_MPTCP_RST) 767 return false; 768 769 *size = TCPOLEN_MPTCP_RST; 770 opts->suboptions |= OPTION_MPTCP_RST; 771 opts->reset_transient = subflow->reset_transient; 772 opts->reset_reason = subflow->reset_reason; 773 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTTX); 774 775 return true; 776 } 777 778 static bool mptcp_established_options_fastclose(struct sock *sk, 779 unsigned int *size, 780 unsigned int remaining, 781 struct mptcp_out_options *opts) 782 { 783 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 784 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 785 786 if (likely(!subflow->send_fastclose)) 787 return false; 788 789 if (remaining < TCPOLEN_MPTCP_FASTCLOSE) 790 return false; 791 792 *size = TCPOLEN_MPTCP_FASTCLOSE; 793 opts->suboptions |= OPTION_MPTCP_FASTCLOSE; 794 opts->rcvr_key = msk->remote_key; 795 796 pr_debug("FASTCLOSE key=%llu", opts->rcvr_key); 797 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSETX); 798 return true; 799 } 800 801 static bool mptcp_established_options_mp_fail(struct sock *sk, 802 unsigned int *size, 803 unsigned int remaining, 804 struct mptcp_out_options *opts) 805 { 806 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 807 808 if (likely(!subflow->send_mp_fail)) 809 return false; 810 811 if (remaining < TCPOLEN_MPTCP_FAIL) 812 return false; 813 814 *size = TCPOLEN_MPTCP_FAIL; 815 opts->suboptions |= OPTION_MPTCP_FAIL; 816 opts->fail_seq = subflow->map_seq; 817 818 pr_debug("MP_FAIL fail_seq=%llu", opts->fail_seq); 819 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILTX); 820 821 return true; 822 } 823 824 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb, 825 unsigned int *size, unsigned int remaining, 826 struct mptcp_out_options *opts) 827 { 828 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 829 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 830 unsigned int opt_size = 0; 831 bool snd_data_fin; 832 bool ret = false; 833 834 opts->suboptions = 0; 835 836 if (unlikely(__mptcp_check_fallback(msk) && !mptcp_check_infinite_map(skb))) 837 return false; 838 839 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) { 840 if (mptcp_established_options_fastclose(sk, &opt_size, remaining, opts) || 841 mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) { 842 *size += opt_size; 843 remaining -= opt_size; 844 } 845 /* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */ 846 if (mptcp_established_options_rst(sk, skb, &opt_size, remaining, opts)) { 847 *size += opt_size; 848 remaining -= opt_size; 849 } 850 return true; 851 } 852 853 snd_data_fin = mptcp_data_fin_enabled(msk); 854 if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, remaining, opts)) 855 ret = true; 856 else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, remaining, opts)) { 857 unsigned int mp_fail_size; 858 859 ret = true; 860 if (mptcp_established_options_mp_fail(sk, &mp_fail_size, 861 remaining - opt_size, opts)) { 862 *size += opt_size + mp_fail_size; 863 remaining -= opt_size - mp_fail_size; 864 return true; 865 } 866 } 867 868 /* we reserved enough space for the above options, and exceeding the 869 * TCP option space would be fatal 870 */ 871 if (WARN_ON_ONCE(opt_size > remaining)) 872 return false; 873 874 *size += opt_size; 875 remaining -= opt_size; 876 if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) { 877 *size += opt_size; 878 remaining -= opt_size; 879 ret = true; 880 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) { 881 *size += opt_size; 882 remaining -= opt_size; 883 ret = true; 884 } 885 886 if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) { 887 *size += opt_size; 888 remaining -= opt_size; 889 ret = true; 890 } 891 892 return ret; 893 } 894 895 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size, 896 struct mptcp_out_options *opts) 897 { 898 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 899 900 if (subflow_req->mp_capable) { 901 opts->suboptions = OPTION_MPTCP_MPC_SYNACK; 902 opts->sndr_key = subflow_req->local_key; 903 opts->csum_reqd = subflow_req->csum_reqd; 904 opts->allow_join_id0 = subflow_req->allow_join_id0; 905 *size = TCPOLEN_MPTCP_MPC_SYNACK; 906 pr_debug("subflow_req=%p, local_key=%llu", 907 subflow_req, subflow_req->local_key); 908 return true; 909 } else if (subflow_req->mp_join) { 910 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK; 911 opts->backup = subflow_req->backup; 912 opts->join_id = subflow_req->local_id; 913 opts->thmac = subflow_req->thmac; 914 opts->nonce = subflow_req->local_nonce; 915 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u", 916 subflow_req, opts->backup, opts->join_id, 917 opts->thmac, opts->nonce); 918 *size = TCPOLEN_MPTCP_MPJ_SYNACK; 919 return true; 920 } 921 return false; 922 } 923 924 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk, 925 struct mptcp_subflow_context *subflow, 926 struct sk_buff *skb, 927 struct mptcp_options_received *mp_opt) 928 { 929 /* here we can process OoO, in-window pkts, only in-sequence 4th ack 930 * will make the subflow fully established 931 */ 932 if (likely(subflow->fully_established)) { 933 /* on passive sockets, check for 3rd ack retransmission 934 * note that msk is always set by subflow_syn_recv_sock() 935 * for mp_join subflows 936 */ 937 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 && 938 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq && 939 subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) && 940 !subflow->request_join) 941 tcp_send_ack(ssk); 942 goto check_notify; 943 } 944 945 /* we must process OoO packets before the first subflow is fully 946 * established. OoO packets are instead a protocol violation 947 * for MP_JOIN subflows as the peer must not send any data 948 * before receiving the forth ack - cfr. RFC 8684 section 3.2. 949 */ 950 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) { 951 if (subflow->mp_join) 952 goto reset; 953 if (subflow->is_mptfo && mp_opt->suboptions & OPTION_MPTCP_MPC_ACK) 954 goto set_fully_established; 955 return subflow->mp_capable; 956 } 957 958 if (subflow->remote_key_valid && 959 (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) || 960 ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) && !mp_opt->echo))) { 961 /* subflows are fully established as soon as we get any 962 * additional ack, including ADD_ADDR. 963 */ 964 subflow->fully_established = 1; 965 WRITE_ONCE(msk->fully_established, true); 966 goto check_notify; 967 } 968 969 /* If the first established packet does not contain MP_CAPABLE + data 970 * then fallback to TCP. Fallback scenarios requires a reset for 971 * MP_JOIN subflows. 972 */ 973 if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) { 974 if (subflow->mp_join) 975 goto reset; 976 subflow->mp_capable = 0; 977 pr_fallback(msk); 978 mptcp_do_fallback(ssk); 979 return false; 980 } 981 982 if (mp_opt->deny_join_id0) 983 WRITE_ONCE(msk->pm.remote_deny_join_id0, true); 984 985 set_fully_established: 986 if (unlikely(!READ_ONCE(msk->pm.server_side))) 987 pr_warn_once("bogus mpc option on established client sk"); 988 mptcp_subflow_fully_established(subflow, mp_opt); 989 990 check_notify: 991 /* if the subflow is not already linked into the conn_list, we can't 992 * notify the PM: this subflow is still on the listener queue 993 * and the PM possibly acquiring the subflow lock could race with 994 * the listener close 995 */ 996 if (likely(subflow->pm_notified) || list_empty(&subflow->node)) 997 return true; 998 999 subflow->pm_notified = 1; 1000 if (subflow->mp_join) { 1001 clear_3rdack_retransmission(ssk); 1002 mptcp_pm_subflow_established(msk); 1003 } else { 1004 mptcp_pm_fully_established(msk, ssk, GFP_ATOMIC); 1005 } 1006 return true; 1007 1008 reset: 1009 mptcp_subflow_reset(ssk); 1010 return false; 1011 } 1012 1013 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq) 1014 { 1015 u32 old_seq32, cur_seq32; 1016 1017 old_seq32 = (u32)old_seq; 1018 cur_seq32 = (u32)cur_seq; 1019 cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32; 1020 if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32))) 1021 return cur_seq + (1LL << 32); 1022 1023 /* reverse wrap could happen, too */ 1024 if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32))) 1025 return cur_seq - (1LL << 32); 1026 return cur_seq; 1027 } 1028 1029 static void ack_update_msk(struct mptcp_sock *msk, 1030 struct sock *ssk, 1031 struct mptcp_options_received *mp_opt) 1032 { 1033 u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt); 1034 struct sock *sk = (struct sock *)msk; 1035 u64 old_snd_una; 1036 1037 mptcp_data_lock(sk); 1038 1039 /* avoid ack expansion on update conflict, to reduce the risk of 1040 * wrongly expanding to a future ack sequence number, which is way 1041 * more dangerous than missing an ack 1042 */ 1043 old_snd_una = msk->snd_una; 1044 new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64); 1045 1046 /* ACK for data not even sent yet? Ignore.*/ 1047 if (unlikely(after64(new_snd_una, snd_nxt))) 1048 new_snd_una = old_snd_una; 1049 1050 new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd; 1051 1052 if (after64(new_wnd_end, msk->wnd_end)) 1053 msk->wnd_end = new_wnd_end; 1054 1055 /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */ 1056 if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt))) 1057 __mptcp_check_push(sk, ssk); 1058 1059 if (after64(new_snd_una, old_snd_una)) { 1060 msk->snd_una = new_snd_una; 1061 __mptcp_data_acked(sk); 1062 } 1063 mptcp_data_unlock(sk); 1064 1065 trace_ack_update_msk(mp_opt->data_ack, 1066 old_snd_una, new_snd_una, 1067 new_wnd_end, msk->wnd_end); 1068 } 1069 1070 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit) 1071 { 1072 /* Skip if DATA_FIN was already received. 1073 * If updating simultaneously with the recvmsg loop, values 1074 * should match. If they mismatch, the peer is misbehaving and 1075 * we will prefer the most recent information. 1076 */ 1077 if (READ_ONCE(msk->rcv_data_fin)) 1078 return false; 1079 1080 WRITE_ONCE(msk->rcv_data_fin_seq, 1081 mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit)); 1082 WRITE_ONCE(msk->rcv_data_fin, 1); 1083 1084 return true; 1085 } 1086 1087 static bool add_addr_hmac_valid(struct mptcp_sock *msk, 1088 struct mptcp_options_received *mp_opt) 1089 { 1090 u64 hmac = 0; 1091 1092 if (mp_opt->echo) 1093 return true; 1094 1095 hmac = add_addr_generate_hmac(msk->remote_key, 1096 msk->local_key, 1097 &mp_opt->addr); 1098 1099 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n", 1100 msk, hmac, mp_opt->ahmac); 1101 1102 return hmac == mp_opt->ahmac; 1103 } 1104 1105 /* Return false if a subflow has been reset, else return true */ 1106 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb) 1107 { 1108 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 1109 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 1110 struct mptcp_options_received mp_opt; 1111 struct mptcp_ext *mpext; 1112 1113 if (__mptcp_check_fallback(msk)) { 1114 /* Keep it simple and unconditionally trigger send data cleanup and 1115 * pending queue spooling. We will need to acquire the data lock 1116 * for more accurate checks, and once the lock is acquired, such 1117 * helpers are cheap. 1118 */ 1119 mptcp_data_lock(subflow->conn); 1120 if (sk_stream_memory_free(sk)) 1121 __mptcp_check_push(subflow->conn, sk); 1122 __mptcp_data_acked(subflow->conn); 1123 mptcp_data_unlock(subflow->conn); 1124 return true; 1125 } 1126 1127 mptcp_get_options(skb, &mp_opt); 1128 1129 /* The subflow can be in close state only if check_fully_established() 1130 * just sent a reset. If so, tell the caller to ignore the current packet. 1131 */ 1132 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt)) 1133 return sk->sk_state != TCP_CLOSE; 1134 1135 if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) { 1136 if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) && 1137 msk->local_key == mp_opt.rcvr_key) { 1138 WRITE_ONCE(msk->rcv_fastclose, true); 1139 mptcp_schedule_work((struct sock *)msk); 1140 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSERX); 1141 } 1142 1143 if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) && 1144 add_addr_hmac_valid(msk, &mp_opt)) { 1145 if (!mp_opt.echo) { 1146 mptcp_pm_add_addr_received(sk, &mp_opt.addr); 1147 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR); 1148 } else { 1149 mptcp_pm_add_addr_echoed(msk, &mp_opt.addr); 1150 mptcp_pm_del_add_timer(msk, &mp_opt.addr, true); 1151 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD); 1152 } 1153 1154 if (mp_opt.addr.port) 1155 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD); 1156 } 1157 1158 if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR) 1159 mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list); 1160 1161 if (mp_opt.suboptions & OPTION_MPTCP_PRIO) { 1162 mptcp_pm_mp_prio_received(sk, mp_opt.backup); 1163 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX); 1164 } 1165 1166 if (mp_opt.suboptions & OPTION_MPTCP_FAIL) { 1167 mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq); 1168 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX); 1169 } 1170 1171 if (mp_opt.suboptions & OPTION_MPTCP_RST) { 1172 subflow->reset_seen = 1; 1173 subflow->reset_reason = mp_opt.reset_reason; 1174 subflow->reset_transient = mp_opt.reset_transient; 1175 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTRX); 1176 } 1177 1178 if (!(mp_opt.suboptions & OPTION_MPTCP_DSS)) 1179 return true; 1180 } 1181 1182 /* we can't wait for recvmsg() to update the ack_seq, otherwise 1183 * monodirectional flows will stuck 1184 */ 1185 if (mp_opt.use_ack) 1186 ack_update_msk(msk, sk, &mp_opt); 1187 1188 /* Zero-data-length packets are dropped by the caller and not 1189 * propagated to the MPTCP layer, so the skb extension does not 1190 * need to be allocated or populated. DATA_FIN information, if 1191 * present, needs to be updated here before the skb is freed. 1192 */ 1193 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) { 1194 if (mp_opt.data_fin && mp_opt.data_len == 1 && 1195 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) && 1196 schedule_work(&msk->work)) 1197 sock_hold(subflow->conn); 1198 1199 return true; 1200 } 1201 1202 mpext = skb_ext_add(skb, SKB_EXT_MPTCP); 1203 if (!mpext) 1204 return true; 1205 1206 memset(mpext, 0, sizeof(*mpext)); 1207 1208 if (likely(mp_opt.use_map)) { 1209 if (mp_opt.mpc_map) { 1210 /* this is an MP_CAPABLE carrying MPTCP data 1211 * we know this map the first chunk of data 1212 */ 1213 mptcp_crypto_key_sha(subflow->remote_key, NULL, 1214 &mpext->data_seq); 1215 mpext->data_seq++; 1216 mpext->subflow_seq = 1; 1217 mpext->dsn64 = 1; 1218 mpext->mpc_map = 1; 1219 mpext->data_fin = 0; 1220 } else { 1221 mpext->data_seq = mp_opt.data_seq; 1222 mpext->subflow_seq = mp_opt.subflow_seq; 1223 mpext->dsn64 = mp_opt.dsn64; 1224 mpext->data_fin = mp_opt.data_fin; 1225 } 1226 mpext->data_len = mp_opt.data_len; 1227 mpext->use_map = 1; 1228 mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD); 1229 1230 if (mpext->csum_reqd) 1231 mpext->csum = mp_opt.csum; 1232 } 1233 1234 return true; 1235 } 1236 1237 static void mptcp_set_rwin(struct tcp_sock *tp, struct tcphdr *th) 1238 { 1239 const struct sock *ssk = (const struct sock *)tp; 1240 struct mptcp_subflow_context *subflow; 1241 u64 ack_seq, rcv_wnd_old, rcv_wnd_new; 1242 struct mptcp_sock *msk; 1243 u32 new_win; 1244 u64 win; 1245 1246 subflow = mptcp_subflow_ctx(ssk); 1247 msk = mptcp_sk(subflow->conn); 1248 1249 ack_seq = READ_ONCE(msk->ack_seq); 1250 rcv_wnd_new = ack_seq + tp->rcv_wnd; 1251 1252 rcv_wnd_old = atomic64_read(&msk->rcv_wnd_sent); 1253 if (after64(rcv_wnd_new, rcv_wnd_old)) { 1254 u64 rcv_wnd; 1255 1256 for (;;) { 1257 rcv_wnd = atomic64_cmpxchg(&msk->rcv_wnd_sent, rcv_wnd_old, rcv_wnd_new); 1258 1259 if (rcv_wnd == rcv_wnd_old) 1260 break; 1261 if (before64(rcv_wnd_new, rcv_wnd)) { 1262 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICTUPDATE); 1263 goto raise_win; 1264 } 1265 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICT); 1266 rcv_wnd_old = rcv_wnd; 1267 } 1268 return; 1269 } 1270 1271 if (rcv_wnd_new != rcv_wnd_old) { 1272 raise_win: 1273 win = rcv_wnd_old - ack_seq; 1274 tp->rcv_wnd = min_t(u64, win, U32_MAX); 1275 new_win = tp->rcv_wnd; 1276 1277 /* Make sure we do not exceed the maximum possible 1278 * scaled window. 1279 */ 1280 if (unlikely(th->syn)) 1281 new_win = min(new_win, 65535U) << tp->rx_opt.rcv_wscale; 1282 if (!tp->rx_opt.rcv_wscale && 1283 READ_ONCE(sock_net(ssk)->ipv4.sysctl_tcp_workaround_signed_windows)) 1284 new_win = min(new_win, MAX_TCP_WINDOW); 1285 else 1286 new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale)); 1287 1288 /* RFC1323 scaling applied */ 1289 new_win >>= tp->rx_opt.rcv_wscale; 1290 th->window = htons(new_win); 1291 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDSHARED); 1292 } 1293 } 1294 1295 __sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum) 1296 { 1297 struct csum_pseudo_header header; 1298 __wsum csum; 1299 1300 /* cfr RFC 8684 3.3.1.: 1301 * the data sequence number used in the pseudo-header is 1302 * always the 64-bit value, irrespective of what length is used in the 1303 * DSS option itself. 1304 */ 1305 header.data_seq = cpu_to_be64(data_seq); 1306 header.subflow_seq = htonl(subflow_seq); 1307 header.data_len = htons(data_len); 1308 header.csum = 0; 1309 1310 csum = csum_partial(&header, sizeof(header), sum); 1311 return csum_fold(csum); 1312 } 1313 1314 static __sum16 mptcp_make_csum(const struct mptcp_ext *mpext) 1315 { 1316 return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len, 1317 ~csum_unfold(mpext->csum)); 1318 } 1319 1320 static void put_len_csum(u16 len, __sum16 csum, void *data) 1321 { 1322 __sum16 *sumptr = data + 2; 1323 __be16 *ptr = data; 1324 1325 put_unaligned_be16(len, ptr); 1326 1327 put_unaligned(csum, sumptr); 1328 } 1329 1330 void mptcp_write_options(struct tcphdr *th, __be32 *ptr, struct tcp_sock *tp, 1331 struct mptcp_out_options *opts) 1332 { 1333 const struct sock *ssk = (const struct sock *)tp; 1334 struct mptcp_subflow_context *subflow; 1335 1336 /* Which options can be used together? 1337 * 1338 * X: mutually exclusive 1339 * O: often used together 1340 * C: can be used together in some cases 1341 * P: could be used together but we prefer not to (optimisations) 1342 * 1343 * Opt: | MPC | MPJ | DSS | ADD | RM | PRIO | FAIL | FC | 1344 * ------|------|------|------|------|------|------|------|------| 1345 * MPC |------|------|------|------|------|------|------|------| 1346 * MPJ | X |------|------|------|------|------|------|------| 1347 * DSS | X | X |------|------|------|------|------|------| 1348 * ADD | X | X | P |------|------|------|------|------| 1349 * RM | C | C | C | P |------|------|------|------| 1350 * PRIO | X | C | C | C | C |------|------|------| 1351 * FAIL | X | X | C | X | X | X |------|------| 1352 * FC | X | X | X | X | X | X | X |------| 1353 * RST | X | X | X | X | X | X | O | O | 1354 * ------|------|------|------|------|------|------|------|------| 1355 * 1356 * The same applies in mptcp_established_options() function. 1357 */ 1358 if (likely(OPTION_MPTCP_DSS & opts->suboptions)) { 1359 struct mptcp_ext *mpext = &opts->ext_copy; 1360 u8 len = TCPOLEN_MPTCP_DSS_BASE; 1361 u8 flags = 0; 1362 1363 if (mpext->use_ack) { 1364 flags = MPTCP_DSS_HAS_ACK; 1365 if (mpext->ack64) { 1366 len += TCPOLEN_MPTCP_DSS_ACK64; 1367 flags |= MPTCP_DSS_ACK64; 1368 } else { 1369 len += TCPOLEN_MPTCP_DSS_ACK32; 1370 } 1371 } 1372 1373 if (mpext->use_map) { 1374 len += TCPOLEN_MPTCP_DSS_MAP64; 1375 1376 /* Use only 64-bit mapping flags for now, add 1377 * support for optional 32-bit mappings later. 1378 */ 1379 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64; 1380 if (mpext->data_fin) 1381 flags |= MPTCP_DSS_DATA_FIN; 1382 1383 if (opts->csum_reqd) 1384 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 1385 } 1386 1387 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags); 1388 1389 if (mpext->use_ack) { 1390 if (mpext->ack64) { 1391 put_unaligned_be64(mpext->data_ack, ptr); 1392 ptr += 2; 1393 } else { 1394 put_unaligned_be32(mpext->data_ack32, ptr); 1395 ptr += 1; 1396 } 1397 } 1398 1399 if (mpext->use_map) { 1400 put_unaligned_be64(mpext->data_seq, ptr); 1401 ptr += 2; 1402 put_unaligned_be32(mpext->subflow_seq, ptr); 1403 ptr += 1; 1404 if (opts->csum_reqd) { 1405 /* data_len == 0 is reserved for the infinite mapping, 1406 * the checksum will also be set to 0. 1407 */ 1408 put_len_csum(mpext->data_len, 1409 (mpext->data_len ? mptcp_make_csum(mpext) : 0), 1410 ptr); 1411 } else { 1412 put_unaligned_be32(mpext->data_len << 16 | 1413 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1414 } 1415 ptr += 1; 1416 } 1417 1418 /* We might need to add MP_FAIL options in rare cases */ 1419 if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) 1420 goto mp_fail; 1421 } else if (OPTIONS_MPTCP_MPC & opts->suboptions) { 1422 u8 len, flag = MPTCP_CAP_HMAC_SHA256; 1423 1424 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) { 1425 len = TCPOLEN_MPTCP_MPC_SYN; 1426 } else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) { 1427 len = TCPOLEN_MPTCP_MPC_SYNACK; 1428 } else if (opts->data_len) { 1429 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 1430 if (opts->csum_reqd) 1431 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 1432 } else { 1433 len = TCPOLEN_MPTCP_MPC_ACK; 1434 } 1435 1436 if (opts->csum_reqd) 1437 flag |= MPTCP_CAP_CHECKSUM_REQD; 1438 1439 if (!opts->allow_join_id0) 1440 flag |= MPTCP_CAP_DENY_JOIN_ID0; 1441 1442 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len, 1443 MPTCP_SUPPORTED_VERSION, 1444 flag); 1445 1446 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) & 1447 opts->suboptions)) 1448 goto mp_capable_done; 1449 1450 put_unaligned_be64(opts->sndr_key, ptr); 1451 ptr += 2; 1452 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions)) 1453 goto mp_capable_done; 1454 1455 put_unaligned_be64(opts->rcvr_key, ptr); 1456 ptr += 2; 1457 if (!opts->data_len) 1458 goto mp_capable_done; 1459 1460 if (opts->csum_reqd) { 1461 put_len_csum(opts->data_len, 1462 __mptcp_make_csum(opts->data_seq, 1463 opts->subflow_seq, 1464 opts->data_len, 1465 ~csum_unfold(opts->csum)), 1466 ptr); 1467 } else { 1468 put_unaligned_be32(opts->data_len << 16 | 1469 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1470 } 1471 ptr += 1; 1472 1473 /* MPC is additionally mutually exclusive with MP_PRIO */ 1474 goto mp_capable_done; 1475 } else if (OPTIONS_MPTCP_MPJ & opts->suboptions) { 1476 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) { 1477 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1478 TCPOLEN_MPTCP_MPJ_SYN, 1479 opts->backup, opts->join_id); 1480 put_unaligned_be32(opts->token, ptr); 1481 ptr += 1; 1482 put_unaligned_be32(opts->nonce, ptr); 1483 ptr += 1; 1484 } else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) { 1485 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1486 TCPOLEN_MPTCP_MPJ_SYNACK, 1487 opts->backup, opts->join_id); 1488 put_unaligned_be64(opts->thmac, ptr); 1489 ptr += 2; 1490 put_unaligned_be32(opts->nonce, ptr); 1491 ptr += 1; 1492 } else { 1493 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1494 TCPOLEN_MPTCP_MPJ_ACK, 0, 0); 1495 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN); 1496 ptr += 5; 1497 } 1498 } else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) { 1499 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE; 1500 u8 echo = MPTCP_ADDR_ECHO; 1501 1502 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1503 if (opts->addr.family == AF_INET6) 1504 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE; 1505 #endif 1506 1507 if (opts->addr.port) 1508 len += TCPOLEN_MPTCP_PORT_LEN; 1509 1510 if (opts->ahmac) { 1511 len += sizeof(opts->ahmac); 1512 echo = 0; 1513 } 1514 1515 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 1516 len, echo, opts->addr.id); 1517 if (opts->addr.family == AF_INET) { 1518 memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4); 1519 ptr += 1; 1520 } 1521 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1522 else if (opts->addr.family == AF_INET6) { 1523 memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16); 1524 ptr += 4; 1525 } 1526 #endif 1527 1528 if (!opts->addr.port) { 1529 if (opts->ahmac) { 1530 put_unaligned_be64(opts->ahmac, ptr); 1531 ptr += 2; 1532 } 1533 } else { 1534 u16 port = ntohs(opts->addr.port); 1535 1536 if (opts->ahmac) { 1537 u8 *bptr = (u8 *)ptr; 1538 1539 put_unaligned_be16(port, bptr); 1540 bptr += 2; 1541 put_unaligned_be64(opts->ahmac, bptr); 1542 bptr += 8; 1543 put_unaligned_be16(TCPOPT_NOP << 8 | 1544 TCPOPT_NOP, bptr); 1545 1546 ptr += 3; 1547 } else { 1548 put_unaligned_be32(port << 16 | 1549 TCPOPT_NOP << 8 | 1550 TCPOPT_NOP, ptr); 1551 ptr += 1; 1552 } 1553 } 1554 } else if (unlikely(OPTION_MPTCP_FASTCLOSE & opts->suboptions)) { 1555 /* FASTCLOSE is mutually exclusive with others except RST */ 1556 *ptr++ = mptcp_option(MPTCPOPT_MP_FASTCLOSE, 1557 TCPOLEN_MPTCP_FASTCLOSE, 1558 0, 0); 1559 put_unaligned_be64(opts->rcvr_key, ptr); 1560 ptr += 2; 1561 1562 if (OPTION_MPTCP_RST & opts->suboptions) 1563 goto mp_rst; 1564 return; 1565 } else if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) { 1566 mp_fail: 1567 /* MP_FAIL is mutually exclusive with others except RST */ 1568 subflow = mptcp_subflow_ctx(ssk); 1569 subflow->send_mp_fail = 0; 1570 1571 *ptr++ = mptcp_option(MPTCPOPT_MP_FAIL, 1572 TCPOLEN_MPTCP_FAIL, 1573 0, 0); 1574 put_unaligned_be64(opts->fail_seq, ptr); 1575 ptr += 2; 1576 1577 if (OPTION_MPTCP_RST & opts->suboptions) 1578 goto mp_rst; 1579 return; 1580 } else if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) { 1581 mp_rst: 1582 *ptr++ = mptcp_option(MPTCPOPT_RST, 1583 TCPOLEN_MPTCP_RST, 1584 opts->reset_transient, 1585 opts->reset_reason); 1586 return; 1587 } 1588 1589 if (OPTION_MPTCP_PRIO & opts->suboptions) { 1590 subflow = mptcp_subflow_ctx(ssk); 1591 subflow->send_mp_prio = 0; 1592 1593 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO, 1594 TCPOLEN_MPTCP_PRIO, 1595 opts->backup, TCPOPT_NOP); 1596 1597 MPTCP_INC_STATS(sock_net((const struct sock *)tp), 1598 MPTCP_MIB_MPPRIOTX); 1599 } 1600 1601 mp_capable_done: 1602 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) { 1603 u8 i = 1; 1604 1605 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR, 1606 TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr, 1607 0, opts->rm_list.ids[0]); 1608 1609 while (i < opts->rm_list.nr) { 1610 u8 id1, id2, id3, id4; 1611 1612 id1 = opts->rm_list.ids[i]; 1613 id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP; 1614 id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP; 1615 id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP; 1616 put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr); 1617 ptr += 1; 1618 i += 4; 1619 } 1620 } 1621 1622 if (tp) 1623 mptcp_set_rwin(tp, th); 1624 } 1625 1626 __be32 mptcp_get_reset_option(const struct sk_buff *skb) 1627 { 1628 const struct mptcp_ext *ext = mptcp_get_ext(skb); 1629 u8 flags, reason; 1630 1631 if (ext) { 1632 flags = ext->reset_transient; 1633 reason = ext->reset_reason; 1634 1635 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST, 1636 flags, reason); 1637 } 1638 1639 return htonl(0u); 1640 } 1641 EXPORT_SYMBOL_GPL(mptcp_get_reset_option); 1642