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