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 /* Only the MPC + ACK can be used with a RM_ADDR */ 54 if (subopt == OPTION_MPTCP_MPC_ACK) { 55 if ((mp_opt->suboptions & ~OPTION_MPTCP_RM_ADDR) != 0) 56 break; 57 } else if (mp_opt->suboptions != 0) { 58 break; 59 } 60 61 /* Cfr RFC 8684 Section 3.3.0: 62 * If a checksum is present but its use had 63 * not been negotiated in the MP_CAPABLE handshake, the receiver MUST 64 * close the subflow with a RST, as it is not behaving as negotiated. 65 * If a checksum is not present when its use has been negotiated, the 66 * receiver MUST close the subflow with a RST, as it is considered 67 * broken 68 * We parse even option with mismatching csum presence, so that 69 * later in subflow_data_ready we can trigger the reset. 70 */ 71 if (opsize != expected_opsize && 72 (expected_opsize != TCPOLEN_MPTCP_MPC_ACK_DATA || 73 opsize != TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM)) 74 break; 75 76 /* try to be gentle vs future versions on the initial syn */ 77 version = *ptr++ & MPTCP_VERSION_MASK; 78 if (opsize != TCPOLEN_MPTCP_MPC_SYN) { 79 if (version != MPTCP_SUPPORTED_VERSION) 80 break; 81 } else if (version < MPTCP_SUPPORTED_VERSION) { 82 break; 83 } 84 85 flags = *ptr++; 86 if (!mptcp_cap_flag_sha256(flags) || 87 (flags & MPTCP_CAP_EXTENSIBILITY)) 88 break; 89 90 /* RFC 6824, Section 3.1: 91 * "For the Checksum Required bit (labeled "A"), if either 92 * host requires the use of checksums, checksums MUST be used. 93 * In other words, the only way for checksums not to be used 94 * is if both hosts in their SYNs set A=0." 95 */ 96 if ((flags & MPTCP_CAP_CHECKSUM_REQD) && 97 opsize < TCPOLEN_MPTCP_MPC_ACK_DATA) 98 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD; 99 100 mp_opt->deny_join_id0 = !!(flags & MPTCP_CAP_DENY_JOIN_ID0); 101 102 mp_opt->suboptions |= subopt; 103 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) { 104 mp_opt->sndr_key = get_unaligned_be64(ptr); 105 ptr += 8; 106 } 107 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) { 108 mp_opt->rcvr_key = get_unaligned_be64(ptr); 109 ptr += 8; 110 } 111 if (opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA) { 112 /* Section 3.1.: 113 * "the data parameters in a MP_CAPABLE are semantically 114 * equivalent to those in a DSS option and can be used 115 * interchangeably." 116 */ 117 mp_opt->suboptions |= OPTION_MPTCP_DSS; 118 mp_opt->use_map = 1; 119 mp_opt->mpc_map = 1; 120 mp_opt->data_len = get_unaligned_be16(ptr); 121 ptr += 2; 122 } 123 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM) { 124 mp_opt->csum = get_unaligned((__force __sum16 *)ptr); 125 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD; 126 ptr += 2; 127 } 128 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d csum=%u\n", 129 version, flags, opsize, 130 opsize >= TCPOLEN_MPTCP_MPC_SYNACK ? mp_opt->sndr_key : 0, 131 opsize >= TCPOLEN_MPTCP_MPC_ACK ? mp_opt->rcvr_key : 0, 132 opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA ? mp_opt->data_len : 0, 133 opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM ? mp_opt->csum : 0); 134 break; 135 136 case MPTCPOPT_MP_JOIN: 137 /* Can be used with a restricted number of other options */ 138 if ((mp_opt->suboptions & ~(OPTION_MPTCP_RM_ADDR | 139 OPTION_MPTCP_PRIO)) != 0) 140 break; 141 142 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) { 143 mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYN; 144 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 145 mp_opt->join_id = *ptr++; 146 mp_opt->token = get_unaligned_be32(ptr); 147 ptr += 4; 148 mp_opt->nonce = get_unaligned_be32(ptr); 149 ptr += 4; 150 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u\n", 151 mp_opt->backup, mp_opt->join_id, 152 mp_opt->token, mp_opt->nonce); 153 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) { 154 mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYNACK; 155 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 156 mp_opt->join_id = *ptr++; 157 mp_opt->thmac = get_unaligned_be64(ptr); 158 ptr += 8; 159 mp_opt->nonce = get_unaligned_be32(ptr); 160 ptr += 4; 161 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u\n", 162 mp_opt->backup, mp_opt->join_id, 163 mp_opt->thmac, mp_opt->nonce); 164 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) { 165 mp_opt->suboptions |= OPTION_MPTCP_MPJ_ACK; 166 ptr += 2; 167 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN); 168 pr_debug("MP_JOIN hmac\n"); 169 } 170 break; 171 172 case MPTCPOPT_DSS: 173 /* Can be used with a restricted number of other options */ 174 if ((mp_opt->suboptions & ~(OPTION_MPTCP_ADD_ADDR | 175 OPTION_MPTCP_RM_ADDR | 176 OPTION_MPTCP_PRIO | 177 OPTION_MPTCP_FASTCLOSE | 178 OPTION_MPTCP_FAIL)) != 0) 179 break; 180 181 pr_debug("DSS\n"); 182 ptr++; 183 184 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK; 185 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0; 186 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0; 187 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0; 188 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK); 189 190 expected_opsize = TCPOLEN_MPTCP_DSS_BASE; 191 192 if (mp_opt->use_ack) { 193 if (mp_opt->ack64) 194 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64; 195 else 196 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32; 197 } 198 199 if (mp_opt->use_map) { 200 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0; 201 if (mp_opt->dsn64) 202 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64; 203 else 204 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32; 205 } 206 207 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d\n", 208 mp_opt->data_fin, mp_opt->dsn64, 209 mp_opt->use_map, mp_opt->ack64, 210 mp_opt->use_ack); 211 212 /* Always parse any csum presence combination, we will enforce 213 * RFC 8684 Section 3.3.0 checks later in subflow_data_ready 214 */ 215 if (opsize != expected_opsize && 216 opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) { 217 mp_opt->dsn64 = 0; 218 mp_opt->use_map = 0; 219 mp_opt->ack64 = 0; 220 mp_opt->use_ack = 0; 221 mp_opt->data_fin = 0; 222 break; 223 } 224 225 mp_opt->suboptions |= OPTION_MPTCP_DSS; 226 if (mp_opt->use_ack) { 227 if (mp_opt->ack64) { 228 mp_opt->data_ack = get_unaligned_be64(ptr); 229 ptr += 8; 230 } else { 231 mp_opt->data_ack = get_unaligned_be32(ptr); 232 ptr += 4; 233 } 234 235 pr_debug("data_ack=%llu\n", mp_opt->data_ack); 236 } 237 238 if (mp_opt->use_map) { 239 if (mp_opt->dsn64) { 240 mp_opt->data_seq = get_unaligned_be64(ptr); 241 ptr += 8; 242 } else { 243 mp_opt->data_seq = get_unaligned_be32(ptr); 244 ptr += 4; 245 } 246 247 mp_opt->subflow_seq = get_unaligned_be32(ptr); 248 ptr += 4; 249 250 mp_opt->data_len = get_unaligned_be16(ptr); 251 ptr += 2; 252 253 if (opsize == expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) { 254 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD; 255 mp_opt->csum = get_unaligned((__force __sum16 *)ptr); 256 ptr += 2; 257 } 258 259 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u\n", 260 mp_opt->data_seq, mp_opt->subflow_seq, 261 mp_opt->data_len, !!(mp_opt->suboptions & OPTION_MPTCP_CSUMREQD), 262 (mp_opt->suboptions & OPTION_MPTCP_CSUMREQD) ? mp_opt->csum : 0); 263 } 264 265 break; 266 267 case MPTCPOPT_ADD_ADDR: 268 /* Can be used with a restricted number of other options */ 269 if ((mp_opt->suboptions & ~(OPTIONS_MPTCP_DSS | 270 OPTION_MPTCP_RM_ADDR | 271 OPTION_MPTCP_PRIO)) != 0) 272 break; 273 274 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO; 275 if (!mp_opt->echo) { 276 if (opsize == TCPOLEN_MPTCP_ADD_ADDR || 277 opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT) 278 mp_opt->addr.family = AF_INET; 279 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 280 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 || 281 opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT) 282 mp_opt->addr.family = AF_INET6; 283 #endif 284 else 285 break; 286 } else { 287 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE || 288 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) 289 mp_opt->addr.family = AF_INET; 290 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 291 else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE || 292 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) 293 mp_opt->addr.family = AF_INET6; 294 #endif 295 else 296 break; 297 } 298 299 mp_opt->suboptions |= OPTION_MPTCP_ADD_ADDR; 300 mp_opt->addr.id = *ptr++; 301 mp_opt->addr.port = 0; 302 mp_opt->ahmac = 0; 303 if (mp_opt->addr.family == AF_INET) { 304 memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4); 305 ptr += 4; 306 if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT || 307 opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) { 308 mp_opt->addr.port = htons(get_unaligned_be16(ptr)); 309 ptr += 2; 310 } 311 } 312 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 313 else { 314 memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16); 315 ptr += 16; 316 if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT || 317 opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) { 318 mp_opt->addr.port = htons(get_unaligned_be16(ptr)); 319 ptr += 2; 320 } 321 } 322 #endif 323 if (!mp_opt->echo) { 324 mp_opt->ahmac = get_unaligned_be64(ptr); 325 ptr += 8; 326 } 327 pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d\n", 328 (mp_opt->addr.family == AF_INET6) ? "6" : "", 329 mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port)); 330 break; 331 332 case MPTCPOPT_RM_ADDR: 333 /* Can be used with a restricted number of other options */ 334 if ((mp_opt->suboptions & ~(OPTION_MPTCP_MPC_ACK | 335 OPTIONS_MPTCP_MPJ | 336 OPTIONS_MPTCP_DSS | 337 OPTION_MPTCP_ADD_ADDR | 338 OPTION_MPTCP_PRIO)) != 0) 339 break; 340 341 if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 || 342 opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX) 343 break; 344 345 ptr++; 346 347 mp_opt->suboptions |= OPTION_MPTCP_RM_ADDR; 348 mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE; 349 for (i = 0; i < mp_opt->rm_list.nr; i++) 350 mp_opt->rm_list.ids[i] = *ptr++; 351 pr_debug("RM_ADDR: rm_list_nr=%d\n", mp_opt->rm_list.nr); 352 break; 353 354 case MPTCPOPT_MP_PRIO: 355 /* Can be used with a restricted number of other options */ 356 if ((mp_opt->suboptions & ~(OPTIONS_MPTCP_MPJ | 357 OPTIONS_MPTCP_DSS | 358 OPTION_MPTCP_ADD_ADDR | 359 OPTION_MPTCP_RM_ADDR)) != 0) 360 break; 361 362 if (opsize != TCPOLEN_MPTCP_PRIO) 363 break; 364 365 mp_opt->suboptions |= OPTION_MPTCP_PRIO; 366 mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP; 367 pr_debug("MP_PRIO: prio=%d\n", mp_opt->backup); 368 break; 369 370 case MPTCPOPT_MP_FASTCLOSE: 371 /* Can be used with a restricted number of other options */ 372 if ((mp_opt->suboptions & ~(OPTIONS_MPTCP_DSS | 373 OPTION_MPTCP_RST)) != 0) 374 break; 375 376 if (opsize != TCPOLEN_MPTCP_FASTCLOSE) 377 break; 378 379 ptr += 2; 380 mp_opt->fc_recv_key = get_unaligned_be64(ptr); 381 ptr += 8; 382 mp_opt->suboptions |= OPTION_MPTCP_FASTCLOSE; 383 pr_debug("MP_FASTCLOSE: recv_key=%llu\n", mp_opt->fc_recv_key); 384 break; 385 386 case MPTCPOPT_RST: 387 /* Can be used with a restricted number of other options */ 388 if ((mp_opt->suboptions & ~(OPTION_MPTCP_FAIL | 389 OPTION_MPTCP_FASTCLOSE)) != 0) 390 break; 391 392 if (opsize != TCPOLEN_MPTCP_RST) 393 break; 394 395 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) 396 break; 397 398 mp_opt->suboptions |= OPTION_MPTCP_RST; 399 flags = *ptr++; 400 mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT; 401 mp_opt->reset_reason = *ptr; 402 pr_debug("MP_RST: transient=%u reason=%u\n", 403 mp_opt->reset_transient, mp_opt->reset_reason); 404 break; 405 406 case MPTCPOPT_MP_FAIL: 407 /* Can be used with a restricted number of other options */ 408 if ((mp_opt->suboptions & ~(OPTIONS_MPTCP_DSS | 409 OPTION_MPTCP_RST)) != 0) 410 break; 411 412 if (opsize != TCPOLEN_MPTCP_FAIL) 413 break; 414 415 ptr += 2; 416 mp_opt->suboptions |= OPTION_MPTCP_FAIL; 417 mp_opt->fail_seq = get_unaligned_be64(ptr); 418 pr_debug("MP_FAIL: data_seq=%llu\n", mp_opt->fail_seq); 419 break; 420 421 default: 422 break; 423 } 424 } 425 426 void mptcp_get_options(const struct sk_buff *skb, 427 struct mptcp_options_received *mp_opt) 428 { 429 const struct tcphdr *th = tcp_hdr(skb); 430 const unsigned char *ptr; 431 int length; 432 433 /* Ensure that casting the whole status to u32 is efficient and safe */ 434 BUILD_BUG_ON(sizeof_field(struct mptcp_options_received, status) != sizeof(u32)); 435 BUILD_BUG_ON(!IS_ALIGNED(offsetof(struct mptcp_options_received, status), 436 sizeof(u32))); 437 *(u32 *)&mp_opt->status = 0; 438 439 length = (th->doff * 4) - sizeof(struct tcphdr); 440 ptr = (const unsigned char *)(th + 1); 441 442 while (length > 0) { 443 int opcode = *ptr++; 444 int opsize; 445 446 switch (opcode) { 447 case TCPOPT_EOL: 448 return; 449 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */ 450 length--; 451 continue; 452 default: 453 if (length < 2) 454 return; 455 opsize = *ptr++; 456 if (opsize < 2) /* "silly options" */ 457 return; 458 if (opsize > length) 459 return; /* don't parse partial options */ 460 if (opcode == TCPOPT_MPTCP) 461 mptcp_parse_option(skb, ptr, opsize, mp_opt); 462 ptr += opsize - 2; 463 length -= opsize; 464 } 465 } 466 } 467 468 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb, 469 unsigned int *size, struct mptcp_out_options *opts) 470 { 471 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 472 473 /* we will use snd_isn to detect first pkt [re]transmission 474 * in mptcp_established_options_mp() 475 */ 476 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq; 477 if (subflow->request_mptcp) { 478 if (unlikely(subflow_simultaneous_connect(sk))) { 479 WARN_ON_ONCE(!mptcp_try_fallback(sk, MPTCP_MIB_SIMULTCONNFALLBACK)); 480 481 /* Ensure mptcp_finish_connect() will not process the 482 * MPC handshake. 483 */ 484 subflow->request_mptcp = 0; 485 return false; 486 } 487 488 opts->suboptions = OPTION_MPTCP_MPC_SYN; 489 opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk)); 490 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk)); 491 *size = TCPOLEN_MPTCP_MPC_SYN; 492 return true; 493 } else if (subflow->request_join) { 494 pr_debug("remote_token=%u, nonce=%u\n", subflow->remote_token, 495 subflow->local_nonce); 496 opts->suboptions = OPTION_MPTCP_MPJ_SYN; 497 opts->join_id = subflow->local_id; 498 opts->token = subflow->remote_token; 499 opts->nonce = subflow->local_nonce; 500 opts->backup = subflow->request_bkup; 501 *size = TCPOLEN_MPTCP_MPJ_SYN; 502 return true; 503 } 504 return false; 505 } 506 507 static void clear_3rdack_retransmission(struct sock *sk) 508 { 509 struct inet_connection_sock *icsk = inet_csk(sk); 510 511 sk_stop_timer(sk, &icsk->icsk_delack_timer); 512 icsk->icsk_ack.ato = 0; 513 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER); 514 } 515 516 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb, 517 bool snd_data_fin_enable, int *size, 518 struct mptcp_out_options *opts) 519 { 520 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 521 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 522 struct mptcp_ext *mpext; 523 unsigned int data_len; 524 u8 len; 525 526 /* When skb is not available, we better over-estimate the emitted 527 * options len. A full DSS option (28 bytes) is longer than 528 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so 529 * tell the caller to defer the estimate to 530 * mptcp_established_options_dss(), which will reserve enough space. 531 */ 532 if (!skb) 533 return false; 534 535 /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */ 536 if (subflow->fully_established || snd_data_fin_enable || 537 subflow->snd_isn != TCP_SKB_CB(skb)->seq || 538 sk->sk_state != TCP_ESTABLISHED) 539 return false; 540 541 if (subflow->mp_capable) { 542 mpext = mptcp_get_ext(skb); 543 data_len = mpext ? mpext->data_len : 0; 544 545 /* we will check ops->data_len in mptcp_write_options() to 546 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and 547 * TCPOLEN_MPTCP_MPC_ACK 548 */ 549 opts->data_len = data_len; 550 opts->suboptions = OPTION_MPTCP_MPC_ACK; 551 opts->sndr_key = subflow->local_key; 552 opts->rcvr_key = subflow->remote_key; 553 opts->csum_reqd = READ_ONCE(msk->csum_enabled); 554 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk)); 555 556 /* Section 3.1. 557 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK 558 * packets that start the first subflow of an MPTCP connection, 559 * as well as the first packet that carries data 560 */ 561 if (data_len > 0) { 562 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 563 if (opts->csum_reqd) { 564 /* we need to propagate more info to csum the pseudo hdr */ 565 opts->data_seq = mpext->data_seq; 566 opts->subflow_seq = mpext->subflow_seq; 567 opts->csum = mpext->csum; 568 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 569 } 570 *size = ALIGN(len, 4); 571 } else { 572 *size = TCPOLEN_MPTCP_MPC_ACK; 573 } 574 575 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d\n", 576 subflow, subflow->local_key, subflow->remote_key, 577 data_len); 578 579 return true; 580 } else if (subflow->mp_join) { 581 opts->suboptions = OPTION_MPTCP_MPJ_ACK; 582 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN); 583 *size = TCPOLEN_MPTCP_MPJ_ACK; 584 pr_debug("subflow=%p\n", subflow); 585 586 /* we can use the full delegate action helper only from BH context 587 * If we are in process context - sk is flushing the backlog at 588 * socket lock release time - just set the appropriate flag, will 589 * be handled by the release callback 590 */ 591 if (sock_owned_by_user(sk)) 592 set_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status); 593 else 594 mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_ACK); 595 return true; 596 } 597 return false; 598 } 599 600 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow, 601 struct sk_buff *skb, struct mptcp_ext *ext) 602 { 603 /* The write_seq value has already been incremented, so the actual 604 * sequence number for the DATA_FIN is one less. 605 */ 606 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1; 607 608 if (!ext->use_map || !skb->len) { 609 /* RFC6824 requires a DSS mapping with specific values 610 * if DATA_FIN is set but no data payload is mapped 611 */ 612 ext->data_fin = 1; 613 ext->use_map = 1; 614 ext->dsn64 = 1; 615 ext->data_seq = data_fin_tx_seq; 616 ext->subflow_seq = 0; 617 ext->data_len = 1; 618 ext->csum = 0; 619 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) { 620 /* If there's an existing DSS mapping and it is the 621 * final mapping, DATA_FIN consumes 1 additional byte of 622 * mapping space. 623 */ 624 ext->data_fin = 1; 625 ext->data_len++; 626 } 627 } 628 629 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb, 630 bool snd_data_fin_enable, int *size, 631 struct mptcp_out_options *opts) 632 { 633 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 634 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 635 struct tcp_sock *tp = tcp_sk(sk); 636 unsigned int dss_size = 0; 637 struct mptcp_ext *mpext; 638 unsigned int ack_size; 639 bool ret = false; 640 641 /* Zero `use_ack` and `use_map` flags with one shot. */ 642 memset(&opts->ext_copy.flags, 0, sizeof(opts->ext_copy.flags)); 643 opts->csum_reqd = READ_ONCE(msk->csum_enabled); 644 mpext = skb ? mptcp_get_ext(skb) : NULL; 645 646 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) { 647 unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64; 648 649 if (mpext) { 650 if (opts->csum_reqd) 651 map_size += TCPOLEN_MPTCP_DSS_CHECKSUM; 652 653 opts->ext_copy = *mpext; 654 } 655 656 dss_size = map_size; 657 if (skb && snd_data_fin_enable) 658 mptcp_write_data_fin(subflow, skb, &opts->ext_copy); 659 opts->suboptions = OPTION_MPTCP_DSS; 660 ret = true; 661 } 662 663 /* passive sockets msk will set the 'can_ack' after accept(), even 664 * if the first subflow may have the already the remote key handy 665 */ 666 if (!READ_ONCE(msk->can_ack)) { 667 *size = ALIGN(dss_size, 4); 668 return ret; 669 } 670 671 if (READ_ONCE(msk->use_64bit_ack)) { 672 ack_size = TCPOLEN_MPTCP_DSS_ACK64; 673 opts->ext_copy.ack64 = 1; 674 } else { 675 ack_size = TCPOLEN_MPTCP_DSS_ACK32; 676 opts->ext_copy.ack64 = 0; 677 } 678 opts->ext_copy.use_ack = 1; 679 opts->suboptions = OPTION_MPTCP_DSS; 680 681 /* Add kind/length/subtype/flag overhead if mapping is not populated */ 682 if (dss_size == 0) 683 ack_size += TCPOLEN_MPTCP_DSS_BASE; 684 685 /* The caller is __tcp_transmit_skb(), and will compute the new rcv 686 * wnd soon: ensure that the window can shrink. 687 */ 688 if (skb) 689 tp->rcv_wnd = tp->rcv_nxt - tp->rcv_wup; 690 691 dss_size += ack_size; 692 693 *size = ALIGN(dss_size, 4); 694 return true; 695 } 696 697 static u64 add_addr_generate_hmac(u64 key1, u64 key2, 698 struct mptcp_addr_info *addr) 699 { 700 u16 port = ntohs(addr->port); 701 u8 hmac[SHA256_DIGEST_SIZE]; 702 u8 msg[19]; 703 int i = 0; 704 705 msg[i++] = addr->id; 706 if (addr->family == AF_INET) { 707 memcpy(&msg[i], &addr->addr.s_addr, 4); 708 i += 4; 709 } 710 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 711 else if (addr->family == AF_INET6) { 712 memcpy(&msg[i], &addr->addr6.s6_addr, 16); 713 i += 16; 714 } 715 #endif 716 msg[i++] = port >> 8; 717 msg[i++] = port & 0xFF; 718 719 mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac); 720 721 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]); 722 } 723 724 static bool mptcp_established_options_add_addr(struct sock *sk, 725 struct sk_buff *skb, int *size, 726 unsigned int remaining, 727 bool has_ts, 728 struct mptcp_out_options *opts) 729 { 730 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 731 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 732 struct mptcp_addr_info addr; 733 bool drop_ts = has_ts; 734 bool echo; 735 736 /* add addr will strip the existing options, be sure to avoid breaking 737 * MPC/MPJ handshakes 738 */ 739 if (!mptcp_pm_should_add_signal(msk) || 740 (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) || 741 !skb || !skb_is_tcp_pure_ack(skb) || 742 !mptcp_pm_add_addr_signal(msk, size, remaining, &addr, &echo, 743 &drop_ts)) 744 return false; 745 746 pr_debug("drop other suboptions\n"); 747 opts->suboptions = OPTION_MPTCP_ADD_ADDR; 748 opts->drop_ts = drop_ts; 749 opts->addr = addr; 750 if (!echo) { 751 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDRTX); 752 opts->ahmac = add_addr_generate_hmac(READ_ONCE(msk->local_key), 753 READ_ONCE(msk->remote_key), 754 &opts->addr); 755 } else { 756 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADDTX); 757 opts->ahmac = 0; 758 } 759 pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d\n", 760 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port)); 761 762 return true; 763 } 764 765 static bool mptcp_established_options_rm_addr(struct sock *sk, int *size, 766 unsigned int remaining, 767 struct mptcp_out_options *opts) 768 { 769 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 770 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 771 struct mptcp_rm_list rm_list; 772 int i; 773 774 if (!mptcp_pm_should_rm_signal(msk) || 775 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list, size))) 776 return false; 777 778 opts->suboptions |= OPTION_MPTCP_RM_ADDR; 779 opts->rm_list = rm_list; 780 781 for (i = 0; i < opts->rm_list.nr; i++) 782 pr_debug("rm_list_ids[%d]=%d\n", i, opts->rm_list.ids[i]); 783 MPTCP_ADD_STATS(sock_net(sk), MPTCP_MIB_RMADDRTX, opts->rm_list.nr); 784 return true; 785 } 786 787 static bool mptcp_established_options_mp_prio(struct sock *sk, int *size, 788 unsigned int remaining, 789 struct mptcp_out_options *opts) 790 { 791 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 792 793 /* can't send MP_PRIO with MPC, as they share the same option space: 794 * 'backup'. Also it makes no sense at all 795 */ 796 if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC)) 797 return false; 798 799 /* account for the trailing 'nop' option */ 800 if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN) 801 return false; 802 803 *size = TCPOLEN_MPTCP_PRIO_ALIGN; 804 opts->suboptions |= OPTION_MPTCP_PRIO; 805 opts->backup = subflow->request_bkup; 806 807 pr_debug("prio=%d\n", opts->backup); 808 809 return true; 810 } 811 812 static noinline bool mptcp_established_options_rst(struct sock *sk, 813 int *size, 814 unsigned int remaining, 815 struct mptcp_out_options *opts) 816 { 817 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 818 819 if (remaining < TCPOLEN_MPTCP_RST) 820 return false; 821 822 *size = TCPOLEN_MPTCP_RST; 823 opts->suboptions |= OPTION_MPTCP_RST; 824 opts->reset_transient = subflow->reset_transient; 825 opts->reset_reason = subflow->reset_reason; 826 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTTX); 827 828 return true; 829 } 830 831 static bool mptcp_established_options_fastclose(struct sock *sk, int *size, 832 unsigned int remaining, 833 struct mptcp_out_options *opts) 834 { 835 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 836 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 837 838 if (likely(!subflow->send_fastclose)) 839 return false; 840 841 if (remaining < TCPOLEN_MPTCP_FASTCLOSE) 842 return false; 843 844 *size = TCPOLEN_MPTCP_FASTCLOSE; 845 opts->suboptions |= OPTION_MPTCP_FASTCLOSE; 846 opts->rcvr_key = READ_ONCE(msk->remote_key); 847 848 pr_debug("FASTCLOSE key=%llu\n", opts->rcvr_key); 849 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSETX); 850 return true; 851 } 852 853 static bool mptcp_established_options_mp_fail(struct sock *sk, int *size, 854 unsigned int remaining, 855 struct mptcp_out_options *opts) 856 { 857 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 858 859 if (likely(!subflow->send_mp_fail)) 860 return false; 861 862 if (remaining < TCPOLEN_MPTCP_FAIL) 863 return false; 864 865 *size = TCPOLEN_MPTCP_FAIL; 866 opts->suboptions |= OPTION_MPTCP_FAIL; 867 opts->fail_seq = subflow->map_seq; 868 869 pr_debug("MP_FAIL fail_seq=%llu\n", opts->fail_seq); 870 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILTX); 871 872 return true; 873 } 874 875 int mptcp_established_options(struct sock *sk, struct sk_buff *skb, 876 unsigned int remaining, bool has_ts, 877 struct mptcp_out_options *opts) 878 { 879 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 880 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 881 int total_size = 0; 882 bool snd_data_fin; 883 bool ret = false; 884 int opt_size = 0; 885 886 opts->suboptions = 0; 887 888 /* Force later mptcp_write_options(), but do not use any actual 889 * option space. 890 */ 891 if (unlikely(__mptcp_check_fallback(msk) && !mptcp_check_infinite_map(skb))) 892 return 0; 893 894 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) { 895 if (mptcp_established_options_fastclose(sk, &opt_size, remaining, opts) || 896 mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) { 897 total_size += opt_size; 898 remaining -= opt_size; 899 } 900 /* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */ 901 if (mptcp_established_options_rst(sk, &opt_size, remaining, opts)) { 902 total_size += opt_size; 903 remaining -= opt_size; 904 } 905 return total_size; 906 } 907 908 snd_data_fin = mptcp_data_fin_enabled(msk); 909 if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, opts)) 910 ret = true; 911 else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, opts)) { 912 int mp_fail_size; 913 914 ret = true; 915 if (mptcp_established_options_mp_fail(sk, &mp_fail_size, 916 remaining - opt_size, opts)) { 917 total_size += opt_size + mp_fail_size; 918 remaining -= opt_size - mp_fail_size; 919 return total_size; 920 } 921 } 922 923 /* we reserved enough space for the above options, and exceeding the 924 * TCP option space would be fatal 925 */ 926 if (WARN_ON_ONCE(opt_size > remaining)) 927 return -1; 928 929 total_size += opt_size; 930 remaining -= opt_size; 931 if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, 932 has_ts, opts)) { 933 total_size += opt_size; 934 remaining -= opt_size; 935 ret = true; 936 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) { 937 total_size += opt_size; 938 remaining -= opt_size; 939 ret = true; 940 } 941 942 if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) { 943 total_size += opt_size; 944 remaining -= opt_size; 945 ret = true; 946 } 947 948 return ret ? total_size : -1; 949 } 950 951 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size, 952 struct mptcp_out_options *opts) 953 { 954 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 955 956 if (subflow_req->mp_capable) { 957 opts->suboptions = OPTION_MPTCP_MPC_SYNACK; 958 opts->sndr_key = subflow_req->local_key; 959 opts->csum_reqd = subflow_req->csum_reqd; 960 opts->allow_join_id0 = subflow_req->allow_join_id0; 961 *size = TCPOLEN_MPTCP_MPC_SYNACK; 962 pr_debug("subflow_req=%p, local_key=%llu\n", 963 subflow_req, subflow_req->local_key); 964 return true; 965 } else if (subflow_req->mp_join) { 966 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK; 967 opts->backup = subflow_req->request_bkup; 968 opts->join_id = subflow_req->local_id; 969 opts->thmac = subflow_req->thmac; 970 opts->nonce = subflow_req->local_nonce; 971 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u\n", 972 subflow_req, opts->backup, opts->join_id, 973 opts->thmac, opts->nonce); 974 *size = TCPOLEN_MPTCP_MPJ_SYNACK; 975 return true; 976 } 977 return false; 978 } 979 980 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk, 981 struct mptcp_subflow_context *subflow, 982 struct sk_buff *skb, 983 struct mptcp_options_received *mp_opt) 984 { 985 /* here we can process OoO, in-window pkts, only in-sequence 4th ack 986 * will make the subflow fully established 987 */ 988 if (likely(subflow->fully_established)) { 989 /* on passive sockets, check for 3rd ack retransmission 990 * note that msk is always set by subflow_syn_recv_sock() 991 * for mp_join subflows 992 */ 993 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 && 994 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq && 995 subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) && 996 !subflow->request_join) 997 tcp_send_ack(ssk); 998 goto check_notify; 999 } 1000 1001 /* we must process OoO packets before the first subflow is fully 1002 * established. OoO packets are instead a protocol violation 1003 * for MP_JOIN subflows as the peer must not send any data 1004 * before receiving the forth ack - cfr. RFC 8684 section 3.2. 1005 */ 1006 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) { 1007 if (subflow->mp_join) 1008 goto reset; 1009 if (subflow->is_mptfo && mp_opt->suboptions & OPTION_MPTCP_MPC_ACK) 1010 goto set_fully_established; 1011 return subflow->mp_capable; 1012 } 1013 1014 if (subflow->remote_key_valid && 1015 (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) || 1016 ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) && 1017 (!mp_opt->echo || subflow->mp_join)))) { 1018 /* subflows are fully established as soon as we get any 1019 * additional ack, including ADD_ADDR. 1020 */ 1021 goto set_fully_established; 1022 } 1023 1024 /* If the first established packet does not contain MP_CAPABLE + data 1025 * then fallback to TCP. Fallback scenarios requires a reset for 1026 * MP_JOIN subflows. 1027 */ 1028 if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) { 1029 if (subflow->mp_join) 1030 goto reset; 1031 subflow->mp_capable = 0; 1032 if (!mptcp_try_fallback(ssk, MPTCP_MIB_MPCAPABLEDATAFALLBACK)) { 1033 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_FALLBACKFAILED); 1034 goto reset; 1035 } 1036 return false; 1037 } 1038 1039 if (unlikely(!READ_ONCE(msk->pm.server_side))) 1040 pr_warn_once("bogus mpc option on established client sk"); 1041 1042 set_fully_established: 1043 if (mp_opt->deny_join_id0) 1044 WRITE_ONCE(msk->pm.remote_deny_join_id0, true); 1045 1046 mptcp_data_lock((struct sock *)msk); 1047 __mptcp_subflow_fully_established(msk, subflow, mp_opt); 1048 mptcp_data_unlock((struct sock *)msk); 1049 1050 check_notify: 1051 /* if the subflow is not already linked into the conn_list, we can't 1052 * notify the PM: this subflow is still on the listener queue 1053 * and the PM possibly acquiring the subflow lock could race with 1054 * the listener close 1055 */ 1056 if (likely(subflow->pm_notified) || list_empty(&subflow->node)) 1057 return true; 1058 1059 subflow->pm_notified = 1; 1060 if (subflow->mp_join) { 1061 clear_3rdack_retransmission(ssk); 1062 mptcp_pm_subflow_established(msk); 1063 } else { 1064 mptcp_pm_fully_established(msk, ssk); 1065 } 1066 return true; 1067 1068 reset: 1069 mptcp_subflow_reset(ssk); 1070 return false; 1071 } 1072 1073 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq) 1074 { 1075 u32 old_seq32, cur_seq32; 1076 1077 old_seq32 = (u32)old_seq; 1078 cur_seq32 = (u32)cur_seq; 1079 cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32; 1080 if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32))) 1081 return cur_seq + (1LL << 32); 1082 1083 /* reverse wrap could happen, too */ 1084 if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32))) 1085 return cur_seq - (1LL << 32); 1086 return cur_seq; 1087 } 1088 1089 static void __mptcp_snd_una_update(struct mptcp_sock *msk, u64 new_snd_una) 1090 { 1091 msk->bytes_acked += new_snd_una - msk->snd_una; 1092 WRITE_ONCE(msk->snd_una, new_snd_una); 1093 } 1094 1095 static void rwin_update(struct mptcp_sock *msk, struct sock *ssk, 1096 struct sk_buff *skb) 1097 { 1098 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 1099 struct tcp_sock *tp = tcp_sk(ssk); 1100 u64 mptcp_rcv_wnd; 1101 1102 /* Avoid touching extra cachelines if TCP is going to accept this 1103 * skb without filling the TCP-level window even with a possibly 1104 * outdated mptcp-level rwin. 1105 */ 1106 if (!skb->len || skb->len < tcp_receive_window(tp)) 1107 return; 1108 1109 mptcp_rcv_wnd = atomic64_read(&msk->rcv_wnd_sent); 1110 if (!after64(mptcp_rcv_wnd, subflow->rcv_wnd_sent)) 1111 return; 1112 1113 /* Some other subflow grew the mptcp-level rwin since rcv_wup, 1114 * resync. 1115 */ 1116 tp->rcv_wnd += mptcp_rcv_wnd - subflow->rcv_wnd_sent; 1117 tcp_update_max_rcv_wnd_seq(tp); 1118 subflow->rcv_wnd_sent = mptcp_rcv_wnd; 1119 } 1120 1121 static void ack_update_msk(struct mptcp_sock *msk, 1122 struct sock *ssk, 1123 struct mptcp_options_received *mp_opt) 1124 { 1125 u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt); 1126 struct sock *sk = (struct sock *)msk; 1127 u64 old_snd_una; 1128 1129 mptcp_data_lock(sk); 1130 1131 /* avoid ack expansion on update conflict, to reduce the risk of 1132 * wrongly expanding to a future ack sequence number, which is way 1133 * more dangerous than missing an ack 1134 */ 1135 old_snd_una = msk->snd_una; 1136 new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64); 1137 1138 /* ACK for data not even sent yet? Ignore.*/ 1139 if (unlikely(after64(new_snd_una, snd_nxt))) 1140 new_snd_una = old_snd_una; 1141 1142 new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd; 1143 1144 if (after64(new_wnd_end, msk->wnd_end)) 1145 WRITE_ONCE(msk->wnd_end, new_wnd_end); 1146 1147 /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */ 1148 if (after64(msk->wnd_end, snd_nxt)) 1149 __mptcp_check_push(sk, ssk); 1150 1151 if (after64(new_snd_una, old_snd_una)) { 1152 __mptcp_snd_una_update(msk, new_snd_una); 1153 __mptcp_data_acked(sk); 1154 } 1155 msk->last_ack_recv = tcp_jiffies32; 1156 mptcp_data_unlock(sk); 1157 1158 trace_ack_update_msk(mp_opt->data_ack, 1159 old_snd_una, new_snd_una, 1160 new_wnd_end, READ_ONCE(msk->wnd_end)); 1161 } 1162 1163 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit) 1164 { 1165 /* Skip if DATA_FIN was already received. 1166 * If updating simultaneously with the recvmsg loop, values 1167 * should match. If they mismatch, the peer is misbehaving and 1168 * we will prefer the most recent information. 1169 */ 1170 if (READ_ONCE(msk->rcv_data_fin)) 1171 return false; 1172 1173 WRITE_ONCE(msk->rcv_data_fin_seq, 1174 mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit)); 1175 WRITE_ONCE(msk->rcv_data_fin, 1); 1176 1177 return true; 1178 } 1179 1180 static bool add_addr_hmac_valid(struct mptcp_sock *msk, 1181 struct mptcp_options_received *mp_opt) 1182 { 1183 u64 hmac = 0; 1184 1185 if (mp_opt->echo) 1186 return true; 1187 1188 hmac = add_addr_generate_hmac(READ_ONCE(msk->remote_key), 1189 READ_ONCE(msk->local_key), 1190 &mp_opt->addr); 1191 1192 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n", 1193 msk, hmac, mp_opt->ahmac); 1194 1195 return hmac == mp_opt->ahmac; 1196 } 1197 1198 static bool mptcp_over_limit(struct sock *sk, struct sock *ssk, 1199 const struct sk_buff *skb) 1200 { 1201 struct mptcp_sock *msk = mptcp_sk(sk); 1202 u32 rcvbuf = READ_ONCE(sk->sk_rcvbuf); 1203 1204 if (likely((u32)sk_rmem_alloc_get(sk) <= rcvbuf && 1205 READ_ONCE(msk->backlog_len) <= rcvbuf)) 1206 return false; 1207 1208 /* Avoid silently dropping pure acks, fin, rst or already-acked segm. */ 1209 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || 1210 TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_FIN | TCPHDR_RST) || 1211 !after(TCP_SKB_CB(skb)->end_seq, tcp_sk(ssk)->rcv_nxt)) 1212 return false; 1213 1214 /* Dropped due to memory constraints, schedule an ack. */ 1215 inet_csk(ssk)->icsk_ack.pending |= ICSK_ACK_NOMEM | ICSK_ACK_NOW; 1216 inet_csk_schedule_ack(ssk); 1217 1218 /* Plain TCP (fallback) and skb is dropped before the TCP recv queue. */ 1219 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPRCVQDROP); 1220 1221 return true; 1222 } 1223 1224 /* Return false when the caller must drop the packet, i.e. in case of error, 1225 * subflow has been reset, or over memory limits. 1226 */ 1227 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb) 1228 { 1229 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 1230 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 1231 struct mptcp_options_received mp_opt; 1232 struct mptcp_ext *mpext; 1233 1234 if (__mptcp_check_fallback(msk)) { 1235 /* Keep it simple and unconditionally trigger send data cleanup and 1236 * pending queue spooling. We will need to acquire the data lock 1237 * for more accurate checks, and once the lock is acquired, such 1238 * helpers are cheap. 1239 */ 1240 mptcp_data_lock(subflow->conn); 1241 if (sk_stream_memory_free(sk)) 1242 __mptcp_check_push(subflow->conn, sk); 1243 1244 /* on fallback we just need to ignore the msk-level snd_una, as 1245 * this is really plain TCP 1246 */ 1247 __mptcp_snd_una_update(msk, READ_ONCE(msk->snd_nxt)); 1248 1249 __mptcp_data_acked(subflow->conn); 1250 mptcp_data_unlock(subflow->conn); 1251 return !mptcp_over_limit(subflow->conn, sk, skb); 1252 } 1253 1254 mptcp_get_options(skb, &mp_opt); 1255 1256 /* The subflow can be in close state only if check_fully_established() 1257 * just sent a reset. If so, tell the caller to ignore the current packet. 1258 */ 1259 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt)) 1260 return sk->sk_state != TCP_CLOSE; 1261 1262 if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) { 1263 if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) && 1264 READ_ONCE(msk->local_key) == mp_opt.fc_recv_key) { 1265 WRITE_ONCE(msk->rcv_fastclose, true); 1266 mptcp_schedule_work((struct sock *)msk); 1267 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSERX); 1268 } 1269 1270 if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) && 1271 add_addr_hmac_valid(msk, &mp_opt)) { 1272 if (!mp_opt.echo) { 1273 mptcp_pm_add_addr_received(sk, &mp_opt.addr); 1274 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR); 1275 } else { 1276 mptcp_pm_add_addr_echoed(msk, &mp_opt.addr); 1277 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD); 1278 } 1279 1280 if (mp_opt.addr.port) 1281 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD); 1282 } 1283 1284 if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR) 1285 mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list); 1286 1287 if (mp_opt.suboptions & OPTION_MPTCP_PRIO) { 1288 mptcp_pm_mp_prio_received(sk, mp_opt.backup); 1289 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX); 1290 } 1291 1292 if (mp_opt.suboptions & OPTION_MPTCP_FAIL) { 1293 mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq); 1294 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX); 1295 } 1296 1297 if (mp_opt.suboptions & OPTION_MPTCP_RST) { 1298 subflow->reset_seen = 1; 1299 subflow->reset_reason = mp_opt.reset_reason; 1300 subflow->reset_transient = mp_opt.reset_transient; 1301 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTRX); 1302 } 1303 1304 if (!(mp_opt.suboptions & OPTION_MPTCP_DSS)) 1305 return true; 1306 } 1307 1308 /* we can't wait for recvmsg() to update the ack_seq, otherwise 1309 * monodirectional flows will stuck 1310 */ 1311 if (mp_opt.use_ack) 1312 ack_update_msk(msk, sk, &mp_opt); 1313 rwin_update(msk, sk, skb); 1314 1315 /* Zero-data-length packets are dropped by the caller and not 1316 * propagated to the MPTCP layer, so the skb extension does not 1317 * need to be allocated or populated. DATA_FIN information, if 1318 * present, needs to be updated here before the skb is freed. 1319 */ 1320 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) { 1321 if (mp_opt.data_fin && mp_opt.data_len == 1 && 1322 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64)) 1323 mptcp_schedule_work((struct sock *)msk); 1324 1325 return true; 1326 } 1327 1328 mpext = skb_ext_add(skb, SKB_EXT_MPTCP); 1329 if (!mpext) 1330 return false; 1331 1332 memset(mpext, 0, sizeof(*mpext)); 1333 1334 if (likely(mp_opt.use_map)) { 1335 if (mp_opt.mpc_map) { 1336 /* this is an MP_CAPABLE carrying MPTCP data 1337 * we know this map the first chunk of data 1338 */ 1339 mptcp_crypto_key_sha(subflow->remote_key, NULL, 1340 &mpext->data_seq); 1341 mpext->data_seq++; 1342 mpext->subflow_seq = 1; 1343 mpext->dsn64 = 1; 1344 mpext->mpc_map = 1; 1345 mpext->data_fin = 0; 1346 } else { 1347 mpext->data_seq = mp_opt.data_seq; 1348 mpext->subflow_seq = mp_opt.subflow_seq; 1349 mpext->dsn64 = mp_opt.dsn64; 1350 mpext->data_fin = mp_opt.data_fin; 1351 } 1352 mpext->data_len = mp_opt.data_len; 1353 mpext->use_map = 1; 1354 mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD); 1355 1356 if (mpext->csum_reqd) 1357 mpext->csum = mp_opt.csum; 1358 } 1359 1360 return true; 1361 } 1362 1363 static u64 mptcp_set_rwin(struct mptcp_sock *msk, struct tcp_sock *tp, 1364 struct tcphdr *th, u64 ack_seq) 1365 { 1366 const struct sock *ssk = (const struct sock *)tp; 1367 u64 rcv_wnd_old, rcv_wnd_new; 1368 u32 new_win; 1369 u64 win; 1370 1371 rcv_wnd_new = ack_seq + tp->rcv_wnd; 1372 1373 rcv_wnd_old = atomic64_read(&msk->rcv_wnd_sent); 1374 if (after64(rcv_wnd_new, rcv_wnd_old)) { 1375 u64 rcv_wnd; 1376 1377 for (;;) { 1378 rcv_wnd = atomic64_cmpxchg(&msk->rcv_wnd_sent, rcv_wnd_old, rcv_wnd_new); 1379 1380 if (rcv_wnd == rcv_wnd_old) 1381 break; 1382 1383 rcv_wnd_old = rcv_wnd; 1384 if (before64(rcv_wnd_new, rcv_wnd_old)) { 1385 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICTUPDATE); 1386 goto raise_win; 1387 } 1388 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICT); 1389 } 1390 goto update_wspace; 1391 } 1392 1393 if (rcv_wnd_new != rcv_wnd_old) { 1394 raise_win: 1395 /* The msk-level rcv wnd is after the tcp level one, 1396 * sync the latter. 1397 */ 1398 rcv_wnd_new = rcv_wnd_old; 1399 win = rcv_wnd_old - ack_seq; 1400 new_win = min_t(u64, win, U32_MAX); 1401 tp->rcv_wnd = new_win; 1402 tcp_update_max_rcv_wnd_seq(tp); 1403 1404 /* Make sure we do not exceed the maximum possible 1405 * scaled window. 1406 */ 1407 if (unlikely(th->syn)) 1408 new_win = min(new_win, 65535U) << tp->rx_opt.rcv_wscale; 1409 if (!tp->rx_opt.rcv_wscale && 1410 READ_ONCE(sock_net(ssk)->ipv4.sysctl_tcp_workaround_signed_windows)) 1411 new_win = min(new_win, MAX_TCP_WINDOW); 1412 else 1413 new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale)); 1414 1415 /* RFC1323 scaling applied */ 1416 new_win >>= tp->rx_opt.rcv_wscale; 1417 th->window = htons(new_win); 1418 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDSHARED); 1419 } 1420 1421 update_wspace: 1422 WRITE_ONCE(msk->old_wspace, tp->rcv_wnd); 1423 return rcv_wnd_new; 1424 } 1425 1426 static void mptcp_track_rwin(struct tcp_sock *tp) 1427 { 1428 const struct sock *ssk = (const struct sock *)tp; 1429 struct mptcp_subflow_context *subflow; 1430 struct mptcp_sock *msk; 1431 1432 if (!ssk) 1433 return; 1434 1435 subflow = mptcp_subflow_ctx(ssk); 1436 msk = mptcp_sk(subflow->conn); 1437 WRITE_ONCE(msk->old_wspace, tp->rcv_wnd); 1438 } 1439 1440 __sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum) 1441 { 1442 struct csum_pseudo_header header; 1443 __wsum csum; 1444 1445 /* cfr RFC 8684 3.3.1.: 1446 * the data sequence number used in the pseudo-header is 1447 * always the 64-bit value, irrespective of what length is used in the 1448 * DSS option itself. 1449 */ 1450 header.data_seq = cpu_to_be64(data_seq); 1451 header.subflow_seq = htonl(subflow_seq); 1452 header.data_len = htons(data_len); 1453 header.csum = 0; 1454 1455 csum = csum_partial(&header, sizeof(header), sum); 1456 return csum_fold(csum); 1457 } 1458 1459 static __sum16 mptcp_make_csum(const struct mptcp_ext *mpext) 1460 { 1461 return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len, 1462 ~csum_unfold(mpext->csum)); 1463 } 1464 1465 static void put_len_csum(u16 len, __sum16 csum, void *data) 1466 { 1467 __sum16 *sumptr = data + 2; 1468 __be16 *ptr = data; 1469 1470 put_unaligned_be16(len, ptr); 1471 1472 put_unaligned(csum, sumptr); 1473 } 1474 1475 void mptcp_write_options(struct tcphdr *th, __be32 *ptr, struct tcp_sock *tp, 1476 struct mptcp_out_options *opts) 1477 { 1478 const struct sock *ssk = (const struct sock *)tp; 1479 struct mptcp_subflow_context *subflow; 1480 1481 /* Which options can be used together? 1482 * 1483 * X: mutually exclusive 1484 * O: often used together 1485 * C: can be used together in some cases 1486 * P: could be used together but we prefer not to (optimisations) 1487 * 1488 * Opt: | MPC | MPJ | DSS | ADD | RM | PRIO | FAIL | FC | 1489 * ------|------|------|------|------|------|------|------|------| 1490 * MPC |------|------|------|------|------|------|------|------| 1491 * MPJ | X |------|------|------|------|------|------|------| 1492 * DSS | X | X |------|------|------|------|------|------| 1493 * ADD | X | X | P |------|------|------|------|------| 1494 * RM | C | C | C | P |------|------|------|------| 1495 * PRIO | X | C | C | C | C |------|------|------| 1496 * FAIL | X | X | C | X | X | X |------|------| 1497 * FC | X | X | P | X | X | X | X |------| 1498 * RST | X | X | X | X | X | X | O | O | 1499 * ------|------|------|------|------|------|------|------|------| 1500 * 1501 * The same applies in mptcp_established_options() function. 1502 */ 1503 if (likely(OPTION_MPTCP_DSS & opts->suboptions)) { 1504 struct mptcp_ext *mpext = &opts->ext_copy; 1505 u8 len = TCPOLEN_MPTCP_DSS_BASE; 1506 u8 flags = 0; 1507 1508 if (mpext->use_ack) { 1509 flags = MPTCP_DSS_HAS_ACK; 1510 if (mpext->ack64) { 1511 len += TCPOLEN_MPTCP_DSS_ACK64; 1512 flags |= MPTCP_DSS_ACK64; 1513 } else { 1514 len += TCPOLEN_MPTCP_DSS_ACK32; 1515 } 1516 } 1517 1518 if (mpext->use_map) { 1519 len += TCPOLEN_MPTCP_DSS_MAP64; 1520 1521 /* Use only 64-bit mapping flags for now, add 1522 * support for optional 32-bit mappings later. 1523 */ 1524 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64; 1525 if (mpext->data_fin) 1526 flags |= MPTCP_DSS_DATA_FIN; 1527 1528 if (opts->csum_reqd) 1529 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 1530 } 1531 1532 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags); 1533 1534 if (mpext->use_ack) { 1535 struct mptcp_sock *msk; 1536 u64 ack_seq; 1537 1538 /* DSS option is set only by mptcp_established_options, 1539 * the caller is __tcp_transmit_skb() and ssk is always 1540 * not NULL. 1541 */ 1542 subflow = mptcp_subflow_ctx(ssk); 1543 msk = mptcp_sk(subflow->conn); 1544 ack_seq = READ_ONCE(msk->ack_seq); 1545 if (mpext->ack64) { 1546 put_unaligned_be64(ack_seq, ptr); 1547 ptr += 2; 1548 } else { 1549 put_unaligned_be32(ack_seq, ptr); 1550 ptr += 1; 1551 } 1552 subflow->rcv_wnd_sent = mptcp_set_rwin(msk, tp, th, 1553 ack_seq); 1554 } 1555 1556 if (mpext->use_map) { 1557 put_unaligned_be64(mpext->data_seq, ptr); 1558 ptr += 2; 1559 put_unaligned_be32(mpext->subflow_seq, ptr); 1560 ptr += 1; 1561 if (opts->csum_reqd) { 1562 /* data_len == 0 is reserved for the infinite mapping, 1563 * the checksum will also be set to 0. 1564 */ 1565 put_len_csum(mpext->data_len, 1566 (mpext->data_len ? mptcp_make_csum(mpext) : 0), 1567 ptr); 1568 } else { 1569 put_unaligned_be32(mpext->data_len << 16 | 1570 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1571 } 1572 ptr += 1; 1573 } 1574 1575 /* We might need to add MP_FAIL options in rare cases */ 1576 if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) 1577 goto mp_fail; 1578 } else if (OPTIONS_MPTCP_MPC & opts->suboptions) { 1579 u8 len, flag = MPTCP_CAP_HMAC_SHA256; 1580 1581 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) { 1582 len = TCPOLEN_MPTCP_MPC_SYN; 1583 } else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) { 1584 len = TCPOLEN_MPTCP_MPC_SYNACK; 1585 } else if (opts->data_len) { 1586 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 1587 if (opts->csum_reqd) 1588 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 1589 } else { 1590 len = TCPOLEN_MPTCP_MPC_ACK; 1591 } 1592 1593 if (opts->csum_reqd) 1594 flag |= MPTCP_CAP_CHECKSUM_REQD; 1595 1596 if (!opts->allow_join_id0) 1597 flag |= MPTCP_CAP_DENY_JOIN_ID0; 1598 1599 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len, 1600 MPTCP_SUPPORTED_VERSION, 1601 flag); 1602 1603 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) & 1604 opts->suboptions)) 1605 goto mp_capable_done; 1606 1607 put_unaligned_be64(opts->sndr_key, ptr); 1608 ptr += 2; 1609 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions)) 1610 goto mp_capable_done; 1611 1612 put_unaligned_be64(opts->rcvr_key, ptr); 1613 ptr += 2; 1614 if (!opts->data_len) 1615 goto mp_capable_done; 1616 1617 if (opts->csum_reqd) { 1618 put_len_csum(opts->data_len, 1619 __mptcp_make_csum(opts->data_seq, 1620 opts->subflow_seq, 1621 opts->data_len, 1622 ~csum_unfold(opts->csum)), 1623 ptr); 1624 } else { 1625 put_unaligned_be32(opts->data_len << 16 | 1626 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1627 } 1628 ptr += 1; 1629 1630 /* MPC is additionally mutually exclusive with MP_PRIO */ 1631 goto mp_capable_done; 1632 } else if (OPTIONS_MPTCP_MPJ & opts->suboptions) { 1633 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) { 1634 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1635 TCPOLEN_MPTCP_MPJ_SYN, 1636 opts->backup, opts->join_id); 1637 put_unaligned_be32(opts->token, ptr); 1638 ptr += 1; 1639 put_unaligned_be32(opts->nonce, ptr); 1640 ptr += 1; 1641 } else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) { 1642 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1643 TCPOLEN_MPTCP_MPJ_SYNACK, 1644 opts->backup, opts->join_id); 1645 put_unaligned_be64(opts->thmac, ptr); 1646 ptr += 2; 1647 put_unaligned_be32(opts->nonce, ptr); 1648 ptr += 1; 1649 } else { 1650 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1651 TCPOLEN_MPTCP_MPJ_ACK, 0, 0); 1652 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN); 1653 ptr += 5; 1654 } 1655 } else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) { 1656 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE; 1657 u8 echo = MPTCP_ADDR_ECHO; 1658 1659 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1660 if (opts->addr.family == AF_INET6) 1661 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE; 1662 #endif 1663 1664 if (opts->addr.port) 1665 len += TCPOLEN_MPTCP_PORT_LEN; 1666 1667 if (opts->ahmac) { 1668 len += sizeof(opts->ahmac); 1669 echo = 0; 1670 } 1671 1672 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 1673 len, echo, opts->addr.id); 1674 if (opts->addr.family == AF_INET) { 1675 memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4); 1676 ptr += 1; 1677 } 1678 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1679 else if (opts->addr.family == AF_INET6) { 1680 memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16); 1681 ptr += 4; 1682 } 1683 #endif 1684 1685 if (!opts->addr.port) { 1686 if (opts->ahmac) { 1687 put_unaligned_be64(opts->ahmac, ptr); 1688 ptr += 2; 1689 } 1690 } else { 1691 u16 port = ntohs(opts->addr.port); 1692 1693 if (opts->ahmac) { 1694 u8 *bptr = (u8 *)ptr; 1695 1696 put_unaligned_be16(port, bptr); 1697 bptr += 2; 1698 put_unaligned_be64(opts->ahmac, bptr); 1699 bptr += 8; 1700 put_unaligned_be16(TCPOPT_NOP << 8 | 1701 TCPOPT_NOP, bptr); 1702 1703 ptr += 3; 1704 } else { 1705 put_unaligned_be32(port << 16 | 1706 TCPOPT_NOP << 8 | 1707 TCPOPT_NOP, ptr); 1708 ptr += 1; 1709 } 1710 } 1711 } else if (unlikely(OPTION_MPTCP_FASTCLOSE & opts->suboptions)) { 1712 /* FASTCLOSE is mutually exclusive with others except RST */ 1713 *ptr++ = mptcp_option(MPTCPOPT_MP_FASTCLOSE, 1714 TCPOLEN_MPTCP_FASTCLOSE, 1715 0, 0); 1716 put_unaligned_be64(opts->rcvr_key, ptr); 1717 ptr += 2; 1718 1719 if (OPTION_MPTCP_RST & opts->suboptions) 1720 goto mp_rst; 1721 return; 1722 } else if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) { 1723 mp_fail: 1724 /* MP_FAIL is mutually exclusive with others except RST */ 1725 subflow = mptcp_subflow_ctx(ssk); 1726 subflow->send_mp_fail = 0; 1727 1728 *ptr++ = mptcp_option(MPTCPOPT_MP_FAIL, 1729 TCPOLEN_MPTCP_FAIL, 1730 0, 0); 1731 put_unaligned_be64(opts->fail_seq, ptr); 1732 ptr += 2; 1733 1734 if (OPTION_MPTCP_RST & opts->suboptions) 1735 goto mp_rst; 1736 return; 1737 } else if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) { 1738 mp_rst: 1739 *ptr++ = mptcp_option(MPTCPOPT_RST, 1740 TCPOLEN_MPTCP_RST, 1741 opts->reset_transient, 1742 opts->reset_reason); 1743 return; 1744 } else if (unlikely(!opts->suboptions)) { 1745 /* Fallback to TCP */ 1746 mptcp_track_rwin(tp); 1747 return; 1748 } 1749 1750 if (OPTION_MPTCP_PRIO & opts->suboptions) { 1751 subflow = mptcp_subflow_ctx(ssk); 1752 subflow->send_mp_prio = 0; 1753 1754 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO, 1755 TCPOLEN_MPTCP_PRIO, 1756 opts->backup, TCPOPT_NOP); 1757 1758 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPPRIOTX); 1759 } 1760 1761 mp_capable_done: 1762 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) { 1763 u8 i = 1; 1764 1765 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR, 1766 TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr, 1767 0, opts->rm_list.ids[0]); 1768 1769 while (i < opts->rm_list.nr) { 1770 u8 id1, id2, id3, id4; 1771 1772 id1 = opts->rm_list.ids[i]; 1773 id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP; 1774 id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP; 1775 id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP; 1776 put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr); 1777 ptr += 1; 1778 i += 4; 1779 } 1780 } 1781 } 1782 1783 __be32 mptcp_get_reset_option(const struct sk_buff *skb) 1784 { 1785 const struct mptcp_ext *ext = mptcp_get_ext(skb); 1786 u8 flags, reason; 1787 1788 if (ext) { 1789 flags = ext->reset_transient; 1790 reason = ext->reset_reason; 1791 1792 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST, 1793 flags, reason); 1794 } 1795 1796 return htonl(0u); 1797 } 1798 EXPORT_SYMBOL_GPL(mptcp_get_reset_option); 1799