1 // SPDX-License-Identifier: GPL-2.0 2 /* Multipath TCP 3 * 4 * Copyright (c) 2017 - 2019, Intel Corporation. 5 */ 6 7 #define pr_fmt(fmt) "MPTCP: " fmt 8 9 #include <linux/kernel.h> 10 #include <crypto/sha2.h> 11 #include <net/tcp.h> 12 #include <net/mptcp.h> 13 #include "protocol.h" 14 #include "mib.h" 15 16 #include <trace/events/mptcp.h> 17 18 static bool mptcp_cap_flag_sha256(u8 flags) 19 { 20 return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256; 21 } 22 23 static void mptcp_parse_option(const struct sk_buff *skb, 24 const unsigned char *ptr, int opsize, 25 struct mptcp_options_received *mp_opt) 26 { 27 u8 subtype = *ptr >> 4; 28 int expected_opsize; 29 u16 subopt; 30 u8 version; 31 u8 flags; 32 u8 i; 33 34 switch (subtype) { 35 case MPTCPOPT_MP_CAPABLE: 36 /* strict size checking */ 37 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) { 38 if (skb->len > tcp_hdr(skb)->doff << 2) 39 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA; 40 else 41 expected_opsize = TCPOLEN_MPTCP_MPC_ACK; 42 subopt = OPTION_MPTCP_MPC_ACK; 43 } else { 44 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK) { 45 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK; 46 subopt = OPTION_MPTCP_MPC_SYNACK; 47 } else { 48 expected_opsize = TCPOLEN_MPTCP_MPC_SYN; 49 subopt = OPTION_MPTCP_MPC_SYN; 50 } 51 } 52 53 /* Cfr RFC 8684 Section 3.3.0: 54 * If a checksum is present but its use had 55 * not been negotiated in the MP_CAPABLE handshake, the receiver MUST 56 * close the subflow with a RST, as it is not behaving as negotiated. 57 * If a checksum is not present when its use has been negotiated, the 58 * receiver MUST close the subflow with a RST, as it is considered 59 * broken 60 * We parse even option with mismatching csum presence, so that 61 * later in subflow_data_ready we can trigger the reset. 62 */ 63 if (opsize != expected_opsize && 64 (expected_opsize != TCPOLEN_MPTCP_MPC_ACK_DATA || 65 opsize != TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM)) 66 break; 67 68 /* try to be gentle vs future versions on the initial syn */ 69 version = *ptr++ & MPTCP_VERSION_MASK; 70 if (opsize != TCPOLEN_MPTCP_MPC_SYN) { 71 if (version != MPTCP_SUPPORTED_VERSION) 72 break; 73 } else if (version < MPTCP_SUPPORTED_VERSION) { 74 break; 75 } 76 77 flags = *ptr++; 78 if (!mptcp_cap_flag_sha256(flags) || 79 (flags & MPTCP_CAP_EXTENSIBILITY)) 80 break; 81 82 /* RFC 6824, Section 3.1: 83 * "For the Checksum Required bit (labeled "A"), if either 84 * host requires the use of checksums, checksums MUST be used. 85 * In other words, the only way for checksums not to be used 86 * is if both hosts in their SYNs set A=0." 87 */ 88 if (flags & MPTCP_CAP_CHECKSUM_REQD) 89 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD; 90 91 mp_opt->deny_join_id0 = !!(flags & MPTCP_CAP_DENY_JOIN_ID0); 92 93 mp_opt->suboptions |= subopt; 94 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) { 95 mp_opt->sndr_key = get_unaligned_be64(ptr); 96 ptr += 8; 97 } 98 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) { 99 mp_opt->rcvr_key = get_unaligned_be64(ptr); 100 ptr += 8; 101 } 102 if (opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA) { 103 /* Section 3.1.: 104 * "the data parameters in a MP_CAPABLE are semantically 105 * equivalent to those in a DSS option and can be used 106 * interchangeably." 107 */ 108 mp_opt->suboptions |= OPTION_MPTCP_DSS; 109 mp_opt->use_map = 1; 110 mp_opt->mpc_map = 1; 111 mp_opt->data_len = get_unaligned_be16(ptr); 112 ptr += 2; 113 } 114 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM) { 115 mp_opt->csum = get_unaligned((__force __sum16 *)ptr); 116 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD; 117 ptr += 2; 118 } 119 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d csum=%u\n", 120 version, flags, opsize, mp_opt->sndr_key, 121 mp_opt->rcvr_key, mp_opt->data_len, mp_opt->csum); 122 break; 123 124 case MPTCPOPT_MP_JOIN: 125 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) { 126 mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYN; 127 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 128 mp_opt->join_id = *ptr++; 129 mp_opt->token = get_unaligned_be32(ptr); 130 ptr += 4; 131 mp_opt->nonce = get_unaligned_be32(ptr); 132 ptr += 4; 133 pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u\n", 134 mp_opt->backup, mp_opt->join_id, 135 mp_opt->token, mp_opt->nonce); 136 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) { 137 mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYNACK; 138 mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP; 139 mp_opt->join_id = *ptr++; 140 mp_opt->thmac = get_unaligned_be64(ptr); 141 ptr += 8; 142 mp_opt->nonce = get_unaligned_be32(ptr); 143 ptr += 4; 144 pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u\n", 145 mp_opt->backup, mp_opt->join_id, 146 mp_opt->thmac, mp_opt->nonce); 147 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) { 148 mp_opt->suboptions |= OPTION_MPTCP_MPJ_ACK; 149 ptr += 2; 150 memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN); 151 pr_debug("MP_JOIN hmac\n"); 152 } 153 break; 154 155 case MPTCPOPT_DSS: 156 pr_debug("DSS\n"); 157 ptr++; 158 159 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\n", 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\n", 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 = get_unaligned((__force __sum16 *)ptr); 225 ptr += 2; 226 } 227 228 pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u\n", 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\n", 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\n", 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\n", 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 pr_debug("MP_FASTCLOSE: recv_key=%llu\n", mp_opt->rcvr_key); 327 break; 328 329 case MPTCPOPT_RST: 330 if (opsize != TCPOLEN_MPTCP_RST) 331 break; 332 333 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) 334 break; 335 336 mp_opt->suboptions |= OPTION_MPTCP_RST; 337 flags = *ptr++; 338 mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT; 339 mp_opt->reset_reason = *ptr; 340 pr_debug("MP_RST: transient=%u reason=%u\n", 341 mp_opt->reset_transient, mp_opt->reset_reason); 342 break; 343 344 case MPTCPOPT_MP_FAIL: 345 if (opsize != TCPOLEN_MPTCP_FAIL) 346 break; 347 348 ptr += 2; 349 mp_opt->suboptions |= OPTION_MPTCP_FAIL; 350 mp_opt->fail_seq = get_unaligned_be64(ptr); 351 pr_debug("MP_FAIL: data_seq=%llu\n", mp_opt->fail_seq); 352 break; 353 354 default: 355 break; 356 } 357 } 358 359 void mptcp_get_options(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 /* Ensure that casting the whole status to u32 is efficient and safe */ 367 BUILD_BUG_ON(sizeof_field(struct mptcp_options_received, status) != sizeof(u32)); 368 BUILD_BUG_ON(!IS_ALIGNED(offsetof(struct mptcp_options_received, status), 369 sizeof(u32))); 370 *(u32 *)&mp_opt->status = 0; 371 372 length = (th->doff * 4) - sizeof(struct tcphdr); 373 ptr = (const unsigned char *)(th + 1); 374 375 while (length > 0) { 376 int opcode = *ptr++; 377 int opsize; 378 379 switch (opcode) { 380 case TCPOPT_EOL: 381 return; 382 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */ 383 length--; 384 continue; 385 default: 386 if (length < 2) 387 return; 388 opsize = *ptr++; 389 if (opsize < 2) /* "silly options" */ 390 return; 391 if (opsize > length) 392 return; /* don't parse partial options */ 393 if (opcode == TCPOPT_MPTCP) 394 mptcp_parse_option(skb, ptr, opsize, mp_opt); 395 ptr += opsize - 2; 396 length -= opsize; 397 } 398 } 399 } 400 401 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb, 402 unsigned int *size, struct mptcp_out_options *opts) 403 { 404 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 405 406 /* we will use snd_isn to detect first pkt [re]transmission 407 * in mptcp_established_options_mp() 408 */ 409 subflow->snd_isn = TCP_SKB_CB(skb)->end_seq; 410 if (subflow->request_mptcp) { 411 opts->suboptions = OPTION_MPTCP_MPC_SYN; 412 opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk)); 413 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk)); 414 *size = TCPOLEN_MPTCP_MPC_SYN; 415 return true; 416 } else if (subflow->request_join) { 417 pr_debug("remote_token=%u, nonce=%u\n", subflow->remote_token, 418 subflow->local_nonce); 419 opts->suboptions = OPTION_MPTCP_MPJ_SYN; 420 opts->join_id = subflow->local_id; 421 opts->token = subflow->remote_token; 422 opts->nonce = subflow->local_nonce; 423 opts->backup = subflow->request_bkup; 424 *size = TCPOLEN_MPTCP_MPJ_SYN; 425 return true; 426 } 427 return false; 428 } 429 430 static void clear_3rdack_retransmission(struct sock *sk) 431 { 432 struct inet_connection_sock *icsk = inet_csk(sk); 433 434 sk_stop_timer(sk, &icsk->icsk_delack_timer); 435 icsk->icsk_ack.timeout = 0; 436 icsk->icsk_ack.ato = 0; 437 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER); 438 } 439 440 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb, 441 bool snd_data_fin_enable, 442 unsigned int *size, 443 struct mptcp_out_options *opts) 444 { 445 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 446 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 447 struct mptcp_ext *mpext; 448 unsigned int data_len; 449 u8 len; 450 451 /* When skb is not available, we better over-estimate the emitted 452 * options len. A full DSS option (28 bytes) is longer than 453 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so 454 * tell the caller to defer the estimate to 455 * mptcp_established_options_dss(), which will reserve enough space. 456 */ 457 if (!skb) 458 return false; 459 460 /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */ 461 if (READ_ONCE(subflow->fully_established) || snd_data_fin_enable || 462 subflow->snd_isn != TCP_SKB_CB(skb)->seq || 463 sk->sk_state != TCP_ESTABLISHED) 464 return false; 465 466 if (subflow->mp_capable) { 467 mpext = mptcp_get_ext(skb); 468 data_len = mpext ? mpext->data_len : 0; 469 470 /* we will check ops->data_len in mptcp_write_options() to 471 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and 472 * TCPOLEN_MPTCP_MPC_ACK 473 */ 474 opts->data_len = data_len; 475 opts->suboptions = OPTION_MPTCP_MPC_ACK; 476 opts->sndr_key = subflow->local_key; 477 opts->rcvr_key = subflow->remote_key; 478 opts->csum_reqd = READ_ONCE(msk->csum_enabled); 479 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk)); 480 481 /* Section 3.1. 482 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK 483 * packets that start the first subflow of an MPTCP connection, 484 * as well as the first packet that carries data 485 */ 486 if (data_len > 0) { 487 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 488 if (opts->csum_reqd) { 489 /* we need to propagate more info to csum the pseudo hdr */ 490 opts->data_seq = mpext->data_seq; 491 opts->subflow_seq = mpext->subflow_seq; 492 opts->csum = mpext->csum; 493 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 494 } 495 *size = ALIGN(len, 4); 496 } else { 497 *size = TCPOLEN_MPTCP_MPC_ACK; 498 } 499 500 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d\n", 501 subflow, subflow->local_key, subflow->remote_key, 502 data_len); 503 504 return true; 505 } else if (subflow->mp_join) { 506 opts->suboptions = OPTION_MPTCP_MPJ_ACK; 507 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN); 508 *size = TCPOLEN_MPTCP_MPJ_ACK; 509 pr_debug("subflow=%p\n", subflow); 510 511 /* we can use the full delegate action helper only from BH context 512 * If we are in process context - sk is flushing the backlog at 513 * socket lock release time - just set the appropriate flag, will 514 * be handled by the release callback 515 */ 516 if (sock_owned_by_user(sk)) 517 set_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status); 518 else 519 mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_ACK); 520 return true; 521 } 522 return false; 523 } 524 525 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow, 526 struct sk_buff *skb, struct mptcp_ext *ext) 527 { 528 /* The write_seq value has already been incremented, so the actual 529 * sequence number for the DATA_FIN is one less. 530 */ 531 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1; 532 533 if (!ext->use_map || !skb->len) { 534 /* RFC6824 requires a DSS mapping with specific values 535 * if DATA_FIN is set but no data payload is mapped 536 */ 537 ext->data_fin = 1; 538 ext->use_map = 1; 539 ext->dsn64 = 1; 540 ext->data_seq = data_fin_tx_seq; 541 ext->subflow_seq = 0; 542 ext->data_len = 1; 543 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) { 544 /* If there's an existing DSS mapping and it is the 545 * final mapping, DATA_FIN consumes 1 additional byte of 546 * mapping space. 547 */ 548 ext->data_fin = 1; 549 ext->data_len++; 550 } 551 } 552 553 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb, 554 bool snd_data_fin_enable, 555 unsigned int *size, 556 struct mptcp_out_options *opts) 557 { 558 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 559 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 560 unsigned int dss_size = 0; 561 struct mptcp_ext *mpext; 562 unsigned int ack_size; 563 bool ret = false; 564 u64 ack_seq; 565 566 opts->csum_reqd = READ_ONCE(msk->csum_enabled); 567 mpext = skb ? mptcp_get_ext(skb) : NULL; 568 569 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) { 570 unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64; 571 572 if (mpext) { 573 if (opts->csum_reqd) 574 map_size += TCPOLEN_MPTCP_DSS_CHECKSUM; 575 576 opts->ext_copy = *mpext; 577 } 578 579 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; 607 608 /* Add kind/length/subtype/flag overhead if mapping is not populated */ 609 if (dss_size == 0) 610 ack_size += TCPOLEN_MPTCP_DSS_BASE; 611 612 dss_size += ack_size; 613 614 *size = ALIGN(dss_size, 4); 615 return true; 616 } 617 618 static u64 add_addr_generate_hmac(u64 key1, u64 key2, 619 struct mptcp_addr_info *addr) 620 { 621 u16 port = ntohs(addr->port); 622 u8 hmac[SHA256_DIGEST_SIZE]; 623 u8 msg[19]; 624 int i = 0; 625 626 msg[i++] = addr->id; 627 if (addr->family == AF_INET) { 628 memcpy(&msg[i], &addr->addr.s_addr, 4); 629 i += 4; 630 } 631 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 632 else if (addr->family == AF_INET6) { 633 memcpy(&msg[i], &addr->addr6.s6_addr, 16); 634 i += 16; 635 } 636 #endif 637 msg[i++] = port >> 8; 638 msg[i++] = port & 0xFF; 639 640 mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac); 641 642 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]); 643 } 644 645 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb, 646 unsigned int *size, 647 unsigned int remaining, 648 struct mptcp_out_options *opts) 649 { 650 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 651 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 652 bool drop_other_suboptions = false; 653 unsigned int opt_size = *size; 654 bool echo; 655 int len; 656 657 /* add addr will strip the existing options, be sure to avoid breaking 658 * MPC/MPJ handshakes 659 */ 660 if (!mptcp_pm_should_add_signal(msk) || 661 (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) || 662 !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, &opts->addr, 663 &echo, &drop_other_suboptions)) 664 return false; 665 666 /* 667 * Later on, mptcp_write_options() will enforce mutually exclusion with 668 * DSS, bail out if such option is set and we can't drop it. 669 */ 670 if (drop_other_suboptions) 671 remaining += opt_size; 672 else if (opts->suboptions & OPTION_MPTCP_DSS) 673 return false; 674 675 len = mptcp_add_addr_len(opts->addr.family, echo, !!opts->addr.port); 676 if (remaining < len) 677 return false; 678 679 *size = len; 680 if (drop_other_suboptions) { 681 pr_debug("drop other suboptions\n"); 682 opts->suboptions = 0; 683 684 /* note that e.g. DSS could have written into the memory 685 * aliased by ahmac, we must reset the field here 686 * to avoid appending the hmac even for ADD_ADDR echo 687 * options 688 */ 689 opts->ahmac = 0; 690 *size -= opt_size; 691 } 692 opts->suboptions |= OPTION_MPTCP_ADD_ADDR; 693 if (!echo) { 694 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDRTX); 695 opts->ahmac = add_addr_generate_hmac(READ_ONCE(msk->local_key), 696 READ_ONCE(msk->remote_key), 697 &opts->addr); 698 } else { 699 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADDTX); 700 } 701 pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d\n", 702 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port)); 703 704 return true; 705 } 706 707 static bool mptcp_established_options_rm_addr(struct sock *sk, 708 unsigned int *size, 709 unsigned int remaining, 710 struct mptcp_out_options *opts) 711 { 712 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 713 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 714 struct mptcp_rm_list rm_list; 715 int i, len; 716 717 if (!mptcp_pm_should_rm_signal(msk) || 718 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list))) 719 return false; 720 721 len = mptcp_rm_addr_len(&rm_list); 722 if (len < 0) 723 return false; 724 if (remaining < len) 725 return false; 726 727 *size = len; 728 opts->suboptions |= OPTION_MPTCP_RM_ADDR; 729 opts->rm_list = rm_list; 730 731 for (i = 0; i < opts->rm_list.nr; i++) 732 pr_debug("rm_list_ids[%d]=%d\n", i, opts->rm_list.ids[i]); 733 MPTCP_ADD_STATS(sock_net(sk), MPTCP_MIB_RMADDRTX, opts->rm_list.nr); 734 return true; 735 } 736 737 static bool mptcp_established_options_mp_prio(struct sock *sk, 738 unsigned int *size, 739 unsigned int remaining, 740 struct mptcp_out_options *opts) 741 { 742 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 743 744 /* can't send MP_PRIO with MPC, as they share the same option space: 745 * 'backup'. Also it makes no sense at all 746 */ 747 if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC)) 748 return false; 749 750 /* account for the trailing 'nop' option */ 751 if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN) 752 return false; 753 754 *size = TCPOLEN_MPTCP_PRIO_ALIGN; 755 opts->suboptions |= OPTION_MPTCP_PRIO; 756 opts->backup = subflow->request_bkup; 757 758 pr_debug("prio=%d\n", opts->backup); 759 760 return true; 761 } 762 763 static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb, 764 unsigned int *size, 765 unsigned int remaining, 766 struct mptcp_out_options *opts) 767 { 768 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 769 770 if (remaining < TCPOLEN_MPTCP_RST) 771 return false; 772 773 *size = TCPOLEN_MPTCP_RST; 774 opts->suboptions |= OPTION_MPTCP_RST; 775 opts->reset_transient = subflow->reset_transient; 776 opts->reset_reason = subflow->reset_reason; 777 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTTX); 778 779 return true; 780 } 781 782 static bool mptcp_established_options_fastclose(struct sock *sk, 783 unsigned int *size, 784 unsigned int remaining, 785 struct mptcp_out_options *opts) 786 { 787 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 788 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 789 790 if (likely(!subflow->send_fastclose)) 791 return false; 792 793 if (remaining < TCPOLEN_MPTCP_FASTCLOSE) 794 return false; 795 796 *size = TCPOLEN_MPTCP_FASTCLOSE; 797 opts->suboptions |= OPTION_MPTCP_FASTCLOSE; 798 opts->rcvr_key = READ_ONCE(msk->remote_key); 799 800 pr_debug("FASTCLOSE key=%llu\n", opts->rcvr_key); 801 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSETX); 802 return true; 803 } 804 805 static bool mptcp_established_options_mp_fail(struct sock *sk, 806 unsigned int *size, 807 unsigned int remaining, 808 struct mptcp_out_options *opts) 809 { 810 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 811 812 if (likely(!subflow->send_mp_fail)) 813 return false; 814 815 if (remaining < TCPOLEN_MPTCP_FAIL) 816 return false; 817 818 *size = TCPOLEN_MPTCP_FAIL; 819 opts->suboptions |= OPTION_MPTCP_FAIL; 820 opts->fail_seq = subflow->map_seq; 821 822 pr_debug("MP_FAIL fail_seq=%llu\n", opts->fail_seq); 823 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILTX); 824 825 return true; 826 } 827 828 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb, 829 unsigned int *size, unsigned int remaining, 830 struct mptcp_out_options *opts) 831 { 832 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 833 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 834 unsigned int opt_size = 0; 835 bool snd_data_fin; 836 bool ret = false; 837 838 opts->suboptions = 0; 839 840 if (unlikely(__mptcp_check_fallback(msk) && !mptcp_check_infinite_map(skb))) 841 return false; 842 843 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) { 844 if (mptcp_established_options_fastclose(sk, &opt_size, remaining, opts) || 845 mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) { 846 *size += opt_size; 847 remaining -= opt_size; 848 } 849 /* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */ 850 if (mptcp_established_options_rst(sk, skb, &opt_size, remaining, opts)) { 851 *size += opt_size; 852 remaining -= opt_size; 853 } 854 return true; 855 } 856 857 snd_data_fin = mptcp_data_fin_enabled(msk); 858 if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, opts)) 859 ret = true; 860 else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, opts)) { 861 unsigned int mp_fail_size; 862 863 ret = true; 864 if (mptcp_established_options_mp_fail(sk, &mp_fail_size, 865 remaining - opt_size, opts)) { 866 *size += opt_size + mp_fail_size; 867 remaining -= opt_size - mp_fail_size; 868 return true; 869 } 870 } 871 872 /* we reserved enough space for the above options, and exceeding the 873 * TCP option space would be fatal 874 */ 875 if (WARN_ON_ONCE(opt_size > remaining)) 876 return false; 877 878 *size += opt_size; 879 remaining -= opt_size; 880 if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) { 881 *size += opt_size; 882 remaining -= opt_size; 883 ret = true; 884 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) { 885 *size += opt_size; 886 remaining -= opt_size; 887 ret = true; 888 } 889 890 if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) { 891 *size += opt_size; 892 remaining -= opt_size; 893 ret = true; 894 } 895 896 return ret; 897 } 898 899 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size, 900 struct mptcp_out_options *opts) 901 { 902 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 903 904 if (subflow_req->mp_capable) { 905 opts->suboptions = OPTION_MPTCP_MPC_SYNACK; 906 opts->sndr_key = subflow_req->local_key; 907 opts->csum_reqd = subflow_req->csum_reqd; 908 opts->allow_join_id0 = subflow_req->allow_join_id0; 909 *size = TCPOLEN_MPTCP_MPC_SYNACK; 910 pr_debug("subflow_req=%p, local_key=%llu\n", 911 subflow_req, subflow_req->local_key); 912 return true; 913 } else if (subflow_req->mp_join) { 914 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK; 915 opts->backup = subflow_req->request_bkup; 916 opts->join_id = subflow_req->local_id; 917 opts->thmac = subflow_req->thmac; 918 opts->nonce = subflow_req->local_nonce; 919 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u\n", 920 subflow_req, opts->backup, opts->join_id, 921 opts->thmac, opts->nonce); 922 *size = TCPOLEN_MPTCP_MPJ_SYNACK; 923 return true; 924 } 925 return false; 926 } 927 928 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk, 929 struct mptcp_subflow_context *subflow, 930 struct sk_buff *skb, 931 struct mptcp_options_received *mp_opt) 932 { 933 /* here we can process OoO, in-window pkts, only in-sequence 4th ack 934 * will make the subflow fully established 935 */ 936 if (likely(READ_ONCE(subflow->fully_established))) { 937 /* on passive sockets, check for 3rd ack retransmission 938 * note that msk is always set by subflow_syn_recv_sock() 939 * for mp_join subflows 940 */ 941 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 && 942 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq && 943 subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) && 944 !subflow->request_join) 945 tcp_send_ack(ssk); 946 goto check_notify; 947 } 948 949 /* we must process OoO packets before the first subflow is fully 950 * established. OoO packets are instead a protocol violation 951 * for MP_JOIN subflows as the peer must not send any data 952 * before receiving the forth ack - cfr. RFC 8684 section 3.2. 953 */ 954 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) { 955 if (subflow->mp_join) 956 goto reset; 957 if (subflow->is_mptfo && mp_opt->suboptions & OPTION_MPTCP_MPC_ACK) 958 goto set_fully_established; 959 return subflow->mp_capable; 960 } 961 962 if (subflow->remote_key_valid && 963 (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) || 964 ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) && 965 (!mp_opt->echo || subflow->mp_join)))) { 966 /* subflows are fully established as soon as we get any 967 * additional ack, including ADD_ADDR. 968 */ 969 goto set_fully_established; 970 } 971 972 /* If the first established packet does not contain MP_CAPABLE + data 973 * then fallback to TCP. Fallback scenarios requires a reset for 974 * MP_JOIN subflows. 975 */ 976 if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) { 977 if (subflow->mp_join) 978 goto reset; 979 subflow->mp_capable = 0; 980 pr_fallback(msk); 981 mptcp_do_fallback(ssk); 982 return false; 983 } 984 985 if (mp_opt->deny_join_id0) 986 WRITE_ONCE(msk->pm.remote_deny_join_id0, true); 987 988 if (unlikely(!READ_ONCE(msk->pm.server_side))) 989 pr_warn_once("bogus mpc option on established client sk"); 990 991 set_fully_established: 992 mptcp_data_lock((struct sock *)msk); 993 __mptcp_subflow_fully_established(msk, subflow, mp_opt); 994 mptcp_data_unlock((struct sock *)msk); 995 996 check_notify: 997 /* if the subflow is not already linked into the conn_list, we can't 998 * notify the PM: this subflow is still on the listener queue 999 * and the PM possibly acquiring the subflow lock could race with 1000 * the listener close 1001 */ 1002 if (likely(subflow->pm_notified) || list_empty(&subflow->node)) 1003 return true; 1004 1005 subflow->pm_notified = 1; 1006 if (subflow->mp_join) { 1007 clear_3rdack_retransmission(ssk); 1008 mptcp_pm_subflow_established(msk); 1009 } else { 1010 mptcp_pm_fully_established(msk, ssk); 1011 } 1012 return true; 1013 1014 reset: 1015 mptcp_subflow_reset(ssk); 1016 return false; 1017 } 1018 1019 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq) 1020 { 1021 u32 old_seq32, cur_seq32; 1022 1023 old_seq32 = (u32)old_seq; 1024 cur_seq32 = (u32)cur_seq; 1025 cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32; 1026 if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32))) 1027 return cur_seq + (1LL << 32); 1028 1029 /* reverse wrap could happen, too */ 1030 if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32))) 1031 return cur_seq - (1LL << 32); 1032 return cur_seq; 1033 } 1034 1035 static void __mptcp_snd_una_update(struct mptcp_sock *msk, u64 new_snd_una) 1036 { 1037 msk->bytes_acked += new_snd_una - msk->snd_una; 1038 WRITE_ONCE(msk->snd_una, new_snd_una); 1039 } 1040 1041 static void ack_update_msk(struct mptcp_sock *msk, 1042 struct sock *ssk, 1043 struct mptcp_options_received *mp_opt) 1044 { 1045 u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt); 1046 struct sock *sk = (struct sock *)msk; 1047 u64 old_snd_una; 1048 1049 mptcp_data_lock(sk); 1050 1051 /* avoid ack expansion on update conflict, to reduce the risk of 1052 * wrongly expanding to a future ack sequence number, which is way 1053 * more dangerous than missing an ack 1054 */ 1055 old_snd_una = msk->snd_una; 1056 new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64); 1057 1058 /* ACK for data not even sent yet? Ignore.*/ 1059 if (unlikely(after64(new_snd_una, snd_nxt))) 1060 new_snd_una = old_snd_una; 1061 1062 new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd; 1063 1064 if (after64(new_wnd_end, msk->wnd_end)) 1065 WRITE_ONCE(msk->wnd_end, new_wnd_end); 1066 1067 /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */ 1068 if (after64(msk->wnd_end, snd_nxt)) 1069 __mptcp_check_push(sk, ssk); 1070 1071 if (after64(new_snd_una, old_snd_una)) { 1072 __mptcp_snd_una_update(msk, new_snd_una); 1073 __mptcp_data_acked(sk); 1074 } 1075 msk->last_ack_recv = tcp_jiffies32; 1076 mptcp_data_unlock(sk); 1077 1078 trace_ack_update_msk(mp_opt->data_ack, 1079 old_snd_una, new_snd_una, 1080 new_wnd_end, READ_ONCE(msk->wnd_end)); 1081 } 1082 1083 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit) 1084 { 1085 /* Skip if DATA_FIN was already received. 1086 * If updating simultaneously with the recvmsg loop, values 1087 * should match. If they mismatch, the peer is misbehaving and 1088 * we will prefer the most recent information. 1089 */ 1090 if (READ_ONCE(msk->rcv_data_fin)) 1091 return false; 1092 1093 WRITE_ONCE(msk->rcv_data_fin_seq, 1094 mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit)); 1095 WRITE_ONCE(msk->rcv_data_fin, 1); 1096 1097 return true; 1098 } 1099 1100 static bool add_addr_hmac_valid(struct mptcp_sock *msk, 1101 struct mptcp_options_received *mp_opt) 1102 { 1103 u64 hmac = 0; 1104 1105 if (mp_opt->echo) 1106 return true; 1107 1108 hmac = add_addr_generate_hmac(READ_ONCE(msk->remote_key), 1109 READ_ONCE(msk->local_key), 1110 &mp_opt->addr); 1111 1112 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n", 1113 msk, hmac, mp_opt->ahmac); 1114 1115 return hmac == mp_opt->ahmac; 1116 } 1117 1118 /* Return false if a subflow has been reset, else return true */ 1119 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb) 1120 { 1121 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 1122 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 1123 struct mptcp_options_received mp_opt; 1124 struct mptcp_ext *mpext; 1125 1126 if (__mptcp_check_fallback(msk)) { 1127 /* Keep it simple and unconditionally trigger send data cleanup and 1128 * pending queue spooling. We will need to acquire the data lock 1129 * for more accurate checks, and once the lock is acquired, such 1130 * helpers are cheap. 1131 */ 1132 mptcp_data_lock(subflow->conn); 1133 if (sk_stream_memory_free(sk)) 1134 __mptcp_check_push(subflow->conn, sk); 1135 1136 /* on fallback we just need to ignore the msk-level snd_una, as 1137 * this is really plain TCP 1138 */ 1139 __mptcp_snd_una_update(msk, READ_ONCE(msk->snd_nxt)); 1140 1141 __mptcp_data_acked(subflow->conn); 1142 mptcp_data_unlock(subflow->conn); 1143 return true; 1144 } 1145 1146 mptcp_get_options(skb, &mp_opt); 1147 1148 /* The subflow can be in close state only if check_fully_established() 1149 * just sent a reset. If so, tell the caller to ignore the current packet. 1150 */ 1151 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt)) 1152 return sk->sk_state != TCP_CLOSE; 1153 1154 if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) { 1155 if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) && 1156 READ_ONCE(msk->local_key) == mp_opt.rcvr_key) { 1157 WRITE_ONCE(msk->rcv_fastclose, true); 1158 mptcp_schedule_work((struct sock *)msk); 1159 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSERX); 1160 } 1161 1162 if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) && 1163 add_addr_hmac_valid(msk, &mp_opt)) { 1164 if (!mp_opt.echo) { 1165 mptcp_pm_add_addr_received(sk, &mp_opt.addr); 1166 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR); 1167 } else { 1168 mptcp_pm_add_addr_echoed(msk, &mp_opt.addr); 1169 mptcp_pm_del_add_timer(msk, &mp_opt.addr, true); 1170 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD); 1171 } 1172 1173 if (mp_opt.addr.port) 1174 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD); 1175 } 1176 1177 if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR) 1178 mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list); 1179 1180 if (mp_opt.suboptions & OPTION_MPTCP_PRIO) { 1181 mptcp_pm_mp_prio_received(sk, mp_opt.backup); 1182 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX); 1183 } 1184 1185 if (mp_opt.suboptions & OPTION_MPTCP_FAIL) { 1186 mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq); 1187 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX); 1188 } 1189 1190 if (mp_opt.suboptions & OPTION_MPTCP_RST) { 1191 subflow->reset_seen = 1; 1192 subflow->reset_reason = mp_opt.reset_reason; 1193 subflow->reset_transient = mp_opt.reset_transient; 1194 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTRX); 1195 } 1196 1197 if (!(mp_opt.suboptions & OPTION_MPTCP_DSS)) 1198 return true; 1199 } 1200 1201 /* we can't wait for recvmsg() to update the ack_seq, otherwise 1202 * monodirectional flows will stuck 1203 */ 1204 if (mp_opt.use_ack) 1205 ack_update_msk(msk, sk, &mp_opt); 1206 1207 /* Zero-data-length packets are dropped by the caller and not 1208 * propagated to the MPTCP layer, so the skb extension does not 1209 * need to be allocated or populated. DATA_FIN information, if 1210 * present, needs to be updated here before the skb is freed. 1211 */ 1212 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) { 1213 if (mp_opt.data_fin && mp_opt.data_len == 1 && 1214 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64)) 1215 mptcp_schedule_work((struct sock *)msk); 1216 1217 return true; 1218 } 1219 1220 mpext = skb_ext_add(skb, SKB_EXT_MPTCP); 1221 if (!mpext) 1222 return true; 1223 1224 memset(mpext, 0, sizeof(*mpext)); 1225 1226 if (likely(mp_opt.use_map)) { 1227 if (mp_opt.mpc_map) { 1228 /* this is an MP_CAPABLE carrying MPTCP data 1229 * we know this map the first chunk of data 1230 */ 1231 mptcp_crypto_key_sha(subflow->remote_key, NULL, 1232 &mpext->data_seq); 1233 mpext->data_seq++; 1234 mpext->subflow_seq = 1; 1235 mpext->dsn64 = 1; 1236 mpext->mpc_map = 1; 1237 mpext->data_fin = 0; 1238 } else { 1239 mpext->data_seq = mp_opt.data_seq; 1240 mpext->subflow_seq = mp_opt.subflow_seq; 1241 mpext->dsn64 = mp_opt.dsn64; 1242 mpext->data_fin = mp_opt.data_fin; 1243 } 1244 mpext->data_len = mp_opt.data_len; 1245 mpext->use_map = 1; 1246 mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD); 1247 1248 if (mpext->csum_reqd) 1249 mpext->csum = mp_opt.csum; 1250 } 1251 1252 return true; 1253 } 1254 1255 static void mptcp_set_rwin(struct tcp_sock *tp, struct tcphdr *th) 1256 { 1257 const struct sock *ssk = (const struct sock *)tp; 1258 struct mptcp_subflow_context *subflow; 1259 u64 ack_seq, rcv_wnd_old, rcv_wnd_new; 1260 struct mptcp_sock *msk; 1261 u32 new_win; 1262 u64 win; 1263 1264 subflow = mptcp_subflow_ctx(ssk); 1265 msk = mptcp_sk(subflow->conn); 1266 1267 ack_seq = READ_ONCE(msk->ack_seq); 1268 rcv_wnd_new = ack_seq + tp->rcv_wnd; 1269 1270 rcv_wnd_old = atomic64_read(&msk->rcv_wnd_sent); 1271 if (after64(rcv_wnd_new, rcv_wnd_old)) { 1272 u64 rcv_wnd; 1273 1274 for (;;) { 1275 rcv_wnd = atomic64_cmpxchg(&msk->rcv_wnd_sent, rcv_wnd_old, rcv_wnd_new); 1276 1277 if (rcv_wnd == rcv_wnd_old) 1278 break; 1279 1280 rcv_wnd_old = rcv_wnd; 1281 if (before64(rcv_wnd_new, rcv_wnd_old)) { 1282 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICTUPDATE); 1283 goto raise_win; 1284 } 1285 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICT); 1286 } 1287 goto update_wspace; 1288 } 1289 1290 if (rcv_wnd_new != rcv_wnd_old) { 1291 raise_win: 1292 win = rcv_wnd_old - ack_seq; 1293 tp->rcv_wnd = min_t(u64, win, U32_MAX); 1294 new_win = tp->rcv_wnd; 1295 1296 /* Make sure we do not exceed the maximum possible 1297 * scaled window. 1298 */ 1299 if (unlikely(th->syn)) 1300 new_win = min(new_win, 65535U) << tp->rx_opt.rcv_wscale; 1301 if (!tp->rx_opt.rcv_wscale && 1302 READ_ONCE(sock_net(ssk)->ipv4.sysctl_tcp_workaround_signed_windows)) 1303 new_win = min(new_win, MAX_TCP_WINDOW); 1304 else 1305 new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale)); 1306 1307 /* RFC1323 scaling applied */ 1308 new_win >>= tp->rx_opt.rcv_wscale; 1309 th->window = htons(new_win); 1310 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDSHARED); 1311 } 1312 1313 update_wspace: 1314 WRITE_ONCE(msk->old_wspace, tp->rcv_wnd); 1315 } 1316 1317 __sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum) 1318 { 1319 struct csum_pseudo_header header; 1320 __wsum csum; 1321 1322 /* cfr RFC 8684 3.3.1.: 1323 * the data sequence number used in the pseudo-header is 1324 * always the 64-bit value, irrespective of what length is used in the 1325 * DSS option itself. 1326 */ 1327 header.data_seq = cpu_to_be64(data_seq); 1328 header.subflow_seq = htonl(subflow_seq); 1329 header.data_len = htons(data_len); 1330 header.csum = 0; 1331 1332 csum = csum_partial(&header, sizeof(header), sum); 1333 return csum_fold(csum); 1334 } 1335 1336 static __sum16 mptcp_make_csum(const struct mptcp_ext *mpext) 1337 { 1338 return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len, 1339 ~csum_unfold(mpext->csum)); 1340 } 1341 1342 static void put_len_csum(u16 len, __sum16 csum, void *data) 1343 { 1344 __sum16 *sumptr = data + 2; 1345 __be16 *ptr = data; 1346 1347 put_unaligned_be16(len, ptr); 1348 1349 put_unaligned(csum, sumptr); 1350 } 1351 1352 void mptcp_write_options(struct tcphdr *th, __be32 *ptr, struct tcp_sock *tp, 1353 struct mptcp_out_options *opts) 1354 { 1355 const struct sock *ssk = (const struct sock *)tp; 1356 struct mptcp_subflow_context *subflow; 1357 1358 /* Which options can be used together? 1359 * 1360 * X: mutually exclusive 1361 * O: often used together 1362 * C: can be used together in some cases 1363 * P: could be used together but we prefer not to (optimisations) 1364 * 1365 * Opt: | MPC | MPJ | DSS | ADD | RM | PRIO | FAIL | FC | 1366 * ------|------|------|------|------|------|------|------|------| 1367 * MPC |------|------|------|------|------|------|------|------| 1368 * MPJ | X |------|------|------|------|------|------|------| 1369 * DSS | X | X |------|------|------|------|------|------| 1370 * ADD | X | X | P |------|------|------|------|------| 1371 * RM | C | C | C | P |------|------|------|------| 1372 * PRIO | X | C | C | C | C |------|------|------| 1373 * FAIL | X | X | C | X | X | X |------|------| 1374 * FC | X | X | X | X | X | X | X |------| 1375 * RST | X | X | X | X | X | X | O | O | 1376 * ------|------|------|------|------|------|------|------|------| 1377 * 1378 * The same applies in mptcp_established_options() function. 1379 */ 1380 if (likely(OPTION_MPTCP_DSS & opts->suboptions)) { 1381 struct mptcp_ext *mpext = &opts->ext_copy; 1382 u8 len = TCPOLEN_MPTCP_DSS_BASE; 1383 u8 flags = 0; 1384 1385 if (mpext->use_ack) { 1386 flags = MPTCP_DSS_HAS_ACK; 1387 if (mpext->ack64) { 1388 len += TCPOLEN_MPTCP_DSS_ACK64; 1389 flags |= MPTCP_DSS_ACK64; 1390 } else { 1391 len += TCPOLEN_MPTCP_DSS_ACK32; 1392 } 1393 } 1394 1395 if (mpext->use_map) { 1396 len += TCPOLEN_MPTCP_DSS_MAP64; 1397 1398 /* Use only 64-bit mapping flags for now, add 1399 * support for optional 32-bit mappings later. 1400 */ 1401 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64; 1402 if (mpext->data_fin) 1403 flags |= MPTCP_DSS_DATA_FIN; 1404 1405 if (opts->csum_reqd) 1406 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 1407 } 1408 1409 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags); 1410 1411 if (mpext->use_ack) { 1412 if (mpext->ack64) { 1413 put_unaligned_be64(mpext->data_ack, ptr); 1414 ptr += 2; 1415 } else { 1416 put_unaligned_be32(mpext->data_ack32, ptr); 1417 ptr += 1; 1418 } 1419 } 1420 1421 if (mpext->use_map) { 1422 put_unaligned_be64(mpext->data_seq, ptr); 1423 ptr += 2; 1424 put_unaligned_be32(mpext->subflow_seq, ptr); 1425 ptr += 1; 1426 if (opts->csum_reqd) { 1427 /* data_len == 0 is reserved for the infinite mapping, 1428 * the checksum will also be set to 0. 1429 */ 1430 put_len_csum(mpext->data_len, 1431 (mpext->data_len ? mptcp_make_csum(mpext) : 0), 1432 ptr); 1433 } else { 1434 put_unaligned_be32(mpext->data_len << 16 | 1435 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1436 } 1437 ptr += 1; 1438 } 1439 1440 /* We might need to add MP_FAIL options in rare cases */ 1441 if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) 1442 goto mp_fail; 1443 } else if (OPTIONS_MPTCP_MPC & opts->suboptions) { 1444 u8 len, flag = MPTCP_CAP_HMAC_SHA256; 1445 1446 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) { 1447 len = TCPOLEN_MPTCP_MPC_SYN; 1448 } else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) { 1449 len = TCPOLEN_MPTCP_MPC_SYNACK; 1450 } else if (opts->data_len) { 1451 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 1452 if (opts->csum_reqd) 1453 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 1454 } else { 1455 len = TCPOLEN_MPTCP_MPC_ACK; 1456 } 1457 1458 if (opts->csum_reqd) 1459 flag |= MPTCP_CAP_CHECKSUM_REQD; 1460 1461 if (!opts->allow_join_id0) 1462 flag |= MPTCP_CAP_DENY_JOIN_ID0; 1463 1464 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len, 1465 MPTCP_SUPPORTED_VERSION, 1466 flag); 1467 1468 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) & 1469 opts->suboptions)) 1470 goto mp_capable_done; 1471 1472 put_unaligned_be64(opts->sndr_key, ptr); 1473 ptr += 2; 1474 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions)) 1475 goto mp_capable_done; 1476 1477 put_unaligned_be64(opts->rcvr_key, ptr); 1478 ptr += 2; 1479 if (!opts->data_len) 1480 goto mp_capable_done; 1481 1482 if (opts->csum_reqd) { 1483 put_len_csum(opts->data_len, 1484 __mptcp_make_csum(opts->data_seq, 1485 opts->subflow_seq, 1486 opts->data_len, 1487 ~csum_unfold(opts->csum)), 1488 ptr); 1489 } else { 1490 put_unaligned_be32(opts->data_len << 16 | 1491 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1492 } 1493 ptr += 1; 1494 1495 /* MPC is additionally mutually exclusive with MP_PRIO */ 1496 goto mp_capable_done; 1497 } else if (OPTIONS_MPTCP_MPJ & opts->suboptions) { 1498 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) { 1499 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1500 TCPOLEN_MPTCP_MPJ_SYN, 1501 opts->backup, opts->join_id); 1502 put_unaligned_be32(opts->token, ptr); 1503 ptr += 1; 1504 put_unaligned_be32(opts->nonce, ptr); 1505 ptr += 1; 1506 } else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) { 1507 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1508 TCPOLEN_MPTCP_MPJ_SYNACK, 1509 opts->backup, opts->join_id); 1510 put_unaligned_be64(opts->thmac, ptr); 1511 ptr += 2; 1512 put_unaligned_be32(opts->nonce, ptr); 1513 ptr += 1; 1514 } else { 1515 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1516 TCPOLEN_MPTCP_MPJ_ACK, 0, 0); 1517 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN); 1518 ptr += 5; 1519 } 1520 } else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) { 1521 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE; 1522 u8 echo = MPTCP_ADDR_ECHO; 1523 1524 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1525 if (opts->addr.family == AF_INET6) 1526 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE; 1527 #endif 1528 1529 if (opts->addr.port) 1530 len += TCPOLEN_MPTCP_PORT_LEN; 1531 1532 if (opts->ahmac) { 1533 len += sizeof(opts->ahmac); 1534 echo = 0; 1535 } 1536 1537 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 1538 len, echo, opts->addr.id); 1539 if (opts->addr.family == AF_INET) { 1540 memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4); 1541 ptr += 1; 1542 } 1543 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1544 else if (opts->addr.family == AF_INET6) { 1545 memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16); 1546 ptr += 4; 1547 } 1548 #endif 1549 1550 if (!opts->addr.port) { 1551 if (opts->ahmac) { 1552 put_unaligned_be64(opts->ahmac, ptr); 1553 ptr += 2; 1554 } 1555 } else { 1556 u16 port = ntohs(opts->addr.port); 1557 1558 if (opts->ahmac) { 1559 u8 *bptr = (u8 *)ptr; 1560 1561 put_unaligned_be16(port, bptr); 1562 bptr += 2; 1563 put_unaligned_be64(opts->ahmac, bptr); 1564 bptr += 8; 1565 put_unaligned_be16(TCPOPT_NOP << 8 | 1566 TCPOPT_NOP, bptr); 1567 1568 ptr += 3; 1569 } else { 1570 put_unaligned_be32(port << 16 | 1571 TCPOPT_NOP << 8 | 1572 TCPOPT_NOP, ptr); 1573 ptr += 1; 1574 } 1575 } 1576 } else if (unlikely(OPTION_MPTCP_FASTCLOSE & opts->suboptions)) { 1577 /* FASTCLOSE is mutually exclusive with others except RST */ 1578 *ptr++ = mptcp_option(MPTCPOPT_MP_FASTCLOSE, 1579 TCPOLEN_MPTCP_FASTCLOSE, 1580 0, 0); 1581 put_unaligned_be64(opts->rcvr_key, ptr); 1582 ptr += 2; 1583 1584 if (OPTION_MPTCP_RST & opts->suboptions) 1585 goto mp_rst; 1586 return; 1587 } else if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) { 1588 mp_fail: 1589 /* MP_FAIL is mutually exclusive with others except RST */ 1590 subflow = mptcp_subflow_ctx(ssk); 1591 subflow->send_mp_fail = 0; 1592 1593 *ptr++ = mptcp_option(MPTCPOPT_MP_FAIL, 1594 TCPOLEN_MPTCP_FAIL, 1595 0, 0); 1596 put_unaligned_be64(opts->fail_seq, ptr); 1597 ptr += 2; 1598 1599 if (OPTION_MPTCP_RST & opts->suboptions) 1600 goto mp_rst; 1601 return; 1602 } else if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) { 1603 mp_rst: 1604 *ptr++ = mptcp_option(MPTCPOPT_RST, 1605 TCPOLEN_MPTCP_RST, 1606 opts->reset_transient, 1607 opts->reset_reason); 1608 return; 1609 } 1610 1611 if (OPTION_MPTCP_PRIO & opts->suboptions) { 1612 subflow = mptcp_subflow_ctx(ssk); 1613 subflow->send_mp_prio = 0; 1614 1615 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO, 1616 TCPOLEN_MPTCP_PRIO, 1617 opts->backup, TCPOPT_NOP); 1618 1619 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPPRIOTX); 1620 } 1621 1622 mp_capable_done: 1623 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) { 1624 u8 i = 1; 1625 1626 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR, 1627 TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr, 1628 0, opts->rm_list.ids[0]); 1629 1630 while (i < opts->rm_list.nr) { 1631 u8 id1, id2, id3, id4; 1632 1633 id1 = opts->rm_list.ids[i]; 1634 id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP; 1635 id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP; 1636 id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP; 1637 put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr); 1638 ptr += 1; 1639 i += 4; 1640 } 1641 } 1642 1643 if (tp) 1644 mptcp_set_rwin(tp, th); 1645 } 1646 1647 __be32 mptcp_get_reset_option(const struct sk_buff *skb) 1648 { 1649 const struct mptcp_ext *ext = mptcp_get_ext(skb); 1650 u8 flags, reason; 1651 1652 if (ext) { 1653 flags = ext->reset_transient; 1654 reason = ext->reset_reason; 1655 1656 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST, 1657 flags, reason); 1658 } 1659 1660 return htonl(0u); 1661 } 1662 EXPORT_SYMBOL_GPL(mptcp_get_reset_option); 1663