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 if (unlikely(subflow_simultaneous_connect(sk))) { 412 WARN_ON_ONCE(!mptcp_try_fallback(sk, MPTCP_MIB_SIMULTCONNFALLBACK)); 413 414 /* Ensure mptcp_finish_connect() will not process the 415 * MPC handshake. 416 */ 417 subflow->request_mptcp = 0; 418 return false; 419 } 420 421 opts->suboptions = OPTION_MPTCP_MPC_SYN; 422 opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk)); 423 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk)); 424 *size = TCPOLEN_MPTCP_MPC_SYN; 425 return true; 426 } else if (subflow->request_join) { 427 pr_debug("remote_token=%u, nonce=%u\n", subflow->remote_token, 428 subflow->local_nonce); 429 opts->suboptions = OPTION_MPTCP_MPJ_SYN; 430 opts->join_id = subflow->local_id; 431 opts->token = subflow->remote_token; 432 opts->nonce = subflow->local_nonce; 433 opts->backup = subflow->request_bkup; 434 *size = TCPOLEN_MPTCP_MPJ_SYN; 435 return true; 436 } 437 return false; 438 } 439 440 static void clear_3rdack_retransmission(struct sock *sk) 441 { 442 struct inet_connection_sock *icsk = inet_csk(sk); 443 444 sk_stop_timer(sk, &icsk->icsk_delack_timer); 445 icsk->icsk_ack.ato = 0; 446 icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER); 447 } 448 449 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb, 450 bool snd_data_fin_enable, int *size, 451 struct mptcp_out_options *opts) 452 { 453 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 454 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 455 struct mptcp_ext *mpext; 456 unsigned int data_len; 457 u8 len; 458 459 /* When skb is not available, we better over-estimate the emitted 460 * options len. A full DSS option (28 bytes) is longer than 461 * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so 462 * tell the caller to defer the estimate to 463 * mptcp_established_options_dss(), which will reserve enough space. 464 */ 465 if (!skb) 466 return false; 467 468 /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */ 469 if (READ_ONCE(subflow->fully_established) || snd_data_fin_enable || 470 subflow->snd_isn != TCP_SKB_CB(skb)->seq || 471 sk->sk_state != TCP_ESTABLISHED) 472 return false; 473 474 if (subflow->mp_capable) { 475 mpext = mptcp_get_ext(skb); 476 data_len = mpext ? mpext->data_len : 0; 477 478 /* we will check ops->data_len in mptcp_write_options() to 479 * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and 480 * TCPOLEN_MPTCP_MPC_ACK 481 */ 482 opts->data_len = data_len; 483 opts->suboptions = OPTION_MPTCP_MPC_ACK; 484 opts->sndr_key = subflow->local_key; 485 opts->rcvr_key = subflow->remote_key; 486 opts->csum_reqd = READ_ONCE(msk->csum_enabled); 487 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk)); 488 489 /* Section 3.1. 490 * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK 491 * packets that start the first subflow of an MPTCP connection, 492 * as well as the first packet that carries data 493 */ 494 if (data_len > 0) { 495 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 496 if (opts->csum_reqd) { 497 /* we need to propagate more info to csum the pseudo hdr */ 498 opts->data_seq = mpext->data_seq; 499 opts->subflow_seq = mpext->subflow_seq; 500 opts->csum = mpext->csum; 501 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 502 } 503 *size = ALIGN(len, 4); 504 } else { 505 *size = TCPOLEN_MPTCP_MPC_ACK; 506 } 507 508 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d\n", 509 subflow, subflow->local_key, subflow->remote_key, 510 data_len); 511 512 return true; 513 } else if (subflow->mp_join) { 514 opts->suboptions = OPTION_MPTCP_MPJ_ACK; 515 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN); 516 *size = TCPOLEN_MPTCP_MPJ_ACK; 517 pr_debug("subflow=%p\n", subflow); 518 519 /* we can use the full delegate action helper only from BH context 520 * If we are in process context - sk is flushing the backlog at 521 * socket lock release time - just set the appropriate flag, will 522 * be handled by the release callback 523 */ 524 if (sock_owned_by_user(sk)) 525 set_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status); 526 else 527 mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_ACK); 528 return true; 529 } 530 return false; 531 } 532 533 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow, 534 struct sk_buff *skb, struct mptcp_ext *ext) 535 { 536 /* The write_seq value has already been incremented, so the actual 537 * sequence number for the DATA_FIN is one less. 538 */ 539 u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1; 540 541 if (!ext->use_map || !skb->len) { 542 /* RFC6824 requires a DSS mapping with specific values 543 * if DATA_FIN is set but no data payload is mapped 544 */ 545 ext->data_fin = 1; 546 ext->use_map = 1; 547 ext->dsn64 = 1; 548 ext->data_seq = data_fin_tx_seq; 549 ext->subflow_seq = 0; 550 ext->data_len = 1; 551 } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) { 552 /* If there's an existing DSS mapping and it is the 553 * final mapping, DATA_FIN consumes 1 additional byte of 554 * mapping space. 555 */ 556 ext->data_fin = 1; 557 ext->data_len++; 558 } 559 } 560 561 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb, 562 bool snd_data_fin_enable, int *size, 563 struct mptcp_out_options *opts) 564 { 565 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 566 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 567 struct tcp_sock *tp = tcp_sk(sk); 568 unsigned int dss_size = 0; 569 struct mptcp_ext *mpext; 570 unsigned int ack_size; 571 bool ret = false; 572 573 /* Zero `use_ack` and `use_map` flags with one shot. */ 574 BUILD_BUG_ON(sizeof_field(struct mptcp_ext, flags) != sizeof(u16)); 575 BUILD_BUG_ON(!IS_ALIGNED(offsetof(struct mptcp_ext, flags), 576 sizeof(u16))); 577 *(u16 *)&opts->ext_copy.flags = 0; 578 opts->csum_reqd = READ_ONCE(msk->csum_enabled); 579 mpext = skb ? mptcp_get_ext(skb) : NULL; 580 581 if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) { 582 unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64; 583 584 if (mpext) { 585 if (opts->csum_reqd) 586 map_size += TCPOLEN_MPTCP_DSS_CHECKSUM; 587 588 opts->ext_copy = *mpext; 589 } 590 591 dss_size = map_size; 592 if (skb && snd_data_fin_enable) 593 mptcp_write_data_fin(subflow, skb, &opts->ext_copy); 594 opts->suboptions = OPTION_MPTCP_DSS; 595 ret = true; 596 } 597 598 /* passive sockets msk will set the 'can_ack' after accept(), even 599 * if the first subflow may have the already the remote key handy 600 */ 601 if (!READ_ONCE(msk->can_ack)) { 602 *size = ALIGN(dss_size, 4); 603 return ret; 604 } 605 606 if (READ_ONCE(msk->use_64bit_ack)) { 607 ack_size = TCPOLEN_MPTCP_DSS_ACK64; 608 opts->ext_copy.ack64 = 1; 609 } else { 610 ack_size = TCPOLEN_MPTCP_DSS_ACK32; 611 opts->ext_copy.ack64 = 0; 612 } 613 opts->ext_copy.use_ack = 1; 614 opts->suboptions = OPTION_MPTCP_DSS; 615 616 /* Add kind/length/subtype/flag overhead if mapping is not populated */ 617 if (dss_size == 0) 618 ack_size += TCPOLEN_MPTCP_DSS_BASE; 619 620 /* The caller is __tcp_transmit_skb(), and will compute the new rcv 621 * wnd soon: ensure that the window can shrink. 622 */ 623 if (skb) 624 tp->rcv_wnd = tp->rcv_nxt - tp->rcv_wup; 625 626 dss_size += ack_size; 627 628 *size = ALIGN(dss_size, 4); 629 return true; 630 } 631 632 static u64 add_addr_generate_hmac(u64 key1, u64 key2, 633 struct mptcp_addr_info *addr) 634 { 635 u16 port = ntohs(addr->port); 636 u8 hmac[SHA256_DIGEST_SIZE]; 637 u8 msg[19]; 638 int i = 0; 639 640 msg[i++] = addr->id; 641 if (addr->family == AF_INET) { 642 memcpy(&msg[i], &addr->addr.s_addr, 4); 643 i += 4; 644 } 645 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 646 else if (addr->family == AF_INET6) { 647 memcpy(&msg[i], &addr->addr6.s6_addr, 16); 648 i += 16; 649 } 650 #endif 651 msg[i++] = port >> 8; 652 msg[i++] = port & 0xFF; 653 654 mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac); 655 656 return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]); 657 } 658 659 static bool mptcp_established_options_add_addr(struct sock *sk, 660 struct sk_buff *skb, int *size, 661 unsigned int remaining, 662 bool has_ts, 663 struct mptcp_out_options *opts) 664 { 665 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 666 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 667 struct mptcp_addr_info addr; 668 bool drop_ts = has_ts; 669 bool echo; 670 671 /* add addr will strip the existing options, be sure to avoid breaking 672 * MPC/MPJ handshakes 673 */ 674 if (!mptcp_pm_should_add_signal(msk) || 675 (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) || 676 !skb || !skb_is_tcp_pure_ack(skb) || 677 !mptcp_pm_add_addr_signal(msk, size, remaining, &addr, &echo, 678 &drop_ts)) 679 return false; 680 681 pr_debug("drop other suboptions\n"); 682 opts->suboptions = OPTION_MPTCP_ADD_ADDR; 683 opts->drop_ts = drop_ts; 684 opts->addr = addr; 685 if (!echo) { 686 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDRTX); 687 opts->ahmac = add_addr_generate_hmac(READ_ONCE(msk->local_key), 688 READ_ONCE(msk->remote_key), 689 &opts->addr); 690 } else { 691 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADDTX); 692 opts->ahmac = 0; 693 } 694 pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d\n", 695 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port)); 696 697 return true; 698 } 699 700 static bool mptcp_established_options_rm_addr(struct sock *sk, int *size, 701 unsigned int remaining, 702 struct mptcp_out_options *opts) 703 { 704 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 705 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 706 struct mptcp_rm_list rm_list; 707 int i; 708 709 if (!mptcp_pm_should_rm_signal(msk) || 710 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list, size))) 711 return false; 712 713 opts->suboptions |= OPTION_MPTCP_RM_ADDR; 714 opts->rm_list = rm_list; 715 716 for (i = 0; i < opts->rm_list.nr; i++) 717 pr_debug("rm_list_ids[%d]=%d\n", i, opts->rm_list.ids[i]); 718 MPTCP_ADD_STATS(sock_net(sk), MPTCP_MIB_RMADDRTX, opts->rm_list.nr); 719 return true; 720 } 721 722 static bool mptcp_established_options_mp_prio(struct sock *sk, int *size, 723 unsigned int remaining, 724 struct mptcp_out_options *opts) 725 { 726 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 727 728 /* can't send MP_PRIO with MPC, as they share the same option space: 729 * 'backup'. Also it makes no sense at all 730 */ 731 if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC)) 732 return false; 733 734 /* account for the trailing 'nop' option */ 735 if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN) 736 return false; 737 738 *size = TCPOLEN_MPTCP_PRIO_ALIGN; 739 opts->suboptions |= OPTION_MPTCP_PRIO; 740 opts->backup = subflow->request_bkup; 741 742 pr_debug("prio=%d\n", opts->backup); 743 744 return true; 745 } 746 747 static noinline bool mptcp_established_options_rst(struct sock *sk, 748 int *size, 749 unsigned int remaining, 750 struct mptcp_out_options *opts) 751 { 752 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 753 754 if (remaining < TCPOLEN_MPTCP_RST) 755 return false; 756 757 *size = TCPOLEN_MPTCP_RST; 758 opts->suboptions |= OPTION_MPTCP_RST; 759 opts->reset_transient = subflow->reset_transient; 760 opts->reset_reason = subflow->reset_reason; 761 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTTX); 762 763 return true; 764 } 765 766 static bool mptcp_established_options_fastclose(struct sock *sk, int *size, 767 unsigned int remaining, 768 struct mptcp_out_options *opts) 769 { 770 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 771 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 772 773 if (likely(!subflow->send_fastclose)) 774 return false; 775 776 if (remaining < TCPOLEN_MPTCP_FASTCLOSE) 777 return false; 778 779 *size = TCPOLEN_MPTCP_FASTCLOSE; 780 opts->suboptions |= OPTION_MPTCP_FASTCLOSE; 781 opts->rcvr_key = READ_ONCE(msk->remote_key); 782 783 pr_debug("FASTCLOSE key=%llu\n", opts->rcvr_key); 784 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSETX); 785 return true; 786 } 787 788 static bool mptcp_established_options_mp_fail(struct sock *sk, int *size, 789 unsigned int remaining, 790 struct mptcp_out_options *opts) 791 { 792 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 793 794 if (likely(!subflow->send_mp_fail)) 795 return false; 796 797 if (remaining < TCPOLEN_MPTCP_FAIL) 798 return false; 799 800 *size = TCPOLEN_MPTCP_FAIL; 801 opts->suboptions |= OPTION_MPTCP_FAIL; 802 opts->fail_seq = subflow->map_seq; 803 804 pr_debug("MP_FAIL fail_seq=%llu\n", opts->fail_seq); 805 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILTX); 806 807 return true; 808 } 809 810 int mptcp_established_options(struct sock *sk, struct sk_buff *skb, 811 unsigned int remaining, bool has_ts, 812 struct mptcp_out_options *opts) 813 { 814 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 815 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 816 int total_size = 0; 817 bool snd_data_fin; 818 bool ret = false; 819 int opt_size = 0; 820 821 opts->suboptions = 0; 822 823 /* Force later mptcp_write_options(), but do not use any actual 824 * option space. 825 */ 826 if (unlikely(__mptcp_check_fallback(msk) && !mptcp_check_infinite_map(skb))) 827 return 0; 828 829 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) { 830 if (mptcp_established_options_fastclose(sk, &opt_size, remaining, opts) || 831 mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) { 832 total_size += opt_size; 833 remaining -= opt_size; 834 } 835 /* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */ 836 if (mptcp_established_options_rst(sk, &opt_size, remaining, opts)) { 837 total_size += opt_size; 838 remaining -= opt_size; 839 } 840 return total_size; 841 } 842 843 snd_data_fin = mptcp_data_fin_enabled(msk); 844 if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, opts)) 845 ret = true; 846 else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, opts)) { 847 int mp_fail_size; 848 849 ret = true; 850 if (mptcp_established_options_mp_fail(sk, &mp_fail_size, 851 remaining - opt_size, opts)) { 852 total_size += opt_size + mp_fail_size; 853 remaining -= opt_size - mp_fail_size; 854 return total_size; 855 } 856 } 857 858 /* we reserved enough space for the above options, and exceeding the 859 * TCP option space would be fatal 860 */ 861 if (WARN_ON_ONCE(opt_size > remaining)) 862 return -1; 863 864 total_size += opt_size; 865 remaining -= opt_size; 866 if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, 867 has_ts, opts)) { 868 total_size += opt_size; 869 remaining -= opt_size; 870 ret = true; 871 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) { 872 total_size += opt_size; 873 remaining -= opt_size; 874 ret = true; 875 } 876 877 if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) { 878 total_size += opt_size; 879 remaining -= opt_size; 880 ret = true; 881 } 882 883 return ret ? total_size : -1; 884 } 885 886 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size, 887 struct mptcp_out_options *opts) 888 { 889 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 890 891 if (subflow_req->mp_capable) { 892 opts->suboptions = OPTION_MPTCP_MPC_SYNACK; 893 opts->sndr_key = subflow_req->local_key; 894 opts->csum_reqd = subflow_req->csum_reqd; 895 opts->allow_join_id0 = subflow_req->allow_join_id0; 896 *size = TCPOLEN_MPTCP_MPC_SYNACK; 897 pr_debug("subflow_req=%p, local_key=%llu\n", 898 subflow_req, subflow_req->local_key); 899 return true; 900 } else if (subflow_req->mp_join) { 901 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK; 902 opts->backup = subflow_req->request_bkup; 903 opts->join_id = subflow_req->local_id; 904 opts->thmac = subflow_req->thmac; 905 opts->nonce = subflow_req->local_nonce; 906 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u\n", 907 subflow_req, opts->backup, opts->join_id, 908 opts->thmac, opts->nonce); 909 *size = TCPOLEN_MPTCP_MPJ_SYNACK; 910 return true; 911 } 912 return false; 913 } 914 915 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk, 916 struct mptcp_subflow_context *subflow, 917 struct sk_buff *skb, 918 struct mptcp_options_received *mp_opt) 919 { 920 /* here we can process OoO, in-window pkts, only in-sequence 4th ack 921 * will make the subflow fully established 922 */ 923 if (likely(READ_ONCE(subflow->fully_established))) { 924 /* on passive sockets, check for 3rd ack retransmission 925 * note that msk is always set by subflow_syn_recv_sock() 926 * for mp_join subflows 927 */ 928 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 && 929 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq && 930 subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) && 931 !subflow->request_join) 932 tcp_send_ack(ssk); 933 goto check_notify; 934 } 935 936 /* we must process OoO packets before the first subflow is fully 937 * established. OoO packets are instead a protocol violation 938 * for MP_JOIN subflows as the peer must not send any data 939 * before receiving the forth ack - cfr. RFC 8684 section 3.2. 940 */ 941 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) { 942 if (subflow->mp_join) 943 goto reset; 944 if (subflow->is_mptfo && mp_opt->suboptions & OPTION_MPTCP_MPC_ACK) 945 goto set_fully_established; 946 return subflow->mp_capable; 947 } 948 949 if (subflow->remote_key_valid && 950 (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) || 951 ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) && 952 (!mp_opt->echo || subflow->mp_join)))) { 953 /* subflows are fully established as soon as we get any 954 * additional ack, including ADD_ADDR. 955 */ 956 goto set_fully_established; 957 } 958 959 /* If the first established packet does not contain MP_CAPABLE + data 960 * then fallback to TCP. Fallback scenarios requires a reset for 961 * MP_JOIN subflows. 962 */ 963 if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) { 964 if (subflow->mp_join) 965 goto reset; 966 subflow->mp_capable = 0; 967 if (!mptcp_try_fallback(ssk, MPTCP_MIB_MPCAPABLEDATAFALLBACK)) { 968 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_FALLBACKFAILED); 969 goto reset; 970 } 971 return false; 972 } 973 974 if (unlikely(!READ_ONCE(msk->pm.server_side))) 975 pr_warn_once("bogus mpc option on established client sk"); 976 977 set_fully_established: 978 if (mp_opt->deny_join_id0) 979 WRITE_ONCE(msk->pm.remote_deny_join_id0, true); 980 981 mptcp_data_lock((struct sock *)msk); 982 __mptcp_subflow_fully_established(msk, subflow, mp_opt); 983 mptcp_data_unlock((struct sock *)msk); 984 985 check_notify: 986 /* if the subflow is not already linked into the conn_list, we can't 987 * notify the PM: this subflow is still on the listener queue 988 * and the PM possibly acquiring the subflow lock could race with 989 * the listener close 990 */ 991 if (likely(subflow->pm_notified) || list_empty(&subflow->node)) 992 return true; 993 994 subflow->pm_notified = 1; 995 if (subflow->mp_join) { 996 clear_3rdack_retransmission(ssk); 997 mptcp_pm_subflow_established(msk); 998 } else { 999 mptcp_pm_fully_established(msk, ssk); 1000 } 1001 return true; 1002 1003 reset: 1004 mptcp_subflow_reset(ssk); 1005 return false; 1006 } 1007 1008 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq) 1009 { 1010 u32 old_seq32, cur_seq32; 1011 1012 old_seq32 = (u32)old_seq; 1013 cur_seq32 = (u32)cur_seq; 1014 cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32; 1015 if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32))) 1016 return cur_seq + (1LL << 32); 1017 1018 /* reverse wrap could happen, too */ 1019 if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32))) 1020 return cur_seq - (1LL << 32); 1021 return cur_seq; 1022 } 1023 1024 static void __mptcp_snd_una_update(struct mptcp_sock *msk, u64 new_snd_una) 1025 { 1026 msk->bytes_acked += new_snd_una - msk->snd_una; 1027 WRITE_ONCE(msk->snd_una, new_snd_una); 1028 } 1029 1030 static void rwin_update(struct mptcp_sock *msk, struct sock *ssk, 1031 struct sk_buff *skb) 1032 { 1033 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 1034 struct tcp_sock *tp = tcp_sk(ssk); 1035 u64 mptcp_rcv_wnd; 1036 1037 /* Avoid touching extra cachelines if TCP is going to accept this 1038 * skb without filling the TCP-level window even with a possibly 1039 * outdated mptcp-level rwin. 1040 */ 1041 if (!skb->len || skb->len < tcp_receive_window(tp)) 1042 return; 1043 1044 mptcp_rcv_wnd = atomic64_read(&msk->rcv_wnd_sent); 1045 if (!after64(mptcp_rcv_wnd, subflow->rcv_wnd_sent)) 1046 return; 1047 1048 /* Some other subflow grew the mptcp-level rwin since rcv_wup, 1049 * resync. 1050 */ 1051 tp->rcv_wnd += mptcp_rcv_wnd - subflow->rcv_wnd_sent; 1052 tcp_update_max_rcv_wnd_seq(tp); 1053 subflow->rcv_wnd_sent = mptcp_rcv_wnd; 1054 } 1055 1056 static void ack_update_msk(struct mptcp_sock *msk, 1057 struct sock *ssk, 1058 struct mptcp_options_received *mp_opt) 1059 { 1060 u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt); 1061 struct sock *sk = (struct sock *)msk; 1062 u64 old_snd_una; 1063 1064 mptcp_data_lock(sk); 1065 1066 /* avoid ack expansion on update conflict, to reduce the risk of 1067 * wrongly expanding to a future ack sequence number, which is way 1068 * more dangerous than missing an ack 1069 */ 1070 old_snd_una = msk->snd_una; 1071 new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64); 1072 1073 /* ACK for data not even sent yet? Ignore.*/ 1074 if (unlikely(after64(new_snd_una, snd_nxt))) 1075 new_snd_una = old_snd_una; 1076 1077 new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd; 1078 1079 if (after64(new_wnd_end, msk->wnd_end)) 1080 WRITE_ONCE(msk->wnd_end, new_wnd_end); 1081 1082 /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */ 1083 if (after64(msk->wnd_end, snd_nxt)) 1084 __mptcp_check_push(sk, ssk); 1085 1086 if (after64(new_snd_una, old_snd_una)) { 1087 __mptcp_snd_una_update(msk, new_snd_una); 1088 __mptcp_data_acked(sk); 1089 } 1090 msk->last_ack_recv = tcp_jiffies32; 1091 mptcp_data_unlock(sk); 1092 1093 trace_ack_update_msk(mp_opt->data_ack, 1094 old_snd_una, new_snd_una, 1095 new_wnd_end, READ_ONCE(msk->wnd_end)); 1096 } 1097 1098 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit) 1099 { 1100 /* Skip if DATA_FIN was already received. 1101 * If updating simultaneously with the recvmsg loop, values 1102 * should match. If they mismatch, the peer is misbehaving and 1103 * we will prefer the most recent information. 1104 */ 1105 if (READ_ONCE(msk->rcv_data_fin)) 1106 return false; 1107 1108 WRITE_ONCE(msk->rcv_data_fin_seq, 1109 mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit)); 1110 WRITE_ONCE(msk->rcv_data_fin, 1); 1111 1112 return true; 1113 } 1114 1115 static bool add_addr_hmac_valid(struct mptcp_sock *msk, 1116 struct mptcp_options_received *mp_opt) 1117 { 1118 u64 hmac = 0; 1119 1120 if (mp_opt->echo) 1121 return true; 1122 1123 hmac = add_addr_generate_hmac(READ_ONCE(msk->remote_key), 1124 READ_ONCE(msk->local_key), 1125 &mp_opt->addr); 1126 1127 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n", 1128 msk, hmac, mp_opt->ahmac); 1129 1130 return hmac == mp_opt->ahmac; 1131 } 1132 1133 /* Return false in case of error (or subflow has been reset), 1134 * else return true. 1135 */ 1136 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb) 1137 { 1138 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 1139 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 1140 struct mptcp_options_received mp_opt; 1141 struct mptcp_ext *mpext; 1142 1143 if (__mptcp_check_fallback(msk)) { 1144 /* Keep it simple and unconditionally trigger send data cleanup and 1145 * pending queue spooling. We will need to acquire the data lock 1146 * for more accurate checks, and once the lock is acquired, such 1147 * helpers are cheap. 1148 */ 1149 mptcp_data_lock(subflow->conn); 1150 if (sk_stream_memory_free(sk)) 1151 __mptcp_check_push(subflow->conn, sk); 1152 1153 /* on fallback we just need to ignore the msk-level snd_una, as 1154 * this is really plain TCP 1155 */ 1156 __mptcp_snd_una_update(msk, READ_ONCE(msk->snd_nxt)); 1157 1158 __mptcp_data_acked(subflow->conn); 1159 mptcp_data_unlock(subflow->conn); 1160 return true; 1161 } 1162 1163 mptcp_get_options(skb, &mp_opt); 1164 1165 /* The subflow can be in close state only if check_fully_established() 1166 * just sent a reset. If so, tell the caller to ignore the current packet. 1167 */ 1168 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt)) 1169 return sk->sk_state != TCP_CLOSE; 1170 1171 if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) { 1172 if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) && 1173 READ_ONCE(msk->local_key) == mp_opt.rcvr_key) { 1174 WRITE_ONCE(msk->rcv_fastclose, true); 1175 mptcp_schedule_work((struct sock *)msk); 1176 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSERX); 1177 } 1178 1179 if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) && 1180 add_addr_hmac_valid(msk, &mp_opt)) { 1181 if (!mp_opt.echo) { 1182 mptcp_pm_add_addr_received(sk, &mp_opt.addr); 1183 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR); 1184 } else { 1185 mptcp_pm_add_addr_echoed(msk, &mp_opt.addr); 1186 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD); 1187 } 1188 1189 if (mp_opt.addr.port) 1190 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD); 1191 } 1192 1193 if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR) 1194 mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list); 1195 1196 if (mp_opt.suboptions & OPTION_MPTCP_PRIO) { 1197 mptcp_pm_mp_prio_received(sk, mp_opt.backup); 1198 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX); 1199 } 1200 1201 if (mp_opt.suboptions & OPTION_MPTCP_FAIL) { 1202 mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq); 1203 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX); 1204 } 1205 1206 if (mp_opt.suboptions & OPTION_MPTCP_RST) { 1207 subflow->reset_seen = 1; 1208 subflow->reset_reason = mp_opt.reset_reason; 1209 subflow->reset_transient = mp_opt.reset_transient; 1210 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTRX); 1211 } 1212 1213 if (!(mp_opt.suboptions & OPTION_MPTCP_DSS)) 1214 return true; 1215 } 1216 1217 /* we can't wait for recvmsg() to update the ack_seq, otherwise 1218 * monodirectional flows will stuck 1219 */ 1220 if (mp_opt.use_ack) 1221 ack_update_msk(msk, sk, &mp_opt); 1222 rwin_update(msk, sk, skb); 1223 1224 /* Zero-data-length packets are dropped by the caller and not 1225 * propagated to the MPTCP layer, so the skb extension does not 1226 * need to be allocated or populated. DATA_FIN information, if 1227 * present, needs to be updated here before the skb is freed. 1228 */ 1229 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) { 1230 if (mp_opt.data_fin && mp_opt.data_len == 1 && 1231 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64)) 1232 mptcp_schedule_work((struct sock *)msk); 1233 1234 return true; 1235 } 1236 1237 mpext = skb_ext_add(skb, SKB_EXT_MPTCP); 1238 if (!mpext) 1239 return false; 1240 1241 memset(mpext, 0, sizeof(*mpext)); 1242 1243 if (likely(mp_opt.use_map)) { 1244 if (mp_opt.mpc_map) { 1245 /* this is an MP_CAPABLE carrying MPTCP data 1246 * we know this map the first chunk of data 1247 */ 1248 mptcp_crypto_key_sha(subflow->remote_key, NULL, 1249 &mpext->data_seq); 1250 mpext->data_seq++; 1251 mpext->subflow_seq = 1; 1252 mpext->dsn64 = 1; 1253 mpext->mpc_map = 1; 1254 mpext->data_fin = 0; 1255 } else { 1256 mpext->data_seq = mp_opt.data_seq; 1257 mpext->subflow_seq = mp_opt.subflow_seq; 1258 mpext->dsn64 = mp_opt.dsn64; 1259 mpext->data_fin = mp_opt.data_fin; 1260 } 1261 mpext->data_len = mp_opt.data_len; 1262 mpext->use_map = 1; 1263 mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD); 1264 1265 if (mpext->csum_reqd) 1266 mpext->csum = mp_opt.csum; 1267 } 1268 1269 return true; 1270 } 1271 1272 static u64 mptcp_set_rwin(struct mptcp_sock *msk, struct tcp_sock *tp, 1273 struct tcphdr *th, u64 ack_seq) 1274 { 1275 const struct sock *ssk = (const struct sock *)tp; 1276 u64 rcv_wnd_old, rcv_wnd_new; 1277 u32 new_win; 1278 u64 win; 1279 1280 rcv_wnd_new = ack_seq + tp->rcv_wnd; 1281 1282 rcv_wnd_old = atomic64_read(&msk->rcv_wnd_sent); 1283 if (after64(rcv_wnd_new, rcv_wnd_old)) { 1284 u64 rcv_wnd; 1285 1286 for (;;) { 1287 rcv_wnd = atomic64_cmpxchg(&msk->rcv_wnd_sent, rcv_wnd_old, rcv_wnd_new); 1288 1289 if (rcv_wnd == rcv_wnd_old) 1290 break; 1291 1292 rcv_wnd_old = rcv_wnd; 1293 if (before64(rcv_wnd_new, rcv_wnd_old)) { 1294 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICTUPDATE); 1295 goto raise_win; 1296 } 1297 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICT); 1298 } 1299 goto update_wspace; 1300 } 1301 1302 if (rcv_wnd_new != rcv_wnd_old) { 1303 raise_win: 1304 /* The msk-level rcv wnd is after the tcp level one, 1305 * sync the latter. 1306 */ 1307 rcv_wnd_new = rcv_wnd_old; 1308 win = rcv_wnd_old - ack_seq; 1309 new_win = min_t(u64, win, U32_MAX); 1310 tp->rcv_wnd = new_win; 1311 tcp_update_max_rcv_wnd_seq(tp); 1312 1313 /* Make sure we do not exceed the maximum possible 1314 * scaled window. 1315 */ 1316 if (unlikely(th->syn)) 1317 new_win = min(new_win, 65535U) << tp->rx_opt.rcv_wscale; 1318 if (!tp->rx_opt.rcv_wscale && 1319 READ_ONCE(sock_net(ssk)->ipv4.sysctl_tcp_workaround_signed_windows)) 1320 new_win = min(new_win, MAX_TCP_WINDOW); 1321 else 1322 new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale)); 1323 1324 /* RFC1323 scaling applied */ 1325 new_win >>= tp->rx_opt.rcv_wscale; 1326 th->window = htons(new_win); 1327 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDSHARED); 1328 } 1329 1330 update_wspace: 1331 WRITE_ONCE(msk->old_wspace, tp->rcv_wnd); 1332 return rcv_wnd_new; 1333 } 1334 1335 static void mptcp_track_rwin(struct tcp_sock *tp) 1336 { 1337 const struct sock *ssk = (const struct sock *)tp; 1338 struct mptcp_subflow_context *subflow; 1339 struct mptcp_sock *msk; 1340 1341 if (!ssk) 1342 return; 1343 1344 subflow = mptcp_subflow_ctx(ssk); 1345 msk = mptcp_sk(subflow->conn); 1346 WRITE_ONCE(msk->old_wspace, tp->rcv_wnd); 1347 } 1348 1349 __sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum) 1350 { 1351 struct csum_pseudo_header header; 1352 __wsum csum; 1353 1354 /* cfr RFC 8684 3.3.1.: 1355 * the data sequence number used in the pseudo-header is 1356 * always the 64-bit value, irrespective of what length is used in the 1357 * DSS option itself. 1358 */ 1359 header.data_seq = cpu_to_be64(data_seq); 1360 header.subflow_seq = htonl(subflow_seq); 1361 header.data_len = htons(data_len); 1362 header.csum = 0; 1363 1364 csum = csum_partial(&header, sizeof(header), sum); 1365 return csum_fold(csum); 1366 } 1367 1368 static __sum16 mptcp_make_csum(const struct mptcp_ext *mpext) 1369 { 1370 return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len, 1371 ~csum_unfold(mpext->csum)); 1372 } 1373 1374 static void put_len_csum(u16 len, __sum16 csum, void *data) 1375 { 1376 __sum16 *sumptr = data + 2; 1377 __be16 *ptr = data; 1378 1379 put_unaligned_be16(len, ptr); 1380 1381 put_unaligned(csum, sumptr); 1382 } 1383 1384 void mptcp_write_options(struct tcphdr *th, __be32 *ptr, struct tcp_sock *tp, 1385 struct mptcp_out_options *opts) 1386 { 1387 const struct sock *ssk = (const struct sock *)tp; 1388 struct mptcp_subflow_context *subflow; 1389 1390 /* Which options can be used together? 1391 * 1392 * X: mutually exclusive 1393 * O: often used together 1394 * C: can be used together in some cases 1395 * P: could be used together but we prefer not to (optimisations) 1396 * 1397 * Opt: | MPC | MPJ | DSS | ADD | RM | PRIO | FAIL | FC | 1398 * ------|------|------|------|------|------|------|------|------| 1399 * MPC |------|------|------|------|------|------|------|------| 1400 * MPJ | X |------|------|------|------|------|------|------| 1401 * DSS | X | X |------|------|------|------|------|------| 1402 * ADD | X | X | P |------|------|------|------|------| 1403 * RM | C | C | C | P |------|------|------|------| 1404 * PRIO | X | C | C | C | C |------|------|------| 1405 * FAIL | X | X | C | X | X | X |------|------| 1406 * FC | X | X | X | X | X | X | X |------| 1407 * RST | X | X | X | X | X | X | O | O | 1408 * ------|------|------|------|------|------|------|------|------| 1409 * 1410 * The same applies in mptcp_established_options() function. 1411 */ 1412 if (likely(OPTION_MPTCP_DSS & opts->suboptions)) { 1413 struct mptcp_ext *mpext = &opts->ext_copy; 1414 u8 len = TCPOLEN_MPTCP_DSS_BASE; 1415 u8 flags = 0; 1416 1417 if (mpext->use_ack) { 1418 flags = MPTCP_DSS_HAS_ACK; 1419 if (mpext->ack64) { 1420 len += TCPOLEN_MPTCP_DSS_ACK64; 1421 flags |= MPTCP_DSS_ACK64; 1422 } else { 1423 len += TCPOLEN_MPTCP_DSS_ACK32; 1424 } 1425 } 1426 1427 if (mpext->use_map) { 1428 len += TCPOLEN_MPTCP_DSS_MAP64; 1429 1430 /* Use only 64-bit mapping flags for now, add 1431 * support for optional 32-bit mappings later. 1432 */ 1433 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64; 1434 if (mpext->data_fin) 1435 flags |= MPTCP_DSS_DATA_FIN; 1436 1437 if (opts->csum_reqd) 1438 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 1439 } 1440 1441 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags); 1442 1443 if (mpext->use_ack) { 1444 struct mptcp_sock *msk; 1445 u64 ack_seq; 1446 1447 /* DSS option is set only by mptcp_established_options, 1448 * the caller is __tcp_transmit_skb() and ssk is always 1449 * not NULL. 1450 */ 1451 subflow = mptcp_subflow_ctx(ssk); 1452 msk = mptcp_sk(subflow->conn); 1453 ack_seq = READ_ONCE(msk->ack_seq); 1454 if (mpext->ack64) { 1455 put_unaligned_be64(ack_seq, ptr); 1456 ptr += 2; 1457 } else { 1458 put_unaligned_be32(ack_seq, ptr); 1459 ptr += 1; 1460 } 1461 subflow->rcv_wnd_sent = mptcp_set_rwin(msk, tp, th, 1462 ack_seq); 1463 } 1464 1465 if (mpext->use_map) { 1466 put_unaligned_be64(mpext->data_seq, ptr); 1467 ptr += 2; 1468 put_unaligned_be32(mpext->subflow_seq, ptr); 1469 ptr += 1; 1470 if (opts->csum_reqd) { 1471 /* data_len == 0 is reserved for the infinite mapping, 1472 * the checksum will also be set to 0. 1473 */ 1474 put_len_csum(mpext->data_len, 1475 (mpext->data_len ? mptcp_make_csum(mpext) : 0), 1476 ptr); 1477 } else { 1478 put_unaligned_be32(mpext->data_len << 16 | 1479 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1480 } 1481 ptr += 1; 1482 } 1483 1484 /* We might need to add MP_FAIL options in rare cases */ 1485 if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) 1486 goto mp_fail; 1487 } else if (OPTIONS_MPTCP_MPC & opts->suboptions) { 1488 u8 len, flag = MPTCP_CAP_HMAC_SHA256; 1489 1490 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) { 1491 len = TCPOLEN_MPTCP_MPC_SYN; 1492 } else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) { 1493 len = TCPOLEN_MPTCP_MPC_SYNACK; 1494 } else if (opts->data_len) { 1495 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 1496 if (opts->csum_reqd) 1497 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 1498 } else { 1499 len = TCPOLEN_MPTCP_MPC_ACK; 1500 } 1501 1502 if (opts->csum_reqd) 1503 flag |= MPTCP_CAP_CHECKSUM_REQD; 1504 1505 if (!opts->allow_join_id0) 1506 flag |= MPTCP_CAP_DENY_JOIN_ID0; 1507 1508 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len, 1509 MPTCP_SUPPORTED_VERSION, 1510 flag); 1511 1512 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) & 1513 opts->suboptions)) 1514 goto mp_capable_done; 1515 1516 put_unaligned_be64(opts->sndr_key, ptr); 1517 ptr += 2; 1518 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions)) 1519 goto mp_capable_done; 1520 1521 put_unaligned_be64(opts->rcvr_key, ptr); 1522 ptr += 2; 1523 if (!opts->data_len) 1524 goto mp_capable_done; 1525 1526 if (opts->csum_reqd) { 1527 put_len_csum(opts->data_len, 1528 __mptcp_make_csum(opts->data_seq, 1529 opts->subflow_seq, 1530 opts->data_len, 1531 ~csum_unfold(opts->csum)), 1532 ptr); 1533 } else { 1534 put_unaligned_be32(opts->data_len << 16 | 1535 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1536 } 1537 ptr += 1; 1538 1539 /* MPC is additionally mutually exclusive with MP_PRIO */ 1540 goto mp_capable_done; 1541 } else if (OPTIONS_MPTCP_MPJ & opts->suboptions) { 1542 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) { 1543 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1544 TCPOLEN_MPTCP_MPJ_SYN, 1545 opts->backup, opts->join_id); 1546 put_unaligned_be32(opts->token, ptr); 1547 ptr += 1; 1548 put_unaligned_be32(opts->nonce, ptr); 1549 ptr += 1; 1550 } else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) { 1551 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1552 TCPOLEN_MPTCP_MPJ_SYNACK, 1553 opts->backup, opts->join_id); 1554 put_unaligned_be64(opts->thmac, ptr); 1555 ptr += 2; 1556 put_unaligned_be32(opts->nonce, ptr); 1557 ptr += 1; 1558 } else { 1559 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1560 TCPOLEN_MPTCP_MPJ_ACK, 0, 0); 1561 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN); 1562 ptr += 5; 1563 } 1564 } else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) { 1565 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE; 1566 u8 echo = MPTCP_ADDR_ECHO; 1567 1568 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1569 if (opts->addr.family == AF_INET6) 1570 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE; 1571 #endif 1572 1573 if (opts->addr.port) 1574 len += TCPOLEN_MPTCP_PORT_LEN; 1575 1576 if (opts->ahmac) { 1577 len += sizeof(opts->ahmac); 1578 echo = 0; 1579 } 1580 1581 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 1582 len, echo, opts->addr.id); 1583 if (opts->addr.family == AF_INET) { 1584 memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4); 1585 ptr += 1; 1586 } 1587 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1588 else if (opts->addr.family == AF_INET6) { 1589 memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16); 1590 ptr += 4; 1591 } 1592 #endif 1593 1594 if (!opts->addr.port) { 1595 if (opts->ahmac) { 1596 put_unaligned_be64(opts->ahmac, ptr); 1597 ptr += 2; 1598 } 1599 } else { 1600 u16 port = ntohs(opts->addr.port); 1601 1602 if (opts->ahmac) { 1603 u8 *bptr = (u8 *)ptr; 1604 1605 put_unaligned_be16(port, bptr); 1606 bptr += 2; 1607 put_unaligned_be64(opts->ahmac, bptr); 1608 bptr += 8; 1609 put_unaligned_be16(TCPOPT_NOP << 8 | 1610 TCPOPT_NOP, bptr); 1611 1612 ptr += 3; 1613 } else { 1614 put_unaligned_be32(port << 16 | 1615 TCPOPT_NOP << 8 | 1616 TCPOPT_NOP, ptr); 1617 ptr += 1; 1618 } 1619 } 1620 } else if (unlikely(OPTION_MPTCP_FASTCLOSE & opts->suboptions)) { 1621 /* FASTCLOSE is mutually exclusive with others except RST */ 1622 *ptr++ = mptcp_option(MPTCPOPT_MP_FASTCLOSE, 1623 TCPOLEN_MPTCP_FASTCLOSE, 1624 0, 0); 1625 put_unaligned_be64(opts->rcvr_key, ptr); 1626 ptr += 2; 1627 1628 if (OPTION_MPTCP_RST & opts->suboptions) 1629 goto mp_rst; 1630 return; 1631 } else if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) { 1632 mp_fail: 1633 /* MP_FAIL is mutually exclusive with others except RST */ 1634 subflow = mptcp_subflow_ctx(ssk); 1635 subflow->send_mp_fail = 0; 1636 1637 *ptr++ = mptcp_option(MPTCPOPT_MP_FAIL, 1638 TCPOLEN_MPTCP_FAIL, 1639 0, 0); 1640 put_unaligned_be64(opts->fail_seq, ptr); 1641 ptr += 2; 1642 1643 if (OPTION_MPTCP_RST & opts->suboptions) 1644 goto mp_rst; 1645 return; 1646 } else if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) { 1647 mp_rst: 1648 *ptr++ = mptcp_option(MPTCPOPT_RST, 1649 TCPOLEN_MPTCP_RST, 1650 opts->reset_transient, 1651 opts->reset_reason); 1652 return; 1653 } else if (unlikely(!opts->suboptions)) { 1654 /* Fallback to TCP */ 1655 mptcp_track_rwin(tp); 1656 return; 1657 } 1658 1659 if (OPTION_MPTCP_PRIO & opts->suboptions) { 1660 subflow = mptcp_subflow_ctx(ssk); 1661 subflow->send_mp_prio = 0; 1662 1663 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO, 1664 TCPOLEN_MPTCP_PRIO, 1665 opts->backup, TCPOPT_NOP); 1666 1667 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPPRIOTX); 1668 } 1669 1670 mp_capable_done: 1671 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) { 1672 u8 i = 1; 1673 1674 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR, 1675 TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr, 1676 0, opts->rm_list.ids[0]); 1677 1678 while (i < opts->rm_list.nr) { 1679 u8 id1, id2, id3, id4; 1680 1681 id1 = opts->rm_list.ids[i]; 1682 id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP; 1683 id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP; 1684 id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP; 1685 put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr); 1686 ptr += 1; 1687 i += 4; 1688 } 1689 } 1690 } 1691 1692 __be32 mptcp_get_reset_option(const struct sk_buff *skb) 1693 { 1694 const struct mptcp_ext *ext = mptcp_get_ext(skb); 1695 u8 flags, reason; 1696 1697 if (ext) { 1698 flags = ext->reset_transient; 1699 reason = ext->reset_reason; 1700 1701 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST, 1702 flags, reason); 1703 } 1704 1705 return htonl(0u); 1706 } 1707 EXPORT_SYMBOL_GPL(mptcp_get_reset_option); 1708