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 struct mptcp_addr_info addr; 655 bool echo; 656 int len; 657 658 /* add addr will strip the existing options, be sure to avoid breaking 659 * MPC/MPJ handshakes 660 */ 661 if (!mptcp_pm_should_add_signal(msk) || 662 (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) || 663 !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, &addr, 664 &echo, &drop_other_suboptions)) 665 return false; 666 667 /* 668 * Later on, mptcp_write_options() will enforce mutually exclusion with 669 * DSS, bail out if such option is set and we can't drop it. 670 */ 671 if (drop_other_suboptions) 672 remaining += opt_size; 673 else if (opts->suboptions & OPTION_MPTCP_DSS) 674 return false; 675 676 len = mptcp_add_addr_len(addr.family, echo, !!addr.port); 677 if (remaining < len) 678 return false; 679 680 *size = len; 681 if (drop_other_suboptions) { 682 pr_debug("drop other suboptions\n"); 683 opts->suboptions = 0; 684 685 /* note that e.g. DSS could have written into the memory 686 * aliased by ahmac, we must reset the field here 687 * to avoid appending the hmac even for ADD_ADDR echo 688 * options 689 */ 690 opts->ahmac = 0; 691 *size -= opt_size; 692 } 693 opts->addr = addr; 694 opts->suboptions |= OPTION_MPTCP_ADD_ADDR; 695 if (!echo) { 696 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDRTX); 697 opts->ahmac = add_addr_generate_hmac(READ_ONCE(msk->local_key), 698 READ_ONCE(msk->remote_key), 699 &opts->addr); 700 } else { 701 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADDTX); 702 } 703 pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d\n", 704 opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port)); 705 706 return true; 707 } 708 709 static bool mptcp_established_options_rm_addr(struct sock *sk, 710 unsigned int *size, 711 unsigned int remaining, 712 struct mptcp_out_options *opts) 713 { 714 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 715 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 716 struct mptcp_rm_list rm_list; 717 int i, len; 718 719 if (!mptcp_pm_should_rm_signal(msk) || 720 !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list))) 721 return false; 722 723 len = mptcp_rm_addr_len(&rm_list); 724 if (len < 0) 725 return false; 726 if (remaining < len) 727 return false; 728 729 *size = len; 730 opts->suboptions |= OPTION_MPTCP_RM_ADDR; 731 opts->rm_list = rm_list; 732 733 for (i = 0; i < opts->rm_list.nr; i++) 734 pr_debug("rm_list_ids[%d]=%d\n", i, opts->rm_list.ids[i]); 735 MPTCP_ADD_STATS(sock_net(sk), MPTCP_MIB_RMADDRTX, opts->rm_list.nr); 736 return true; 737 } 738 739 static bool mptcp_established_options_mp_prio(struct sock *sk, 740 unsigned int *size, 741 unsigned int remaining, 742 struct mptcp_out_options *opts) 743 { 744 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 745 746 /* can't send MP_PRIO with MPC, as they share the same option space: 747 * 'backup'. Also it makes no sense at all 748 */ 749 if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC)) 750 return false; 751 752 /* account for the trailing 'nop' option */ 753 if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN) 754 return false; 755 756 *size = TCPOLEN_MPTCP_PRIO_ALIGN; 757 opts->suboptions |= OPTION_MPTCP_PRIO; 758 opts->backup = subflow->request_bkup; 759 760 pr_debug("prio=%d\n", opts->backup); 761 762 return true; 763 } 764 765 static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb, 766 unsigned int *size, 767 unsigned int remaining, 768 struct mptcp_out_options *opts) 769 { 770 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 771 772 if (remaining < TCPOLEN_MPTCP_RST) 773 return false; 774 775 *size = TCPOLEN_MPTCP_RST; 776 opts->suboptions |= OPTION_MPTCP_RST; 777 opts->reset_transient = subflow->reset_transient; 778 opts->reset_reason = subflow->reset_reason; 779 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTTX); 780 781 return true; 782 } 783 784 static bool mptcp_established_options_fastclose(struct sock *sk, 785 unsigned int *size, 786 unsigned int remaining, 787 struct mptcp_out_options *opts) 788 { 789 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 790 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 791 792 if (likely(!subflow->send_fastclose)) 793 return false; 794 795 if (remaining < TCPOLEN_MPTCP_FASTCLOSE) 796 return false; 797 798 *size = TCPOLEN_MPTCP_FASTCLOSE; 799 opts->suboptions |= OPTION_MPTCP_FASTCLOSE; 800 opts->rcvr_key = READ_ONCE(msk->remote_key); 801 802 pr_debug("FASTCLOSE key=%llu\n", opts->rcvr_key); 803 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSETX); 804 return true; 805 } 806 807 static bool mptcp_established_options_mp_fail(struct sock *sk, 808 unsigned int *size, 809 unsigned int remaining, 810 struct mptcp_out_options *opts) 811 { 812 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 813 814 if (likely(!subflow->send_mp_fail)) 815 return false; 816 817 if (remaining < TCPOLEN_MPTCP_FAIL) 818 return false; 819 820 *size = TCPOLEN_MPTCP_FAIL; 821 opts->suboptions |= OPTION_MPTCP_FAIL; 822 opts->fail_seq = subflow->map_seq; 823 824 pr_debug("MP_FAIL fail_seq=%llu\n", opts->fail_seq); 825 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILTX); 826 827 return true; 828 } 829 830 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb, 831 unsigned int *size, unsigned int remaining, 832 struct mptcp_out_options *opts) 833 { 834 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 835 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 836 unsigned int opt_size = 0; 837 bool snd_data_fin; 838 bool ret = false; 839 840 opts->suboptions = 0; 841 842 if (unlikely(__mptcp_check_fallback(msk) && !mptcp_check_infinite_map(skb))) 843 return false; 844 845 if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) { 846 if (mptcp_established_options_fastclose(sk, &opt_size, remaining, opts) || 847 mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) { 848 *size += opt_size; 849 remaining -= opt_size; 850 } 851 /* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */ 852 if (mptcp_established_options_rst(sk, skb, &opt_size, remaining, opts)) { 853 *size += opt_size; 854 remaining -= opt_size; 855 } 856 return true; 857 } 858 859 snd_data_fin = mptcp_data_fin_enabled(msk); 860 if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, opts)) 861 ret = true; 862 else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, opts)) { 863 unsigned int mp_fail_size; 864 865 ret = true; 866 if (mptcp_established_options_mp_fail(sk, &mp_fail_size, 867 remaining - opt_size, opts)) { 868 *size += opt_size + mp_fail_size; 869 remaining -= opt_size - mp_fail_size; 870 return true; 871 } 872 } 873 874 /* we reserved enough space for the above options, and exceeding the 875 * TCP option space would be fatal 876 */ 877 if (WARN_ON_ONCE(opt_size > remaining)) 878 return false; 879 880 *size += opt_size; 881 remaining -= opt_size; 882 if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) { 883 *size += opt_size; 884 remaining -= opt_size; 885 ret = true; 886 } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) { 887 *size += opt_size; 888 remaining -= opt_size; 889 ret = true; 890 } 891 892 if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) { 893 *size += opt_size; 894 remaining -= opt_size; 895 ret = true; 896 } 897 898 return ret; 899 } 900 901 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size, 902 struct mptcp_out_options *opts) 903 { 904 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 905 906 if (subflow_req->mp_capable) { 907 opts->suboptions = OPTION_MPTCP_MPC_SYNACK; 908 opts->sndr_key = subflow_req->local_key; 909 opts->csum_reqd = subflow_req->csum_reqd; 910 opts->allow_join_id0 = subflow_req->allow_join_id0; 911 *size = TCPOLEN_MPTCP_MPC_SYNACK; 912 pr_debug("subflow_req=%p, local_key=%llu\n", 913 subflow_req, subflow_req->local_key); 914 return true; 915 } else if (subflow_req->mp_join) { 916 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK; 917 opts->backup = subflow_req->request_bkup; 918 opts->join_id = subflow_req->local_id; 919 opts->thmac = subflow_req->thmac; 920 opts->nonce = subflow_req->local_nonce; 921 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u\n", 922 subflow_req, opts->backup, opts->join_id, 923 opts->thmac, opts->nonce); 924 *size = TCPOLEN_MPTCP_MPJ_SYNACK; 925 return true; 926 } 927 return false; 928 } 929 930 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk, 931 struct mptcp_subflow_context *subflow, 932 struct sk_buff *skb, 933 struct mptcp_options_received *mp_opt) 934 { 935 /* here we can process OoO, in-window pkts, only in-sequence 4th ack 936 * will make the subflow fully established 937 */ 938 if (likely(READ_ONCE(subflow->fully_established))) { 939 /* on passive sockets, check for 3rd ack retransmission 940 * note that msk is always set by subflow_syn_recv_sock() 941 * for mp_join subflows 942 */ 943 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 && 944 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq && 945 subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) && 946 !subflow->request_join) 947 tcp_send_ack(ssk); 948 goto check_notify; 949 } 950 951 /* we must process OoO packets before the first subflow is fully 952 * established. OoO packets are instead a protocol violation 953 * for MP_JOIN subflows as the peer must not send any data 954 * before receiving the forth ack - cfr. RFC 8684 section 3.2. 955 */ 956 if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) { 957 if (subflow->mp_join) 958 goto reset; 959 if (subflow->is_mptfo && mp_opt->suboptions & OPTION_MPTCP_MPC_ACK) 960 goto set_fully_established; 961 return subflow->mp_capable; 962 } 963 964 if (subflow->remote_key_valid && 965 (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) || 966 ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) && 967 (!mp_opt->echo || subflow->mp_join)))) { 968 /* subflows are fully established as soon as we get any 969 * additional ack, including ADD_ADDR. 970 */ 971 goto set_fully_established; 972 } 973 974 /* If the first established packet does not contain MP_CAPABLE + data 975 * then fallback to TCP. Fallback scenarios requires a reset for 976 * MP_JOIN subflows. 977 */ 978 if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) { 979 if (subflow->mp_join) 980 goto reset; 981 subflow->mp_capable = 0; 982 pr_fallback(msk); 983 mptcp_do_fallback(ssk); 984 return false; 985 } 986 987 if (mp_opt->deny_join_id0) 988 WRITE_ONCE(msk->pm.remote_deny_join_id0, true); 989 990 if (unlikely(!READ_ONCE(msk->pm.server_side))) 991 pr_warn_once("bogus mpc option on established client sk"); 992 993 set_fully_established: 994 mptcp_data_lock((struct sock *)msk); 995 __mptcp_subflow_fully_established(msk, subflow, mp_opt); 996 mptcp_data_unlock((struct sock *)msk); 997 998 check_notify: 999 /* if the subflow is not already linked into the conn_list, we can't 1000 * notify the PM: this subflow is still on the listener queue 1001 * and the PM possibly acquiring the subflow lock could race with 1002 * the listener close 1003 */ 1004 if (likely(subflow->pm_notified) || list_empty(&subflow->node)) 1005 return true; 1006 1007 subflow->pm_notified = 1; 1008 if (subflow->mp_join) { 1009 clear_3rdack_retransmission(ssk); 1010 mptcp_pm_subflow_established(msk); 1011 } else { 1012 mptcp_pm_fully_established(msk, ssk); 1013 } 1014 return true; 1015 1016 reset: 1017 mptcp_subflow_reset(ssk); 1018 return false; 1019 } 1020 1021 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq) 1022 { 1023 u32 old_seq32, cur_seq32; 1024 1025 old_seq32 = (u32)old_seq; 1026 cur_seq32 = (u32)cur_seq; 1027 cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32; 1028 if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32))) 1029 return cur_seq + (1LL << 32); 1030 1031 /* reverse wrap could happen, too */ 1032 if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32))) 1033 return cur_seq - (1LL << 32); 1034 return cur_seq; 1035 } 1036 1037 static void __mptcp_snd_una_update(struct mptcp_sock *msk, u64 new_snd_una) 1038 { 1039 msk->bytes_acked += new_snd_una - msk->snd_una; 1040 WRITE_ONCE(msk->snd_una, new_snd_una); 1041 } 1042 1043 static void ack_update_msk(struct mptcp_sock *msk, 1044 struct sock *ssk, 1045 struct mptcp_options_received *mp_opt) 1046 { 1047 u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt); 1048 struct sock *sk = (struct sock *)msk; 1049 u64 old_snd_una; 1050 1051 mptcp_data_lock(sk); 1052 1053 /* avoid ack expansion on update conflict, to reduce the risk of 1054 * wrongly expanding to a future ack sequence number, which is way 1055 * more dangerous than missing an ack 1056 */ 1057 old_snd_una = msk->snd_una; 1058 new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64); 1059 1060 /* ACK for data not even sent yet? Ignore.*/ 1061 if (unlikely(after64(new_snd_una, snd_nxt))) 1062 new_snd_una = old_snd_una; 1063 1064 new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd; 1065 1066 if (after64(new_wnd_end, msk->wnd_end)) 1067 WRITE_ONCE(msk->wnd_end, new_wnd_end); 1068 1069 /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */ 1070 if (after64(msk->wnd_end, snd_nxt)) 1071 __mptcp_check_push(sk, ssk); 1072 1073 if (after64(new_snd_una, old_snd_una)) { 1074 __mptcp_snd_una_update(msk, new_snd_una); 1075 __mptcp_data_acked(sk); 1076 } 1077 msk->last_ack_recv = tcp_jiffies32; 1078 mptcp_data_unlock(sk); 1079 1080 trace_ack_update_msk(mp_opt->data_ack, 1081 old_snd_una, new_snd_una, 1082 new_wnd_end, READ_ONCE(msk->wnd_end)); 1083 } 1084 1085 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit) 1086 { 1087 /* Skip if DATA_FIN was already received. 1088 * If updating simultaneously with the recvmsg loop, values 1089 * should match. If they mismatch, the peer is misbehaving and 1090 * we will prefer the most recent information. 1091 */ 1092 if (READ_ONCE(msk->rcv_data_fin)) 1093 return false; 1094 1095 WRITE_ONCE(msk->rcv_data_fin_seq, 1096 mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit)); 1097 WRITE_ONCE(msk->rcv_data_fin, 1); 1098 1099 return true; 1100 } 1101 1102 static bool add_addr_hmac_valid(struct mptcp_sock *msk, 1103 struct mptcp_options_received *mp_opt) 1104 { 1105 u64 hmac = 0; 1106 1107 if (mp_opt->echo) 1108 return true; 1109 1110 hmac = add_addr_generate_hmac(READ_ONCE(msk->remote_key), 1111 READ_ONCE(msk->local_key), 1112 &mp_opt->addr); 1113 1114 pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n", 1115 msk, hmac, mp_opt->ahmac); 1116 1117 return hmac == mp_opt->ahmac; 1118 } 1119 1120 /* Return false if a subflow has been reset, else return true */ 1121 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb) 1122 { 1123 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 1124 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 1125 struct mptcp_options_received mp_opt; 1126 struct mptcp_ext *mpext; 1127 1128 if (__mptcp_check_fallback(msk)) { 1129 /* Keep it simple and unconditionally trigger send data cleanup and 1130 * pending queue spooling. We will need to acquire the data lock 1131 * for more accurate checks, and once the lock is acquired, such 1132 * helpers are cheap. 1133 */ 1134 mptcp_data_lock(subflow->conn); 1135 if (sk_stream_memory_free(sk)) 1136 __mptcp_check_push(subflow->conn, sk); 1137 1138 /* on fallback we just need to ignore the msk-level snd_una, as 1139 * this is really plain TCP 1140 */ 1141 __mptcp_snd_una_update(msk, READ_ONCE(msk->snd_nxt)); 1142 1143 __mptcp_data_acked(subflow->conn); 1144 mptcp_data_unlock(subflow->conn); 1145 return true; 1146 } 1147 1148 mptcp_get_options(skb, &mp_opt); 1149 1150 /* The subflow can be in close state only if check_fully_established() 1151 * just sent a reset. If so, tell the caller to ignore the current packet. 1152 */ 1153 if (!check_fully_established(msk, sk, subflow, skb, &mp_opt)) 1154 return sk->sk_state != TCP_CLOSE; 1155 1156 if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) { 1157 if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) && 1158 READ_ONCE(msk->local_key) == mp_opt.rcvr_key) { 1159 WRITE_ONCE(msk->rcv_fastclose, true); 1160 mptcp_schedule_work((struct sock *)msk); 1161 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSERX); 1162 } 1163 1164 if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) && 1165 add_addr_hmac_valid(msk, &mp_opt)) { 1166 if (!mp_opt.echo) { 1167 mptcp_pm_add_addr_received(sk, &mp_opt.addr); 1168 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR); 1169 } else { 1170 mptcp_pm_add_addr_echoed(msk, &mp_opt.addr); 1171 mptcp_pm_del_add_timer(msk, &mp_opt.addr, true); 1172 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD); 1173 } 1174 1175 if (mp_opt.addr.port) 1176 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD); 1177 } 1178 1179 if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR) 1180 mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list); 1181 1182 if (mp_opt.suboptions & OPTION_MPTCP_PRIO) { 1183 mptcp_pm_mp_prio_received(sk, mp_opt.backup); 1184 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX); 1185 } 1186 1187 if (mp_opt.suboptions & OPTION_MPTCP_FAIL) { 1188 mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq); 1189 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX); 1190 } 1191 1192 if (mp_opt.suboptions & OPTION_MPTCP_RST) { 1193 subflow->reset_seen = 1; 1194 subflow->reset_reason = mp_opt.reset_reason; 1195 subflow->reset_transient = mp_opt.reset_transient; 1196 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTRX); 1197 } 1198 1199 if (!(mp_opt.suboptions & OPTION_MPTCP_DSS)) 1200 return true; 1201 } 1202 1203 /* we can't wait for recvmsg() to update the ack_seq, otherwise 1204 * monodirectional flows will stuck 1205 */ 1206 if (mp_opt.use_ack) 1207 ack_update_msk(msk, sk, &mp_opt); 1208 1209 /* Zero-data-length packets are dropped by the caller and not 1210 * propagated to the MPTCP layer, so the skb extension does not 1211 * need to be allocated or populated. DATA_FIN information, if 1212 * present, needs to be updated here before the skb is freed. 1213 */ 1214 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) { 1215 if (mp_opt.data_fin && mp_opt.data_len == 1 && 1216 mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64)) 1217 mptcp_schedule_work((struct sock *)msk); 1218 1219 return true; 1220 } 1221 1222 mpext = skb_ext_add(skb, SKB_EXT_MPTCP); 1223 if (!mpext) 1224 return true; 1225 1226 memset(mpext, 0, sizeof(*mpext)); 1227 1228 if (likely(mp_opt.use_map)) { 1229 if (mp_opt.mpc_map) { 1230 /* this is an MP_CAPABLE carrying MPTCP data 1231 * we know this map the first chunk of data 1232 */ 1233 mptcp_crypto_key_sha(subflow->remote_key, NULL, 1234 &mpext->data_seq); 1235 mpext->data_seq++; 1236 mpext->subflow_seq = 1; 1237 mpext->dsn64 = 1; 1238 mpext->mpc_map = 1; 1239 mpext->data_fin = 0; 1240 } else { 1241 mpext->data_seq = mp_opt.data_seq; 1242 mpext->subflow_seq = mp_opt.subflow_seq; 1243 mpext->dsn64 = mp_opt.dsn64; 1244 mpext->data_fin = mp_opt.data_fin; 1245 } 1246 mpext->data_len = mp_opt.data_len; 1247 mpext->use_map = 1; 1248 mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD); 1249 1250 if (mpext->csum_reqd) 1251 mpext->csum = mp_opt.csum; 1252 } 1253 1254 return true; 1255 } 1256 1257 static void mptcp_set_rwin(struct tcp_sock *tp, struct tcphdr *th) 1258 { 1259 const struct sock *ssk = (const struct sock *)tp; 1260 struct mptcp_subflow_context *subflow; 1261 u64 ack_seq, rcv_wnd_old, rcv_wnd_new; 1262 struct mptcp_sock *msk; 1263 u32 new_win; 1264 u64 win; 1265 1266 subflow = mptcp_subflow_ctx(ssk); 1267 msk = mptcp_sk(subflow->conn); 1268 1269 ack_seq = READ_ONCE(msk->ack_seq); 1270 rcv_wnd_new = ack_seq + tp->rcv_wnd; 1271 1272 rcv_wnd_old = atomic64_read(&msk->rcv_wnd_sent); 1273 if (after64(rcv_wnd_new, rcv_wnd_old)) { 1274 u64 rcv_wnd; 1275 1276 for (;;) { 1277 rcv_wnd = atomic64_cmpxchg(&msk->rcv_wnd_sent, rcv_wnd_old, rcv_wnd_new); 1278 1279 if (rcv_wnd == rcv_wnd_old) 1280 break; 1281 1282 rcv_wnd_old = rcv_wnd; 1283 if (before64(rcv_wnd_new, rcv_wnd_old)) { 1284 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICTUPDATE); 1285 goto raise_win; 1286 } 1287 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICT); 1288 } 1289 goto update_wspace; 1290 } 1291 1292 if (rcv_wnd_new != rcv_wnd_old) { 1293 raise_win: 1294 win = rcv_wnd_old - ack_seq; 1295 tp->rcv_wnd = min_t(u64, win, U32_MAX); 1296 new_win = tp->rcv_wnd; 1297 1298 /* Make sure we do not exceed the maximum possible 1299 * scaled window. 1300 */ 1301 if (unlikely(th->syn)) 1302 new_win = min(new_win, 65535U) << tp->rx_opt.rcv_wscale; 1303 if (!tp->rx_opt.rcv_wscale && 1304 READ_ONCE(sock_net(ssk)->ipv4.sysctl_tcp_workaround_signed_windows)) 1305 new_win = min(new_win, MAX_TCP_WINDOW); 1306 else 1307 new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale)); 1308 1309 /* RFC1323 scaling applied */ 1310 new_win >>= tp->rx_opt.rcv_wscale; 1311 th->window = htons(new_win); 1312 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDSHARED); 1313 } 1314 1315 update_wspace: 1316 WRITE_ONCE(msk->old_wspace, tp->rcv_wnd); 1317 } 1318 1319 __sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum) 1320 { 1321 struct csum_pseudo_header header; 1322 __wsum csum; 1323 1324 /* cfr RFC 8684 3.3.1.: 1325 * the data sequence number used in the pseudo-header is 1326 * always the 64-bit value, irrespective of what length is used in the 1327 * DSS option itself. 1328 */ 1329 header.data_seq = cpu_to_be64(data_seq); 1330 header.subflow_seq = htonl(subflow_seq); 1331 header.data_len = htons(data_len); 1332 header.csum = 0; 1333 1334 csum = csum_partial(&header, sizeof(header), sum); 1335 return csum_fold(csum); 1336 } 1337 1338 static __sum16 mptcp_make_csum(const struct mptcp_ext *mpext) 1339 { 1340 return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len, 1341 ~csum_unfold(mpext->csum)); 1342 } 1343 1344 static void put_len_csum(u16 len, __sum16 csum, void *data) 1345 { 1346 __sum16 *sumptr = data + 2; 1347 __be16 *ptr = data; 1348 1349 put_unaligned_be16(len, ptr); 1350 1351 put_unaligned(csum, sumptr); 1352 } 1353 1354 void mptcp_write_options(struct tcphdr *th, __be32 *ptr, struct tcp_sock *tp, 1355 struct mptcp_out_options *opts) 1356 { 1357 const struct sock *ssk = (const struct sock *)tp; 1358 struct mptcp_subflow_context *subflow; 1359 1360 /* Which options can be used together? 1361 * 1362 * X: mutually exclusive 1363 * O: often used together 1364 * C: can be used together in some cases 1365 * P: could be used together but we prefer not to (optimisations) 1366 * 1367 * Opt: | MPC | MPJ | DSS | ADD | RM | PRIO | FAIL | FC | 1368 * ------|------|------|------|------|------|------|------|------| 1369 * MPC |------|------|------|------|------|------|------|------| 1370 * MPJ | X |------|------|------|------|------|------|------| 1371 * DSS | X | X |------|------|------|------|------|------| 1372 * ADD | X | X | P |------|------|------|------|------| 1373 * RM | C | C | C | P |------|------|------|------| 1374 * PRIO | X | C | C | C | C |------|------|------| 1375 * FAIL | X | X | C | X | X | X |------|------| 1376 * FC | X | X | X | X | X | X | X |------| 1377 * RST | X | X | X | X | X | X | O | O | 1378 * ------|------|------|------|------|------|------|------|------| 1379 * 1380 * The same applies in mptcp_established_options() function. 1381 */ 1382 if (likely(OPTION_MPTCP_DSS & opts->suboptions)) { 1383 struct mptcp_ext *mpext = &opts->ext_copy; 1384 u8 len = TCPOLEN_MPTCP_DSS_BASE; 1385 u8 flags = 0; 1386 1387 if (mpext->use_ack) { 1388 flags = MPTCP_DSS_HAS_ACK; 1389 if (mpext->ack64) { 1390 len += TCPOLEN_MPTCP_DSS_ACK64; 1391 flags |= MPTCP_DSS_ACK64; 1392 } else { 1393 len += TCPOLEN_MPTCP_DSS_ACK32; 1394 } 1395 } 1396 1397 if (mpext->use_map) { 1398 len += TCPOLEN_MPTCP_DSS_MAP64; 1399 1400 /* Use only 64-bit mapping flags for now, add 1401 * support for optional 32-bit mappings later. 1402 */ 1403 flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64; 1404 if (mpext->data_fin) 1405 flags |= MPTCP_DSS_DATA_FIN; 1406 1407 if (opts->csum_reqd) 1408 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 1409 } 1410 1411 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags); 1412 1413 if (mpext->use_ack) { 1414 if (mpext->ack64) { 1415 put_unaligned_be64(mpext->data_ack, ptr); 1416 ptr += 2; 1417 } else { 1418 put_unaligned_be32(mpext->data_ack32, ptr); 1419 ptr += 1; 1420 } 1421 } 1422 1423 if (mpext->use_map) { 1424 put_unaligned_be64(mpext->data_seq, ptr); 1425 ptr += 2; 1426 put_unaligned_be32(mpext->subflow_seq, ptr); 1427 ptr += 1; 1428 if (opts->csum_reqd) { 1429 /* data_len == 0 is reserved for the infinite mapping, 1430 * the checksum will also be set to 0. 1431 */ 1432 put_len_csum(mpext->data_len, 1433 (mpext->data_len ? mptcp_make_csum(mpext) : 0), 1434 ptr); 1435 } else { 1436 put_unaligned_be32(mpext->data_len << 16 | 1437 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1438 } 1439 ptr += 1; 1440 } 1441 1442 /* We might need to add MP_FAIL options in rare cases */ 1443 if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) 1444 goto mp_fail; 1445 } else if (OPTIONS_MPTCP_MPC & opts->suboptions) { 1446 u8 len, flag = MPTCP_CAP_HMAC_SHA256; 1447 1448 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) { 1449 len = TCPOLEN_MPTCP_MPC_SYN; 1450 } else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) { 1451 len = TCPOLEN_MPTCP_MPC_SYNACK; 1452 } else if (opts->data_len) { 1453 len = TCPOLEN_MPTCP_MPC_ACK_DATA; 1454 if (opts->csum_reqd) 1455 len += TCPOLEN_MPTCP_DSS_CHECKSUM; 1456 } else { 1457 len = TCPOLEN_MPTCP_MPC_ACK; 1458 } 1459 1460 if (opts->csum_reqd) 1461 flag |= MPTCP_CAP_CHECKSUM_REQD; 1462 1463 if (!opts->allow_join_id0) 1464 flag |= MPTCP_CAP_DENY_JOIN_ID0; 1465 1466 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len, 1467 MPTCP_SUPPORTED_VERSION, 1468 flag); 1469 1470 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) & 1471 opts->suboptions)) 1472 goto mp_capable_done; 1473 1474 put_unaligned_be64(opts->sndr_key, ptr); 1475 ptr += 2; 1476 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions)) 1477 goto mp_capable_done; 1478 1479 put_unaligned_be64(opts->rcvr_key, ptr); 1480 ptr += 2; 1481 if (!opts->data_len) 1482 goto mp_capable_done; 1483 1484 if (opts->csum_reqd) { 1485 put_len_csum(opts->data_len, 1486 __mptcp_make_csum(opts->data_seq, 1487 opts->subflow_seq, 1488 opts->data_len, 1489 ~csum_unfold(opts->csum)), 1490 ptr); 1491 } else { 1492 put_unaligned_be32(opts->data_len << 16 | 1493 TCPOPT_NOP << 8 | TCPOPT_NOP, ptr); 1494 } 1495 ptr += 1; 1496 1497 /* MPC is additionally mutually exclusive with MP_PRIO */ 1498 goto mp_capable_done; 1499 } else if (OPTIONS_MPTCP_MPJ & opts->suboptions) { 1500 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) { 1501 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1502 TCPOLEN_MPTCP_MPJ_SYN, 1503 opts->backup, opts->join_id); 1504 put_unaligned_be32(opts->token, ptr); 1505 ptr += 1; 1506 put_unaligned_be32(opts->nonce, ptr); 1507 ptr += 1; 1508 } else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) { 1509 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1510 TCPOLEN_MPTCP_MPJ_SYNACK, 1511 opts->backup, opts->join_id); 1512 put_unaligned_be64(opts->thmac, ptr); 1513 ptr += 2; 1514 put_unaligned_be32(opts->nonce, ptr); 1515 ptr += 1; 1516 } else { 1517 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN, 1518 TCPOLEN_MPTCP_MPJ_ACK, 0, 0); 1519 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN); 1520 ptr += 5; 1521 } 1522 } else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) { 1523 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE; 1524 u8 echo = MPTCP_ADDR_ECHO; 1525 1526 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1527 if (opts->addr.family == AF_INET6) 1528 len = TCPOLEN_MPTCP_ADD_ADDR6_BASE; 1529 #endif 1530 1531 if (opts->addr.port) 1532 len += TCPOLEN_MPTCP_PORT_LEN; 1533 1534 if (opts->ahmac) { 1535 len += sizeof(opts->ahmac); 1536 echo = 0; 1537 } 1538 1539 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR, 1540 len, echo, opts->addr.id); 1541 if (opts->addr.family == AF_INET) { 1542 memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4); 1543 ptr += 1; 1544 } 1545 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1546 else if (opts->addr.family == AF_INET6) { 1547 memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16); 1548 ptr += 4; 1549 } 1550 #endif 1551 1552 if (!opts->addr.port) { 1553 if (opts->ahmac) { 1554 put_unaligned_be64(opts->ahmac, ptr); 1555 ptr += 2; 1556 } 1557 } else { 1558 u16 port = ntohs(opts->addr.port); 1559 1560 if (opts->ahmac) { 1561 u8 *bptr = (u8 *)ptr; 1562 1563 put_unaligned_be16(port, bptr); 1564 bptr += 2; 1565 put_unaligned_be64(opts->ahmac, bptr); 1566 bptr += 8; 1567 put_unaligned_be16(TCPOPT_NOP << 8 | 1568 TCPOPT_NOP, bptr); 1569 1570 ptr += 3; 1571 } else { 1572 put_unaligned_be32(port << 16 | 1573 TCPOPT_NOP << 8 | 1574 TCPOPT_NOP, ptr); 1575 ptr += 1; 1576 } 1577 } 1578 } else if (unlikely(OPTION_MPTCP_FASTCLOSE & opts->suboptions)) { 1579 /* FASTCLOSE is mutually exclusive with others except RST */ 1580 *ptr++ = mptcp_option(MPTCPOPT_MP_FASTCLOSE, 1581 TCPOLEN_MPTCP_FASTCLOSE, 1582 0, 0); 1583 put_unaligned_be64(opts->rcvr_key, ptr); 1584 ptr += 2; 1585 1586 if (OPTION_MPTCP_RST & opts->suboptions) 1587 goto mp_rst; 1588 return; 1589 } else if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) { 1590 mp_fail: 1591 /* MP_FAIL is mutually exclusive with others except RST */ 1592 subflow = mptcp_subflow_ctx(ssk); 1593 subflow->send_mp_fail = 0; 1594 1595 *ptr++ = mptcp_option(MPTCPOPT_MP_FAIL, 1596 TCPOLEN_MPTCP_FAIL, 1597 0, 0); 1598 put_unaligned_be64(opts->fail_seq, ptr); 1599 ptr += 2; 1600 1601 if (OPTION_MPTCP_RST & opts->suboptions) 1602 goto mp_rst; 1603 return; 1604 } else if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) { 1605 mp_rst: 1606 *ptr++ = mptcp_option(MPTCPOPT_RST, 1607 TCPOLEN_MPTCP_RST, 1608 opts->reset_transient, 1609 opts->reset_reason); 1610 return; 1611 } 1612 1613 if (OPTION_MPTCP_PRIO & opts->suboptions) { 1614 subflow = mptcp_subflow_ctx(ssk); 1615 subflow->send_mp_prio = 0; 1616 1617 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO, 1618 TCPOLEN_MPTCP_PRIO, 1619 opts->backup, TCPOPT_NOP); 1620 1621 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPPRIOTX); 1622 } 1623 1624 mp_capable_done: 1625 if (OPTION_MPTCP_RM_ADDR & opts->suboptions) { 1626 u8 i = 1; 1627 1628 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR, 1629 TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr, 1630 0, opts->rm_list.ids[0]); 1631 1632 while (i < opts->rm_list.nr) { 1633 u8 id1, id2, id3, id4; 1634 1635 id1 = opts->rm_list.ids[i]; 1636 id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP; 1637 id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP; 1638 id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP; 1639 put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr); 1640 ptr += 1; 1641 i += 4; 1642 } 1643 } 1644 1645 if (tp) 1646 mptcp_set_rwin(tp, th); 1647 } 1648 1649 __be32 mptcp_get_reset_option(const struct sk_buff *skb) 1650 { 1651 const struct mptcp_ext *ext = mptcp_get_ext(skb); 1652 u8 flags, reason; 1653 1654 if (ext) { 1655 flags = ext->reset_transient; 1656 reason = ext->reset_reason; 1657 1658 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST, 1659 flags, reason); 1660 } 1661 1662 return htonl(0u); 1663 } 1664 EXPORT_SYMBOL_GPL(mptcp_get_reset_option); 1665