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 <linux/module.h> 11 #include <linux/netdevice.h> 12 #include <crypto/sha2.h> 13 #include <crypto/utils.h> 14 #include <net/sock.h> 15 #include <net/inet_common.h> 16 #include <net/inet_hashtables.h> 17 #include <net/protocol.h> 18 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 19 #include <net/ip6_route.h> 20 #include <net/transp_v6.h> 21 #endif 22 #include <net/mptcp.h> 23 24 #include "protocol.h" 25 #include "mib.h" 26 27 #include <trace/events/mptcp.h> 28 #include <trace/events/sock.h> 29 30 static void mptcp_subflow_ops_undo_override(struct sock *ssk); 31 32 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req, 33 enum linux_mptcp_mib_field field) 34 { 35 MPTCP_INC_STATS(sock_net(req_to_sk(req)), field); 36 } 37 38 static void subflow_req_destructor(struct request_sock *req) 39 { 40 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 41 42 pr_debug("subflow_req=%p\n", subflow_req); 43 44 if (subflow_req->msk) 45 sock_put((struct sock *)subflow_req->msk); 46 47 mptcp_token_destroy_request(req); 48 } 49 50 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2, 51 void *hmac) 52 { 53 u8 msg[8]; 54 55 put_unaligned_be32(nonce1, &msg[0]); 56 put_unaligned_be32(nonce2, &msg[4]); 57 58 mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac); 59 } 60 61 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk) 62 { 63 return mptcp_is_fully_established((void *)msk) && 64 ((mptcp_pm_is_userspace(msk) && 65 mptcp_userspace_pm_active(msk)) || 66 READ_ONCE(msk->pm.accept_subflow)); 67 } 68 69 /* validate received token and create truncated hmac and nonce for SYN-ACK */ 70 static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req) 71 { 72 struct mptcp_sock *msk = subflow_req->msk; 73 u8 hmac[SHA256_DIGEST_SIZE]; 74 75 get_random_bytes(&subflow_req->local_nonce, sizeof(u32)); 76 77 subflow_generate_hmac(READ_ONCE(msk->local_key), 78 READ_ONCE(msk->remote_key), 79 subflow_req->local_nonce, 80 subflow_req->remote_nonce, hmac); 81 82 subflow_req->thmac = get_unaligned_be64(hmac); 83 } 84 85 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req) 86 { 87 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 88 struct mptcp_sock *msk; 89 int local_id; 90 91 msk = mptcp_token_get_sock(sock_net(req_to_sk(req)), subflow_req->token); 92 if (!msk) { 93 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN); 94 return NULL; 95 } 96 97 local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req); 98 if (local_id < 0) { 99 sock_put((struct sock *)msk); 100 return NULL; 101 } 102 subflow_req->local_id = local_id; 103 subflow_req->request_bkup = mptcp_pm_is_backup(msk, (struct sock_common *)req); 104 105 return msk; 106 } 107 108 static void subflow_init_req(struct request_sock *req, const struct sock *sk_listener) 109 { 110 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 111 112 subflow_req->mp_capable = 0; 113 subflow_req->mp_join = 0; 114 subflow_req->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk_listener)); 115 subflow_req->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk_listener)); 116 subflow_req->msk = NULL; 117 mptcp_token_init_request(req); 118 } 119 120 static bool subflow_use_different_sport(struct mptcp_sock *msk, const struct sock *sk) 121 { 122 return inet_sk(sk)->inet_sport != inet_sk((struct sock *)msk)->inet_sport; 123 } 124 125 static void subflow_add_reset_reason(struct sk_buff *skb, u8 reason) 126 { 127 struct mptcp_ext *mpext = skb_ext_add(skb, SKB_EXT_MPTCP); 128 129 if (mpext) { 130 memset(mpext, 0, sizeof(*mpext)); 131 mpext->reset_reason = reason; 132 } 133 } 134 135 static int subflow_reset_req_endp(struct request_sock *req, struct sk_buff *skb) 136 { 137 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEENDPATTEMPT); 138 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT); 139 return -EPERM; 140 } 141 142 /* Init mptcp request socket. 143 * 144 * Returns an error code if a JOIN has failed and a TCP reset 145 * should be sent. 146 */ 147 static int subflow_check_req(struct request_sock *req, 148 const struct sock *sk_listener, 149 struct sk_buff *skb) 150 { 151 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener); 152 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 153 struct mptcp_options_received mp_opt; 154 bool opt_mp_capable, opt_mp_join; 155 156 pr_debug("subflow_req=%p, listener=%p\n", subflow_req, listener); 157 158 #ifdef CONFIG_TCP_MD5SIG 159 /* no MPTCP if MD5SIG is enabled on this socket or we may run out of 160 * TCP option space. 161 */ 162 if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info)) { 163 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP); 164 return -EINVAL; 165 } 166 #endif 167 168 mptcp_get_options(skb, &mp_opt); 169 170 opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYN); 171 opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYN); 172 if (opt_mp_capable) { 173 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE); 174 175 if (unlikely(listener->pm_listener)) 176 return subflow_reset_req_endp(req, skb); 177 if (opt_mp_join) 178 return 0; 179 } else if (opt_mp_join) { 180 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX); 181 182 if (mp_opt.backup) 183 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNBACKUPRX); 184 } else if (unlikely(listener->pm_listener)) { 185 return subflow_reset_req_endp(req, skb); 186 } 187 188 if (opt_mp_capable && listener->request_mptcp) { 189 int err, retries = MPTCP_TOKEN_MAX_RETRIES; 190 191 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq; 192 again: 193 do { 194 get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key)); 195 } while (subflow_req->local_key == 0); 196 197 if (unlikely(req->syncookie)) { 198 mptcp_crypto_key_sha(subflow_req->local_key, 199 &subflow_req->token, 200 &subflow_req->idsn); 201 if (mptcp_token_exists(subflow_req->token)) { 202 if (retries-- > 0) 203 goto again; 204 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT); 205 } else { 206 subflow_req->mp_capable = 1; 207 } 208 return 0; 209 } 210 211 err = mptcp_token_new_request(req); 212 if (err == 0) 213 subflow_req->mp_capable = 1; 214 else if (retries-- > 0) 215 goto again; 216 else 217 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT); 218 219 } else if (opt_mp_join && listener->request_mptcp) { 220 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq; 221 subflow_req->mp_join = 1; 222 subflow_req->backup = mp_opt.backup; 223 subflow_req->remote_id = mp_opt.join_id; 224 subflow_req->token = mp_opt.token; 225 subflow_req->remote_nonce = mp_opt.nonce; 226 subflow_req->msk = subflow_token_join_request(req); 227 228 /* Can't fall back to TCP in this case. */ 229 if (!subflow_req->msk) { 230 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP); 231 return -EPERM; 232 } 233 234 if (subflow_use_different_sport(subflow_req->msk, sk_listener)) { 235 pr_debug("syn inet_sport=%d %d\n", 236 ntohs(inet_sk(sk_listener)->inet_sport), 237 ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport)); 238 if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) { 239 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX); 240 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT); 241 return -EPERM; 242 } 243 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX); 244 } 245 246 subflow_req_create_thmac(subflow_req); 247 248 if (unlikely(req->syncookie)) { 249 if (!mptcp_can_accept_new_subflow(subflow_req->msk)) { 250 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT); 251 return -EPERM; 252 } 253 254 subflow_init_req_cookie_join_save(subflow_req, skb); 255 } 256 257 pr_debug("token=%u, remote_nonce=%u msk=%p\n", subflow_req->token, 258 subflow_req->remote_nonce, subflow_req->msk); 259 } 260 261 return 0; 262 } 263 264 int mptcp_subflow_init_cookie_req(struct request_sock *req, 265 const struct sock *sk_listener, 266 struct sk_buff *skb) 267 { 268 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener); 269 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 270 struct mptcp_options_received mp_opt; 271 bool opt_mp_capable, opt_mp_join; 272 int err; 273 274 subflow_init_req(req, sk_listener); 275 mptcp_get_options(skb, &mp_opt); 276 277 opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_ACK); 278 opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK); 279 if (opt_mp_capable && opt_mp_join) 280 return -EINVAL; 281 282 if (opt_mp_capable && listener->request_mptcp) { 283 if (mp_opt.sndr_key == 0) 284 return -EINVAL; 285 286 subflow_req->local_key = mp_opt.rcvr_key; 287 err = mptcp_token_new_request(req); 288 if (err) 289 return err; 290 291 subflow_req->mp_capable = 1; 292 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1; 293 } else if (opt_mp_join && listener->request_mptcp) { 294 if (!mptcp_token_join_cookie_init_state(subflow_req, skb)) 295 return -EINVAL; 296 297 subflow_req->mp_join = 1; 298 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1; 299 } 300 301 return 0; 302 } 303 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req); 304 305 static enum sk_rst_reason mptcp_get_rst_reason(const struct sk_buff *skb) 306 { 307 const struct mptcp_ext *mpext = mptcp_get_ext(skb); 308 309 if (!mpext) 310 return SK_RST_REASON_NOT_SPECIFIED; 311 312 return sk_rst_convert_mptcp_reason(mpext->reset_reason); 313 } 314 315 static struct dst_entry *subflow_v4_route_req(const struct sock *sk, 316 struct sk_buff *skb, 317 struct flowi *fl, 318 struct request_sock *req, 319 u32 tw_isn) 320 { 321 struct dst_entry *dst; 322 int err; 323 324 tcp_rsk(req)->is_mptcp = 1; 325 subflow_init_req(req, sk); 326 327 dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req, tw_isn); 328 if (!dst) 329 return NULL; 330 331 err = subflow_check_req(req, sk, skb); 332 if (err == 0) 333 return dst; 334 335 dst_release(dst); 336 if (!req->syncookie) 337 tcp_request_sock_ops.send_reset(sk, skb, 338 mptcp_get_rst_reason(skb)); 339 return NULL; 340 } 341 342 static void subflow_prep_synack(const struct sock *sk, struct request_sock *req, 343 struct tcp_fastopen_cookie *foc, 344 enum tcp_synack_type synack_type) 345 { 346 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 347 struct inet_request_sock *ireq = inet_rsk(req); 348 349 /* clear tstamp_ok, as needed depending on cookie */ 350 if (foc && foc->len > -1) 351 ireq->tstamp_ok = 0; 352 353 if (synack_type == TCP_SYNACK_FASTOPEN) 354 mptcp_fastopen_subflow_synack_set_params(subflow, req); 355 } 356 357 static int subflow_v4_send_synack(const struct sock *sk, struct dst_entry *dst, 358 struct flowi *fl, 359 struct request_sock *req, 360 struct tcp_fastopen_cookie *foc, 361 enum tcp_synack_type synack_type, 362 struct sk_buff *syn_skb) 363 { 364 subflow_prep_synack(sk, req, foc, synack_type); 365 366 return tcp_request_sock_ipv4_ops.send_synack(sk, dst, fl, req, foc, 367 synack_type, syn_skb); 368 } 369 370 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 371 static int subflow_v6_send_synack(const struct sock *sk, struct dst_entry *dst, 372 struct flowi *fl, 373 struct request_sock *req, 374 struct tcp_fastopen_cookie *foc, 375 enum tcp_synack_type synack_type, 376 struct sk_buff *syn_skb) 377 { 378 subflow_prep_synack(sk, req, foc, synack_type); 379 380 return tcp_request_sock_ipv6_ops.send_synack(sk, dst, fl, req, foc, 381 synack_type, syn_skb); 382 } 383 384 static struct dst_entry *subflow_v6_route_req(const struct sock *sk, 385 struct sk_buff *skb, 386 struct flowi *fl, 387 struct request_sock *req, 388 u32 tw_isn) 389 { 390 struct dst_entry *dst; 391 int err; 392 393 tcp_rsk(req)->is_mptcp = 1; 394 subflow_init_req(req, sk); 395 396 dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req, tw_isn); 397 if (!dst) 398 return NULL; 399 400 err = subflow_check_req(req, sk, skb); 401 if (err == 0) 402 return dst; 403 404 dst_release(dst); 405 if (!req->syncookie) 406 tcp6_request_sock_ops.send_reset(sk, skb, 407 mptcp_get_rst_reason(skb)); 408 return NULL; 409 } 410 #endif 411 412 /* validate received truncated hmac and create hmac for third ACK */ 413 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow) 414 { 415 u8 hmac[SHA256_DIGEST_SIZE]; 416 u64 thmac; 417 418 subflow_generate_hmac(subflow->remote_key, subflow->local_key, 419 subflow->remote_nonce, subflow->local_nonce, 420 hmac); 421 422 thmac = get_unaligned_be64(hmac); 423 pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n", 424 subflow, subflow->token, thmac, subflow->thmac); 425 426 return thmac == subflow->thmac; 427 } 428 429 void mptcp_subflow_reset(struct sock *ssk) 430 { 431 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 432 struct sock *sk = subflow->conn; 433 434 /* mptcp_mp_fail_no_response() can reach here on an already closed 435 * socket 436 */ 437 if (ssk->sk_state == TCP_CLOSE) 438 return; 439 440 /* must hold: tcp_done() could drop last reference on parent */ 441 sock_hold(sk); 442 443 mptcp_send_active_reset_reason(ssk); 444 tcp_done(ssk); 445 if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags)) 446 mptcp_schedule_work(sk); 447 448 sock_put(sk); 449 } 450 451 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk) 452 { 453 return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport; 454 } 455 456 void __mptcp_sync_state(struct sock *sk, int state) 457 { 458 struct mptcp_subflow_context *subflow; 459 struct mptcp_sock *msk = mptcp_sk(sk); 460 struct sock *ssk = msk->first; 461 462 subflow = mptcp_subflow_ctx(ssk); 463 __mptcp_propagate_sndbuf(sk, ssk); 464 if (!msk->rcvspace_init) 465 mptcp_rcv_space_init(msk, ssk); 466 467 if (sk->sk_state == TCP_SYN_SENT) { 468 /* subflow->idsn is always available is TCP_SYN_SENT state, 469 * even for the FASTOPEN scenarios 470 */ 471 WRITE_ONCE(msk->write_seq, subflow->idsn + 1); 472 WRITE_ONCE(msk->snd_nxt, msk->write_seq); 473 mptcp_set_state(sk, state); 474 sk->sk_state_change(sk); 475 } 476 } 477 478 static void subflow_set_remote_key(struct mptcp_sock *msk, 479 struct mptcp_subflow_context *subflow, 480 const struct mptcp_options_received *mp_opt) 481 { 482 /* active MPC subflow will reach here multiple times: 483 * at subflow_finish_connect() time and at 4th ack time 484 */ 485 if (subflow->remote_key_valid) 486 return; 487 488 subflow->remote_key_valid = 1; 489 subflow->remote_key = mp_opt->sndr_key; 490 mptcp_crypto_key_sha(subflow->remote_key, NULL, &subflow->iasn); 491 subflow->iasn++; 492 493 WRITE_ONCE(msk->remote_key, subflow->remote_key); 494 WRITE_ONCE(msk->ack_seq, subflow->iasn); 495 WRITE_ONCE(msk->can_ack, true); 496 atomic64_set(&msk->rcv_wnd_sent, subflow->iasn); 497 } 498 499 static void mptcp_propagate_state(struct sock *sk, struct sock *ssk, 500 struct mptcp_subflow_context *subflow, 501 const struct mptcp_options_received *mp_opt) 502 { 503 struct mptcp_sock *msk = mptcp_sk(sk); 504 505 mptcp_data_lock(sk); 506 if (mp_opt) { 507 /* Options are available only in the non fallback cases 508 * avoid updating rx path fields otherwise 509 */ 510 WRITE_ONCE(msk->snd_una, subflow->idsn + 1); 511 WRITE_ONCE(msk->wnd_end, subflow->idsn + 1 + tcp_sk(ssk)->snd_wnd); 512 subflow_set_remote_key(msk, subflow, mp_opt); 513 } 514 515 if (!sock_owned_by_user(sk)) { 516 __mptcp_sync_state(sk, ssk->sk_state); 517 } else { 518 msk->pending_state = ssk->sk_state; 519 __set_bit(MPTCP_SYNC_STATE, &msk->cb_flags); 520 } 521 mptcp_data_unlock(sk); 522 } 523 524 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb) 525 { 526 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 527 struct mptcp_options_received mp_opt; 528 struct sock *parent = subflow->conn; 529 struct mptcp_sock *msk; 530 531 subflow->icsk_af_ops->sk_rx_dst_set(sk, skb); 532 533 /* be sure no special action on any packet other than syn-ack */ 534 if (subflow->conn_finished) 535 return; 536 537 msk = mptcp_sk(parent); 538 subflow->rel_write_seq = 1; 539 subflow->conn_finished = 1; 540 subflow->ssn_offset = TCP_SKB_CB(skb)->seq; 541 pr_debug("subflow=%p synack seq=%x\n", subflow, subflow->ssn_offset); 542 543 mptcp_get_options(skb, &mp_opt); 544 if (subflow->request_mptcp) { 545 if (!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYNACK)) { 546 MPTCP_INC_STATS(sock_net(sk), 547 MPTCP_MIB_MPCAPABLEACTIVEFALLBACK); 548 mptcp_do_fallback(sk); 549 pr_fallback(msk); 550 goto fallback; 551 } 552 553 if (mp_opt.suboptions & OPTION_MPTCP_CSUMREQD) 554 WRITE_ONCE(msk->csum_enabled, true); 555 if (mp_opt.deny_join_id0) 556 WRITE_ONCE(msk->pm.remote_deny_join_id0, true); 557 subflow->mp_capable = 1; 558 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK); 559 mptcp_finish_connect(sk); 560 mptcp_active_enable(parent); 561 mptcp_propagate_state(parent, sk, subflow, &mp_opt); 562 } else if (subflow->request_join) { 563 u8 hmac[SHA256_DIGEST_SIZE]; 564 565 if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYNACK)) { 566 subflow->reset_reason = MPTCP_RST_EMPTCP; 567 goto do_reset; 568 } 569 570 subflow->backup = mp_opt.backup; 571 subflow->thmac = mp_opt.thmac; 572 subflow->remote_nonce = mp_opt.nonce; 573 WRITE_ONCE(subflow->remote_id, mp_opt.join_id); 574 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u backup=%d\n", 575 subflow, subflow->thmac, subflow->remote_nonce, 576 subflow->backup); 577 578 if (!subflow_thmac_valid(subflow)) { 579 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC); 580 subflow->reset_reason = MPTCP_RST_EMPTCP; 581 goto do_reset; 582 } 583 584 if (!mptcp_finish_join(sk)) 585 goto do_reset; 586 587 subflow_generate_hmac(subflow->local_key, subflow->remote_key, 588 subflow->local_nonce, 589 subflow->remote_nonce, 590 hmac); 591 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN); 592 593 subflow->mp_join = 1; 594 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX); 595 596 if (subflow->backup) 597 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKBACKUPRX); 598 599 if (subflow_use_different_dport(msk, sk)) { 600 pr_debug("synack inet_dport=%d %d\n", 601 ntohs(inet_sk(sk)->inet_dport), 602 ntohs(inet_sk(parent)->inet_dport)); 603 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX); 604 } 605 } else if (mptcp_check_fallback(sk)) { 606 /* It looks like MPTCP is blocked, while TCP is not */ 607 if (subflow->mpc_drop) 608 mptcp_active_disable(parent); 609 fallback: 610 mptcp_propagate_state(parent, sk, subflow, NULL); 611 } 612 return; 613 614 do_reset: 615 subflow->reset_transient = 0; 616 mptcp_subflow_reset(sk); 617 } 618 619 static void subflow_set_local_id(struct mptcp_subflow_context *subflow, int local_id) 620 { 621 WARN_ON_ONCE(local_id < 0 || local_id > 255); 622 WRITE_ONCE(subflow->local_id, local_id); 623 } 624 625 static int subflow_chk_local_id(struct sock *sk) 626 { 627 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 628 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 629 int err; 630 631 if (likely(subflow->local_id >= 0)) 632 return 0; 633 634 err = mptcp_pm_get_local_id(msk, (struct sock_common *)sk); 635 if (err < 0) 636 return err; 637 638 subflow_set_local_id(subflow, err); 639 subflow->request_bkup = mptcp_pm_is_backup(msk, (struct sock_common *)sk); 640 641 return 0; 642 } 643 644 static int subflow_rebuild_header(struct sock *sk) 645 { 646 int err = subflow_chk_local_id(sk); 647 648 if (unlikely(err < 0)) 649 return err; 650 651 return inet_sk_rebuild_header(sk); 652 } 653 654 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 655 static int subflow_v6_rebuild_header(struct sock *sk) 656 { 657 int err = subflow_chk_local_id(sk); 658 659 if (unlikely(err < 0)) 660 return err; 661 662 return inet6_sk_rebuild_header(sk); 663 } 664 #endif 665 666 static struct request_sock_ops mptcp_subflow_v4_request_sock_ops __ro_after_init; 667 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops __ro_after_init; 668 669 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb) 670 { 671 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 672 673 pr_debug("subflow=%p\n", subflow); 674 675 /* Never answer to SYNs sent to broadcast or multicast */ 676 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) 677 goto drop; 678 679 return tcp_conn_request(&mptcp_subflow_v4_request_sock_ops, 680 &subflow_request_sock_ipv4_ops, 681 sk, skb); 682 drop: 683 tcp_listendrop(sk); 684 return 0; 685 } 686 687 static void subflow_v4_req_destructor(struct request_sock *req) 688 { 689 subflow_req_destructor(req); 690 tcp_request_sock_ops.destructor(req); 691 } 692 693 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 694 static struct request_sock_ops mptcp_subflow_v6_request_sock_ops __ro_after_init; 695 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops __ro_after_init; 696 static struct inet_connection_sock_af_ops subflow_v6_specific __ro_after_init; 697 static struct inet_connection_sock_af_ops subflow_v6m_specific __ro_after_init; 698 static struct proto tcpv6_prot_override __ro_after_init; 699 700 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb) 701 { 702 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 703 704 pr_debug("subflow=%p\n", subflow); 705 706 if (skb->protocol == htons(ETH_P_IP)) 707 return subflow_v4_conn_request(sk, skb); 708 709 if (!ipv6_unicast_destination(skb)) 710 goto drop; 711 712 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) { 713 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS); 714 return 0; 715 } 716 717 return tcp_conn_request(&mptcp_subflow_v6_request_sock_ops, 718 &subflow_request_sock_ipv6_ops, sk, skb); 719 720 drop: 721 tcp_listendrop(sk); 722 return 0; /* don't send reset */ 723 } 724 725 static void subflow_v6_req_destructor(struct request_sock *req) 726 { 727 subflow_req_destructor(req); 728 tcp6_request_sock_ops.destructor(req); 729 } 730 #endif 731 732 struct request_sock *mptcp_subflow_reqsk_alloc(const struct request_sock_ops *ops, 733 struct sock *sk_listener, 734 bool attach_listener) 735 { 736 if (ops->family == AF_INET) 737 ops = &mptcp_subflow_v4_request_sock_ops; 738 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 739 else if (ops->family == AF_INET6) 740 ops = &mptcp_subflow_v6_request_sock_ops; 741 #endif 742 743 return inet_reqsk_alloc(ops, sk_listener, attach_listener); 744 } 745 EXPORT_SYMBOL(mptcp_subflow_reqsk_alloc); 746 747 /* validate hmac received in third ACK */ 748 static bool subflow_hmac_valid(const struct request_sock *req, 749 const struct mptcp_options_received *mp_opt) 750 { 751 const struct mptcp_subflow_request_sock *subflow_req; 752 u8 hmac[SHA256_DIGEST_SIZE]; 753 struct mptcp_sock *msk; 754 755 subflow_req = mptcp_subflow_rsk(req); 756 msk = subflow_req->msk; 757 if (!msk) 758 return false; 759 760 subflow_generate_hmac(READ_ONCE(msk->remote_key), 761 READ_ONCE(msk->local_key), 762 subflow_req->remote_nonce, 763 subflow_req->local_nonce, hmac); 764 765 return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN); 766 } 767 768 static void subflow_ulp_fallback(struct sock *sk, 769 struct mptcp_subflow_context *old_ctx) 770 { 771 struct inet_connection_sock *icsk = inet_csk(sk); 772 773 mptcp_subflow_tcp_fallback(sk, old_ctx); 774 icsk->icsk_ulp_ops = NULL; 775 rcu_assign_pointer(icsk->icsk_ulp_data, NULL); 776 tcp_sk(sk)->is_mptcp = 0; 777 778 mptcp_subflow_ops_undo_override(sk); 779 } 780 781 void mptcp_subflow_drop_ctx(struct sock *ssk) 782 { 783 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk); 784 785 if (!ctx) 786 return; 787 788 list_del(&mptcp_subflow_ctx(ssk)->node); 789 if (inet_csk(ssk)->icsk_ulp_ops) { 790 subflow_ulp_fallback(ssk, ctx); 791 if (ctx->conn) 792 sock_put(ctx->conn); 793 } 794 795 kfree_rcu(ctx, rcu); 796 } 797 798 void __mptcp_subflow_fully_established(struct mptcp_sock *msk, 799 struct mptcp_subflow_context *subflow, 800 const struct mptcp_options_received *mp_opt) 801 { 802 subflow_set_remote_key(msk, subflow, mp_opt); 803 subflow->fully_established = 1; 804 WRITE_ONCE(msk->fully_established, true); 805 806 if (subflow->is_mptfo) 807 __mptcp_fastopen_gen_msk_ackseq(msk, subflow, mp_opt); 808 } 809 810 static struct sock *subflow_syn_recv_sock(const struct sock *sk, 811 struct sk_buff *skb, 812 struct request_sock *req, 813 struct dst_entry *dst, 814 struct request_sock *req_unhash, 815 bool *own_req) 816 { 817 struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk); 818 struct mptcp_subflow_request_sock *subflow_req; 819 struct mptcp_options_received mp_opt; 820 bool fallback, fallback_is_fatal; 821 enum sk_rst_reason reason; 822 struct mptcp_sock *owner; 823 struct sock *child; 824 825 pr_debug("listener=%p, req=%p, conn=%p\n", listener, req, listener->conn); 826 827 /* After child creation we must look for MPC even when options 828 * are not parsed 829 */ 830 mp_opt.suboptions = 0; 831 832 /* hopefully temporary handling for MP_JOIN+syncookie */ 833 subflow_req = mptcp_subflow_rsk(req); 834 fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join; 835 fallback = !tcp_rsk(req)->is_mptcp; 836 if (fallback) 837 goto create_child; 838 839 /* if the sk is MP_CAPABLE, we try to fetch the client key */ 840 if (subflow_req->mp_capable) { 841 /* we can receive and accept an in-window, out-of-order pkt, 842 * which may not carry the MP_CAPABLE opt even on mptcp enabled 843 * paths: always try to extract the peer key, and fallback 844 * for packets missing it. 845 * Even OoO DSS packets coming legitly after dropped or 846 * reordered MPC will cause fallback, but we don't have other 847 * options. 848 */ 849 mptcp_get_options(skb, &mp_opt); 850 if (!(mp_opt.suboptions & 851 (OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_ACK))) 852 fallback = true; 853 854 } else if (subflow_req->mp_join) { 855 mptcp_get_options(skb, &mp_opt); 856 if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK) || 857 !subflow_hmac_valid(req, &mp_opt) || 858 !mptcp_can_accept_new_subflow(subflow_req->msk)) { 859 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC); 860 fallback = true; 861 } 862 } 863 864 create_child: 865 child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst, 866 req_unhash, own_req); 867 868 if (child && *own_req) { 869 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child); 870 871 tcp_rsk(req)->drop_req = false; 872 873 /* we need to fallback on ctx allocation failure and on pre-reqs 874 * checking above. In the latter scenario we additionally need 875 * to reset the context to non MPTCP status. 876 */ 877 if (!ctx || fallback) { 878 if (fallback_is_fatal) { 879 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP); 880 goto dispose_child; 881 } 882 goto fallback; 883 } 884 885 /* ssk inherits options of listener sk */ 886 ctx->setsockopt_seq = listener->setsockopt_seq; 887 888 if (ctx->mp_capable) { 889 ctx->conn = mptcp_sk_clone_init(listener->conn, &mp_opt, child, req); 890 if (!ctx->conn) 891 goto fallback; 892 893 ctx->subflow_id = 1; 894 owner = mptcp_sk(ctx->conn); 895 mptcp_pm_new_connection(owner, child, 1); 896 897 /* with OoO packets we can reach here without ingress 898 * mpc option 899 */ 900 if (mp_opt.suboptions & OPTION_MPTCP_MPC_ACK) { 901 mptcp_pm_fully_established(owner, child); 902 ctx->pm_notified = 1; 903 } 904 } else if (ctx->mp_join) { 905 owner = subflow_req->msk; 906 if (!owner) { 907 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT); 908 goto dispose_child; 909 } 910 911 /* move the msk reference ownership to the subflow */ 912 subflow_req->msk = NULL; 913 ctx->conn = (struct sock *)owner; 914 915 if (subflow_use_different_sport(owner, sk)) { 916 pr_debug("ack inet_sport=%d %d\n", 917 ntohs(inet_sk(sk)->inet_sport), 918 ntohs(inet_sk((struct sock *)owner)->inet_sport)); 919 if (!mptcp_pm_sport_in_anno_list(owner, sk)) { 920 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX); 921 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT); 922 goto dispose_child; 923 } 924 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX); 925 } 926 927 if (!mptcp_finish_join(child)) { 928 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(child); 929 930 subflow_add_reset_reason(skb, subflow->reset_reason); 931 goto dispose_child; 932 } 933 934 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX); 935 tcp_rsk(req)->drop_req = true; 936 } 937 } 938 939 /* check for expected invariant - should never trigger, just help 940 * catching earlier subtle bugs 941 */ 942 WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp && 943 (!mptcp_subflow_ctx(child) || 944 !mptcp_subflow_ctx(child)->conn)); 945 return child; 946 947 dispose_child: 948 mptcp_subflow_drop_ctx(child); 949 tcp_rsk(req)->drop_req = true; 950 inet_csk_prepare_for_destroy_sock(child); 951 tcp_done(child); 952 reason = mptcp_get_rst_reason(skb); 953 req->rsk_ops->send_reset(sk, skb, reason); 954 955 /* The last child reference will be released by the caller */ 956 return child; 957 958 fallback: 959 if (fallback) 960 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK); 961 mptcp_subflow_drop_ctx(child); 962 return child; 963 } 964 965 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init; 966 static struct proto tcp_prot_override __ro_after_init; 967 968 enum mapping_status { 969 MAPPING_OK, 970 MAPPING_INVALID, 971 MAPPING_EMPTY, 972 MAPPING_DATA_FIN, 973 MAPPING_DUMMY, 974 MAPPING_BAD_CSUM 975 }; 976 977 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn) 978 { 979 pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d\n", 980 ssn, subflow->map_subflow_seq, subflow->map_data_len); 981 } 982 983 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb) 984 { 985 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 986 unsigned int skb_consumed; 987 988 skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq; 989 if (unlikely(skb_consumed >= skb->len)) { 990 DEBUG_NET_WARN_ON_ONCE(1); 991 return true; 992 } 993 994 return skb->len - skb_consumed <= subflow->map_data_len - 995 mptcp_subflow_get_map_offset(subflow); 996 } 997 998 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb) 999 { 1000 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 1001 u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset; 1002 1003 if (unlikely(before(ssn, subflow->map_subflow_seq))) { 1004 /* Mapping covers data later in the subflow stream, 1005 * currently unsupported. 1006 */ 1007 dbg_bad_map(subflow, ssn); 1008 return false; 1009 } 1010 if (unlikely(!before(ssn, subflow->map_subflow_seq + 1011 subflow->map_data_len))) { 1012 /* Mapping does covers past subflow data, invalid */ 1013 dbg_bad_map(subflow, ssn); 1014 return false; 1015 } 1016 return true; 1017 } 1018 1019 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb, 1020 bool csum_reqd) 1021 { 1022 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 1023 u32 offset, seq, delta; 1024 __sum16 csum; 1025 int len; 1026 1027 if (!csum_reqd) 1028 return MAPPING_OK; 1029 1030 /* mapping already validated on previous traversal */ 1031 if (subflow->map_csum_len == subflow->map_data_len) 1032 return MAPPING_OK; 1033 1034 /* traverse the receive queue, ensuring it contains a full 1035 * DSS mapping and accumulating the related csum. 1036 * Preserve the accoumlate csum across multiple calls, to compute 1037 * the csum only once 1038 */ 1039 delta = subflow->map_data_len - subflow->map_csum_len; 1040 for (;;) { 1041 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len; 1042 offset = seq - TCP_SKB_CB(skb)->seq; 1043 1044 /* if the current skb has not been accounted yet, csum its contents 1045 * up to the amount covered by the current DSS 1046 */ 1047 if (offset < skb->len) { 1048 __wsum csum; 1049 1050 len = min(skb->len - offset, delta); 1051 csum = skb_checksum(skb, offset, len, 0); 1052 subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum, 1053 subflow->map_csum_len); 1054 1055 delta -= len; 1056 subflow->map_csum_len += len; 1057 } 1058 if (delta == 0) 1059 break; 1060 1061 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) { 1062 /* if this subflow is closed, the partial mapping 1063 * will be never completed; flush the pending skbs, so 1064 * that subflow_sched_work_if_closed() can kick in 1065 */ 1066 if (unlikely(ssk->sk_state == TCP_CLOSE)) 1067 while ((skb = skb_peek(&ssk->sk_receive_queue))) 1068 sk_eat_skb(ssk, skb); 1069 1070 /* not enough data to validate the csum */ 1071 return MAPPING_EMPTY; 1072 } 1073 1074 /* the DSS mapping for next skbs will be validated later, 1075 * when a get_mapping_status call will process such skb 1076 */ 1077 skb = skb->next; 1078 } 1079 1080 /* note that 'map_data_len' accounts only for the carried data, does 1081 * not include the eventual seq increment due to the data fin, 1082 * while the pseudo header requires the original DSS data len, 1083 * including that 1084 */ 1085 csum = __mptcp_make_csum(subflow->map_seq, 1086 subflow->map_subflow_seq, 1087 subflow->map_data_len + subflow->map_data_fin, 1088 subflow->map_data_csum); 1089 if (unlikely(csum)) { 1090 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR); 1091 return MAPPING_BAD_CSUM; 1092 } 1093 1094 subflow->valid_csum_seen = 1; 1095 return MAPPING_OK; 1096 } 1097 1098 static enum mapping_status get_mapping_status(struct sock *ssk, 1099 struct mptcp_sock *msk) 1100 { 1101 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 1102 bool csum_reqd = READ_ONCE(msk->csum_enabled); 1103 struct mptcp_ext *mpext; 1104 struct sk_buff *skb; 1105 u16 data_len; 1106 u64 map_seq; 1107 1108 skb = skb_peek(&ssk->sk_receive_queue); 1109 if (!skb) 1110 return MAPPING_EMPTY; 1111 1112 if (mptcp_check_fallback(ssk)) 1113 return MAPPING_DUMMY; 1114 1115 mpext = mptcp_get_ext(skb); 1116 if (!mpext || !mpext->use_map) { 1117 if (!subflow->map_valid && !skb->len) { 1118 /* the TCP stack deliver 0 len FIN pkt to the receive 1119 * queue, that is the only 0len pkts ever expected here, 1120 * and we can admit no mapping only for 0 len pkts 1121 */ 1122 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) 1123 WARN_ONCE(1, "0len seq %d:%d flags %x", 1124 TCP_SKB_CB(skb)->seq, 1125 TCP_SKB_CB(skb)->end_seq, 1126 TCP_SKB_CB(skb)->tcp_flags); 1127 sk_eat_skb(ssk, skb); 1128 return MAPPING_EMPTY; 1129 } 1130 1131 if (!subflow->map_valid) 1132 return MAPPING_INVALID; 1133 1134 goto validate_seq; 1135 } 1136 1137 trace_get_mapping_status(mpext); 1138 1139 data_len = mpext->data_len; 1140 if (data_len == 0) { 1141 pr_debug("infinite mapping received\n"); 1142 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX); 1143 subflow->map_data_len = 0; 1144 return MAPPING_INVALID; 1145 } 1146 1147 if (mpext->data_fin == 1) { 1148 u64 data_fin_seq; 1149 1150 if (data_len == 1) { 1151 bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq, 1152 mpext->dsn64); 1153 pr_debug("DATA_FIN with no payload seq=%llu\n", mpext->data_seq); 1154 if (subflow->map_valid) { 1155 /* A DATA_FIN might arrive in a DSS 1156 * option before the previous mapping 1157 * has been fully consumed. Continue 1158 * handling the existing mapping. 1159 */ 1160 skb_ext_del(skb, SKB_EXT_MPTCP); 1161 return MAPPING_OK; 1162 } 1163 1164 if (updated) 1165 mptcp_schedule_work((struct sock *)msk); 1166 1167 return MAPPING_DATA_FIN; 1168 } 1169 1170 data_fin_seq = mpext->data_seq + data_len - 1; 1171 1172 /* If mpext->data_seq is a 32-bit value, data_fin_seq must also 1173 * be limited to 32 bits. 1174 */ 1175 if (!mpext->dsn64) 1176 data_fin_seq &= GENMASK_ULL(31, 0); 1177 1178 mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64); 1179 pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d\n", 1180 data_fin_seq, mpext->dsn64); 1181 1182 /* Adjust for DATA_FIN using 1 byte of sequence space */ 1183 data_len--; 1184 } 1185 1186 map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64); 1187 WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64); 1188 1189 if (subflow->map_valid) { 1190 /* Allow replacing only with an identical map */ 1191 if (subflow->map_seq == map_seq && 1192 subflow->map_subflow_seq == mpext->subflow_seq && 1193 subflow->map_data_len == data_len && 1194 subflow->map_csum_reqd == mpext->csum_reqd) { 1195 skb_ext_del(skb, SKB_EXT_MPTCP); 1196 goto validate_csum; 1197 } 1198 1199 /* If this skb data are fully covered by the current mapping, 1200 * the new map would need caching, which is not supported 1201 */ 1202 if (skb_is_fully_mapped(ssk, skb)) { 1203 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH); 1204 return MAPPING_INVALID; 1205 } 1206 1207 /* will validate the next map after consuming the current one */ 1208 goto validate_csum; 1209 } 1210 1211 subflow->map_seq = map_seq; 1212 subflow->map_subflow_seq = mpext->subflow_seq; 1213 subflow->map_data_len = data_len; 1214 subflow->map_valid = 1; 1215 subflow->map_data_fin = mpext->data_fin; 1216 subflow->mpc_map = mpext->mpc_map; 1217 subflow->map_csum_reqd = mpext->csum_reqd; 1218 subflow->map_csum_len = 0; 1219 subflow->map_data_csum = csum_unfold(mpext->csum); 1220 1221 /* Cfr RFC 8684 Section 3.3.0 */ 1222 if (unlikely(subflow->map_csum_reqd != csum_reqd)) 1223 return MAPPING_INVALID; 1224 1225 pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u\n", 1226 subflow->map_seq, subflow->map_subflow_seq, 1227 subflow->map_data_len, subflow->map_csum_reqd, 1228 subflow->map_data_csum); 1229 1230 validate_seq: 1231 /* we revalidate valid mapping on new skb, because we must ensure 1232 * the current skb is completely covered by the available mapping 1233 */ 1234 if (!validate_mapping(ssk, skb)) { 1235 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH); 1236 return MAPPING_INVALID; 1237 } 1238 1239 skb_ext_del(skb, SKB_EXT_MPTCP); 1240 1241 validate_csum: 1242 return validate_data_csum(ssk, skb, csum_reqd); 1243 } 1244 1245 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb, 1246 u64 limit) 1247 { 1248 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 1249 bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN; 1250 struct tcp_sock *tp = tcp_sk(ssk); 1251 u32 offset, incr, avail_len; 1252 1253 offset = tp->copied_seq - TCP_SKB_CB(skb)->seq; 1254 if (WARN_ON_ONCE(offset > skb->len)) 1255 goto out; 1256 1257 avail_len = skb->len - offset; 1258 incr = limit >= avail_len ? avail_len + fin : limit; 1259 1260 pr_debug("discarding=%d len=%d offset=%d seq=%d\n", incr, skb->len, 1261 offset, subflow->map_subflow_seq); 1262 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA); 1263 tcp_sk(ssk)->copied_seq += incr; 1264 1265 out: 1266 if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq)) 1267 sk_eat_skb(ssk, skb); 1268 if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) 1269 subflow->map_valid = 0; 1270 } 1271 1272 /* sched mptcp worker to remove the subflow if no more data is pending */ 1273 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk) 1274 { 1275 struct sock *sk = (struct sock *)msk; 1276 1277 if (likely(ssk->sk_state != TCP_CLOSE && 1278 (ssk->sk_state != TCP_CLOSE_WAIT || 1279 inet_sk_state_load(sk) != TCP_ESTABLISHED))) 1280 return; 1281 1282 if (skb_queue_empty(&ssk->sk_receive_queue) && 1283 !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags)) 1284 mptcp_schedule_work(sk); 1285 } 1286 1287 static bool subflow_can_fallback(struct mptcp_subflow_context *subflow) 1288 { 1289 struct mptcp_sock *msk = mptcp_sk(subflow->conn); 1290 1291 if (subflow->mp_join) 1292 return false; 1293 else if (READ_ONCE(msk->csum_enabled)) 1294 return !subflow->valid_csum_seen; 1295 else 1296 return READ_ONCE(msk->allow_infinite_fallback); 1297 } 1298 1299 static void mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk) 1300 { 1301 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 1302 unsigned long fail_tout; 1303 1304 /* graceful failure can happen only on the MPC subflow */ 1305 if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first))) 1306 return; 1307 1308 /* since the close timeout take precedence on the fail one, 1309 * no need to start the latter when the first is already set 1310 */ 1311 if (sock_flag((struct sock *)msk, SOCK_DEAD)) 1312 return; 1313 1314 /* we don't need extreme accuracy here, use a zero fail_tout as special 1315 * value meaning no fail timeout at all; 1316 */ 1317 fail_tout = jiffies + TCP_RTO_MAX; 1318 if (!fail_tout) 1319 fail_tout = 1; 1320 WRITE_ONCE(subflow->fail_tout, fail_tout); 1321 tcp_send_ack(ssk); 1322 1323 mptcp_reset_tout_timer(msk, subflow->fail_tout); 1324 } 1325 1326 static bool subflow_check_data_avail(struct sock *ssk) 1327 { 1328 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 1329 enum mapping_status status; 1330 struct mptcp_sock *msk; 1331 struct sk_buff *skb; 1332 1333 if (!skb_peek(&ssk->sk_receive_queue)) 1334 WRITE_ONCE(subflow->data_avail, false); 1335 if (subflow->data_avail) 1336 return true; 1337 1338 msk = mptcp_sk(subflow->conn); 1339 for (;;) { 1340 u64 ack_seq; 1341 u64 old_ack; 1342 1343 status = get_mapping_status(ssk, msk); 1344 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue)); 1345 if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY || 1346 status == MAPPING_BAD_CSUM)) 1347 goto fallback; 1348 1349 if (status != MAPPING_OK) 1350 goto no_data; 1351 1352 skb = skb_peek(&ssk->sk_receive_queue); 1353 if (WARN_ON_ONCE(!skb)) 1354 goto no_data; 1355 1356 if (unlikely(!READ_ONCE(msk->can_ack))) 1357 goto fallback; 1358 1359 old_ack = READ_ONCE(msk->ack_seq); 1360 ack_seq = mptcp_subflow_get_mapped_dsn(subflow); 1361 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx\n", old_ack, 1362 ack_seq); 1363 if (unlikely(before64(ack_seq, old_ack))) { 1364 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq); 1365 continue; 1366 } 1367 1368 WRITE_ONCE(subflow->data_avail, true); 1369 break; 1370 } 1371 return true; 1372 1373 no_data: 1374 subflow_sched_work_if_closed(msk, ssk); 1375 return false; 1376 1377 fallback: 1378 if (!__mptcp_check_fallback(msk)) { 1379 /* RFC 8684 section 3.7. */ 1380 if (status == MAPPING_BAD_CSUM && 1381 (subflow->mp_join || subflow->valid_csum_seen)) { 1382 subflow->send_mp_fail = 1; 1383 1384 if (!READ_ONCE(msk->allow_infinite_fallback)) { 1385 subflow->reset_transient = 0; 1386 subflow->reset_reason = MPTCP_RST_EMIDDLEBOX; 1387 goto reset; 1388 } 1389 mptcp_subflow_fail(msk, ssk); 1390 WRITE_ONCE(subflow->data_avail, true); 1391 return true; 1392 } 1393 1394 if (!subflow_can_fallback(subflow) && subflow->map_data_len) { 1395 /* fatal protocol error, close the socket. 1396 * subflow_error_report() will introduce the appropriate barriers 1397 */ 1398 subflow->reset_transient = 0; 1399 subflow->reset_reason = MPTCP_RST_EMPTCP; 1400 1401 reset: 1402 WRITE_ONCE(ssk->sk_err, EBADMSG); 1403 tcp_set_state(ssk, TCP_CLOSE); 1404 while ((skb = skb_peek(&ssk->sk_receive_queue))) 1405 sk_eat_skb(ssk, skb); 1406 mptcp_send_active_reset_reason(ssk); 1407 WRITE_ONCE(subflow->data_avail, false); 1408 return false; 1409 } 1410 1411 mptcp_do_fallback(ssk); 1412 } 1413 1414 skb = skb_peek(&ssk->sk_receive_queue); 1415 subflow->map_valid = 1; 1416 subflow->map_seq = READ_ONCE(msk->ack_seq); 1417 subflow->map_data_len = skb->len; 1418 subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset; 1419 WRITE_ONCE(subflow->data_avail, true); 1420 return true; 1421 } 1422 1423 bool mptcp_subflow_data_available(struct sock *sk) 1424 { 1425 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 1426 1427 /* check if current mapping is still valid */ 1428 if (subflow->map_valid && 1429 mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) { 1430 subflow->map_valid = 0; 1431 WRITE_ONCE(subflow->data_avail, false); 1432 1433 pr_debug("Done with mapping: seq=%u data_len=%u\n", 1434 subflow->map_subflow_seq, 1435 subflow->map_data_len); 1436 } 1437 1438 return subflow_check_data_avail(sk); 1439 } 1440 1441 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy, 1442 * not the ssk one. 1443 * 1444 * In mptcp, rwin is about the mptcp-level connection data. 1445 * 1446 * Data that is still on the ssk rx queue can thus be ignored, 1447 * as far as mptcp peer is concerned that data is still inflight. 1448 * DSS ACK is updated when skb is moved to the mptcp rx queue. 1449 */ 1450 void mptcp_space(const struct sock *ssk, int *space, int *full_space) 1451 { 1452 const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 1453 const struct sock *sk = subflow->conn; 1454 1455 *space = __mptcp_space(sk); 1456 *full_space = mptcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); 1457 } 1458 1459 static void subflow_error_report(struct sock *ssk) 1460 { 1461 struct sock *sk = mptcp_subflow_ctx(ssk)->conn; 1462 1463 /* bail early if this is a no-op, so that we avoid introducing a 1464 * problematic lockdep dependency between TCP accept queue lock 1465 * and msk socket spinlock 1466 */ 1467 if (!sk->sk_socket) 1468 return; 1469 1470 mptcp_data_lock(sk); 1471 if (!sock_owned_by_user(sk)) 1472 __mptcp_error_report(sk); 1473 else 1474 __set_bit(MPTCP_ERROR_REPORT, &mptcp_sk(sk)->cb_flags); 1475 mptcp_data_unlock(sk); 1476 } 1477 1478 static void subflow_data_ready(struct sock *sk) 1479 { 1480 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 1481 u16 state = 1 << inet_sk_state_load(sk); 1482 struct sock *parent = subflow->conn; 1483 struct mptcp_sock *msk; 1484 1485 trace_sk_data_ready(sk); 1486 1487 msk = mptcp_sk(parent); 1488 if (state & TCPF_LISTEN) { 1489 /* MPJ subflow are removed from accept queue before reaching here, 1490 * avoid stray wakeups 1491 */ 1492 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue)) 1493 return; 1494 1495 parent->sk_data_ready(parent); 1496 return; 1497 } 1498 1499 WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable && 1500 !subflow->mp_join && !(state & TCPF_CLOSE)); 1501 1502 if (mptcp_subflow_data_available(sk)) { 1503 mptcp_data_ready(parent, sk); 1504 1505 /* subflow-level lowat test are not relevant. 1506 * respect the msk-level threshold eventually mandating an immediate ack 1507 */ 1508 if (mptcp_data_avail(msk) < parent->sk_rcvlowat && 1509 (tcp_sk(sk)->rcv_nxt - tcp_sk(sk)->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss) 1510 inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW; 1511 } else if (unlikely(sk->sk_err)) { 1512 subflow_error_report(sk); 1513 } 1514 } 1515 1516 static void subflow_write_space(struct sock *ssk) 1517 { 1518 struct sock *sk = mptcp_subflow_ctx(ssk)->conn; 1519 1520 mptcp_propagate_sndbuf(sk, ssk); 1521 mptcp_write_space(sk); 1522 } 1523 1524 static const struct inet_connection_sock_af_ops * 1525 subflow_default_af_ops(struct sock *sk) 1526 { 1527 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1528 if (sk->sk_family == AF_INET6) 1529 return &subflow_v6_specific; 1530 #endif 1531 return &subflow_specific; 1532 } 1533 1534 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1535 void mptcpv6_handle_mapped(struct sock *sk, bool mapped) 1536 { 1537 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 1538 struct inet_connection_sock *icsk = inet_csk(sk); 1539 const struct inet_connection_sock_af_ops *target; 1540 1541 target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk); 1542 1543 pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d\n", 1544 subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped); 1545 1546 if (likely(icsk->icsk_af_ops == target)) 1547 return; 1548 1549 subflow->icsk_af_ops = icsk->icsk_af_ops; 1550 icsk->icsk_af_ops = target; 1551 } 1552 #endif 1553 1554 void mptcp_info2sockaddr(const struct mptcp_addr_info *info, 1555 struct sockaddr_storage *addr, 1556 unsigned short family) 1557 { 1558 memset(addr, 0, sizeof(*addr)); 1559 addr->ss_family = family; 1560 if (addr->ss_family == AF_INET) { 1561 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr; 1562 1563 if (info->family == AF_INET) 1564 in_addr->sin_addr = info->addr; 1565 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1566 else if (ipv6_addr_v4mapped(&info->addr6)) 1567 in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3]; 1568 #endif 1569 in_addr->sin_port = info->port; 1570 } 1571 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1572 else if (addr->ss_family == AF_INET6) { 1573 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr; 1574 1575 if (info->family == AF_INET) 1576 ipv6_addr_set_v4mapped(info->addr.s_addr, 1577 &in6_addr->sin6_addr); 1578 else 1579 in6_addr->sin6_addr = info->addr6; 1580 in6_addr->sin6_port = info->port; 1581 } 1582 #endif 1583 } 1584 1585 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_pm_local *local, 1586 const struct mptcp_addr_info *remote) 1587 { 1588 struct mptcp_sock *msk = mptcp_sk(sk); 1589 struct mptcp_subflow_context *subflow; 1590 int local_id = local->addr.id; 1591 struct sockaddr_storage addr; 1592 int remote_id = remote->id; 1593 int err = -ENOTCONN; 1594 struct socket *sf; 1595 struct sock *ssk; 1596 u32 remote_token; 1597 int addrlen; 1598 1599 /* The userspace PM sent the request too early? */ 1600 if (!mptcp_is_fully_established(sk)) 1601 goto err_out; 1602 1603 err = mptcp_subflow_create_socket(sk, local->addr.family, &sf); 1604 if (err) { 1605 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTXCREATSKERR); 1606 pr_debug("msk=%p local=%d remote=%d create sock error: %d\n", 1607 msk, local_id, remote_id, err); 1608 goto err_out; 1609 } 1610 1611 ssk = sf->sk; 1612 subflow = mptcp_subflow_ctx(ssk); 1613 do { 1614 get_random_bytes(&subflow->local_nonce, sizeof(u32)); 1615 } while (!subflow->local_nonce); 1616 1617 /* if 'IPADDRANY', the ID will be set later, after the routing */ 1618 if (local->addr.family == AF_INET) { 1619 if (!local->addr.addr.s_addr) 1620 local_id = -1; 1621 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1622 } else if (sk->sk_family == AF_INET6) { 1623 if (ipv6_addr_any(&local->addr.addr6)) 1624 local_id = -1; 1625 #endif 1626 } 1627 1628 if (local_id >= 0) 1629 subflow_set_local_id(subflow, local_id); 1630 1631 subflow->remote_key_valid = 1; 1632 subflow->remote_key = READ_ONCE(msk->remote_key); 1633 subflow->local_key = READ_ONCE(msk->local_key); 1634 subflow->token = msk->token; 1635 mptcp_info2sockaddr(&local->addr, &addr, ssk->sk_family); 1636 1637 addrlen = sizeof(struct sockaddr_in); 1638 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1639 if (addr.ss_family == AF_INET6) 1640 addrlen = sizeof(struct sockaddr_in6); 1641 #endif 1642 ssk->sk_bound_dev_if = local->ifindex; 1643 err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen); 1644 if (err) { 1645 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTXBINDERR); 1646 pr_debug("msk=%p local=%d remote=%d bind error: %d\n", 1647 msk, local_id, remote_id, err); 1648 goto failed; 1649 } 1650 1651 mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL); 1652 pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d\n", msk, 1653 remote_token, local_id, remote_id); 1654 subflow->remote_token = remote_token; 1655 WRITE_ONCE(subflow->remote_id, remote_id); 1656 subflow->request_join = 1; 1657 subflow->request_bkup = !!(local->flags & MPTCP_PM_ADDR_FLAG_BACKUP); 1658 subflow->subflow_id = msk->subflow_id++; 1659 mptcp_info2sockaddr(remote, &addr, ssk->sk_family); 1660 1661 sock_hold(ssk); 1662 list_add_tail(&subflow->node, &msk->conn_list); 1663 err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK); 1664 if (err && err != -EINPROGRESS) { 1665 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTXCONNECTERR); 1666 pr_debug("msk=%p local=%d remote=%d connect error: %d\n", 1667 msk, local_id, remote_id, err); 1668 goto failed_unlink; 1669 } 1670 1671 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNTX); 1672 1673 /* discard the subflow socket */ 1674 mptcp_sock_graft(ssk, sk->sk_socket); 1675 iput(SOCK_INODE(sf)); 1676 WRITE_ONCE(msk->allow_infinite_fallback, false); 1677 mptcp_stop_tout_timer(sk); 1678 return 0; 1679 1680 failed_unlink: 1681 list_del(&subflow->node); 1682 sock_put(mptcp_subflow_tcp_sock(subflow)); 1683 1684 failed: 1685 subflow->disposable = 1; 1686 sock_release(sf); 1687 1688 err_out: 1689 /* we account subflows before the creation, and this failures will not 1690 * be caught by sk_state_change() 1691 */ 1692 mptcp_pm_close_subflow(msk); 1693 return err; 1694 } 1695 1696 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child) 1697 { 1698 #ifdef CONFIG_SOCK_CGROUP_DATA 1699 struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data, 1700 *child_skcd = &child->sk_cgrp_data; 1701 1702 /* only the additional subflows created by kworkers have to be modified */ 1703 if (cgroup_id(sock_cgroup_ptr(parent_skcd)) != 1704 cgroup_id(sock_cgroup_ptr(child_skcd))) { 1705 #ifdef CONFIG_MEMCG 1706 struct mem_cgroup *memcg = parent->sk_memcg; 1707 1708 mem_cgroup_sk_free(child); 1709 if (memcg && css_tryget(&memcg->css)) 1710 child->sk_memcg = memcg; 1711 #endif /* CONFIG_MEMCG */ 1712 1713 cgroup_sk_free(child_skcd); 1714 *child_skcd = *parent_skcd; 1715 cgroup_sk_clone(child_skcd); 1716 } 1717 #endif /* CONFIG_SOCK_CGROUP_DATA */ 1718 } 1719 1720 static void mptcp_subflow_ops_override(struct sock *ssk) 1721 { 1722 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1723 if (ssk->sk_prot == &tcpv6_prot) 1724 ssk->sk_prot = &tcpv6_prot_override; 1725 else 1726 #endif 1727 ssk->sk_prot = &tcp_prot_override; 1728 } 1729 1730 static void mptcp_subflow_ops_undo_override(struct sock *ssk) 1731 { 1732 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 1733 if (ssk->sk_prot == &tcpv6_prot_override) 1734 ssk->sk_prot = &tcpv6_prot; 1735 else 1736 #endif 1737 ssk->sk_prot = &tcp_prot; 1738 } 1739 1740 int mptcp_subflow_create_socket(struct sock *sk, unsigned short family, 1741 struct socket **new_sock) 1742 { 1743 struct mptcp_subflow_context *subflow; 1744 struct net *net = sock_net(sk); 1745 struct socket *sf; 1746 int err; 1747 1748 /* un-accepted server sockets can reach here - on bad configuration 1749 * bail early to avoid greater trouble later 1750 */ 1751 if (unlikely(!sk->sk_socket)) 1752 return -EINVAL; 1753 1754 err = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, &sf); 1755 if (err) 1756 return err; 1757 1758 lock_sock_nested(sf->sk, SINGLE_DEPTH_NESTING); 1759 1760 err = security_mptcp_add_subflow(sk, sf->sk); 1761 if (err) 1762 goto err_free; 1763 1764 /* the newly created socket has to be in the same cgroup as its parent */ 1765 mptcp_attach_cgroup(sk, sf->sk); 1766 1767 /* kernel sockets do not by default acquire net ref, but TCP timer 1768 * needs it. 1769 * Update ns_tracker to current stack trace and refcounted tracker. 1770 */ 1771 __netns_tracker_free(net, &sf->sk->ns_tracker, false); 1772 sf->sk->sk_net_refcnt = 1; 1773 get_net_track(net, &sf->sk->ns_tracker, GFP_KERNEL); 1774 sock_inuse_add(net, 1); 1775 err = tcp_set_ulp(sf->sk, "mptcp"); 1776 if (err) 1777 goto err_free; 1778 1779 mptcp_sockopt_sync_locked(mptcp_sk(sk), sf->sk); 1780 release_sock(sf->sk); 1781 1782 /* the newly created socket really belongs to the owning MPTCP 1783 * socket, even if for additional subflows the allocation is performed 1784 * by a kernel workqueue. Adjust inode references, so that the 1785 * procfs/diag interfaces really show this one belonging to the correct 1786 * user. 1787 */ 1788 SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino; 1789 SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid; 1790 SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid; 1791 1792 subflow = mptcp_subflow_ctx(sf->sk); 1793 pr_debug("subflow=%p\n", subflow); 1794 1795 *new_sock = sf; 1796 sock_hold(sk); 1797 subflow->conn = sk; 1798 mptcp_subflow_ops_override(sf->sk); 1799 1800 return 0; 1801 1802 err_free: 1803 release_sock(sf->sk); 1804 sock_release(sf); 1805 return err; 1806 } 1807 1808 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk, 1809 gfp_t priority) 1810 { 1811 struct inet_connection_sock *icsk = inet_csk(sk); 1812 struct mptcp_subflow_context *ctx; 1813 1814 ctx = kzalloc(sizeof(*ctx), priority); 1815 if (!ctx) 1816 return NULL; 1817 1818 rcu_assign_pointer(icsk->icsk_ulp_data, ctx); 1819 INIT_LIST_HEAD(&ctx->node); 1820 INIT_LIST_HEAD(&ctx->delegated_node); 1821 1822 pr_debug("subflow=%p\n", ctx); 1823 1824 ctx->tcp_sock = sk; 1825 WRITE_ONCE(ctx->local_id, -1); 1826 1827 return ctx; 1828 } 1829 1830 static void __subflow_state_change(struct sock *sk) 1831 { 1832 struct socket_wq *wq; 1833 1834 rcu_read_lock(); 1835 wq = rcu_dereference(sk->sk_wq); 1836 if (skwq_has_sleeper(wq)) 1837 wake_up_interruptible_all(&wq->wait); 1838 rcu_read_unlock(); 1839 } 1840 1841 static bool subflow_is_done(const struct sock *sk) 1842 { 1843 return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE; 1844 } 1845 1846 static void subflow_state_change(struct sock *sk) 1847 { 1848 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk); 1849 struct sock *parent = subflow->conn; 1850 struct mptcp_sock *msk; 1851 1852 __subflow_state_change(sk); 1853 1854 msk = mptcp_sk(parent); 1855 if (subflow_simultaneous_connect(sk)) { 1856 mptcp_do_fallback(sk); 1857 pr_fallback(msk); 1858 subflow->conn_finished = 1; 1859 mptcp_propagate_state(parent, sk, subflow, NULL); 1860 } 1861 1862 /* as recvmsg() does not acquire the subflow socket for ssk selection 1863 * a fin packet carrying a DSS can be unnoticed if we don't trigger 1864 * the data available machinery here. 1865 */ 1866 if (mptcp_subflow_data_available(sk)) 1867 mptcp_data_ready(parent, sk); 1868 else if (unlikely(sk->sk_err)) 1869 subflow_error_report(sk); 1870 1871 subflow_sched_work_if_closed(mptcp_sk(parent), sk); 1872 1873 /* when the fallback subflow closes the rx side, trigger a 'dummy' 1874 * ingress data fin, so that the msk state will follow along 1875 */ 1876 if (__mptcp_check_fallback(msk) && subflow_is_done(sk) && msk->first == sk && 1877 mptcp_update_rcv_data_fin(msk, READ_ONCE(msk->ack_seq), true)) 1878 mptcp_schedule_work(parent); 1879 } 1880 1881 void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_ssk) 1882 { 1883 struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue; 1884 struct request_sock *req, *head, *tail; 1885 struct mptcp_subflow_context *subflow; 1886 struct sock *sk, *ssk; 1887 1888 /* Due to lock dependencies no relevant lock can be acquired under rskq_lock. 1889 * Splice the req list, so that accept() can not reach the pending ssk after 1890 * the listener socket is released below. 1891 */ 1892 spin_lock_bh(&queue->rskq_lock); 1893 head = queue->rskq_accept_head; 1894 tail = queue->rskq_accept_tail; 1895 queue->rskq_accept_head = NULL; 1896 queue->rskq_accept_tail = NULL; 1897 spin_unlock_bh(&queue->rskq_lock); 1898 1899 if (!head) 1900 return; 1901 1902 /* can't acquire the msk socket lock under the subflow one, 1903 * or will cause ABBA deadlock 1904 */ 1905 release_sock(listener_ssk); 1906 1907 for (req = head; req; req = req->dl_next) { 1908 ssk = req->sk; 1909 if (!sk_is_mptcp(ssk)) 1910 continue; 1911 1912 subflow = mptcp_subflow_ctx(ssk); 1913 if (!subflow || !subflow->conn) 1914 continue; 1915 1916 sk = subflow->conn; 1917 sock_hold(sk); 1918 1919 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 1920 __mptcp_unaccepted_force_close(sk); 1921 release_sock(sk); 1922 1923 /* lockdep will report a false positive ABBA deadlock 1924 * between cancel_work_sync and the listener socket. 1925 * The involved locks belong to different sockets WRT 1926 * the existing AB chain. 1927 * Using a per socket key is problematic as key 1928 * deregistration requires process context and must be 1929 * performed at socket disposal time, in atomic 1930 * context. 1931 * Just tell lockdep to consider the listener socket 1932 * released here. 1933 */ 1934 mutex_release(&listener_sk->sk_lock.dep_map, _RET_IP_); 1935 mptcp_cancel_work(sk); 1936 mutex_acquire(&listener_sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1937 1938 sock_put(sk); 1939 } 1940 1941 /* we are still under the listener msk socket lock */ 1942 lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING); 1943 1944 /* restore the listener queue, to let the TCP code clean it up */ 1945 spin_lock_bh(&queue->rskq_lock); 1946 WARN_ON_ONCE(queue->rskq_accept_head); 1947 queue->rskq_accept_head = head; 1948 queue->rskq_accept_tail = tail; 1949 spin_unlock_bh(&queue->rskq_lock); 1950 } 1951 1952 static int subflow_ulp_init(struct sock *sk) 1953 { 1954 struct inet_connection_sock *icsk = inet_csk(sk); 1955 struct mptcp_subflow_context *ctx; 1956 struct tcp_sock *tp = tcp_sk(sk); 1957 int err = 0; 1958 1959 /* disallow attaching ULP to a socket unless it has been 1960 * created with sock_create_kern() 1961 */ 1962 if (!sk->sk_kern_sock) { 1963 err = -EOPNOTSUPP; 1964 goto out; 1965 } 1966 1967 ctx = subflow_create_ctx(sk, GFP_KERNEL); 1968 if (!ctx) { 1969 err = -ENOMEM; 1970 goto out; 1971 } 1972 1973 pr_debug("subflow=%p, family=%d\n", ctx, sk->sk_family); 1974 1975 tp->is_mptcp = 1; 1976 ctx->icsk_af_ops = icsk->icsk_af_ops; 1977 icsk->icsk_af_ops = subflow_default_af_ops(sk); 1978 ctx->tcp_state_change = sk->sk_state_change; 1979 ctx->tcp_error_report = sk->sk_error_report; 1980 1981 WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable); 1982 WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space); 1983 1984 sk->sk_data_ready = subflow_data_ready; 1985 sk->sk_write_space = subflow_write_space; 1986 sk->sk_state_change = subflow_state_change; 1987 sk->sk_error_report = subflow_error_report; 1988 out: 1989 return err; 1990 } 1991 1992 static void subflow_ulp_release(struct sock *ssk) 1993 { 1994 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk); 1995 bool release = true; 1996 struct sock *sk; 1997 1998 if (!ctx) 1999 return; 2000 2001 sk = ctx->conn; 2002 if (sk) { 2003 /* if the msk has been orphaned, keep the ctx 2004 * alive, will be freed by __mptcp_close_ssk(), 2005 * when the subflow is still unaccepted 2006 */ 2007 release = ctx->disposable || list_empty(&ctx->node); 2008 2009 /* inet_child_forget() does not call sk_state_change(), 2010 * explicitly trigger the socket close machinery 2011 */ 2012 if (!release && !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, 2013 &mptcp_sk(sk)->flags)) 2014 mptcp_schedule_work(sk); 2015 sock_put(sk); 2016 } 2017 2018 mptcp_subflow_ops_undo_override(ssk); 2019 if (release) 2020 kfree_rcu(ctx, rcu); 2021 } 2022 2023 static void subflow_ulp_clone(const struct request_sock *req, 2024 struct sock *newsk, 2025 const gfp_t priority) 2026 { 2027 struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req); 2028 struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk); 2029 struct mptcp_subflow_context *new_ctx; 2030 2031 if (!tcp_rsk(req)->is_mptcp || 2032 (!subflow_req->mp_capable && !subflow_req->mp_join)) { 2033 subflow_ulp_fallback(newsk, old_ctx); 2034 return; 2035 } 2036 2037 new_ctx = subflow_create_ctx(newsk, priority); 2038 if (!new_ctx) { 2039 subflow_ulp_fallback(newsk, old_ctx); 2040 return; 2041 } 2042 2043 new_ctx->conn_finished = 1; 2044 new_ctx->icsk_af_ops = old_ctx->icsk_af_ops; 2045 new_ctx->tcp_state_change = old_ctx->tcp_state_change; 2046 new_ctx->tcp_error_report = old_ctx->tcp_error_report; 2047 new_ctx->rel_write_seq = 1; 2048 new_ctx->tcp_sock = newsk; 2049 2050 if (subflow_req->mp_capable) { 2051 /* see comments in subflow_syn_recv_sock(), MPTCP connection 2052 * is fully established only after we receive the remote key 2053 */ 2054 new_ctx->mp_capable = 1; 2055 new_ctx->local_key = subflow_req->local_key; 2056 new_ctx->token = subflow_req->token; 2057 new_ctx->ssn_offset = subflow_req->ssn_offset; 2058 new_ctx->idsn = subflow_req->idsn; 2059 2060 /* this is the first subflow, id is always 0 */ 2061 subflow_set_local_id(new_ctx, 0); 2062 } else if (subflow_req->mp_join) { 2063 new_ctx->ssn_offset = subflow_req->ssn_offset; 2064 new_ctx->mp_join = 1; 2065 new_ctx->fully_established = 1; 2066 new_ctx->remote_key_valid = 1; 2067 new_ctx->backup = subflow_req->backup; 2068 new_ctx->request_bkup = subflow_req->request_bkup; 2069 WRITE_ONCE(new_ctx->remote_id, subflow_req->remote_id); 2070 new_ctx->token = subflow_req->token; 2071 new_ctx->thmac = subflow_req->thmac; 2072 2073 /* the subflow req id is valid, fetched via subflow_check_req() 2074 * and subflow_token_join_request() 2075 */ 2076 subflow_set_local_id(new_ctx, subflow_req->local_id); 2077 } 2078 } 2079 2080 static void tcp_release_cb_override(struct sock *ssk) 2081 { 2082 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk); 2083 long status; 2084 2085 /* process and clear all the pending actions, but leave the subflow into 2086 * the napi queue. To respect locking, only the same CPU that originated 2087 * the action can touch the list. mptcp_napi_poll will take care of it. 2088 */ 2089 status = set_mask_bits(&subflow->delegated_status, MPTCP_DELEGATE_ACTIONS_MASK, 0); 2090 if (status) 2091 mptcp_subflow_process_delegated(ssk, status); 2092 2093 tcp_release_cb(ssk); 2094 } 2095 2096 static int tcp_abort_override(struct sock *ssk, int err) 2097 { 2098 /* closing a listener subflow requires a great deal of care. 2099 * keep it simple and just prevent such operation 2100 */ 2101 if (inet_sk_state_load(ssk) == TCP_LISTEN) 2102 return -EINVAL; 2103 2104 return tcp_abort(ssk, err); 2105 } 2106 2107 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = { 2108 .name = "mptcp", 2109 .owner = THIS_MODULE, 2110 .init = subflow_ulp_init, 2111 .release = subflow_ulp_release, 2112 .clone = subflow_ulp_clone, 2113 }; 2114 2115 static int subflow_ops_init(struct request_sock_ops *subflow_ops) 2116 { 2117 subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock); 2118 2119 subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name, 2120 subflow_ops->obj_size, 0, 2121 SLAB_ACCOUNT | 2122 SLAB_TYPESAFE_BY_RCU, 2123 NULL); 2124 if (!subflow_ops->slab) 2125 return -ENOMEM; 2126 2127 return 0; 2128 } 2129 2130 void __init mptcp_subflow_init(void) 2131 { 2132 mptcp_subflow_v4_request_sock_ops = tcp_request_sock_ops; 2133 mptcp_subflow_v4_request_sock_ops.slab_name = "request_sock_subflow_v4"; 2134 mptcp_subflow_v4_request_sock_ops.destructor = subflow_v4_req_destructor; 2135 2136 if (subflow_ops_init(&mptcp_subflow_v4_request_sock_ops) != 0) 2137 panic("MPTCP: failed to init subflow v4 request sock ops\n"); 2138 2139 subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops; 2140 subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req; 2141 subflow_request_sock_ipv4_ops.send_synack = subflow_v4_send_synack; 2142 2143 subflow_specific = ipv4_specific; 2144 subflow_specific.conn_request = subflow_v4_conn_request; 2145 subflow_specific.syn_recv_sock = subflow_syn_recv_sock; 2146 subflow_specific.sk_rx_dst_set = subflow_finish_connect; 2147 subflow_specific.rebuild_header = subflow_rebuild_header; 2148 2149 tcp_prot_override = tcp_prot; 2150 tcp_prot_override.release_cb = tcp_release_cb_override; 2151 tcp_prot_override.diag_destroy = tcp_abort_override; 2152 2153 #if IS_ENABLED(CONFIG_MPTCP_IPV6) 2154 /* In struct mptcp_subflow_request_sock, we assume the TCP request sock 2155 * structures for v4 and v6 have the same size. It should not changed in 2156 * the future but better to make sure to be warned if it is no longer 2157 * the case. 2158 */ 2159 BUILD_BUG_ON(sizeof(struct tcp_request_sock) != sizeof(struct tcp6_request_sock)); 2160 2161 mptcp_subflow_v6_request_sock_ops = tcp6_request_sock_ops; 2162 mptcp_subflow_v6_request_sock_ops.slab_name = "request_sock_subflow_v6"; 2163 mptcp_subflow_v6_request_sock_ops.destructor = subflow_v6_req_destructor; 2164 2165 if (subflow_ops_init(&mptcp_subflow_v6_request_sock_ops) != 0) 2166 panic("MPTCP: failed to init subflow v6 request sock ops\n"); 2167 2168 subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops; 2169 subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req; 2170 subflow_request_sock_ipv6_ops.send_synack = subflow_v6_send_synack; 2171 2172 subflow_v6_specific = ipv6_specific; 2173 subflow_v6_specific.conn_request = subflow_v6_conn_request; 2174 subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock; 2175 subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect; 2176 subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header; 2177 2178 subflow_v6m_specific = subflow_v6_specific; 2179 subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit; 2180 subflow_v6m_specific.send_check = ipv4_specific.send_check; 2181 subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len; 2182 subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced; 2183 subflow_v6m_specific.rebuild_header = subflow_rebuild_header; 2184 2185 tcpv6_prot_override = tcpv6_prot; 2186 tcpv6_prot_override.release_cb = tcp_release_cb_override; 2187 tcpv6_prot_override.diag_destroy = tcp_abort_override; 2188 #endif 2189 2190 mptcp_diag_subflow_init(&subflow_ulp_ops); 2191 2192 if (tcp_register_ulp(&subflow_ulp_ops) != 0) 2193 panic("MPTCP: failed to register subflows to ULP\n"); 2194 } 2195