net/mptcp/subflow.c

   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/algapi.h>
  13 #include <crypto/sha2.h>
  14 #include <net/sock.h>
  15 #include <net/inet_common.h>
  16 #include <net/inet_hashtables.h>
  17 #include <net/protocol.h>
  18 #include <net/tcp.h>
  19 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
  20 #include <net/ip6_route.h>
  21 #include <net/transp_v6.h>
  22 #endif
  23 #include <net/mptcp.h>
  24 #include <uapi/linux/mptcp.h>
  25 #include "protocol.h"
  26 #include "mib.h"
  27
  28 #include <trace/events/mptcp.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", subflow_req);
  43
  44         if (subflow_req->msk)
  45                 sock_put((struct sock *)subflow_req->msk);
  46
  47         mptcp_token_destroy_request(req);
  48         tcp_request_sock_ops.destructor(req);
  49 }
  50
  51 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
  52                                   void *hmac)
  53 {
  54         u8 msg[8];
  55
  56         put_unaligned_be32(nonce1, &msg[0]);
  57         put_unaligned_be32(nonce2, &msg[4]);
  58
  59         mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
  60 }
  61
  62 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
  63 {
  64         return mptcp_is_fully_established((void *)msk) &&
  65                READ_ONCE(msk->pm.accept_subflow);
  66 }
  67
  68 /* validate received token and create truncated hmac and nonce for SYN-ACK */
  69 static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req)
  70 {
  71         struct mptcp_sock *msk = subflow_req->msk;
  72         u8 hmac[SHA256_DIGEST_SIZE];
  73
  74         get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
  75
  76         subflow_generate_hmac(msk->local_key, msk->remote_key,
  77                               subflow_req->local_nonce,
  78                               subflow_req->remote_nonce, hmac);
  79
  80         subflow_req->thmac = get_unaligned_be64(hmac);
  81 }
  82
  83 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req)
  84 {
  85         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  86         struct mptcp_sock *msk;
  87         int local_id;
  88
  89         msk = mptcp_token_get_sock(subflow_req->token);
  90         if (!msk) {
  91                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
  92                 return NULL;
  93         }
  94
  95         local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
  96         if (local_id < 0) {
  97                 sock_put((struct sock *)msk);
  98                 return NULL;
  99         }
 100         subflow_req->local_id = local_id;
 101
 102         return msk;
 103 }
 104
 105 static void subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
 106 {
 107         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 108
 109         subflow_req->mp_capable = 0;
 110         subflow_req->mp_join = 0;
 111         subflow_req->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk_listener));
 112         subflow_req->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk_listener));
 113         subflow_req->msk = NULL;
 114         mptcp_token_init_request(req);
 115 }
 116
 117 static bool subflow_use_different_sport(struct mptcp_sock *msk, const struct sock *sk)
 118 {
 119         return inet_sk(sk)->inet_sport != inet_sk((struct sock *)msk)->inet_sport;
 120 }
 121
 122 static void subflow_add_reset_reason(struct sk_buff *skb, u8 reason)
 123 {
 124         struct mptcp_ext *mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
 125
 126         if (mpext) {
 127                 memset(mpext, 0, sizeof(*mpext));
 128                 mpext->reset_reason = reason;
 129         }
 130 }
 131
 132 /* Init mptcp request socket.
 133  *
 134  * Returns an error code if a JOIN has failed and a TCP reset
 135  * should be sent.
 136  */
 137 static int subflow_check_req(struct request_sock *req,
 138                              const struct sock *sk_listener,
 139                              struct sk_buff *skb)
 140 {
 141         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 142         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 143         struct mptcp_options_received mp_opt;
 144
 145         pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
 146
 147 #ifdef CONFIG_TCP_MD5SIG
 148         /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
 149          * TCP option space.
 150          */
 151         if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
 152                 return -EINVAL;
 153 #endif
 154
 155         mptcp_get_options(sk_listener, skb, &mp_opt);
 156
 157         if (mp_opt.mp_capable) {
 158                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
 159
 160                 if (mp_opt.mp_join)
 161                         return 0;
 162         } else if (mp_opt.mp_join) {
 163                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
 164         }
 165
 166         if (mp_opt.mp_capable && listener->request_mptcp) {
 167                 int err, retries = MPTCP_TOKEN_MAX_RETRIES;
 168
 169                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 170 again:
 171                 do {
 172                         get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
 173                 } while (subflow_req->local_key == 0);
 174
 175                 if (unlikely(req->syncookie)) {
 176                         mptcp_crypto_key_sha(subflow_req->local_key,
 177                                              &subflow_req->token,
 178                                              &subflow_req->idsn);
 179                         if (mptcp_token_exists(subflow_req->token)) {
 180                                 if (retries-- > 0)
 181                                         goto again;
 182                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
 183                         } else {
 184                                 subflow_req->mp_capable = 1;
 185                         }
 186                         return 0;
 187                 }
 188
 189                 err = mptcp_token_new_request(req);
 190                 if (err == 0)
 191                         subflow_req->mp_capable = 1;
 192                 else if (retries-- > 0)
 193                         goto again;
 194                 else
 195                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
 196
 197         } else if (mp_opt.mp_join && listener->request_mptcp) {
 198                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 199                 subflow_req->mp_join = 1;
 200                 subflow_req->backup = mp_opt.backup;
 201                 subflow_req->remote_id = mp_opt.join_id;
 202                 subflow_req->token = mp_opt.token;
 203                 subflow_req->remote_nonce = mp_opt.nonce;
 204                 subflow_req->msk = subflow_token_join_request(req);
 205
 206                 /* Can't fall back to TCP in this case. */
 207                 if (!subflow_req->msk) {
 208                         subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 209                         return -EPERM;
 210                 }
 211
 212                 if (subflow_use_different_sport(subflow_req->msk, sk_listener)) {
 213                         pr_debug("syn inet_sport=%d %d",
 214                                  ntohs(inet_sk(sk_listener)->inet_sport),
 215                                  ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport));
 216                         if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) {
 217                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX);
 218                                 return -EPERM;
 219                         }
 220                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX);
 221                 }
 222
 223                 subflow_req_create_thmac(subflow_req);
 224
 225                 if (unlikely(req->syncookie)) {
 226                         if (mptcp_can_accept_new_subflow(subflow_req->msk))
 227                                 subflow_init_req_cookie_join_save(subflow_req, skb);
 228                         else
 229                                 return -EPERM;
 230                 }
 231
 232                 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
 233                          subflow_req->remote_nonce, subflow_req->msk);
 234         }
 235
 236         return 0;
 237 }
 238
 239 int mptcp_subflow_init_cookie_req(struct request_sock *req,
 240                                   const struct sock *sk_listener,
 241                                   struct sk_buff *skb)
 242 {
 243         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 244         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 245         struct mptcp_options_received mp_opt;
 246         int err;
 247
 248         subflow_init_req(req, sk_listener);
 249         mptcp_get_options(sk_listener, skb, &mp_opt);
 250
 251         if (mp_opt.mp_capable && mp_opt.mp_join)
 252                 return -EINVAL;
 253
 254         if (mp_opt.mp_capable && listener->request_mptcp) {
 255                 if (mp_opt.sndr_key == 0)
 256                         return -EINVAL;
 257
 258                 subflow_req->local_key = mp_opt.rcvr_key;
 259                 err = mptcp_token_new_request(req);
 260                 if (err)
 261                         return err;
 262
 263                 subflow_req->mp_capable = 1;
 264                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 265         } else if (mp_opt.mp_join && listener->request_mptcp) {
 266                 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
 267                         return -EINVAL;
 268
 269                 subflow_req->mp_join = 1;
 270                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 271         }
 272
 273         return 0;
 274 }
 275 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
 276
 277 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
 278                                               struct sk_buff *skb,
 279                                               struct flowi *fl,
 280                                               struct request_sock *req)
 281 {
 282         struct dst_entry *dst;
 283         int err;
 284
 285         tcp_rsk(req)->is_mptcp = 1;
 286         subflow_init_req(req, sk);
 287
 288         dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req);
 289         if (!dst)
 290                 return NULL;
 291
 292         err = subflow_check_req(req, sk, skb);
 293         if (err == 0)
 294                 return dst;
 295
 296         dst_release(dst);
 297         if (!req->syncookie)
 298                 tcp_request_sock_ops.send_reset(sk, skb);
 299         return NULL;
 300 }
 301
 302 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 303 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
 304                                               struct sk_buff *skb,
 305                                               struct flowi *fl,
 306                                               struct request_sock *req)
 307 {
 308         struct dst_entry *dst;
 309         int err;
 310
 311         tcp_rsk(req)->is_mptcp = 1;
 312         subflow_init_req(req, sk);
 313
 314         dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req);
 315         if (!dst)
 316                 return NULL;
 317
 318         err = subflow_check_req(req, sk, skb);
 319         if (err == 0)
 320                 return dst;
 321
 322         dst_release(dst);
 323         if (!req->syncookie)
 324                 tcp6_request_sock_ops.send_reset(sk, skb);
 325         return NULL;
 326 }
 327 #endif
 328
 329 /* validate received truncated hmac and create hmac for third ACK */
 330 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
 331 {
 332         u8 hmac[SHA256_DIGEST_SIZE];
 333         u64 thmac;
 334
 335         subflow_generate_hmac(subflow->remote_key, subflow->local_key,
 336                               subflow->remote_nonce, subflow->local_nonce,
 337                               hmac);
 338
 339         thmac = get_unaligned_be64(hmac);
 340         pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
 341                  subflow, subflow->token,
 342                  (unsigned long long)thmac,
 343                  (unsigned long long)subflow->thmac);
 344
 345         return thmac == subflow->thmac;
 346 }
 347
 348 void mptcp_subflow_reset(struct sock *ssk)
 349 {
 350         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 351         struct sock *sk = subflow->conn;
 352
 353         /* must hold: tcp_done() could drop last reference on parent */
 354         sock_hold(sk);
 355
 356         tcp_set_state(ssk, TCP_CLOSE);
 357         tcp_send_active_reset(ssk, GFP_ATOMIC);
 358         tcp_done(ssk);
 359         if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags) &&
 360             schedule_work(&mptcp_sk(sk)->work))
 361                 return; /* worker will put sk for us */
 362
 363         sock_put(sk);
 364 }
 365
 366 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
 367 {
 368         return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
 369 }
 370
 371 void __mptcp_set_connected(struct sock *sk)
 372 {
 373         if (sk->sk_state == TCP_SYN_SENT) {
 374                 inet_sk_state_store(sk, TCP_ESTABLISHED);
 375                 sk->sk_state_change(sk);
 376         }
 377 }
 378
 379 static void mptcp_set_connected(struct sock *sk)
 380 {
 381         mptcp_data_lock(sk);
 382         if (!sock_owned_by_user(sk))
 383                 __mptcp_set_connected(sk);
 384         else
 385                 set_bit(MPTCP_CONNECTED, &mptcp_sk(sk)->flags);
 386         mptcp_data_unlock(sk);
 387 }
 388
 389 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 390 {
 391         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 392         struct mptcp_options_received mp_opt;
 393         struct sock *parent = subflow->conn;
 394
 395         subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 396
 397
 398         /* be sure no special action on any packet other than syn-ack */
 399         if (subflow->conn_finished)
 400                 return;
 401
 402         mptcp_propagate_sndbuf(parent, sk);
 403         subflow->rel_write_seq = 1;
 404         subflow->conn_finished = 1;
 405         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 406         pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
 407
 408         mptcp_get_options(sk, skb, &mp_opt);
 409         if (subflow->request_mptcp) {
 410                 if (!mp_opt.mp_capable) {
 411                         MPTCP_INC_STATS(sock_net(sk),
 412                                         MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
 413                         mptcp_do_fallback(sk);
 414                         pr_fallback(mptcp_sk(subflow->conn));
 415                         goto fallback;
 416                 }
 417
 418                 if (mp_opt.csum_reqd)
 419                         WRITE_ONCE(mptcp_sk(parent)->csum_enabled, true);
 420                 if (mp_opt.deny_join_id0)
 421                         WRITE_ONCE(mptcp_sk(parent)->pm.remote_deny_join_id0, true);
 422                 subflow->mp_capable = 1;
 423                 subflow->can_ack = 1;
 424                 subflow->remote_key = mp_opt.sndr_key;
 425                 pr_debug("subflow=%p, remote_key=%llu", subflow,
 426                          subflow->remote_key);
 427                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
 428                 mptcp_finish_connect(sk);
 429                 mptcp_set_connected(parent);
 430         } else if (subflow->request_join) {
 431                 u8 hmac[SHA256_DIGEST_SIZE];
 432
 433                 if (!mp_opt.mp_join) {
 434                         subflow->reset_reason = MPTCP_RST_EMPTCP;
 435                         goto do_reset;
 436                 }
 437
 438                 subflow->thmac = mp_opt.thmac;
 439                 subflow->remote_nonce = mp_opt.nonce;
 440                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
 441                          subflow->thmac, subflow->remote_nonce);
 442
 443                 if (!subflow_thmac_valid(subflow)) {
 444                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 445                         subflow->reset_reason = MPTCP_RST_EMPTCP;
 446                         goto do_reset;
 447                 }
 448
 449                 if (!mptcp_finish_join(sk))
 450                         goto do_reset;
 451
 452                 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 453                                       subflow->local_nonce,
 454                                       subflow->remote_nonce,
 455                                       hmac);
 456                 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
 457
 458                 subflow->mp_join = 1;
 459                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 460
 461                 if (subflow_use_different_dport(mptcp_sk(parent), sk)) {
 462                         pr_debug("synack inet_dport=%d %d",
 463                                  ntohs(inet_sk(sk)->inet_dport),
 464                                  ntohs(inet_sk(parent)->inet_dport));
 465                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
 466                 }
 467         } else if (mptcp_check_fallback(sk)) {
 468 fallback:
 469                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
 470                 mptcp_set_connected(parent);
 471         }
 472         return;
 473
 474 do_reset:
 475         subflow->reset_transient = 0;
 476         mptcp_subflow_reset(sk);
 477 }
 478
 479 struct request_sock_ops mptcp_subflow_request_sock_ops;
 480 EXPORT_SYMBOL_GPL(mptcp_subflow_request_sock_ops);
 481 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
 482
 483 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 484 {
 485         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 486
 487         pr_debug("subflow=%p", subflow);
 488
 489         /* Never answer to SYNs sent to broadcast or multicast */
 490         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 491                 goto drop;
 492
 493         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 494                                 &subflow_request_sock_ipv4_ops,
 495                                 sk, skb);
 496 drop:
 497         tcp_listendrop(sk);
 498         return 0;
 499 }
 500
 501 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 502 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
 503 static struct inet_connection_sock_af_ops subflow_v6_specific;
 504 static struct inet_connection_sock_af_ops subflow_v6m_specific;
 505 static struct proto tcpv6_prot_override;
 506
 507 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 508 {
 509         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 510
 511         pr_debug("subflow=%p", subflow);
 512
 513         if (skb->protocol == htons(ETH_P_IP))
 514                 return subflow_v4_conn_request(sk, skb);
 515
 516         if (!ipv6_unicast_destination(skb))
 517                 goto drop;
 518
 519         if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
 520                 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
 521                 return 0;
 522         }
 523
 524         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 525                                 &subflow_request_sock_ipv6_ops, sk, skb);
 526
 527 drop:
 528         tcp_listendrop(sk);
 529         return 0; /* don't send reset */
 530 }
 531 #endif
 532
 533 /* validate hmac received in third ACK */
 534 static bool subflow_hmac_valid(const struct request_sock *req,
 535                                const struct mptcp_options_received *mp_opt)
 536 {
 537         const struct mptcp_subflow_request_sock *subflow_req;
 538         u8 hmac[SHA256_DIGEST_SIZE];
 539         struct mptcp_sock *msk;
 540
 541         subflow_req = mptcp_subflow_rsk(req);
 542         msk = subflow_req->msk;
 543         if (!msk)
 544                 return false;
 545
 546         subflow_generate_hmac(msk->remote_key, msk->local_key,
 547                               subflow_req->remote_nonce,
 548                               subflow_req->local_nonce, hmac);
 549
 550         return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
 551 }
 552
 553 static void mptcp_sock_destruct(struct sock *sk)
 554 {
 555         /* if new mptcp socket isn't accepted, it is free'd
 556          * from the tcp listener sockets request queue, linked
 557          * from req->sk.  The tcp socket is released.
 558          * This calls the ULP release function which will
 559          * also remove the mptcp socket, via
 560          * sock_put(ctx->conn).
 561          *
 562          * Problem is that the mptcp socket will be in
 563          * ESTABLISHED state and will not have the SOCK_DEAD flag.
 564          * Both result in warnings from inet_sock_destruct.
 565          */
 566         if ((1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
 567                 sk->sk_state = TCP_CLOSE;
 568                 WARN_ON_ONCE(sk->sk_socket);
 569                 sock_orphan(sk);
 570         }
 571
 572         mptcp_destroy_common(mptcp_sk(sk));
 573         inet_sock_destruct(sk);
 574 }
 575
 576 static void mptcp_force_close(struct sock *sk)
 577 {
 578         /* the msk is not yet exposed to user-space */
 579         inet_sk_state_store(sk, TCP_CLOSE);
 580         sk_common_release(sk);
 581 }
 582
 583 static void subflow_ulp_fallback(struct sock *sk,
 584                                  struct mptcp_subflow_context *old_ctx)
 585 {
 586         struct inet_connection_sock *icsk = inet_csk(sk);
 587
 588         mptcp_subflow_tcp_fallback(sk, old_ctx);
 589         icsk->icsk_ulp_ops = NULL;
 590         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 591         tcp_sk(sk)->is_mptcp = 0;
 592
 593         mptcp_subflow_ops_undo_override(sk);
 594 }
 595
 596 static void subflow_drop_ctx(struct sock *ssk)
 597 {
 598         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 599
 600         if (!ctx)
 601                 return;
 602
 603         subflow_ulp_fallback(ssk, ctx);
 604         if (ctx->conn)
 605                 sock_put(ctx->conn);
 606
 607         kfree_rcu(ctx, rcu);
 608 }
 609
 610 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 611                                      struct mptcp_options_received *mp_opt)
 612 {
 613         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 614
 615         subflow->remote_key = mp_opt->sndr_key;
 616         subflow->fully_established = 1;
 617         subflow->can_ack = 1;
 618         WRITE_ONCE(msk->fully_established, true);
 619 }
 620
 621 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 622                                           struct sk_buff *skb,
 623                                           struct request_sock *req,
 624                                           struct dst_entry *dst,
 625                                           struct request_sock *req_unhash,
 626                                           bool *own_req)
 627 {
 628         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 629         struct mptcp_subflow_request_sock *subflow_req;
 630         struct mptcp_options_received mp_opt;
 631         bool fallback, fallback_is_fatal;
 632         struct sock *new_msk = NULL;
 633         struct sock *child;
 634
 635         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 636
 637         /* After child creation we must look for 'mp_capable' even when options
 638          * are not parsed
 639          */
 640         mp_opt.mp_capable = 0;
 641
 642         /* hopefully temporary handling for MP_JOIN+syncookie */
 643         subflow_req = mptcp_subflow_rsk(req);
 644         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 645         fallback = !tcp_rsk(req)->is_mptcp;
 646         if (fallback)
 647                 goto create_child;
 648
 649         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 650         if (subflow_req->mp_capable) {
 651                 /* we can receive and accept an in-window, out-of-order pkt,
 652                  * which may not carry the MP_CAPABLE opt even on mptcp enabled
 653                  * paths: always try to extract the peer key, and fallback
 654                  * for packets missing it.
 655                  * Even OoO DSS packets coming legitly after dropped or
 656                  * reordered MPC will cause fallback, but we don't have other
 657                  * options.
 658                  */
 659                 mptcp_get_options(sk, skb, &mp_opt);
 660                 if (!mp_opt.mp_capable) {
 661                         fallback = true;
 662                         goto create_child;
 663                 }
 664
 665                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 666                 if (!new_msk)
 667                         fallback = true;
 668         } else if (subflow_req->mp_join) {
 669                 mptcp_get_options(sk, skb, &mp_opt);
 670                 if (!mp_opt.mp_join || !subflow_hmac_valid(req, &mp_opt) ||
 671                     !mptcp_can_accept_new_subflow(subflow_req->msk)) {
 672                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 673                         fallback = true;
 674                 }
 675         }
 676
 677 create_child:
 678         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 679                                                      req_unhash, own_req);
 680
 681         if (child && *own_req) {
 682                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 683
 684                 tcp_rsk(req)->drop_req = false;
 685
 686                 /* we need to fallback on ctx allocation failure and on pre-reqs
 687                  * checking above. In the latter scenario we additionally need
 688                  * to reset the context to non MPTCP status.
 689                  */
 690                 if (!ctx || fallback) {
 691                         if (fallback_is_fatal) {
 692                                 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 693                                 goto dispose_child;
 694                         }
 695
 696                         subflow_drop_ctx(child);
 697                         goto out;
 698                 }
 699
 700                 /* ssk inherits options of listener sk */
 701                 ctx->setsockopt_seq = listener->setsockopt_seq;
 702
 703                 if (ctx->mp_capable) {
 704                         /* this can't race with mptcp_close(), as the msk is
 705                          * not yet exposted to user-space
 706                          */
 707                         inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
 708
 709                         /* record the newly created socket as the first msk
 710                          * subflow, but don't link it yet into conn_list
 711                          */
 712                         WRITE_ONCE(mptcp_sk(new_msk)->first, child);
 713
 714                         /* new mpc subflow takes ownership of the newly
 715                          * created mptcp socket
 716                          */
 717                         new_msk->sk_destruct = mptcp_sock_destruct;
 718                         mptcp_sk(new_msk)->setsockopt_seq = ctx->setsockopt_seq;
 719                         mptcp_pm_new_connection(mptcp_sk(new_msk), child, 1);
 720                         mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
 721                         ctx->conn = new_msk;
 722                         new_msk = NULL;
 723
 724                         /* with OoO packets we can reach here without ingress
 725                          * mpc option
 726                          */
 727                         if (mp_opt.mp_capable)
 728                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 729                 } else if (ctx->mp_join) {
 730                         struct mptcp_sock *owner;
 731
 732                         owner = subflow_req->msk;
 733                         if (!owner) {
 734                                 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 735                                 goto dispose_child;
 736                         }
 737
 738                         /* move the msk reference ownership to the subflow */
 739                         subflow_req->msk = NULL;
 740                         ctx->conn = (struct sock *)owner;
 741
 742                         if (subflow_use_different_sport(owner, sk)) {
 743                                 pr_debug("ack inet_sport=%d %d",
 744                                          ntohs(inet_sk(sk)->inet_sport),
 745                                          ntohs(inet_sk((struct sock *)owner)->inet_sport));
 746                                 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
 747                                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
 748                                         goto dispose_child;
 749                                 }
 750                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
 751                         }
 752
 753                         if (!mptcp_finish_join(child))
 754                                 goto dispose_child;
 755
 756                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 757                         tcp_rsk(req)->drop_req = true;
 758                 }
 759         }
 760
 761 out:
 762         /* dispose of the left over mptcp master, if any */
 763         if (unlikely(new_msk))
 764                 mptcp_force_close(new_msk);
 765
 766         /* check for expected invariant - should never trigger, just help
 767          * catching eariler subtle bugs
 768          */
 769         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 770                      (!mptcp_subflow_ctx(child) ||
 771                       !mptcp_subflow_ctx(child)->conn));
 772         return child;
 773
 774 dispose_child:
 775         subflow_drop_ctx(child);
 776         tcp_rsk(req)->drop_req = true;
 777         inet_csk_prepare_for_destroy_sock(child);
 778         tcp_done(child);
 779         req->rsk_ops->send_reset(sk, skb);
 780
 781         /* The last child reference will be released by the caller */
 782         return child;
 783 }
 784
 785 static struct inet_connection_sock_af_ops subflow_specific;
 786 static struct proto tcp_prot_override;
 787
 788 enum mapping_status {
 789         MAPPING_OK,
 790         MAPPING_INVALID,
 791         MAPPING_EMPTY,
 792         MAPPING_DATA_FIN,
 793         MAPPING_DUMMY
 794 };
 795
 796 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 797 {
 798         pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 799                  ssn, subflow->map_subflow_seq, subflow->map_data_len);
 800 }
 801
 802 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 803 {
 804         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 805         unsigned int skb_consumed;
 806
 807         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 808         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 809                 return true;
 810
 811         return skb->len - skb_consumed <= subflow->map_data_len -
 812                                           mptcp_subflow_get_map_offset(subflow);
 813 }
 814
 815 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 816 {
 817         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 818         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 819
 820         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 821                 /* Mapping covers data later in the subflow stream,
 822                  * currently unsupported.
 823                  */
 824                 dbg_bad_map(subflow, ssn);
 825                 return false;
 826         }
 827         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 828                                   subflow->map_data_len))) {
 829                 /* Mapping does covers past subflow data, invalid */
 830                 dbg_bad_map(subflow, ssn);
 831                 return false;
 832         }
 833         return true;
 834 }
 835
 836 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
 837                                               bool csum_reqd)
 838 {
 839         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 840         struct csum_pseudo_header header;
 841         u32 offset, seq, delta;
 842         __wsum csum;
 843         int len;
 844
 845         if (!csum_reqd)
 846                 return MAPPING_OK;
 847
 848         /* mapping already validated on previous traversal */
 849         if (subflow->map_csum_len == subflow->map_data_len)
 850                 return MAPPING_OK;
 851
 852         /* traverse the receive queue, ensuring it contains a full
 853          * DSS mapping and accumulating the related csum.
 854          * Preserve the accoumlate csum across multiple calls, to compute
 855          * the csum only once
 856          */
 857         delta = subflow->map_data_len - subflow->map_csum_len;
 858         for (;;) {
 859                 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
 860                 offset = seq - TCP_SKB_CB(skb)->seq;
 861
 862                 /* if the current skb has not been accounted yet, csum its contents
 863                  * up to the amount covered by the current DSS
 864                  */
 865                 if (offset < skb->len) {
 866                         __wsum csum;
 867
 868                         len = min(skb->len - offset, delta);
 869                         csum = skb_checksum(skb, offset, len, 0);
 870                         subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
 871                                                                 subflow->map_csum_len);
 872
 873                         delta -= len;
 874                         subflow->map_csum_len += len;
 875                 }
 876                 if (delta == 0)
 877                         break;
 878
 879                 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
 880                         /* if this subflow is closed, the partial mapping
 881                          * will be never completed; flush the pending skbs, so
 882                          * that subflow_sched_work_if_closed() can kick in
 883                          */
 884                         if (unlikely(ssk->sk_state == TCP_CLOSE))
 885                                 while ((skb = skb_peek(&ssk->sk_receive_queue)))
 886                                         sk_eat_skb(ssk, skb);
 887
 888                         /* not enough data to validate the csum */
 889                         return MAPPING_EMPTY;
 890                 }
 891
 892                 /* the DSS mapping for next skbs will be validated later,
 893                  * when a get_mapping_status call will process such skb
 894                  */
 895                 skb = skb->next;
 896         }
 897
 898         /* note that 'map_data_len' accounts only for the carried data, does
 899          * not include the eventual seq increment due to the data fin,
 900          * while the pseudo header requires the original DSS data len,
 901          * including that
 902          */
 903         header.data_seq = cpu_to_be64(subflow->map_seq);
 904         header.subflow_seq = htonl(subflow->map_subflow_seq);
 905         header.data_len = htons(subflow->map_data_len + subflow->map_data_fin);
 906         header.csum = 0;
 907
 908         csum = csum_partial(&header, sizeof(header), subflow->map_data_csum);
 909         if (unlikely(csum_fold(csum))) {
 910                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
 911                 return subflow->mp_join ? MAPPING_INVALID : MAPPING_DUMMY;
 912         }
 913
 914         return MAPPING_OK;
 915 }
 916
 917 static enum mapping_status get_mapping_status(struct sock *ssk,
 918                                               struct mptcp_sock *msk)
 919 {
 920         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 921         bool csum_reqd = READ_ONCE(msk->csum_enabled);
 922         struct mptcp_ext *mpext;
 923         struct sk_buff *skb;
 924         u16 data_len;
 925         u64 map_seq;
 926
 927         skb = skb_peek(&ssk->sk_receive_queue);
 928         if (!skb)
 929                 return MAPPING_EMPTY;
 930
 931         if (mptcp_check_fallback(ssk))
 932                 return MAPPING_DUMMY;
 933
 934         mpext = mptcp_get_ext(skb);
 935         if (!mpext || !mpext->use_map) {
 936                 if (!subflow->map_valid && !skb->len) {
 937                         /* the TCP stack deliver 0 len FIN pkt to the receive
 938                          * queue, that is the only 0len pkts ever expected here,
 939                          * and we can admit no mapping only for 0 len pkts
 940                          */
 941                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 942                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 943                                           TCP_SKB_CB(skb)->seq,
 944                                           TCP_SKB_CB(skb)->end_seq,
 945                                           TCP_SKB_CB(skb)->tcp_flags);
 946                         sk_eat_skb(ssk, skb);
 947                         return MAPPING_EMPTY;
 948                 }
 949
 950                 if (!subflow->map_valid)
 951                         return MAPPING_INVALID;
 952
 953                 goto validate_seq;
 954         }
 955
 956         trace_get_mapping_status(mpext);
 957
 958         data_len = mpext->data_len;
 959         if (data_len == 0) {
 960                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 961                 return MAPPING_INVALID;
 962         }
 963
 964         if (mpext->data_fin == 1) {
 965                 if (data_len == 1) {
 966                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
 967                                                                  mpext->dsn64);
 968                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
 969                         if (subflow->map_valid) {
 970                                 /* A DATA_FIN might arrive in a DSS
 971                                  * option before the previous mapping
 972                                  * has been fully consumed. Continue
 973                                  * handling the existing mapping.
 974                                  */
 975                                 skb_ext_del(skb, SKB_EXT_MPTCP);
 976                                 return MAPPING_OK;
 977                         } else {
 978                                 if (updated && schedule_work(&msk->work))
 979                                         sock_hold((struct sock *)msk);
 980
 981                                 return MAPPING_DATA_FIN;
 982                         }
 983                 } else {
 984                         u64 data_fin_seq = mpext->data_seq + data_len - 1;
 985
 986                         /* If mpext->data_seq is a 32-bit value, data_fin_seq
 987                          * must also be limited to 32 bits.
 988                          */
 989                         if (!mpext->dsn64)
 990                                 data_fin_seq &= GENMASK_ULL(31, 0);
 991
 992                         mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
 993                         pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
 994                                  data_fin_seq, mpext->dsn64);
 995                 }
 996
 997                 /* Adjust for DATA_FIN using 1 byte of sequence space */
 998                 data_len--;
 999         }
1000
1001         map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1002         WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1003
1004         if (subflow->map_valid) {
1005                 /* Allow replacing only with an identical map */
1006                 if (subflow->map_seq == map_seq &&
1007                     subflow->map_subflow_seq == mpext->subflow_seq &&
1008                     subflow->map_data_len == data_len &&
1009                     subflow->map_csum_reqd == mpext->csum_reqd) {
1010                         skb_ext_del(skb, SKB_EXT_MPTCP);
1011                         goto validate_csum;
1012                 }
1013
1014                 /* If this skb data are fully covered by the current mapping,
1015                  * the new map would need caching, which is not supported
1016                  */
1017                 if (skb_is_fully_mapped(ssk, skb)) {
1018                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1019                         return MAPPING_INVALID;
1020                 }
1021
1022                 /* will validate the next map after consuming the current one */
1023                 goto validate_csum;
1024         }
1025
1026         subflow->map_seq = map_seq;
1027         subflow->map_subflow_seq = mpext->subflow_seq;
1028         subflow->map_data_len = data_len;
1029         subflow->map_valid = 1;
1030         subflow->map_data_fin = mpext->data_fin;
1031         subflow->mpc_map = mpext->mpc_map;
1032         subflow->map_csum_reqd = mpext->csum_reqd;
1033         subflow->map_csum_len = 0;
1034         subflow->map_data_csum = csum_unfold(mpext->csum);
1035
1036         /* Cfr RFC 8684 Section 3.3.0 */
1037         if (unlikely(subflow->map_csum_reqd != csum_reqd))
1038                 return MAPPING_INVALID;
1039
1040         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
1041                  subflow->map_seq, subflow->map_subflow_seq,
1042                  subflow->map_data_len, subflow->map_csum_reqd,
1043                  subflow->map_data_csum);
1044
1045 validate_seq:
1046         /* we revalidate valid mapping on new skb, because we must ensure
1047          * the current skb is completely covered by the available mapping
1048          */
1049         if (!validate_mapping(ssk, skb)) {
1050                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1051                 return MAPPING_INVALID;
1052         }
1053
1054         skb_ext_del(skb, SKB_EXT_MPTCP);
1055
1056 validate_csum:
1057         return validate_data_csum(ssk, skb, csum_reqd);
1058 }
1059
1060 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1061                                        u64 limit)
1062 {
1063         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1064         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1065         u32 incr;
1066
1067         incr = limit >= skb->len ? skb->len + fin : limit;
1068
1069         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
1070                  subflow->map_subflow_seq);
1071         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1072         tcp_sk(ssk)->copied_seq += incr;
1073         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1074                 sk_eat_skb(ssk, skb);
1075         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1076                 subflow->map_valid = 0;
1077 }
1078
1079 /* sched mptcp worker to remove the subflow if no more data is pending */
1080 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1081 {
1082         struct sock *sk = (struct sock *)msk;
1083
1084         if (likely(ssk->sk_state != TCP_CLOSE))
1085                 return;
1086
1087         if (skb_queue_empty(&ssk->sk_receive_queue) &&
1088             !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags)) {
1089                 sock_hold(sk);
1090                 if (!schedule_work(&msk->work))
1091                         sock_put(sk);
1092         }
1093 }
1094
1095 static bool subflow_check_data_avail(struct sock *ssk)
1096 {
1097         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1098         enum mapping_status status;
1099         struct mptcp_sock *msk;
1100         struct sk_buff *skb;
1101
1102         if (!skb_peek(&ssk->sk_receive_queue))
1103                 WRITE_ONCE(subflow->data_avail, 0);
1104         if (subflow->data_avail)
1105                 return true;
1106
1107         msk = mptcp_sk(subflow->conn);
1108         for (;;) {
1109                 u64 ack_seq;
1110                 u64 old_ack;
1111
1112                 status = get_mapping_status(ssk, msk);
1113                 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1114                 if (unlikely(status == MAPPING_INVALID))
1115                         goto fallback;
1116
1117                 if (unlikely(status == MAPPING_DUMMY))
1118                         goto fallback;
1119
1120                 if (status != MAPPING_OK)
1121                         goto no_data;
1122
1123                 skb = skb_peek(&ssk->sk_receive_queue);
1124                 if (WARN_ON_ONCE(!skb))
1125                         goto no_data;
1126
1127                 /* if msk lacks the remote key, this subflow must provide an
1128                  * MP_CAPABLE-based mapping
1129                  */
1130                 if (unlikely(!READ_ONCE(msk->can_ack))) {
1131                         if (!subflow->mpc_map)
1132                                 goto fallback;
1133                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
1134                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
1135                         WRITE_ONCE(msk->can_ack, true);
1136                 }
1137
1138                 old_ack = READ_ONCE(msk->ack_seq);
1139                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1140                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1141                          ack_seq);
1142                 if (unlikely(before64(ack_seq, old_ack))) {
1143                         mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1144                         continue;
1145                 }
1146
1147                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1148                 break;
1149         }
1150         return true;
1151
1152 no_data:
1153         subflow_sched_work_if_closed(msk, ssk);
1154         return false;
1155
1156 fallback:
1157         /* RFC 8684 section 3.7. */
1158         if (subflow->mp_join || subflow->fully_established) {
1159                 /* fatal protocol error, close the socket.
1160                  * subflow_error_report() will introduce the appropriate barriers
1161                  */
1162                 ssk->sk_err = EBADMSG;
1163                 tcp_set_state(ssk, TCP_CLOSE);
1164                 subflow->reset_transient = 0;
1165                 subflow->reset_reason = MPTCP_RST_EMPTCP;
1166                 tcp_send_active_reset(ssk, GFP_ATOMIC);
1167                 WRITE_ONCE(subflow->data_avail, 0);
1168                 return false;
1169         }
1170
1171         __mptcp_do_fallback(msk);
1172         skb = skb_peek(&ssk->sk_receive_queue);
1173         subflow->map_valid = 1;
1174         subflow->map_seq = READ_ONCE(msk->ack_seq);
1175         subflow->map_data_len = skb->len;
1176         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1177         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1178         return true;
1179 }
1180
1181 bool mptcp_subflow_data_available(struct sock *sk)
1182 {
1183         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1184
1185         /* check if current mapping is still valid */
1186         if (subflow->map_valid &&
1187             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1188                 subflow->map_valid = 0;
1189                 WRITE_ONCE(subflow->data_avail, 0);
1190
1191                 pr_debug("Done with mapping: seq=%u data_len=%u",
1192                          subflow->map_subflow_seq,
1193                          subflow->map_data_len);
1194         }
1195
1196         return subflow_check_data_avail(sk);
1197 }
1198
1199 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1200  * not the ssk one.
1201  *
1202  * In mptcp, rwin is about the mptcp-level connection data.
1203  *
1204  * Data that is still on the ssk rx queue can thus be ignored,
1205  * as far as mptcp peer is concerned that data is still inflight.
1206  * DSS ACK is updated when skb is moved to the mptcp rx queue.
1207  */
1208 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1209 {
1210         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1211         const struct sock *sk = subflow->conn;
1212
1213         *space = __mptcp_space(sk);
1214         *full_space = tcp_full_space(sk);
1215 }
1216
1217 void __mptcp_error_report(struct sock *sk)
1218 {
1219         struct mptcp_subflow_context *subflow;
1220         struct mptcp_sock *msk = mptcp_sk(sk);
1221
1222         mptcp_for_each_subflow(msk, subflow) {
1223                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1224                 int err = sock_error(ssk);
1225
1226                 if (!err)
1227                         continue;
1228
1229                 /* only propagate errors on fallen-back sockets or
1230                  * on MPC connect
1231                  */
1232                 if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(msk))
1233                         continue;
1234
1235                 inet_sk_state_store(sk, inet_sk_state_load(ssk));
1236                 sk->sk_err = -err;
1237
1238                 /* This barrier is coupled with smp_rmb() in mptcp_poll() */
1239                 smp_wmb();
1240                 sk_error_report(sk);
1241                 break;
1242         }
1243 }
1244
1245 static void subflow_error_report(struct sock *ssk)
1246 {
1247         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1248
1249         mptcp_data_lock(sk);
1250         if (!sock_owned_by_user(sk))
1251                 __mptcp_error_report(sk);
1252         else
1253                 set_bit(MPTCP_ERROR_REPORT,  &mptcp_sk(sk)->flags);
1254         mptcp_data_unlock(sk);
1255 }
1256
1257 static void subflow_data_ready(struct sock *sk)
1258 {
1259         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1260         u16 state = 1 << inet_sk_state_load(sk);
1261         struct sock *parent = subflow->conn;
1262         struct mptcp_sock *msk;
1263
1264         msk = mptcp_sk(parent);
1265         if (state & TCPF_LISTEN) {
1266                 /* MPJ subflow are removed from accept queue before reaching here,
1267                  * avoid stray wakeups
1268                  */
1269                 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1270                         return;
1271
1272                 set_bit(MPTCP_DATA_READY, &msk->flags);
1273                 parent->sk_data_ready(parent);
1274                 return;
1275         }
1276
1277         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1278                      !subflow->mp_join && !(state & TCPF_CLOSE));
1279
1280         if (mptcp_subflow_data_available(sk))
1281                 mptcp_data_ready(parent, sk);
1282         else if (unlikely(sk->sk_err))
1283                 subflow_error_report(sk);
1284 }
1285
1286 static void subflow_write_space(struct sock *ssk)
1287 {
1288         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1289
1290         mptcp_propagate_sndbuf(sk, ssk);
1291         mptcp_write_space(sk);
1292 }
1293
1294 static struct inet_connection_sock_af_ops *
1295 subflow_default_af_ops(struct sock *sk)
1296 {
1297 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1298         if (sk->sk_family == AF_INET6)
1299                 return &subflow_v6_specific;
1300 #endif
1301         return &subflow_specific;
1302 }
1303
1304 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1305 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1306 {
1307         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1308         struct inet_connection_sock *icsk = inet_csk(sk);
1309         struct inet_connection_sock_af_ops *target;
1310
1311         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1312
1313         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1314                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1315
1316         if (likely(icsk->icsk_af_ops == target))
1317                 return;
1318
1319         subflow->icsk_af_ops = icsk->icsk_af_ops;
1320         icsk->icsk_af_ops = target;
1321 }
1322 #endif
1323
1324 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1325                          struct sockaddr_storage *addr,
1326                          unsigned short family)
1327 {
1328         memset(addr, 0, sizeof(*addr));
1329         addr->ss_family = family;
1330         if (addr->ss_family == AF_INET) {
1331                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1332
1333                 if (info->family == AF_INET)
1334                         in_addr->sin_addr = info->addr;
1335 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1336                 else if (ipv6_addr_v4mapped(&info->addr6))
1337                         in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1338 #endif
1339                 in_addr->sin_port = info->port;
1340         }
1341 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1342         else if (addr->ss_family == AF_INET6) {
1343                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1344
1345                 if (info->family == AF_INET)
1346                         ipv6_addr_set_v4mapped(info->addr.s_addr,
1347                                                &in6_addr->sin6_addr);
1348                 else
1349                         in6_addr->sin6_addr = info->addr6;
1350                 in6_addr->sin6_port = info->port;
1351         }
1352 #endif
1353 }
1354
1355 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1356                             const struct mptcp_addr_info *remote,
1357                             u8 flags, int ifindex)
1358 {
1359         struct mptcp_sock *msk = mptcp_sk(sk);
1360         struct mptcp_subflow_context *subflow;
1361         struct sockaddr_storage addr;
1362         int remote_id = remote->id;
1363         int local_id = loc->id;
1364         struct socket *sf;
1365         struct sock *ssk;
1366         u32 remote_token;
1367         int addrlen;
1368         int err;
1369
1370         if (!mptcp_is_fully_established(sk))
1371                 return -ENOTCONN;
1372
1373         err = mptcp_subflow_create_socket(sk, &sf);
1374         if (err)
1375                 return err;
1376
1377         ssk = sf->sk;
1378         subflow = mptcp_subflow_ctx(ssk);
1379         do {
1380                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1381         } while (!subflow->local_nonce);
1382
1383         if (!local_id) {
1384                 err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
1385                 if (err < 0)
1386                         goto failed;
1387
1388                 local_id = err;
1389         }
1390
1391         subflow->remote_key = msk->remote_key;
1392         subflow->local_key = msk->local_key;
1393         subflow->token = msk->token;
1394         mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1395
1396         addrlen = sizeof(struct sockaddr_in);
1397 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1398         if (addr.ss_family == AF_INET6)
1399                 addrlen = sizeof(struct sockaddr_in6);
1400 #endif
1401         ssk->sk_bound_dev_if = ifindex;
1402         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1403         if (err)
1404                 goto failed;
1405
1406         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1407         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1408                  remote_token, local_id, remote_id);
1409         subflow->remote_token = remote_token;
1410         subflow->local_id = local_id;
1411         subflow->remote_id = remote_id;
1412         subflow->request_join = 1;
1413         subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1414         mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1415
1416         mptcp_add_pending_subflow(msk, subflow);
1417         mptcp_sockopt_sync(msk, ssk);
1418         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1419         if (err && err != -EINPROGRESS)
1420                 goto failed_unlink;
1421
1422         /* discard the subflow socket */
1423         mptcp_sock_graft(ssk, sk->sk_socket);
1424         iput(SOCK_INODE(sf));
1425         return err;
1426
1427 failed_unlink:
1428         spin_lock_bh(&msk->join_list_lock);
1429         list_del(&subflow->node);
1430         spin_unlock_bh(&msk->join_list_lock);
1431         sock_put(mptcp_subflow_tcp_sock(subflow));
1432
1433 failed:
1434         subflow->disposable = 1;
1435         sock_release(sf);
1436         return err;
1437 }
1438
1439 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1440 {
1441 #ifdef CONFIG_SOCK_CGROUP_DATA
1442         struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1443                                 *child_skcd = &child->sk_cgrp_data;
1444
1445         /* only the additional subflows created by kworkers have to be modified */
1446         if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1447             cgroup_id(sock_cgroup_ptr(child_skcd))) {
1448 #ifdef CONFIG_MEMCG
1449                 struct mem_cgroup *memcg = parent->sk_memcg;
1450
1451                 mem_cgroup_sk_free(child);
1452                 if (memcg && css_tryget(&memcg->css))
1453                         child->sk_memcg = memcg;
1454 #endif /* CONFIG_MEMCG */
1455
1456                 cgroup_sk_free(child_skcd);
1457                 *child_skcd = *parent_skcd;
1458                 cgroup_sk_clone(child_skcd);
1459         }
1460 #endif /* CONFIG_SOCK_CGROUP_DATA */
1461 }
1462
1463 static void mptcp_subflow_ops_override(struct sock *ssk)
1464 {
1465 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1466         if (ssk->sk_prot == &tcpv6_prot)
1467                 ssk->sk_prot = &tcpv6_prot_override;
1468         else
1469 #endif
1470                 ssk->sk_prot = &tcp_prot_override;
1471 }
1472
1473 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1474 {
1475 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1476         if (ssk->sk_prot == &tcpv6_prot_override)
1477                 ssk->sk_prot = &tcpv6_prot;
1478         else
1479 #endif
1480                 ssk->sk_prot = &tcp_prot;
1481 }
1482 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1483 {
1484         struct mptcp_subflow_context *subflow;
1485         struct net *net = sock_net(sk);
1486         struct socket *sf;
1487         int err;
1488
1489         /* un-accepted server sockets can reach here - on bad configuration
1490          * bail early to avoid greater trouble later
1491          */
1492         if (unlikely(!sk->sk_socket))
1493                 return -EINVAL;
1494
1495         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1496                                &sf);
1497         if (err)
1498                 return err;
1499
1500         lock_sock(sf->sk);
1501
1502         /* the newly created socket has to be in the same cgroup as its parent */
1503         mptcp_attach_cgroup(sk, sf->sk);
1504
1505         /* kernel sockets do not by default acquire net ref, but TCP timer
1506          * needs it.
1507          */
1508         sf->sk->sk_net_refcnt = 1;
1509         get_net(net);
1510 #ifdef CONFIG_PROC_FS
1511         this_cpu_add(*net->core.sock_inuse, 1);
1512 #endif
1513         err = tcp_set_ulp(sf->sk, "mptcp");
1514         release_sock(sf->sk);
1515
1516         if (err) {
1517                 sock_release(sf);
1518                 return err;
1519         }
1520
1521         /* the newly created socket really belongs to the owning MPTCP master
1522          * socket, even if for additional subflows the allocation is performed
1523          * by a kernel workqueue. Adjust inode references, so that the
1524          * procfs/diag interaces really show this one belonging to the correct
1525          * user.
1526          */
1527         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1528         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1529         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1530
1531         subflow = mptcp_subflow_ctx(sf->sk);
1532         pr_debug("subflow=%p", subflow);
1533
1534         *new_sock = sf;
1535         sock_hold(sk);
1536         subflow->conn = sk;
1537         mptcp_subflow_ops_override(sf->sk);
1538
1539         return 0;
1540 }
1541
1542 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1543                                                         gfp_t priority)
1544 {
1545         struct inet_connection_sock *icsk = inet_csk(sk);
1546         struct mptcp_subflow_context *ctx;
1547
1548         ctx = kzalloc(sizeof(*ctx), priority);
1549         if (!ctx)
1550                 return NULL;
1551
1552         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1553         INIT_LIST_HEAD(&ctx->node);
1554         INIT_LIST_HEAD(&ctx->delegated_node);
1555
1556         pr_debug("subflow=%p", ctx);
1557
1558         ctx->tcp_sock = sk;
1559
1560         return ctx;
1561 }
1562
1563 static void __subflow_state_change(struct sock *sk)
1564 {
1565         struct socket_wq *wq;
1566
1567         rcu_read_lock();
1568         wq = rcu_dereference(sk->sk_wq);
1569         if (skwq_has_sleeper(wq))
1570                 wake_up_interruptible_all(&wq->wait);
1571         rcu_read_unlock();
1572 }
1573
1574 static bool subflow_is_done(const struct sock *sk)
1575 {
1576         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1577 }
1578
1579 static void subflow_state_change(struct sock *sk)
1580 {
1581         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1582         struct sock *parent = subflow->conn;
1583
1584         __subflow_state_change(sk);
1585
1586         if (subflow_simultaneous_connect(sk)) {
1587                 mptcp_propagate_sndbuf(parent, sk);
1588                 mptcp_do_fallback(sk);
1589                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1590                 pr_fallback(mptcp_sk(parent));
1591                 subflow->conn_finished = 1;
1592                 mptcp_set_connected(parent);
1593         }
1594
1595         /* as recvmsg() does not acquire the subflow socket for ssk selection
1596          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1597          * the data available machinery here.
1598          */
1599         if (mptcp_subflow_data_available(sk))
1600                 mptcp_data_ready(parent, sk);
1601         else if (unlikely(sk->sk_err))
1602                 subflow_error_report(sk);
1603
1604         subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1605
1606         if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1607             !subflow->rx_eof && subflow_is_done(sk)) {
1608                 subflow->rx_eof = 1;
1609                 mptcp_subflow_eof(parent);
1610         }
1611 }
1612
1613 static int subflow_ulp_init(struct sock *sk)
1614 {
1615         struct inet_connection_sock *icsk = inet_csk(sk);
1616         struct mptcp_subflow_context *ctx;
1617         struct tcp_sock *tp = tcp_sk(sk);
1618         int err = 0;
1619
1620         /* disallow attaching ULP to a socket unless it has been
1621          * created with sock_create_kern()
1622          */
1623         if (!sk->sk_kern_sock) {
1624                 err = -EOPNOTSUPP;
1625                 goto out;
1626         }
1627
1628         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1629         if (!ctx) {
1630                 err = -ENOMEM;
1631                 goto out;
1632         }
1633
1634         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1635
1636         tp->is_mptcp = 1;
1637         ctx->icsk_af_ops = icsk->icsk_af_ops;
1638         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1639         ctx->tcp_data_ready = sk->sk_data_ready;
1640         ctx->tcp_state_change = sk->sk_state_change;
1641         ctx->tcp_write_space = sk->sk_write_space;
1642         ctx->tcp_error_report = sk->sk_error_report;
1643         sk->sk_data_ready = subflow_data_ready;
1644         sk->sk_write_space = subflow_write_space;
1645         sk->sk_state_change = subflow_state_change;
1646         sk->sk_error_report = subflow_error_report;
1647 out:
1648         return err;
1649 }
1650
1651 static void subflow_ulp_release(struct sock *ssk)
1652 {
1653         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1654         bool release = true;
1655         struct sock *sk;
1656
1657         if (!ctx)
1658                 return;
1659
1660         sk = ctx->conn;
1661         if (sk) {
1662                 /* if the msk has been orphaned, keep the ctx
1663                  * alive, will be freed by __mptcp_close_ssk(),
1664                  * when the subflow is still unaccepted
1665                  */
1666                 release = ctx->disposable || list_empty(&ctx->node);
1667                 sock_put(sk);
1668         }
1669
1670         mptcp_subflow_ops_undo_override(ssk);
1671         if (release)
1672                 kfree_rcu(ctx, rcu);
1673 }
1674
1675 static void subflow_ulp_clone(const struct request_sock *req,
1676                               struct sock *newsk,
1677                               const gfp_t priority)
1678 {
1679         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1680         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1681         struct mptcp_subflow_context *new_ctx;
1682
1683         if (!tcp_rsk(req)->is_mptcp ||
1684             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1685                 subflow_ulp_fallback(newsk, old_ctx);
1686                 return;
1687         }
1688
1689         new_ctx = subflow_create_ctx(newsk, priority);
1690         if (!new_ctx) {
1691                 subflow_ulp_fallback(newsk, old_ctx);
1692                 return;
1693         }
1694
1695         new_ctx->conn_finished = 1;
1696         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1697         new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1698         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1699         new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1700         new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1701         new_ctx->rel_write_seq = 1;
1702         new_ctx->tcp_sock = newsk;
1703
1704         if (subflow_req->mp_capable) {
1705                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1706                  * is fully established only after we receive the remote key
1707                  */
1708                 new_ctx->mp_capable = 1;
1709                 new_ctx->local_key = subflow_req->local_key;
1710                 new_ctx->token = subflow_req->token;
1711                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1712                 new_ctx->idsn = subflow_req->idsn;
1713         } else if (subflow_req->mp_join) {
1714                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1715                 new_ctx->mp_join = 1;
1716                 new_ctx->fully_established = 1;
1717                 new_ctx->backup = subflow_req->backup;
1718                 new_ctx->local_id = subflow_req->local_id;
1719                 new_ctx->remote_id = subflow_req->remote_id;
1720                 new_ctx->token = subflow_req->token;
1721                 new_ctx->thmac = subflow_req->thmac;
1722         }
1723 }
1724
1725 static void tcp_release_cb_override(struct sock *ssk)
1726 {
1727         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1728
1729         if (mptcp_subflow_has_delegated_action(subflow))
1730                 mptcp_subflow_process_delegated(ssk);
1731
1732         tcp_release_cb(ssk);
1733 }
1734
1735 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1736         .name           = "mptcp",
1737         .owner          = THIS_MODULE,
1738         .init           = subflow_ulp_init,
1739         .release        = subflow_ulp_release,
1740         .clone          = subflow_ulp_clone,
1741 };
1742
1743 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1744 {
1745         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1746         subflow_ops->slab_name = "request_sock_subflow";
1747
1748         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1749                                               subflow_ops->obj_size, 0,
1750                                               SLAB_ACCOUNT |
1751                                               SLAB_TYPESAFE_BY_RCU,
1752                                               NULL);
1753         if (!subflow_ops->slab)
1754                 return -ENOMEM;
1755
1756         subflow_ops->destructor = subflow_req_destructor;
1757
1758         return 0;
1759 }
1760
1761 void __init mptcp_subflow_init(void)
1762 {
1763         mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1764         if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1765                 panic("MPTCP: failed to init subflow request sock ops\n");
1766
1767         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1768         subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
1769
1770         subflow_specific = ipv4_specific;
1771         subflow_specific.conn_request = subflow_v4_conn_request;
1772         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1773         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1774
1775         tcp_prot_override = tcp_prot;
1776         tcp_prot_override.release_cb = tcp_release_cb_override;
1777
1778 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1779         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1780         subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
1781
1782         subflow_v6_specific = ipv6_specific;
1783         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1784         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1785         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1786
1787         subflow_v6m_specific = subflow_v6_specific;
1788         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1789         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1790         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1791         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1792         subflow_v6m_specific.net_frag_header_len = 0;
1793
1794         tcpv6_prot_override = tcpv6_prot;
1795         tcpv6_prot_override.release_cb = tcp_release_cb_override;
1796 #endif
1797
1798         mptcp_diag_subflow_init(&subflow_ulp_ops);
1799
1800         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1801                 panic("MPTCP: failed to register subflows to ULP\n");
1802 }