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/sha.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 #endif
  22 #include <net/mptcp.h>
  23 #include "protocol.h"
  24 #include "mib.h"
  25
  26 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
  27                                   enum linux_mptcp_mib_field field)
  28 {
  29         MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
  30 }
  31
  32 static int subflow_rebuild_header(struct sock *sk)
  33 {
  34         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
  35         int local_id, err = 0;
  36
  37         if (subflow->request_mptcp && !subflow->token) {
  38                 pr_debug("subflow=%p", sk);
  39                 err = mptcp_token_new_connect(sk);
  40         } else if (subflow->request_join && !subflow->local_nonce) {
  41                 struct mptcp_sock *msk = (struct mptcp_sock *)subflow->conn;
  42
  43                 pr_debug("subflow=%p", sk);
  44
  45                 do {
  46                         get_random_bytes(&subflow->local_nonce, sizeof(u32));
  47                 } while (!subflow->local_nonce);
  48
  49                 if (subflow->local_id)
  50                         goto out;
  51
  52                 local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
  53                 if (local_id < 0)
  54                         return -EINVAL;
  55
  56                 subflow->local_id = local_id;
  57         }
  58
  59 out:
  60         if (err)
  61                 return err;
  62
  63         return subflow->icsk_af_ops->rebuild_header(sk);
  64 }
  65
  66 static void subflow_req_destructor(struct request_sock *req)
  67 {
  68         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  69
  70         pr_debug("subflow_req=%p", subflow_req);
  71
  72         if (subflow_req->mp_capable)
  73                 mptcp_token_destroy_request(subflow_req->token);
  74         tcp_request_sock_ops.destructor(req);
  75 }
  76
  77 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
  78                                   void *hmac)
  79 {
  80         u8 msg[8];
  81
  82         put_unaligned_be32(nonce1, &msg[0]);
  83         put_unaligned_be32(nonce2, &msg[4]);
  84
  85         mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
  86 }
  87
  88 /* validate received token and create truncated hmac and nonce for SYN-ACK */
  89 static bool subflow_token_join_request(struct request_sock *req,
  90                                        const struct sk_buff *skb)
  91 {
  92         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  93         u8 hmac[SHA256_DIGEST_SIZE];
  94         struct mptcp_sock *msk;
  95         int local_id;
  96
  97         msk = mptcp_token_get_sock(subflow_req->token);
  98         if (!msk) {
  99                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
 100                 return false;
 101         }
 102
 103         local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
 104         if (local_id < 0) {
 105                 sock_put((struct sock *)msk);
 106                 return false;
 107         }
 108         subflow_req->local_id = local_id;
 109
 110         get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
 111
 112         subflow_generate_hmac(msk->local_key, msk->remote_key,
 113                               subflow_req->local_nonce,
 114                               subflow_req->remote_nonce, hmac);
 115
 116         subflow_req->thmac = get_unaligned_be64(hmac);
 117
 118         sock_put((struct sock *)msk);
 119         return true;
 120 }
 121
 122 static void subflow_init_req(struct request_sock *req,
 123                              const struct sock *sk_listener,
 124                              struct sk_buff *skb)
 125 {
 126         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 127         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 128         struct mptcp_options_received mp_opt;
 129
 130         pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
 131
 132         mptcp_get_options(skb, &mp_opt);
 133
 134         subflow_req->mp_capable = 0;
 135         subflow_req->mp_join = 0;
 136
 137 #ifdef CONFIG_TCP_MD5SIG
 138         /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
 139          * TCP option space.
 140          */
 141         if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
 142                 return;
 143 #endif
 144
 145         if (mp_opt.mp_capable) {
 146                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
 147
 148                 if (mp_opt.mp_join)
 149                         return;
 150         } else if (mp_opt.mp_join) {
 151                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
 152         }
 153
 154         if (mp_opt.mp_capable && listener->request_mptcp) {
 155                 int err;
 156
 157                 err = mptcp_token_new_request(req);
 158                 if (err == 0)
 159                         subflow_req->mp_capable = 1;
 160
 161                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 162         } else if (mp_opt.mp_join && listener->request_mptcp) {
 163                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 164                 subflow_req->mp_join = 1;
 165                 subflow_req->backup = mp_opt.backup;
 166                 subflow_req->remote_id = mp_opt.join_id;
 167                 subflow_req->token = mp_opt.token;
 168                 subflow_req->remote_nonce = mp_opt.nonce;
 169                 pr_debug("token=%u, remote_nonce=%u", subflow_req->token,
 170                          subflow_req->remote_nonce);
 171                 if (!subflow_token_join_request(req, skb)) {
 172                         subflow_req->mp_join = 0;
 173                         // @@ need to trigger RST
 174                 }
 175         }
 176 }
 177
 178 static void subflow_v4_init_req(struct request_sock *req,
 179                                 const struct sock *sk_listener,
 180                                 struct sk_buff *skb)
 181 {
 182         tcp_rsk(req)->is_mptcp = 1;
 183
 184         tcp_request_sock_ipv4_ops.init_req(req, sk_listener, skb);
 185
 186         subflow_init_req(req, sk_listener, skb);
 187 }
 188
 189 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 190 static void subflow_v6_init_req(struct request_sock *req,
 191                                 const struct sock *sk_listener,
 192                                 struct sk_buff *skb)
 193 {
 194         tcp_rsk(req)->is_mptcp = 1;
 195
 196         tcp_request_sock_ipv6_ops.init_req(req, sk_listener, skb);
 197
 198         subflow_init_req(req, sk_listener, skb);
 199 }
 200 #endif
 201
 202 /* validate received truncated hmac and create hmac for third ACK */
 203 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
 204 {
 205         u8 hmac[SHA256_DIGEST_SIZE];
 206         u64 thmac;
 207
 208         subflow_generate_hmac(subflow->remote_key, subflow->local_key,
 209                               subflow->remote_nonce, subflow->local_nonce,
 210                               hmac);
 211
 212         thmac = get_unaligned_be64(hmac);
 213         pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
 214                  subflow, subflow->token,
 215                  (unsigned long long)thmac,
 216                  (unsigned long long)subflow->thmac);
 217
 218         return thmac == subflow->thmac;
 219 }
 220
 221 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 222 {
 223         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 224         struct mptcp_options_received mp_opt;
 225         struct sock *parent = subflow->conn;
 226         struct tcp_sock *tp = tcp_sk(sk);
 227
 228         subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 229
 230         if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
 231                 inet_sk_state_store(parent, TCP_ESTABLISHED);
 232                 parent->sk_state_change(parent);
 233         }
 234
 235         /* be sure no special action on any packet other than syn-ack */
 236         if (subflow->conn_finished)
 237                 return;
 238
 239         subflow->conn_finished = 1;
 240
 241         mptcp_get_options(skb, &mp_opt);
 242         if (subflow->request_mptcp && mp_opt.mp_capable) {
 243                 subflow->mp_capable = 1;
 244                 subflow->can_ack = 1;
 245                 subflow->remote_key = mp_opt.sndr_key;
 246                 pr_debug("subflow=%p, remote_key=%llu", subflow,
 247                          subflow->remote_key);
 248         } else if (subflow->request_join && mp_opt.mp_join) {
 249                 subflow->mp_join = 1;
 250                 subflow->thmac = mp_opt.thmac;
 251                 subflow->remote_nonce = mp_opt.nonce;
 252                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
 253                          subflow->thmac, subflow->remote_nonce);
 254         } else if (subflow->request_mptcp) {
 255                 tp->is_mptcp = 0;
 256         }
 257
 258         if (!tp->is_mptcp)
 259                 return;
 260
 261         if (subflow->mp_capable) {
 262                 pr_debug("subflow=%p, remote_key=%llu", mptcp_subflow_ctx(sk),
 263                          subflow->remote_key);
 264                 mptcp_finish_connect(sk);
 265
 266                 if (skb) {
 267                         pr_debug("synack seq=%u", TCP_SKB_CB(skb)->seq);
 268                         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 269                 }
 270         } else if (subflow->mp_join) {
 271                 u8 hmac[SHA256_DIGEST_SIZE];
 272
 273                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u",
 274                          subflow, subflow->thmac,
 275                          subflow->remote_nonce);
 276                 if (!subflow_thmac_valid(subflow)) {
 277                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 278                         subflow->mp_join = 0;
 279                         goto do_reset;
 280                 }
 281
 282                 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 283                                       subflow->local_nonce,
 284                                       subflow->remote_nonce,
 285                                       hmac);
 286
 287                 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
 288
 289                 if (skb)
 290                         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 291
 292                 if (!mptcp_finish_join(sk))
 293                         goto do_reset;
 294
 295                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 296         } else {
 297 do_reset:
 298                 tcp_send_active_reset(sk, GFP_ATOMIC);
 299                 tcp_done(sk);
 300         }
 301 }
 302
 303 static struct request_sock_ops subflow_request_sock_ops;
 304 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
 305
 306 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 307 {
 308         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 309
 310         pr_debug("subflow=%p", subflow);
 311
 312         /* Never answer to SYNs sent to broadcast or multicast */
 313         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 314                 goto drop;
 315
 316         return tcp_conn_request(&subflow_request_sock_ops,
 317                                 &subflow_request_sock_ipv4_ops,
 318                                 sk, skb);
 319 drop:
 320         tcp_listendrop(sk);
 321         return 0;
 322 }
 323
 324 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 325 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
 326 static struct inet_connection_sock_af_ops subflow_v6_specific;
 327 static struct inet_connection_sock_af_ops subflow_v6m_specific;
 328
 329 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 330 {
 331         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 332
 333         pr_debug("subflow=%p", subflow);
 334
 335         if (skb->protocol == htons(ETH_P_IP))
 336                 return subflow_v4_conn_request(sk, skb);
 337
 338         if (!ipv6_unicast_destination(skb))
 339                 goto drop;
 340
 341         return tcp_conn_request(&subflow_request_sock_ops,
 342                                 &subflow_request_sock_ipv6_ops, sk, skb);
 343
 344 drop:
 345         tcp_listendrop(sk);
 346         return 0; /* don't send reset */
 347 }
 348 #endif
 349
 350 /* validate hmac received in third ACK */
 351 static bool subflow_hmac_valid(const struct request_sock *req,
 352                                const struct mptcp_options_received *mp_opt)
 353 {
 354         const struct mptcp_subflow_request_sock *subflow_req;
 355         u8 hmac[SHA256_DIGEST_SIZE];
 356         struct mptcp_sock *msk;
 357         bool ret;
 358
 359         subflow_req = mptcp_subflow_rsk(req);
 360         msk = mptcp_token_get_sock(subflow_req->token);
 361         if (!msk)
 362                 return false;
 363
 364         subflow_generate_hmac(msk->remote_key, msk->local_key,
 365                               subflow_req->remote_nonce,
 366                               subflow_req->local_nonce, hmac);
 367
 368         ret = true;
 369         if (crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN))
 370                 ret = false;
 371
 372         sock_put((struct sock *)msk);
 373         return ret;
 374 }
 375
 376 static void mptcp_sock_destruct(struct sock *sk)
 377 {
 378         /* if new mptcp socket isn't accepted, it is free'd
 379          * from the tcp listener sockets request queue, linked
 380          * from req->sk.  The tcp socket is released.
 381          * This calls the ULP release function which will
 382          * also remove the mptcp socket, via
 383          * sock_put(ctx->conn).
 384          *
 385          * Problem is that the mptcp socket will not be in
 386          * SYN_RECV state and doesn't have SOCK_DEAD flag.
 387          * Both result in warnings from inet_sock_destruct.
 388          */
 389
 390         if (sk->sk_state == TCP_SYN_RECV) {
 391                 sk->sk_state = TCP_CLOSE;
 392                 WARN_ON_ONCE(sk->sk_socket);
 393                 sock_orphan(sk);
 394         }
 395
 396         inet_sock_destruct(sk);
 397 }
 398
 399 static void mptcp_force_close(struct sock *sk)
 400 {
 401         inet_sk_state_store(sk, TCP_CLOSE);
 402         sk_common_release(sk);
 403 }
 404
 405 static void subflow_ulp_fallback(struct sock *sk,
 406                                  struct mptcp_subflow_context *old_ctx)
 407 {
 408         struct inet_connection_sock *icsk = inet_csk(sk);
 409
 410         mptcp_subflow_tcp_fallback(sk, old_ctx);
 411         icsk->icsk_ulp_ops = NULL;
 412         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 413         tcp_sk(sk)->is_mptcp = 0;
 414 }
 415
 416 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 417                                           struct sk_buff *skb,
 418                                           struct request_sock *req,
 419                                           struct dst_entry *dst,
 420                                           struct request_sock *req_unhash,
 421                                           bool *own_req)
 422 {
 423         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 424         struct mptcp_subflow_request_sock *subflow_req;
 425         struct mptcp_options_received mp_opt;
 426         bool fallback_is_fatal = false;
 427         struct sock *new_msk = NULL;
 428         bool fallback = false;
 429         struct sock *child;
 430
 431         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 432
 433         /* we need later a valid 'mp_capable' value even when options are not
 434          * parsed
 435          */
 436         mp_opt.mp_capable = 0;
 437         if (tcp_rsk(req)->is_mptcp == 0)
 438                 goto create_child;
 439
 440         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 441         subflow_req = mptcp_subflow_rsk(req);
 442         if (subflow_req->mp_capable) {
 443                 if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
 444                         /* here we can receive and accept an in-window,
 445                          * out-of-order pkt, which will not carry the MP_CAPABLE
 446                          * opt even on mptcp enabled paths
 447                          */
 448                         goto create_msk;
 449                 }
 450
 451                 mptcp_get_options(skb, &mp_opt);
 452                 if (!mp_opt.mp_capable) {
 453                         fallback = true;
 454                         goto create_child;
 455                 }
 456
 457 create_msk:
 458                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 459                 if (!new_msk)
 460                         fallback = true;
 461         } else if (subflow_req->mp_join) {
 462                 fallback_is_fatal = true;
 463                 mptcp_get_options(skb, &mp_opt);
 464                 if (!mp_opt.mp_join ||
 465                     !subflow_hmac_valid(req, &mp_opt)) {
 466                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 467                         return NULL;
 468                 }
 469         }
 470
 471 create_child:
 472         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 473                                                      req_unhash, own_req);
 474
 475         if (child && *own_req) {
 476                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 477
 478                 /* we need to fallback on ctx allocation failure and on pre-reqs
 479                  * checking above. In the latter scenario we additionally need
 480                  * to reset the context to non MPTCP status.
 481                  */
 482                 if (!ctx || fallback) {
 483                         if (fallback_is_fatal)
 484                                 goto close_child;
 485
 486                         if (ctx) {
 487                                 subflow_ulp_fallback(child, ctx);
 488                                 kfree_rcu(ctx, rcu);
 489                         }
 490                         goto out;
 491                 }
 492
 493                 if (ctx->mp_capable) {
 494                         /* new mpc subflow takes ownership of the newly
 495                          * created mptcp socket
 496                          */
 497                         new_msk->sk_destruct = mptcp_sock_destruct;
 498                         mptcp_pm_new_connection(mptcp_sk(new_msk), 1);
 499                         ctx->conn = new_msk;
 500                         new_msk = NULL;
 501
 502                         /* with OoO packets we can reach here without ingress
 503                          * mpc option
 504                          */
 505                         ctx->remote_key = mp_opt.sndr_key;
 506                         ctx->fully_established = mp_opt.mp_capable;
 507                         ctx->can_ack = mp_opt.mp_capable;
 508                 } else if (ctx->mp_join) {
 509                         struct mptcp_sock *owner;
 510
 511                         owner = mptcp_token_get_sock(ctx->token);
 512                         if (!owner)
 513                                 goto close_child;
 514
 515                         ctx->conn = (struct sock *)owner;
 516                         if (!mptcp_finish_join(child))
 517                                 goto close_child;
 518
 519                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 520                 }
 521         }
 522
 523 out:
 524         /* dispose of the left over mptcp master, if any */
 525         if (unlikely(new_msk))
 526                 mptcp_force_close(new_msk);
 527
 528         /* check for expected invariant - should never trigger, just help
 529          * catching eariler subtle bugs
 530          */
 531         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 532                      (!mptcp_subflow_ctx(child) ||
 533                       !mptcp_subflow_ctx(child)->conn));
 534         return child;
 535
 536 close_child:
 537         tcp_send_active_reset(child, GFP_ATOMIC);
 538         inet_csk_prepare_forced_close(child);
 539         tcp_done(child);
 540         return NULL;
 541 }
 542
 543 static struct inet_connection_sock_af_ops subflow_specific;
 544
 545 enum mapping_status {
 546         MAPPING_OK,
 547         MAPPING_INVALID,
 548         MAPPING_EMPTY,
 549         MAPPING_DATA_FIN
 550 };
 551
 552 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
 553 {
 554         if ((u32)seq == (u32)old_seq)
 555                 return old_seq;
 556
 557         /* Assume map covers data not mapped yet. */
 558         return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
 559 }
 560
 561 static void warn_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 562 {
 563         WARN_ONCE(1, "Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 564                   ssn, subflow->map_subflow_seq, subflow->map_data_len);
 565 }
 566
 567 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 568 {
 569         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 570         unsigned int skb_consumed;
 571
 572         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 573         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 574                 return true;
 575
 576         return skb->len - skb_consumed <= subflow->map_data_len -
 577                                           mptcp_subflow_get_map_offset(subflow);
 578 }
 579
 580 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 581 {
 582         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 583         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 584
 585         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 586                 /* Mapping covers data later in the subflow stream,
 587                  * currently unsupported.
 588                  */
 589                 warn_bad_map(subflow, ssn);
 590                 return false;
 591         }
 592         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 593                                   subflow->map_data_len))) {
 594                 /* Mapping does covers past subflow data, invalid */
 595                 warn_bad_map(subflow, ssn + skb->len);
 596                 return false;
 597         }
 598         return true;
 599 }
 600
 601 static enum mapping_status get_mapping_status(struct sock *ssk)
 602 {
 603         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 604         struct mptcp_ext *mpext;
 605         struct sk_buff *skb;
 606         u16 data_len;
 607         u64 map_seq;
 608
 609         skb = skb_peek(&ssk->sk_receive_queue);
 610         if (!skb)
 611                 return MAPPING_EMPTY;
 612
 613         mpext = mptcp_get_ext(skb);
 614         if (!mpext || !mpext->use_map) {
 615                 if (!subflow->map_valid && !skb->len) {
 616                         /* the TCP stack deliver 0 len FIN pkt to the receive
 617                          * queue, that is the only 0len pkts ever expected here,
 618                          * and we can admit no mapping only for 0 len pkts
 619                          */
 620                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 621                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 622                                           TCP_SKB_CB(skb)->seq,
 623                                           TCP_SKB_CB(skb)->end_seq,
 624                                           TCP_SKB_CB(skb)->tcp_flags);
 625                         sk_eat_skb(ssk, skb);
 626                         return MAPPING_EMPTY;
 627                 }
 628
 629                 if (!subflow->map_valid)
 630                         return MAPPING_INVALID;
 631
 632                 goto validate_seq;
 633         }
 634
 635         pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
 636                  mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
 637                  mpext->data_len, mpext->data_fin);
 638
 639         data_len = mpext->data_len;
 640         if (data_len == 0) {
 641                 pr_err("Infinite mapping not handled");
 642                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 643                 return MAPPING_INVALID;
 644         }
 645
 646         if (mpext->data_fin == 1) {
 647                 if (data_len == 1) {
 648                         pr_debug("DATA_FIN with no payload");
 649                         if (subflow->map_valid) {
 650                                 /* A DATA_FIN might arrive in a DSS
 651                                  * option before the previous mapping
 652                                  * has been fully consumed. Continue
 653                                  * handling the existing mapping.
 654                                  */
 655                                 skb_ext_del(skb, SKB_EXT_MPTCP);
 656                                 return MAPPING_OK;
 657                         } else {
 658                                 return MAPPING_DATA_FIN;
 659                         }
 660                 }
 661
 662                 /* Adjust for DATA_FIN using 1 byte of sequence space */
 663                 data_len--;
 664         }
 665
 666         if (!mpext->dsn64) {
 667                 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
 668                                      mpext->data_seq);
 669                 pr_debug("expanded seq=%llu", subflow->map_seq);
 670         } else {
 671                 map_seq = mpext->data_seq;
 672         }
 673
 674         if (subflow->map_valid) {
 675                 /* Allow replacing only with an identical map */
 676                 if (subflow->map_seq == map_seq &&
 677                     subflow->map_subflow_seq == mpext->subflow_seq &&
 678                     subflow->map_data_len == data_len) {
 679                         skb_ext_del(skb, SKB_EXT_MPTCP);
 680                         return MAPPING_OK;
 681                 }
 682
 683                 /* If this skb data are fully covered by the current mapping,
 684                  * the new map would need caching, which is not supported
 685                  */
 686                 if (skb_is_fully_mapped(ssk, skb)) {
 687                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
 688                         return MAPPING_INVALID;
 689                 }
 690
 691                 /* will validate the next map after consuming the current one */
 692                 return MAPPING_OK;
 693         }
 694
 695         subflow->map_seq = map_seq;
 696         subflow->map_subflow_seq = mpext->subflow_seq;
 697         subflow->map_data_len = data_len;
 698         subflow->map_valid = 1;
 699         subflow->mpc_map = mpext->mpc_map;
 700         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
 701                  subflow->map_seq, subflow->map_subflow_seq,
 702                  subflow->map_data_len);
 703
 704 validate_seq:
 705         /* we revalidate valid mapping on new skb, because we must ensure
 706          * the current skb is completely covered by the available mapping
 707          */
 708         if (!validate_mapping(ssk, skb))
 709                 return MAPPING_INVALID;
 710
 711         skb_ext_del(skb, SKB_EXT_MPTCP);
 712         return MAPPING_OK;
 713 }
 714
 715 static int subflow_read_actor(read_descriptor_t *desc,
 716                               struct sk_buff *skb,
 717                               unsigned int offset, size_t len)
 718 {
 719         size_t copy_len = min(desc->count, len);
 720
 721         desc->count -= copy_len;
 722
 723         pr_debug("flushed %zu bytes, %zu left", copy_len, desc->count);
 724         return copy_len;
 725 }
 726
 727 static bool subflow_check_data_avail(struct sock *ssk)
 728 {
 729         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 730         enum mapping_status status;
 731         struct mptcp_sock *msk;
 732         struct sk_buff *skb;
 733
 734         pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
 735                  subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
 736         if (subflow->data_avail)
 737                 return true;
 738
 739         msk = mptcp_sk(subflow->conn);
 740         for (;;) {
 741                 u32 map_remaining;
 742                 size_t delta;
 743                 u64 ack_seq;
 744                 u64 old_ack;
 745
 746                 status = get_mapping_status(ssk);
 747                 pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
 748                 if (status == MAPPING_INVALID) {
 749                         ssk->sk_err = EBADMSG;
 750                         goto fatal;
 751                 }
 752
 753                 if (status != MAPPING_OK)
 754                         return false;
 755
 756                 skb = skb_peek(&ssk->sk_receive_queue);
 757                 if (WARN_ON_ONCE(!skb))
 758                         return false;
 759
 760                 /* if msk lacks the remote key, this subflow must provide an
 761                  * MP_CAPABLE-based mapping
 762                  */
 763                 if (unlikely(!READ_ONCE(msk->can_ack))) {
 764                         if (!subflow->mpc_map) {
 765                                 ssk->sk_err = EBADMSG;
 766                                 goto fatal;
 767                         }
 768                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
 769                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
 770                         WRITE_ONCE(msk->can_ack, true);
 771                 }
 772
 773                 old_ack = READ_ONCE(msk->ack_seq);
 774                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
 775                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
 776                          ack_seq);
 777                 if (ack_seq == old_ack)
 778                         break;
 779
 780                 /* only accept in-sequence mapping. Old values are spurious
 781                  * retransmission; we can hit "future" values on active backup
 782                  * subflow switch, we relay on retransmissions to get
 783                  * in-sequence data.
 784                  * Cuncurrent subflows support will require subflow data
 785                  * reordering
 786                  */
 787                 map_remaining = subflow->map_data_len -
 788                                 mptcp_subflow_get_map_offset(subflow);
 789                 if (before64(ack_seq, old_ack))
 790                         delta = min_t(size_t, old_ack - ack_seq, map_remaining);
 791                 else
 792                         delta = min_t(size_t, ack_seq - old_ack, map_remaining);
 793
 794                 /* discard mapped data */
 795                 pr_debug("discarding %zu bytes, current map len=%d", delta,
 796                          map_remaining);
 797                 if (delta) {
 798                         read_descriptor_t desc = {
 799                                 .count = delta,
 800                         };
 801                         int ret;
 802
 803                         ret = tcp_read_sock(ssk, &desc, subflow_read_actor);
 804                         if (ret < 0) {
 805                                 ssk->sk_err = -ret;
 806                                 goto fatal;
 807                         }
 808                         if (ret < delta)
 809                                 return false;
 810                         if (delta == map_remaining)
 811                                 subflow->map_valid = 0;
 812                 }
 813         }
 814         return true;
 815
 816 fatal:
 817         /* fatal protocol error, close the socket */
 818         /* This barrier is coupled with smp_rmb() in tcp_poll() */
 819         smp_wmb();
 820         ssk->sk_error_report(ssk);
 821         tcp_set_state(ssk, TCP_CLOSE);
 822         tcp_send_active_reset(ssk, GFP_ATOMIC);
 823         return false;
 824 }
 825
 826 bool mptcp_subflow_data_available(struct sock *sk)
 827 {
 828         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 829         struct sk_buff *skb;
 830
 831         /* check if current mapping is still valid */
 832         if (subflow->map_valid &&
 833             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
 834                 subflow->map_valid = 0;
 835                 subflow->data_avail = 0;
 836
 837                 pr_debug("Done with mapping: seq=%u data_len=%u",
 838                          subflow->map_subflow_seq,
 839                          subflow->map_data_len);
 840         }
 841
 842         if (!subflow_check_data_avail(sk)) {
 843                 subflow->data_avail = 0;
 844                 return false;
 845         }
 846
 847         skb = skb_peek(&sk->sk_receive_queue);
 848         subflow->data_avail = skb &&
 849                        before(tcp_sk(sk)->copied_seq, TCP_SKB_CB(skb)->end_seq);
 850         return subflow->data_avail;
 851 }
 852
 853 static void subflow_data_ready(struct sock *sk)
 854 {
 855         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 856         struct sock *parent = subflow->conn;
 857
 858         if (!subflow->mp_capable && !subflow->mp_join) {
 859                 subflow->tcp_data_ready(sk);
 860
 861                 parent->sk_data_ready(parent);
 862                 return;
 863         }
 864
 865         if (mptcp_subflow_data_available(sk))
 866                 mptcp_data_ready(parent, sk);
 867 }
 868
 869 static void subflow_write_space(struct sock *sk)
 870 {
 871         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 872         struct sock *parent = subflow->conn;
 873
 874         sk_stream_write_space(sk);
 875         if (sk_stream_is_writeable(sk)) {
 876                 set_bit(MPTCP_SEND_SPACE, &mptcp_sk(parent)->flags);
 877                 smp_mb__after_atomic();
 878                 /* set SEND_SPACE before sk_stream_write_space clears NOSPACE */
 879                 sk_stream_write_space(parent);
 880         }
 881 }
 882
 883 static struct inet_connection_sock_af_ops *
 884 subflow_default_af_ops(struct sock *sk)
 885 {
 886 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 887         if (sk->sk_family == AF_INET6)
 888                 return &subflow_v6_specific;
 889 #endif
 890         return &subflow_specific;
 891 }
 892
 893 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 894 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
 895 {
 896         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 897         struct inet_connection_sock *icsk = inet_csk(sk);
 898         struct inet_connection_sock_af_ops *target;
 899
 900         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
 901
 902         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
 903                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
 904
 905         if (likely(icsk->icsk_af_ops == target))
 906                 return;
 907
 908         subflow->icsk_af_ops = icsk->icsk_af_ops;
 909         icsk->icsk_af_ops = target;
 910 }
 911 #endif
 912
 913 static void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
 914                                 struct sockaddr_storage *addr)
 915 {
 916         memset(addr, 0, sizeof(*addr));
 917         addr->ss_family = info->family;
 918         if (addr->ss_family == AF_INET) {
 919                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
 920
 921                 in_addr->sin_addr = info->addr;
 922                 in_addr->sin_port = info->port;
 923         }
 924 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 925         else if (addr->ss_family == AF_INET6) {
 926                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
 927
 928                 in6_addr->sin6_addr = info->addr6;
 929                 in6_addr->sin6_port = info->port;
 930         }
 931 #endif
 932 }
 933
 934 int __mptcp_subflow_connect(struct sock *sk, int ifindex,
 935                             const struct mptcp_addr_info *loc,
 936                             const struct mptcp_addr_info *remote)
 937 {
 938         struct mptcp_sock *msk = mptcp_sk(sk);
 939         struct mptcp_subflow_context *subflow;
 940         struct sockaddr_storage addr;
 941         struct socket *sf;
 942         u32 remote_token;
 943         int addrlen;
 944         int err;
 945
 946         if (sk->sk_state != TCP_ESTABLISHED)
 947                 return -ENOTCONN;
 948
 949         err = mptcp_subflow_create_socket(sk, &sf);
 950         if (err)
 951                 return err;
 952
 953         subflow = mptcp_subflow_ctx(sf->sk);
 954         subflow->remote_key = msk->remote_key;
 955         subflow->local_key = msk->local_key;
 956         subflow->token = msk->token;
 957         mptcp_info2sockaddr(loc, &addr);
 958
 959         addrlen = sizeof(struct sockaddr_in);
 960 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 961         if (loc->family == AF_INET6)
 962                 addrlen = sizeof(struct sockaddr_in6);
 963 #endif
 964         sf->sk->sk_bound_dev_if = ifindex;
 965         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
 966         if (err)
 967                 goto failed;
 968
 969         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
 970         pr_debug("msk=%p remote_token=%u", msk, remote_token);
 971         subflow->remote_token = remote_token;
 972         subflow->local_id = loc->id;
 973         subflow->request_join = 1;
 974         subflow->request_bkup = 1;
 975         mptcp_info2sockaddr(remote, &addr);
 976
 977         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
 978         if (err && err != -EINPROGRESS)
 979                 goto failed;
 980
 981         spin_lock_bh(&msk->join_list_lock);
 982         list_add_tail(&subflow->node, &msk->join_list);
 983         spin_unlock_bh(&msk->join_list_lock);
 984
 985         return err;
 986
 987 failed:
 988         sock_release(sf);
 989         return err;
 990 }
 991
 992 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
 993 {
 994         struct mptcp_subflow_context *subflow;
 995         struct net *net = sock_net(sk);
 996         struct socket *sf;
 997         int err;
 998
 999         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1000                                &sf);
1001         if (err)
1002                 return err;
1003
1004         lock_sock(sf->sk);
1005
1006         /* kernel sockets do not by default acquire net ref, but TCP timer
1007          * needs it.
1008          */
1009         sf->sk->sk_net_refcnt = 1;
1010         get_net(net);
1011 #ifdef CONFIG_PROC_FS
1012         this_cpu_add(*net->core.sock_inuse, 1);
1013 #endif
1014         err = tcp_set_ulp(sf->sk, "mptcp");
1015         release_sock(sf->sk);
1016
1017         if (err)
1018                 return err;
1019
1020         /* the newly created socket really belongs to the owning MPTCP master
1021          * socket, even if for additional subflows the allocation is performed
1022          * by a kernel workqueue. Adjust inode references, so that the
1023          * procfs/diag interaces really show this one belonging to the correct
1024          * user.
1025          */
1026         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1027         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1028         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1029
1030         subflow = mptcp_subflow_ctx(sf->sk);
1031         pr_debug("subflow=%p", subflow);
1032
1033         *new_sock = sf;
1034         sock_hold(sk);
1035         subflow->conn = sk;
1036
1037         return 0;
1038 }
1039
1040 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1041                                                         gfp_t priority)
1042 {
1043         struct inet_connection_sock *icsk = inet_csk(sk);
1044         struct mptcp_subflow_context *ctx;
1045
1046         ctx = kzalloc(sizeof(*ctx), priority);
1047         if (!ctx)
1048                 return NULL;
1049
1050         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1051         INIT_LIST_HEAD(&ctx->node);
1052
1053         pr_debug("subflow=%p", ctx);
1054
1055         ctx->tcp_sock = sk;
1056
1057         return ctx;
1058 }
1059
1060 static void __subflow_state_change(struct sock *sk)
1061 {
1062         struct socket_wq *wq;
1063
1064         rcu_read_lock();
1065         wq = rcu_dereference(sk->sk_wq);
1066         if (skwq_has_sleeper(wq))
1067                 wake_up_interruptible_all(&wq->wait);
1068         rcu_read_unlock();
1069 }
1070
1071 static bool subflow_is_done(const struct sock *sk)
1072 {
1073         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1074 }
1075
1076 static void subflow_state_change(struct sock *sk)
1077 {
1078         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1079         struct sock *parent = subflow->conn;
1080
1081         __subflow_state_change(sk);
1082
1083         /* as recvmsg() does not acquire the subflow socket for ssk selection
1084          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1085          * the data available machinery here.
1086          */
1087         if (subflow->mp_capable && mptcp_subflow_data_available(sk))
1088                 mptcp_data_ready(parent, sk);
1089
1090         if (!(parent->sk_shutdown & RCV_SHUTDOWN) &&
1091             !subflow->rx_eof && subflow_is_done(sk)) {
1092                 subflow->rx_eof = 1;
1093                 mptcp_subflow_eof(parent);
1094         }
1095 }
1096
1097 static int subflow_ulp_init(struct sock *sk)
1098 {
1099         struct inet_connection_sock *icsk = inet_csk(sk);
1100         struct mptcp_subflow_context *ctx;
1101         struct tcp_sock *tp = tcp_sk(sk);
1102         int err = 0;
1103
1104         /* disallow attaching ULP to a socket unless it has been
1105          * created with sock_create_kern()
1106          */
1107         if (!sk->sk_kern_sock) {
1108                 err = -EOPNOTSUPP;
1109                 goto out;
1110         }
1111
1112         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1113         if (!ctx) {
1114                 err = -ENOMEM;
1115                 goto out;
1116         }
1117
1118         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1119
1120         tp->is_mptcp = 1;
1121         ctx->icsk_af_ops = icsk->icsk_af_ops;
1122         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1123         ctx->tcp_data_ready = sk->sk_data_ready;
1124         ctx->tcp_state_change = sk->sk_state_change;
1125         ctx->tcp_write_space = sk->sk_write_space;
1126         sk->sk_data_ready = subflow_data_ready;
1127         sk->sk_write_space = subflow_write_space;
1128         sk->sk_state_change = subflow_state_change;
1129 out:
1130         return err;
1131 }
1132
1133 static void subflow_ulp_release(struct sock *sk)
1134 {
1135         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(sk);
1136
1137         if (!ctx)
1138                 return;
1139
1140         if (ctx->conn)
1141                 sock_put(ctx->conn);
1142
1143         kfree_rcu(ctx, rcu);
1144 }
1145
1146 static void subflow_ulp_clone(const struct request_sock *req,
1147                               struct sock *newsk,
1148                               const gfp_t priority)
1149 {
1150         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1151         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1152         struct mptcp_subflow_context *new_ctx;
1153
1154         if (!tcp_rsk(req)->is_mptcp ||
1155             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1156                 subflow_ulp_fallback(newsk, old_ctx);
1157                 return;
1158         }
1159
1160         new_ctx = subflow_create_ctx(newsk, priority);
1161         if (!new_ctx) {
1162                 subflow_ulp_fallback(newsk, old_ctx);
1163                 return;
1164         }
1165
1166         new_ctx->conn_finished = 1;
1167         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1168         new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1169         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1170         new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1171         new_ctx->rel_write_seq = 1;
1172         new_ctx->tcp_sock = newsk;
1173
1174         if (subflow_req->mp_capable) {
1175                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1176                  * is fully established only after we receive the remote key
1177                  */
1178                 new_ctx->mp_capable = 1;
1179                 new_ctx->local_key = subflow_req->local_key;
1180                 new_ctx->token = subflow_req->token;
1181                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1182                 new_ctx->idsn = subflow_req->idsn;
1183         } else if (subflow_req->mp_join) {
1184                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1185                 new_ctx->mp_join = 1;
1186                 new_ctx->fully_established = 1;
1187                 new_ctx->backup = subflow_req->backup;
1188                 new_ctx->local_id = subflow_req->local_id;
1189                 new_ctx->token = subflow_req->token;
1190                 new_ctx->thmac = subflow_req->thmac;
1191         }
1192 }
1193
1194 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1195         .name           = "mptcp",
1196         .owner          = THIS_MODULE,
1197         .init           = subflow_ulp_init,
1198         .release        = subflow_ulp_release,
1199         .clone          = subflow_ulp_clone,
1200 };
1201
1202 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1203 {
1204         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1205         subflow_ops->slab_name = "request_sock_subflow";
1206
1207         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1208                                               subflow_ops->obj_size, 0,
1209                                               SLAB_ACCOUNT |
1210                                               SLAB_TYPESAFE_BY_RCU,
1211                                               NULL);
1212         if (!subflow_ops->slab)
1213                 return -ENOMEM;
1214
1215         subflow_ops->destructor = subflow_req_destructor;
1216
1217         return 0;
1218 }
1219
1220 void mptcp_subflow_init(void)
1221 {
1222         subflow_request_sock_ops = tcp_request_sock_ops;
1223         if (subflow_ops_init(&subflow_request_sock_ops) != 0)
1224                 panic("MPTCP: failed to init subflow request sock ops\n");
1225
1226         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1227         subflow_request_sock_ipv4_ops.init_req = subflow_v4_init_req;
1228
1229         subflow_specific = ipv4_specific;
1230         subflow_specific.conn_request = subflow_v4_conn_request;
1231         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1232         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1233         subflow_specific.rebuild_header = subflow_rebuild_header;
1234
1235 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1236         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1237         subflow_request_sock_ipv6_ops.init_req = subflow_v6_init_req;
1238
1239         subflow_v6_specific = ipv6_specific;
1240         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1241         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1242         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1243         subflow_v6_specific.rebuild_header = subflow_rebuild_header;
1244
1245         subflow_v6m_specific = subflow_v6_specific;
1246         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1247         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1248         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1249         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1250         subflow_v6m_specific.net_frag_header_len = 0;
1251 #endif
1252
1253         mptcp_diag_subflow_init(&subflow_ulp_ops);
1254
1255         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1256                 panic("MPTCP: failed to register subflows to ULP\n");
1257 }