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         mptcp_token_destroy(mptcp_sk(sk)->token);
 397         inet_sock_destruct(sk);
 398 }
 399
 400 static void mptcp_force_close(struct sock *sk)
 401 {
 402         inet_sk_state_store(sk, TCP_CLOSE);
 403         sk_common_release(sk);
 404 }
 405
 406 static void subflow_ulp_fallback(struct sock *sk,
 407                                  struct mptcp_subflow_context *old_ctx)
 408 {
 409         struct inet_connection_sock *icsk = inet_csk(sk);
 410
 411         mptcp_subflow_tcp_fallback(sk, old_ctx);
 412         icsk->icsk_ulp_ops = NULL;
 413         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 414         tcp_sk(sk)->is_mptcp = 0;
 415 }
 416
 417 static void subflow_drop_ctx(struct sock *ssk)
 418 {
 419         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 420
 421         if (!ctx)
 422                 return;
 423
 424         subflow_ulp_fallback(ssk, ctx);
 425         if (ctx->conn)
 426                 sock_put(ctx->conn);
 427
 428         kfree_rcu(ctx, rcu);
 429 }
 430
 431 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 432                                           struct sk_buff *skb,
 433                                           struct request_sock *req,
 434                                           struct dst_entry *dst,
 435                                           struct request_sock *req_unhash,
 436                                           bool *own_req)
 437 {
 438         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 439         struct mptcp_subflow_request_sock *subflow_req;
 440         struct mptcp_options_received mp_opt;
 441         bool fallback_is_fatal = false;
 442         struct sock *new_msk = NULL;
 443         bool fallback = false;
 444         struct sock *child;
 445
 446         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 447
 448         /* we need later a valid 'mp_capable' value even when options are not
 449          * parsed
 450          */
 451         mp_opt.mp_capable = 0;
 452         if (tcp_rsk(req)->is_mptcp == 0)
 453                 goto create_child;
 454
 455         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 456         subflow_req = mptcp_subflow_rsk(req);
 457         if (subflow_req->mp_capable) {
 458                 if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
 459                         /* here we can receive and accept an in-window,
 460                          * out-of-order pkt, which will not carry the MP_CAPABLE
 461                          * opt even on mptcp enabled paths
 462                          */
 463                         goto create_msk;
 464                 }
 465
 466                 mptcp_get_options(skb, &mp_opt);
 467                 if (!mp_opt.mp_capable) {
 468                         fallback = true;
 469                         goto create_child;
 470                 }
 471
 472 create_msk:
 473                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 474                 if (!new_msk)
 475                         fallback = true;
 476         } else if (subflow_req->mp_join) {
 477                 fallback_is_fatal = true;
 478                 mptcp_get_options(skb, &mp_opt);
 479                 if (!mp_opt.mp_join ||
 480                     !subflow_hmac_valid(req, &mp_opt)) {
 481                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 482                         return NULL;
 483                 }
 484         }
 485
 486 create_child:
 487         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 488                                                      req_unhash, own_req);
 489
 490         if (child && *own_req) {
 491                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 492
 493                 tcp_rsk(req)->drop_req = false;
 494
 495                 /* we need to fallback on ctx allocation failure and on pre-reqs
 496                  * checking above. In the latter scenario we additionally need
 497                  * to reset the context to non MPTCP status.
 498                  */
 499                 if (!ctx || fallback) {
 500                         if (fallback_is_fatal)
 501                                 goto dispose_child;
 502
 503                         subflow_drop_ctx(child);
 504                         goto out;
 505                 }
 506
 507                 if (ctx->mp_capable) {
 508                         /* new mpc subflow takes ownership of the newly
 509                          * created mptcp socket
 510                          */
 511                         new_msk->sk_destruct = mptcp_sock_destruct;
 512                         mptcp_pm_new_connection(mptcp_sk(new_msk), 1);
 513                         ctx->conn = new_msk;
 514                         new_msk = NULL;
 515
 516                         /* with OoO packets we can reach here without ingress
 517                          * mpc option
 518                          */
 519                         ctx->remote_key = mp_opt.sndr_key;
 520                         ctx->fully_established = mp_opt.mp_capable;
 521                         ctx->can_ack = mp_opt.mp_capable;
 522                 } else if (ctx->mp_join) {
 523                         struct mptcp_sock *owner;
 524
 525                         owner = mptcp_token_get_sock(ctx->token);
 526                         if (!owner)
 527                                 goto dispose_child;
 528
 529                         ctx->conn = (struct sock *)owner;
 530                         if (!mptcp_finish_join(child))
 531                                 goto dispose_child;
 532
 533                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 534                         tcp_rsk(req)->drop_req = true;
 535                 }
 536         }
 537
 538 out:
 539         /* dispose of the left over mptcp master, if any */
 540         if (unlikely(new_msk))
 541                 mptcp_force_close(new_msk);
 542
 543         /* check for expected invariant - should never trigger, just help
 544          * catching eariler subtle bugs
 545          */
 546         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 547                      (!mptcp_subflow_ctx(child) ||
 548                       !mptcp_subflow_ctx(child)->conn));
 549         return child;
 550
 551 dispose_child:
 552         subflow_drop_ctx(child);
 553         tcp_rsk(req)->drop_req = true;
 554         tcp_send_active_reset(child, GFP_ATOMIC);
 555         inet_csk_prepare_for_destroy_sock(child);
 556         tcp_done(child);
 557
 558         /* The last child reference will be released by the caller */
 559         return child;
 560 }
 561
 562 static struct inet_connection_sock_af_ops subflow_specific;
 563
 564 enum mapping_status {
 565         MAPPING_OK,
 566         MAPPING_INVALID,
 567         MAPPING_EMPTY,
 568         MAPPING_DATA_FIN
 569 };
 570
 571 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
 572 {
 573         if ((u32)seq == (u32)old_seq)
 574                 return old_seq;
 575
 576         /* Assume map covers data not mapped yet. */
 577         return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
 578 }
 579
 580 static void warn_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 581 {
 582         WARN_ONCE(1, "Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 583                   ssn, subflow->map_subflow_seq, subflow->map_data_len);
 584 }
 585
 586 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 587 {
 588         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 589         unsigned int skb_consumed;
 590
 591         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 592         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 593                 return true;
 594
 595         return skb->len - skb_consumed <= subflow->map_data_len -
 596                                           mptcp_subflow_get_map_offset(subflow);
 597 }
 598
 599 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 600 {
 601         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 602         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 603
 604         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 605                 /* Mapping covers data later in the subflow stream,
 606                  * currently unsupported.
 607                  */
 608                 warn_bad_map(subflow, ssn);
 609                 return false;
 610         }
 611         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 612                                   subflow->map_data_len))) {
 613                 /* Mapping does covers past subflow data, invalid */
 614                 warn_bad_map(subflow, ssn + skb->len);
 615                 return false;
 616         }
 617         return true;
 618 }
 619
 620 static enum mapping_status get_mapping_status(struct sock *ssk)
 621 {
 622         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 623         struct mptcp_ext *mpext;
 624         struct sk_buff *skb;
 625         u16 data_len;
 626         u64 map_seq;
 627
 628         skb = skb_peek(&ssk->sk_receive_queue);
 629         if (!skb)
 630                 return MAPPING_EMPTY;
 631
 632         mpext = mptcp_get_ext(skb);
 633         if (!mpext || !mpext->use_map) {
 634                 if (!subflow->map_valid && !skb->len) {
 635                         /* the TCP stack deliver 0 len FIN pkt to the receive
 636                          * queue, that is the only 0len pkts ever expected here,
 637                          * and we can admit no mapping only for 0 len pkts
 638                          */
 639                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 640                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 641                                           TCP_SKB_CB(skb)->seq,
 642                                           TCP_SKB_CB(skb)->end_seq,
 643                                           TCP_SKB_CB(skb)->tcp_flags);
 644                         sk_eat_skb(ssk, skb);
 645                         return MAPPING_EMPTY;
 646                 }
 647
 648                 if (!subflow->map_valid)
 649                         return MAPPING_INVALID;
 650
 651                 goto validate_seq;
 652         }
 653
 654         pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
 655                  mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
 656                  mpext->data_len, mpext->data_fin);
 657
 658         data_len = mpext->data_len;
 659         if (data_len == 0) {
 660                 pr_err("Infinite mapping not handled");
 661                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 662                 return MAPPING_INVALID;
 663         }
 664
 665         if (mpext->data_fin == 1) {
 666                 if (data_len == 1) {
 667                         pr_debug("DATA_FIN with no payload");
 668                         if (subflow->map_valid) {
 669                                 /* A DATA_FIN might arrive in a DSS
 670                                  * option before the previous mapping
 671                                  * has been fully consumed. Continue
 672                                  * handling the existing mapping.
 673                                  */
 674                                 skb_ext_del(skb, SKB_EXT_MPTCP);
 675                                 return MAPPING_OK;
 676                         } else {
 677                                 return MAPPING_DATA_FIN;
 678                         }
 679                 }
 680
 681                 /* Adjust for DATA_FIN using 1 byte of sequence space */
 682                 data_len--;
 683         }
 684
 685         if (!mpext->dsn64) {
 686                 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
 687                                      mpext->data_seq);
 688                 subflow->use_64bit_ack = 0;
 689                 pr_debug("expanded seq=%llu", subflow->map_seq);
 690         } else {
 691                 map_seq = mpext->data_seq;
 692                 subflow->use_64bit_ack = 1;
 693         }
 694
 695         if (subflow->map_valid) {
 696                 /* Allow replacing only with an identical map */
 697                 if (subflow->map_seq == map_seq &&
 698                     subflow->map_subflow_seq == mpext->subflow_seq &&
 699                     subflow->map_data_len == data_len) {
 700                         skb_ext_del(skb, SKB_EXT_MPTCP);
 701                         return MAPPING_OK;
 702                 }
 703
 704                 /* If this skb data are fully covered by the current mapping,
 705                  * the new map would need caching, which is not supported
 706                  */
 707                 if (skb_is_fully_mapped(ssk, skb)) {
 708                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
 709                         return MAPPING_INVALID;
 710                 }
 711
 712                 /* will validate the next map after consuming the current one */
 713                 return MAPPING_OK;
 714         }
 715
 716         subflow->map_seq = map_seq;
 717         subflow->map_subflow_seq = mpext->subflow_seq;
 718         subflow->map_data_len = data_len;
 719         subflow->map_valid = 1;
 720         subflow->mpc_map = mpext->mpc_map;
 721         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
 722                  subflow->map_seq, subflow->map_subflow_seq,
 723                  subflow->map_data_len);
 724
 725 validate_seq:
 726         /* we revalidate valid mapping on new skb, because we must ensure
 727          * the current skb is completely covered by the available mapping
 728          */
 729         if (!validate_mapping(ssk, skb))
 730                 return MAPPING_INVALID;
 731
 732         skb_ext_del(skb, SKB_EXT_MPTCP);
 733         return MAPPING_OK;
 734 }
 735
 736 static int subflow_read_actor(read_descriptor_t *desc,
 737                               struct sk_buff *skb,
 738                               unsigned int offset, size_t len)
 739 {
 740         size_t copy_len = min(desc->count, len);
 741
 742         desc->count -= copy_len;
 743
 744         pr_debug("flushed %zu bytes, %zu left", copy_len, desc->count);
 745         return copy_len;
 746 }
 747
 748 static bool subflow_check_data_avail(struct sock *ssk)
 749 {
 750         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 751         enum mapping_status status;
 752         struct mptcp_sock *msk;
 753         struct sk_buff *skb;
 754
 755         pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
 756                  subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
 757         if (subflow->data_avail)
 758                 return true;
 759
 760         msk = mptcp_sk(subflow->conn);
 761         for (;;) {
 762                 u32 map_remaining;
 763                 size_t delta;
 764                 u64 ack_seq;
 765                 u64 old_ack;
 766
 767                 status = get_mapping_status(ssk);
 768                 pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
 769                 if (status == MAPPING_INVALID) {
 770                         ssk->sk_err = EBADMSG;
 771                         goto fatal;
 772                 }
 773
 774                 if (status != MAPPING_OK)
 775                         return false;
 776
 777                 skb = skb_peek(&ssk->sk_receive_queue);
 778                 if (WARN_ON_ONCE(!skb))
 779                         return false;
 780
 781                 /* if msk lacks the remote key, this subflow must provide an
 782                  * MP_CAPABLE-based mapping
 783                  */
 784                 if (unlikely(!READ_ONCE(msk->can_ack))) {
 785                         if (!subflow->mpc_map) {
 786                                 ssk->sk_err = EBADMSG;
 787                                 goto fatal;
 788                         }
 789                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
 790                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
 791                         WRITE_ONCE(msk->can_ack, true);
 792                 }
 793
 794                 old_ack = READ_ONCE(msk->ack_seq);
 795                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
 796                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
 797                          ack_seq);
 798                 if (ack_seq == old_ack)
 799                         break;
 800
 801                 /* only accept in-sequence mapping. Old values are spurious
 802                  * retransmission; we can hit "future" values on active backup
 803                  * subflow switch, we relay on retransmissions to get
 804                  * in-sequence data.
 805                  * Cuncurrent subflows support will require subflow data
 806                  * reordering
 807                  */
 808                 map_remaining = subflow->map_data_len -
 809                                 mptcp_subflow_get_map_offset(subflow);
 810                 if (before64(ack_seq, old_ack))
 811                         delta = min_t(size_t, old_ack - ack_seq, map_remaining);
 812                 else
 813                         delta = min_t(size_t, ack_seq - old_ack, map_remaining);
 814
 815                 /* discard mapped data */
 816                 pr_debug("discarding %zu bytes, current map len=%d", delta,
 817                          map_remaining);
 818                 if (delta) {
 819                         read_descriptor_t desc = {
 820                                 .count = delta,
 821                         };
 822                         int ret;
 823
 824                         ret = tcp_read_sock(ssk, &desc, subflow_read_actor);
 825                         if (ret < 0) {
 826                                 ssk->sk_err = -ret;
 827                                 goto fatal;
 828                         }
 829                         if (ret < delta)
 830                                 return false;
 831                         if (delta == map_remaining)
 832                                 subflow->map_valid = 0;
 833                 }
 834         }
 835         return true;
 836
 837 fatal:
 838         /* fatal protocol error, close the socket */
 839         /* This barrier is coupled with smp_rmb() in tcp_poll() */
 840         smp_wmb();
 841         ssk->sk_error_report(ssk);
 842         tcp_set_state(ssk, TCP_CLOSE);
 843         tcp_send_active_reset(ssk, GFP_ATOMIC);
 844         return false;
 845 }
 846
 847 bool mptcp_subflow_data_available(struct sock *sk)
 848 {
 849         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 850         struct sk_buff *skb;
 851
 852         /* check if current mapping is still valid */
 853         if (subflow->map_valid &&
 854             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
 855                 subflow->map_valid = 0;
 856                 subflow->data_avail = 0;
 857
 858                 pr_debug("Done with mapping: seq=%u data_len=%u",
 859                          subflow->map_subflow_seq,
 860                          subflow->map_data_len);
 861         }
 862
 863         if (!subflow_check_data_avail(sk)) {
 864                 subflow->data_avail = 0;
 865                 return false;
 866         }
 867
 868         skb = skb_peek(&sk->sk_receive_queue);
 869         subflow->data_avail = skb &&
 870                        before(tcp_sk(sk)->copied_seq, TCP_SKB_CB(skb)->end_seq);
 871         return subflow->data_avail;
 872 }
 873
 874 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
 875  * not the ssk one.
 876  *
 877  * In mptcp, rwin is about the mptcp-level connection data.
 878  *
 879  * Data that is still on the ssk rx queue can thus be ignored,
 880  * as far as mptcp peer is concerened that data is still inflight.
 881  * DSS ACK is updated when skb is moved to the mptcp rx queue.
 882  */
 883 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
 884 {
 885         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 886         const struct sock *sk = subflow->conn;
 887
 888         *space = tcp_space(sk);
 889         *full_space = tcp_full_space(sk);
 890 }
 891
 892 static void subflow_data_ready(struct sock *sk)
 893 {
 894         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 895         struct sock *parent = subflow->conn;
 896
 897         if (!subflow->mp_capable && !subflow->mp_join) {
 898                 subflow->tcp_data_ready(sk);
 899
 900                 parent->sk_data_ready(parent);
 901                 return;
 902         }
 903
 904         if (mptcp_subflow_data_available(sk))
 905                 mptcp_data_ready(parent, sk);
 906 }
 907
 908 static void subflow_write_space(struct sock *sk)
 909 {
 910         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 911         struct sock *parent = subflow->conn;
 912
 913         sk_stream_write_space(sk);
 914         if (sk_stream_is_writeable(sk)) {
 915                 set_bit(MPTCP_SEND_SPACE, &mptcp_sk(parent)->flags);
 916                 smp_mb__after_atomic();
 917                 /* set SEND_SPACE before sk_stream_write_space clears NOSPACE */
 918                 sk_stream_write_space(parent);
 919         }
 920 }
 921
 922 static struct inet_connection_sock_af_ops *
 923 subflow_default_af_ops(struct sock *sk)
 924 {
 925 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 926         if (sk->sk_family == AF_INET6)
 927                 return &subflow_v6_specific;
 928 #endif
 929         return &subflow_specific;
 930 }
 931
 932 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 933 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
 934 {
 935         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 936         struct inet_connection_sock *icsk = inet_csk(sk);
 937         struct inet_connection_sock_af_ops *target;
 938
 939         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
 940
 941         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
 942                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
 943
 944         if (likely(icsk->icsk_af_ops == target))
 945                 return;
 946
 947         subflow->icsk_af_ops = icsk->icsk_af_ops;
 948         icsk->icsk_af_ops = target;
 949 }
 950 #endif
 951
 952 static void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
 953                                 struct sockaddr_storage *addr)
 954 {
 955         memset(addr, 0, sizeof(*addr));
 956         addr->ss_family = info->family;
 957         if (addr->ss_family == AF_INET) {
 958                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
 959
 960                 in_addr->sin_addr = info->addr;
 961                 in_addr->sin_port = info->port;
 962         }
 963 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 964         else if (addr->ss_family == AF_INET6) {
 965                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
 966
 967                 in6_addr->sin6_addr = info->addr6;
 968                 in6_addr->sin6_port = info->port;
 969         }
 970 #endif
 971 }
 972
 973 int __mptcp_subflow_connect(struct sock *sk, int ifindex,
 974                             const struct mptcp_addr_info *loc,
 975                             const struct mptcp_addr_info *remote)
 976 {
 977         struct mptcp_sock *msk = mptcp_sk(sk);
 978         struct mptcp_subflow_context *subflow;
 979         struct sockaddr_storage addr;
 980         struct socket *sf;
 981         u32 remote_token;
 982         int addrlen;
 983         int err;
 984
 985         if (sk->sk_state != TCP_ESTABLISHED)
 986                 return -ENOTCONN;
 987
 988         err = mptcp_subflow_create_socket(sk, &sf);
 989         if (err)
 990                 return err;
 991
 992         subflow = mptcp_subflow_ctx(sf->sk);
 993         subflow->remote_key = msk->remote_key;
 994         subflow->local_key = msk->local_key;
 995         subflow->token = msk->token;
 996         mptcp_info2sockaddr(loc, &addr);
 997
 998         addrlen = sizeof(struct sockaddr_in);
 999 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1000         if (loc->family == AF_INET6)
1001                 addrlen = sizeof(struct sockaddr_in6);
1002 #endif
1003         sf->sk->sk_bound_dev_if = ifindex;
1004         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1005         if (err)
1006                 goto failed;
1007
1008         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1009         pr_debug("msk=%p remote_token=%u", msk, remote_token);
1010         subflow->remote_token = remote_token;
1011         subflow->local_id = loc->id;
1012         subflow->request_join = 1;
1013         subflow->request_bkup = 1;
1014         mptcp_info2sockaddr(remote, &addr);
1015
1016         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1017         if (err && err != -EINPROGRESS)
1018                 goto failed;
1019
1020         spin_lock_bh(&msk->join_list_lock);
1021         list_add_tail(&subflow->node, &msk->join_list);
1022         spin_unlock_bh(&msk->join_list_lock);
1023
1024         return err;
1025
1026 failed:
1027         sock_release(sf);
1028         return err;
1029 }
1030
1031 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1032 {
1033         struct mptcp_subflow_context *subflow;
1034         struct net *net = sock_net(sk);
1035         struct socket *sf;
1036         int err;
1037
1038         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1039                                &sf);
1040         if (err)
1041                 return err;
1042
1043         lock_sock(sf->sk);
1044
1045         /* kernel sockets do not by default acquire net ref, but TCP timer
1046          * needs it.
1047          */
1048         sf->sk->sk_net_refcnt = 1;
1049         get_net(net);
1050 #ifdef CONFIG_PROC_FS
1051         this_cpu_add(*net->core.sock_inuse, 1);
1052 #endif
1053         err = tcp_set_ulp(sf->sk, "mptcp");
1054         release_sock(sf->sk);
1055
1056         if (err)
1057                 return err;
1058
1059         /* the newly created socket really belongs to the owning MPTCP master
1060          * socket, even if for additional subflows the allocation is performed
1061          * by a kernel workqueue. Adjust inode references, so that the
1062          * procfs/diag interaces really show this one belonging to the correct
1063          * user.
1064          */
1065         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1066         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1067         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1068
1069         subflow = mptcp_subflow_ctx(sf->sk);
1070         pr_debug("subflow=%p", subflow);
1071
1072         *new_sock = sf;
1073         sock_hold(sk);
1074         subflow->conn = sk;
1075
1076         return 0;
1077 }
1078
1079 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1080                                                         gfp_t priority)
1081 {
1082         struct inet_connection_sock *icsk = inet_csk(sk);
1083         struct mptcp_subflow_context *ctx;
1084
1085         ctx = kzalloc(sizeof(*ctx), priority);
1086         if (!ctx)
1087                 return NULL;
1088
1089         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1090         INIT_LIST_HEAD(&ctx->node);
1091
1092         pr_debug("subflow=%p", ctx);
1093
1094         ctx->tcp_sock = sk;
1095
1096         return ctx;
1097 }
1098
1099 static void __subflow_state_change(struct sock *sk)
1100 {
1101         struct socket_wq *wq;
1102
1103         rcu_read_lock();
1104         wq = rcu_dereference(sk->sk_wq);
1105         if (skwq_has_sleeper(wq))
1106                 wake_up_interruptible_all(&wq->wait);
1107         rcu_read_unlock();
1108 }
1109
1110 static bool subflow_is_done(const struct sock *sk)
1111 {
1112         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1113 }
1114
1115 static void subflow_state_change(struct sock *sk)
1116 {
1117         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1118         struct sock *parent = subflow->conn;
1119
1120         __subflow_state_change(sk);
1121
1122         /* as recvmsg() does not acquire the subflow socket for ssk selection
1123          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1124          * the data available machinery here.
1125          */
1126         if (subflow->mp_capable && mptcp_subflow_data_available(sk))
1127                 mptcp_data_ready(parent, sk);
1128
1129         if (!(parent->sk_shutdown & RCV_SHUTDOWN) &&
1130             !subflow->rx_eof && subflow_is_done(sk)) {
1131                 subflow->rx_eof = 1;
1132                 mptcp_subflow_eof(parent);
1133         }
1134 }
1135
1136 static int subflow_ulp_init(struct sock *sk)
1137 {
1138         struct inet_connection_sock *icsk = inet_csk(sk);
1139         struct mptcp_subflow_context *ctx;
1140         struct tcp_sock *tp = tcp_sk(sk);
1141         int err = 0;
1142
1143         /* disallow attaching ULP to a socket unless it has been
1144          * created with sock_create_kern()
1145          */
1146         if (!sk->sk_kern_sock) {
1147                 err = -EOPNOTSUPP;
1148                 goto out;
1149         }
1150
1151         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1152         if (!ctx) {
1153                 err = -ENOMEM;
1154                 goto out;
1155         }
1156
1157         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1158
1159         tp->is_mptcp = 1;
1160         ctx->icsk_af_ops = icsk->icsk_af_ops;
1161         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1162         ctx->tcp_data_ready = sk->sk_data_ready;
1163         ctx->tcp_state_change = sk->sk_state_change;
1164         ctx->tcp_write_space = sk->sk_write_space;
1165         sk->sk_data_ready = subflow_data_ready;
1166         sk->sk_write_space = subflow_write_space;
1167         sk->sk_state_change = subflow_state_change;
1168 out:
1169         return err;
1170 }
1171
1172 static void subflow_ulp_release(struct sock *sk)
1173 {
1174         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(sk);
1175
1176         if (!ctx)
1177                 return;
1178
1179         if (ctx->conn)
1180                 sock_put(ctx->conn);
1181
1182         kfree_rcu(ctx, rcu);
1183 }
1184
1185 static void subflow_ulp_clone(const struct request_sock *req,
1186                               struct sock *newsk,
1187                               const gfp_t priority)
1188 {
1189         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1190         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1191         struct mptcp_subflow_context *new_ctx;
1192
1193         if (!tcp_rsk(req)->is_mptcp ||
1194             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1195                 subflow_ulp_fallback(newsk, old_ctx);
1196                 return;
1197         }
1198
1199         new_ctx = subflow_create_ctx(newsk, priority);
1200         if (!new_ctx) {
1201                 subflow_ulp_fallback(newsk, old_ctx);
1202                 return;
1203         }
1204
1205         new_ctx->conn_finished = 1;
1206         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1207         new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1208         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1209         new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1210         new_ctx->rel_write_seq = 1;
1211         new_ctx->tcp_sock = newsk;
1212
1213         if (subflow_req->mp_capable) {
1214                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1215                  * is fully established only after we receive the remote key
1216                  */
1217                 new_ctx->mp_capable = 1;
1218                 new_ctx->local_key = subflow_req->local_key;
1219                 new_ctx->token = subflow_req->token;
1220                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1221                 new_ctx->idsn = subflow_req->idsn;
1222         } else if (subflow_req->mp_join) {
1223                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1224                 new_ctx->mp_join = 1;
1225                 new_ctx->fully_established = 1;
1226                 new_ctx->backup = subflow_req->backup;
1227                 new_ctx->local_id = subflow_req->local_id;
1228                 new_ctx->token = subflow_req->token;
1229                 new_ctx->thmac = subflow_req->thmac;
1230         }
1231 }
1232
1233 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1234         .name           = "mptcp",
1235         .owner          = THIS_MODULE,
1236         .init           = subflow_ulp_init,
1237         .release        = subflow_ulp_release,
1238         .clone          = subflow_ulp_clone,
1239 };
1240
1241 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1242 {
1243         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1244         subflow_ops->slab_name = "request_sock_subflow";
1245
1246         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1247                                               subflow_ops->obj_size, 0,
1248                                               SLAB_ACCOUNT |
1249                                               SLAB_TYPESAFE_BY_RCU,
1250                                               NULL);
1251         if (!subflow_ops->slab)
1252                 return -ENOMEM;
1253
1254         subflow_ops->destructor = subflow_req_destructor;
1255
1256         return 0;
1257 }
1258
1259 void mptcp_subflow_init(void)
1260 {
1261         subflow_request_sock_ops = tcp_request_sock_ops;
1262         if (subflow_ops_init(&subflow_request_sock_ops) != 0)
1263                 panic("MPTCP: failed to init subflow request sock ops\n");
1264
1265         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1266         subflow_request_sock_ipv4_ops.init_req = subflow_v4_init_req;
1267
1268         subflow_specific = ipv4_specific;
1269         subflow_specific.conn_request = subflow_v4_conn_request;
1270         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1271         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1272         subflow_specific.rebuild_header = subflow_rebuild_header;
1273
1274 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1275         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1276         subflow_request_sock_ipv6_ops.init_req = subflow_v6_init_req;
1277
1278         subflow_v6_specific = ipv6_specific;
1279         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1280         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1281         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1282         subflow_v6_specific.rebuild_header = subflow_rebuild_header;
1283
1284         subflow_v6m_specific = subflow_v6_specific;
1285         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1286         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1287         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1288         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1289         subflow_v6m_specific.net_frag_header_len = 0;
1290 #endif
1291
1292         mptcp_diag_subflow_init(&subflow_ulp_ops);
1293
1294         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1295                 panic("MPTCP: failed to register subflows to ULP\n");
1296 }