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