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 void subflow_req_destructor(struct request_sock *req)
  33 {
  34         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  35
  36         pr_debug("subflow_req=%p", subflow_req);
  37
  38         if (subflow_req->msk)
  39                 sock_put((struct sock *)subflow_req->msk);
  40
  41         mptcp_token_destroy_request(req);
  42         tcp_request_sock_ops.destructor(req);
  43 }
  44
  45 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
  46                                   void *hmac)
  47 {
  48         u8 msg[8];
  49
  50         put_unaligned_be32(nonce1, &msg[0]);
  51         put_unaligned_be32(nonce2, &msg[4]);
  52
  53         mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
  54 }
  55
  56 /* validate received token and create truncated hmac and nonce for SYN-ACK */
  57 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req,
  58                                                      const struct sk_buff *skb)
  59 {
  60         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  61         u8 hmac[SHA256_DIGEST_SIZE];
  62         struct mptcp_sock *msk;
  63         int local_id;
  64
  65         msk = mptcp_token_get_sock(subflow_req->token);
  66         if (!msk) {
  67                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
  68                 return NULL;
  69         }
  70
  71         local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
  72         if (local_id < 0) {
  73                 sock_put((struct sock *)msk);
  74                 return NULL;
  75         }
  76         subflow_req->local_id = local_id;
  77
  78         get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
  79
  80         subflow_generate_hmac(msk->local_key, msk->remote_key,
  81                               subflow_req->local_nonce,
  82                               subflow_req->remote_nonce, hmac);
  83
  84         subflow_req->thmac = get_unaligned_be64(hmac);
  85         return msk;
  86 }
  87
  88 static void subflow_init_req(struct request_sock *req,
  89                              const struct sock *sk_listener,
  90                              struct sk_buff *skb)
  91 {
  92         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
  93         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  94         struct mptcp_options_received mp_opt;
  95
  96         pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
  97
  98         mptcp_get_options(skb, &mp_opt);
  99
 100         subflow_req->mp_capable = 0;
 101         subflow_req->mp_join = 0;
 102         subflow_req->msk = NULL;
 103         mptcp_token_init_request(req);
 104
 105 #ifdef CONFIG_TCP_MD5SIG
 106         /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
 107          * TCP option space.
 108          */
 109         if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
 110                 return;
 111 #endif
 112
 113         if (mp_opt.mp_capable) {
 114                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
 115
 116                 if (mp_opt.mp_join)
 117                         return;
 118         } else if (mp_opt.mp_join) {
 119                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
 120         }
 121
 122         if (mp_opt.mp_capable && listener->request_mptcp) {
 123                 int err;
 124
 125                 err = mptcp_token_new_request(req);
 126                 if (err == 0)
 127                         subflow_req->mp_capable = 1;
 128
 129                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 130         } else if (mp_opt.mp_join && listener->request_mptcp) {
 131                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 132                 subflow_req->mp_join = 1;
 133                 subflow_req->backup = mp_opt.backup;
 134                 subflow_req->remote_id = mp_opt.join_id;
 135                 subflow_req->token = mp_opt.token;
 136                 subflow_req->remote_nonce = mp_opt.nonce;
 137                 subflow_req->msk = subflow_token_join_request(req, skb);
 138                 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
 139                          subflow_req->remote_nonce, subflow_req->msk);
 140         }
 141 }
 142
 143 static void subflow_v4_init_req(struct request_sock *req,
 144                                 const struct sock *sk_listener,
 145                                 struct sk_buff *skb)
 146 {
 147         tcp_rsk(req)->is_mptcp = 1;
 148
 149         tcp_request_sock_ipv4_ops.init_req(req, sk_listener, skb);
 150
 151         subflow_init_req(req, sk_listener, skb);
 152 }
 153
 154 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 155 static void subflow_v6_init_req(struct request_sock *req,
 156                                 const struct sock *sk_listener,
 157                                 struct sk_buff *skb)
 158 {
 159         tcp_rsk(req)->is_mptcp = 1;
 160
 161         tcp_request_sock_ipv6_ops.init_req(req, sk_listener, skb);
 162
 163         subflow_init_req(req, sk_listener, skb);
 164 }
 165 #endif
 166
 167 /* validate received truncated hmac and create hmac for third ACK */
 168 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
 169 {
 170         u8 hmac[SHA256_DIGEST_SIZE];
 171         u64 thmac;
 172
 173         subflow_generate_hmac(subflow->remote_key, subflow->local_key,
 174                               subflow->remote_nonce, subflow->local_nonce,
 175                               hmac);
 176
 177         thmac = get_unaligned_be64(hmac);
 178         pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
 179                  subflow, subflow->token,
 180                  (unsigned long long)thmac,
 181                  (unsigned long long)subflow->thmac);
 182
 183         return thmac == subflow->thmac;
 184 }
 185
 186 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 187 {
 188         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 189         struct mptcp_options_received mp_opt;
 190         struct sock *parent = subflow->conn;
 191
 192         subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 193
 194         if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
 195                 inet_sk_state_store(parent, TCP_ESTABLISHED);
 196                 parent->sk_state_change(parent);
 197         }
 198
 199         /* be sure no special action on any packet other than syn-ack */
 200         if (subflow->conn_finished)
 201                 return;
 202
 203         subflow->rel_write_seq = 1;
 204         subflow->conn_finished = 1;
 205         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 206         pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
 207
 208         mptcp_get_options(skb, &mp_opt);
 209         if (subflow->request_mptcp) {
 210                 if (!mp_opt.mp_capable) {
 211                         MPTCP_INC_STATS(sock_net(sk),
 212                                         MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
 213                         mptcp_do_fallback(sk);
 214                         pr_fallback(mptcp_sk(subflow->conn));
 215                         goto fallback;
 216                 }
 217
 218                 subflow->mp_capable = 1;
 219                 subflow->can_ack = 1;
 220                 subflow->remote_key = mp_opt.sndr_key;
 221                 pr_debug("subflow=%p, remote_key=%llu", subflow,
 222                          subflow->remote_key);
 223                 mptcp_finish_connect(sk);
 224         } else if (subflow->request_join) {
 225                 u8 hmac[SHA256_DIGEST_SIZE];
 226
 227                 if (!mp_opt.mp_join)
 228                         goto do_reset;
 229
 230                 subflow->thmac = mp_opt.thmac;
 231                 subflow->remote_nonce = mp_opt.nonce;
 232                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
 233                          subflow->thmac, subflow->remote_nonce);
 234
 235                 if (!subflow_thmac_valid(subflow)) {
 236                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 237                         goto do_reset;
 238                 }
 239
 240                 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 241                                       subflow->local_nonce,
 242                                       subflow->remote_nonce,
 243                                       hmac);
 244                 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
 245
 246                 if (!mptcp_finish_join(sk))
 247                         goto do_reset;
 248
 249                 subflow->mp_join = 1;
 250                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 251         } else if (mptcp_check_fallback(sk)) {
 252 fallback:
 253                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
 254         }
 255         return;
 256
 257 do_reset:
 258         tcp_send_active_reset(sk, GFP_ATOMIC);
 259         tcp_done(sk);
 260 }
 261
 262 static struct request_sock_ops subflow_request_sock_ops;
 263 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
 264
 265 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 266 {
 267         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 268
 269         pr_debug("subflow=%p", subflow);
 270
 271         /* Never answer to SYNs sent to broadcast or multicast */
 272         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 273                 goto drop;
 274
 275         return tcp_conn_request(&subflow_request_sock_ops,
 276                                 &subflow_request_sock_ipv4_ops,
 277                                 sk, skb);
 278 drop:
 279         tcp_listendrop(sk);
 280         return 0;
 281 }
 282
 283 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 284 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
 285 static struct inet_connection_sock_af_ops subflow_v6_specific;
 286 static struct inet_connection_sock_af_ops subflow_v6m_specific;
 287
 288 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 289 {
 290         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 291
 292         pr_debug("subflow=%p", subflow);
 293
 294         if (skb->protocol == htons(ETH_P_IP))
 295                 return subflow_v4_conn_request(sk, skb);
 296
 297         if (!ipv6_unicast_destination(skb))
 298                 goto drop;
 299
 300         return tcp_conn_request(&subflow_request_sock_ops,
 301                                 &subflow_request_sock_ipv6_ops, sk, skb);
 302
 303 drop:
 304         tcp_listendrop(sk);
 305         return 0; /* don't send reset */
 306 }
 307 #endif
 308
 309 /* validate hmac received in third ACK */
 310 static bool subflow_hmac_valid(const struct request_sock *req,
 311                                const struct mptcp_options_received *mp_opt)
 312 {
 313         const struct mptcp_subflow_request_sock *subflow_req;
 314         u8 hmac[SHA256_DIGEST_SIZE];
 315         struct mptcp_sock *msk;
 316
 317         subflow_req = mptcp_subflow_rsk(req);
 318         msk = subflow_req->msk;
 319         if (!msk)
 320                 return false;
 321
 322         subflow_generate_hmac(msk->remote_key, msk->local_key,
 323                               subflow_req->remote_nonce,
 324                               subflow_req->local_nonce, hmac);
 325
 326         return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
 327 }
 328
 329 static void mptcp_sock_destruct(struct sock *sk)
 330 {
 331         /* if new mptcp socket isn't accepted, it is free'd
 332          * from the tcp listener sockets request queue, linked
 333          * from req->sk.  The tcp socket is released.
 334          * This calls the ULP release function which will
 335          * also remove the mptcp socket, via
 336          * sock_put(ctx->conn).
 337          *
 338          * Problem is that the mptcp socket will not be in
 339          * SYN_RECV state and doesn't have SOCK_DEAD flag.
 340          * Both result in warnings from inet_sock_destruct.
 341          */
 342
 343         if (sk->sk_state == TCP_SYN_RECV) {
 344                 sk->sk_state = TCP_CLOSE;
 345                 WARN_ON_ONCE(sk->sk_socket);
 346                 sock_orphan(sk);
 347         }
 348
 349         mptcp_token_destroy(mptcp_sk(sk));
 350         inet_sock_destruct(sk);
 351 }
 352
 353 static void mptcp_force_close(struct sock *sk)
 354 {
 355         inet_sk_state_store(sk, TCP_CLOSE);
 356         sk_common_release(sk);
 357 }
 358
 359 static void subflow_ulp_fallback(struct sock *sk,
 360                                  struct mptcp_subflow_context *old_ctx)
 361 {
 362         struct inet_connection_sock *icsk = inet_csk(sk);
 363
 364         mptcp_subflow_tcp_fallback(sk, old_ctx);
 365         icsk->icsk_ulp_ops = NULL;
 366         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 367         tcp_sk(sk)->is_mptcp = 0;
 368 }
 369
 370 static void subflow_drop_ctx(struct sock *ssk)
 371 {
 372         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 373
 374         if (!ctx)
 375                 return;
 376
 377         subflow_ulp_fallback(ssk, ctx);
 378         if (ctx->conn)
 379                 sock_put(ctx->conn);
 380
 381         kfree_rcu(ctx, rcu);
 382 }
 383
 384 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 385                                      struct mptcp_options_received *mp_opt)
 386 {
 387         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 388
 389         subflow->remote_key = mp_opt->sndr_key;
 390         subflow->fully_established = 1;
 391         subflow->can_ack = 1;
 392         WRITE_ONCE(msk->fully_established, true);
 393 }
 394
 395 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 396                                           struct sk_buff *skb,
 397                                           struct request_sock *req,
 398                                           struct dst_entry *dst,
 399                                           struct request_sock *req_unhash,
 400                                           bool *own_req)
 401 {
 402         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 403         struct mptcp_subflow_request_sock *subflow_req;
 404         struct mptcp_options_received mp_opt;
 405         bool fallback, fallback_is_fatal;
 406         struct sock *new_msk = NULL;
 407         struct sock *child;
 408
 409         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 410
 411         /* After child creation we must look for 'mp_capable' even when options
 412          * are not parsed
 413          */
 414         mp_opt.mp_capable = 0;
 415
 416         /* hopefully temporary handling for MP_JOIN+syncookie */
 417         subflow_req = mptcp_subflow_rsk(req);
 418         fallback_is_fatal = subflow_req->mp_join;
 419         fallback = !tcp_rsk(req)->is_mptcp;
 420         if (fallback)
 421                 goto create_child;
 422
 423         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 424         if (subflow_req->mp_capable) {
 425                 if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
 426                         /* here we can receive and accept an in-window,
 427                          * out-of-order pkt, which will not carry the MP_CAPABLE
 428                          * opt even on mptcp enabled paths
 429                          */
 430                         goto create_msk;
 431                 }
 432
 433                 mptcp_get_options(skb, &mp_opt);
 434                 if (!mp_opt.mp_capable) {
 435                         fallback = true;
 436                         goto create_child;
 437                 }
 438
 439 create_msk:
 440                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 441                 if (!new_msk)
 442                         fallback = true;
 443         } else if (subflow_req->mp_join) {
 444                 mptcp_get_options(skb, &mp_opt);
 445                 if (!mp_opt.mp_join ||
 446                     !subflow_hmac_valid(req, &mp_opt)) {
 447                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 448                         fallback = true;
 449                 }
 450         }
 451
 452 create_child:
 453         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 454                                                      req_unhash, own_req);
 455
 456         if (child && *own_req) {
 457                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 458
 459                 tcp_rsk(req)->drop_req = false;
 460
 461                 /* we need to fallback on ctx allocation failure and on pre-reqs
 462                  * checking above. In the latter scenario we additionally need
 463                  * to reset the context to non MPTCP status.
 464                  */
 465                 if (!ctx || fallback) {
 466                         if (fallback_is_fatal)
 467                                 goto dispose_child;
 468
 469                         subflow_drop_ctx(child);
 470                         goto out;
 471                 }
 472
 473                 if (ctx->mp_capable) {
 474                         /* this can't race with mptcp_close(), as the msk is
 475                          * not yet exposted to user-space
 476                          */
 477                         inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
 478
 479                         /* new mpc subflow takes ownership of the newly
 480                          * created mptcp socket
 481                          */
 482                         new_msk->sk_destruct = mptcp_sock_destruct;
 483                         mptcp_pm_new_connection(mptcp_sk(new_msk), 1);
 484                         mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
 485                         ctx->conn = new_msk;
 486                         new_msk = NULL;
 487
 488                         /* with OoO packets we can reach here without ingress
 489                          * mpc option
 490                          */
 491                         if (mp_opt.mp_capable)
 492                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 493                 } else if (ctx->mp_join) {
 494                         struct mptcp_sock *owner;
 495
 496                         owner = subflow_req->msk;
 497                         if (!owner)
 498                                 goto dispose_child;
 499
 500                         /* move the msk reference ownership to the subflow */
 501                         subflow_req->msk = NULL;
 502                         ctx->conn = (struct sock *)owner;
 503                         if (!mptcp_finish_join(child))
 504                                 goto dispose_child;
 505
 506                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 507                         tcp_rsk(req)->drop_req = true;
 508                 }
 509         }
 510
 511 out:
 512         /* dispose of the left over mptcp master, if any */
 513         if (unlikely(new_msk))
 514                 mptcp_force_close(new_msk);
 515
 516         /* check for expected invariant - should never trigger, just help
 517          * catching eariler subtle bugs
 518          */
 519         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 520                      (!mptcp_subflow_ctx(child) ||
 521                       !mptcp_subflow_ctx(child)->conn));
 522         return child;
 523
 524 dispose_child:
 525         subflow_drop_ctx(child);
 526         tcp_rsk(req)->drop_req = true;
 527         tcp_send_active_reset(child, GFP_ATOMIC);
 528         inet_csk_prepare_for_destroy_sock(child);
 529         tcp_done(child);
 530
 531         /* The last child reference will be released by the caller */
 532         return child;
 533 }
 534
 535 static struct inet_connection_sock_af_ops subflow_specific;
 536
 537 enum mapping_status {
 538         MAPPING_OK,
 539         MAPPING_INVALID,
 540         MAPPING_EMPTY,
 541         MAPPING_DATA_FIN,
 542         MAPPING_DUMMY
 543 };
 544
 545 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
 546 {
 547         if ((u32)seq == (u32)old_seq)
 548                 return old_seq;
 549
 550         /* Assume map covers data not mapped yet. */
 551         return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
 552 }
 553
 554 static void warn_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 555 {
 556         WARN_ONCE(1, "Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 557                   ssn, subflow->map_subflow_seq, subflow->map_data_len);
 558 }
 559
 560 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 561 {
 562         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 563         unsigned int skb_consumed;
 564
 565         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 566         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 567                 return true;
 568
 569         return skb->len - skb_consumed <= subflow->map_data_len -
 570                                           mptcp_subflow_get_map_offset(subflow);
 571 }
 572
 573 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 574 {
 575         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 576         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 577
 578         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 579                 /* Mapping covers data later in the subflow stream,
 580                  * currently unsupported.
 581                  */
 582                 warn_bad_map(subflow, ssn);
 583                 return false;
 584         }
 585         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 586                                   subflow->map_data_len))) {
 587                 /* Mapping does covers past subflow data, invalid */
 588                 warn_bad_map(subflow, ssn + skb->len);
 589                 return false;
 590         }
 591         return true;
 592 }
 593
 594 static enum mapping_status get_mapping_status(struct sock *ssk)
 595 {
 596         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 597         struct mptcp_ext *mpext;
 598         struct sk_buff *skb;
 599         u16 data_len;
 600         u64 map_seq;
 601
 602         skb = skb_peek(&ssk->sk_receive_queue);
 603         if (!skb)
 604                 return MAPPING_EMPTY;
 605
 606         if (mptcp_check_fallback(ssk))
 607                 return MAPPING_DUMMY;
 608
 609         mpext = mptcp_get_ext(skb);
 610         if (!mpext || !mpext->use_map) {
 611                 if (!subflow->map_valid && !skb->len) {
 612                         /* the TCP stack deliver 0 len FIN pkt to the receive
 613                          * queue, that is the only 0len pkts ever expected here,
 614                          * and we can admit no mapping only for 0 len pkts
 615                          */
 616                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 617                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 618                                           TCP_SKB_CB(skb)->seq,
 619                                           TCP_SKB_CB(skb)->end_seq,
 620                                           TCP_SKB_CB(skb)->tcp_flags);
 621                         sk_eat_skb(ssk, skb);
 622                         return MAPPING_EMPTY;
 623                 }
 624
 625                 if (!subflow->map_valid)
 626                         return MAPPING_INVALID;
 627
 628                 goto validate_seq;
 629         }
 630
 631         pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
 632                  mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
 633                  mpext->data_len, mpext->data_fin);
 634
 635         data_len = mpext->data_len;
 636         if (data_len == 0) {
 637                 pr_err("Infinite mapping not handled");
 638                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 639                 return MAPPING_INVALID;
 640         }
 641
 642         if (mpext->data_fin == 1) {
 643                 if (data_len == 1) {
 644                         pr_debug("DATA_FIN with no payload");
 645                         if (subflow->map_valid) {
 646                                 /* A DATA_FIN might arrive in a DSS
 647                                  * option before the previous mapping
 648                                  * has been fully consumed. Continue
 649                                  * handling the existing mapping.
 650                                  */
 651                                 skb_ext_del(skb, SKB_EXT_MPTCP);
 652                                 return MAPPING_OK;
 653                         } else {
 654                                 return MAPPING_DATA_FIN;
 655                         }
 656                 }
 657
 658                 /* Adjust for DATA_FIN using 1 byte of sequence space */
 659                 data_len--;
 660         }
 661
 662         if (!mpext->dsn64) {
 663                 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
 664                                      mpext->data_seq);
 665                 subflow->use_64bit_ack = 0;
 666                 pr_debug("expanded seq=%llu", subflow->map_seq);
 667         } else {
 668                 map_seq = mpext->data_seq;
 669                 subflow->use_64bit_ack = 1;
 670         }
 671
 672         if (subflow->map_valid) {
 673                 /* Allow replacing only with an identical map */
 674                 if (subflow->map_seq == map_seq &&
 675                     subflow->map_subflow_seq == mpext->subflow_seq &&
 676                     subflow->map_data_len == data_len) {
 677                         skb_ext_del(skb, SKB_EXT_MPTCP);
 678                         return MAPPING_OK;
 679                 }
 680
 681                 /* If this skb data are fully covered by the current mapping,
 682                  * the new map would need caching, which is not supported
 683                  */
 684                 if (skb_is_fully_mapped(ssk, skb)) {
 685                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
 686                         return MAPPING_INVALID;
 687                 }
 688
 689                 /* will validate the next map after consuming the current one */
 690                 return MAPPING_OK;
 691         }
 692
 693         subflow->map_seq = map_seq;
 694         subflow->map_subflow_seq = mpext->subflow_seq;
 695         subflow->map_data_len = data_len;
 696         subflow->map_valid = 1;
 697         subflow->mpc_map = mpext->mpc_map;
 698         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
 699                  subflow->map_seq, subflow->map_subflow_seq,
 700                  subflow->map_data_len);
 701
 702 validate_seq:
 703         /* we revalidate valid mapping on new skb, because we must ensure
 704          * the current skb is completely covered by the available mapping
 705          */
 706         if (!validate_mapping(ssk, skb))
 707                 return MAPPING_INVALID;
 708
 709         skb_ext_del(skb, SKB_EXT_MPTCP);
 710         return MAPPING_OK;
 711 }
 712
 713 static int subflow_read_actor(read_descriptor_t *desc,
 714                               struct sk_buff *skb,
 715                               unsigned int offset, size_t len)
 716 {
 717         size_t copy_len = min(desc->count, len);
 718
 719         desc->count -= copy_len;
 720
 721         pr_debug("flushed %zu bytes, %zu left", copy_len, desc->count);
 722         return copy_len;
 723 }
 724
 725 static bool subflow_check_data_avail(struct sock *ssk)
 726 {
 727         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 728         enum mapping_status status;
 729         struct mptcp_sock *msk;
 730         struct sk_buff *skb;
 731
 732         pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
 733                  subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
 734         if (subflow->data_avail)
 735                 return true;
 736
 737         msk = mptcp_sk(subflow->conn);
 738         for (;;) {
 739                 u32 map_remaining;
 740                 size_t delta;
 741                 u64 ack_seq;
 742                 u64 old_ack;
 743
 744                 status = get_mapping_status(ssk);
 745                 pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
 746                 if (status == MAPPING_INVALID) {
 747                         ssk->sk_err = EBADMSG;
 748                         goto fatal;
 749                 }
 750                 if (status == MAPPING_DUMMY) {
 751                         __mptcp_do_fallback(msk);
 752                         skb = skb_peek(&ssk->sk_receive_queue);
 753                         subflow->map_valid = 1;
 754                         subflow->map_seq = READ_ONCE(msk->ack_seq);
 755                         subflow->map_data_len = skb->len;
 756                         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq -
 757                                                    subflow->ssn_offset;
 758                         return true;
 759                 }
 760
 761                 if (status != MAPPING_OK)
 762                         return false;
 763
 764                 skb = skb_peek(&ssk->sk_receive_queue);
 765                 if (WARN_ON_ONCE(!skb))
 766                         return false;
 767
 768                 /* if msk lacks the remote key, this subflow must provide an
 769                  * MP_CAPABLE-based mapping
 770                  */
 771                 if (unlikely(!READ_ONCE(msk->can_ack))) {
 772                         if (!subflow->mpc_map) {
 773                                 ssk->sk_err = EBADMSG;
 774                                 goto fatal;
 775                         }
 776                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
 777                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
 778                         WRITE_ONCE(msk->can_ack, true);
 779                 }
 780
 781                 old_ack = READ_ONCE(msk->ack_seq);
 782                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
 783                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
 784                          ack_seq);
 785                 if (ack_seq == old_ack)
 786                         break;
 787
 788                 /* only accept in-sequence mapping. Old values are spurious
 789                  * retransmission; we can hit "future" values on active backup
 790                  * subflow switch, we relay on retransmissions to get
 791                  * in-sequence data.
 792                  * Cuncurrent subflows support will require subflow data
 793                  * reordering
 794                  */
 795                 map_remaining = subflow->map_data_len -
 796                                 mptcp_subflow_get_map_offset(subflow);
 797                 if (before64(ack_seq, old_ack))
 798                         delta = min_t(size_t, old_ack - ack_seq, map_remaining);
 799                 else
 800                         delta = min_t(size_t, ack_seq - old_ack, map_remaining);
 801
 802                 /* discard mapped data */
 803                 pr_debug("discarding %zu bytes, current map len=%d", delta,
 804                          map_remaining);
 805                 if (delta) {
 806                         read_descriptor_t desc = {
 807                                 .count = delta,
 808                         };
 809                         int ret;
 810
 811                         ret = tcp_read_sock(ssk, &desc, subflow_read_actor);
 812                         if (ret < 0) {
 813                                 ssk->sk_err = -ret;
 814                                 goto fatal;
 815                         }
 816                         if (ret < delta)
 817                                 return false;
 818                         if (delta == map_remaining)
 819                                 subflow->map_valid = 0;
 820                 }
 821         }
 822         return true;
 823
 824 fatal:
 825         /* fatal protocol error, close the socket */
 826         /* This barrier is coupled with smp_rmb() in tcp_poll() */
 827         smp_wmb();
 828         ssk->sk_error_report(ssk);
 829         tcp_set_state(ssk, TCP_CLOSE);
 830         tcp_send_active_reset(ssk, GFP_ATOMIC);
 831         return false;
 832 }
 833
 834 bool mptcp_subflow_data_available(struct sock *sk)
 835 {
 836         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 837         struct sk_buff *skb;
 838
 839         /* check if current mapping is still valid */
 840         if (subflow->map_valid &&
 841             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
 842                 subflow->map_valid = 0;
 843                 subflow->data_avail = 0;
 844
 845                 pr_debug("Done with mapping: seq=%u data_len=%u",
 846                          subflow->map_subflow_seq,
 847                          subflow->map_data_len);
 848         }
 849
 850         if (!subflow_check_data_avail(sk)) {
 851                 subflow->data_avail = 0;
 852                 return false;
 853         }
 854
 855         skb = skb_peek(&sk->sk_receive_queue);
 856         subflow->data_avail = skb &&
 857                        before(tcp_sk(sk)->copied_seq, TCP_SKB_CB(skb)->end_seq);
 858         return subflow->data_avail;
 859 }
 860
 861 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
 862  * not the ssk one.
 863  *
 864  * In mptcp, rwin is about the mptcp-level connection data.
 865  *
 866  * Data that is still on the ssk rx queue can thus be ignored,
 867  * as far as mptcp peer is concerened that data is still inflight.
 868  * DSS ACK is updated when skb is moved to the mptcp rx queue.
 869  */
 870 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
 871 {
 872         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 873         const struct sock *sk = subflow->conn;
 874
 875         *space = tcp_space(sk);
 876         *full_space = tcp_full_space(sk);
 877 }
 878
 879 static void subflow_data_ready(struct sock *sk)
 880 {
 881         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 882         u16 state = 1 << inet_sk_state_load(sk);
 883         struct sock *parent = subflow->conn;
 884         struct mptcp_sock *msk;
 885
 886         msk = mptcp_sk(parent);
 887         if (state & TCPF_LISTEN) {
 888                 set_bit(MPTCP_DATA_READY, &msk->flags);
 889                 parent->sk_data_ready(parent);
 890                 return;
 891         }
 892
 893         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
 894                      !subflow->mp_join && !(state & TCPF_CLOSE));
 895
 896         if (mptcp_subflow_data_available(sk))
 897                 mptcp_data_ready(parent, sk);
 898 }
 899
 900 static void subflow_write_space(struct sock *sk)
 901 {
 902         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 903         struct sock *parent = subflow->conn;
 904
 905         sk_stream_write_space(sk);
 906         if (sk_stream_is_writeable(sk)) {
 907                 set_bit(MPTCP_SEND_SPACE, &mptcp_sk(parent)->flags);
 908                 smp_mb__after_atomic();
 909                 /* set SEND_SPACE before sk_stream_write_space clears NOSPACE */
 910                 sk_stream_write_space(parent);
 911         }
 912 }
 913
 914 static struct inet_connection_sock_af_ops *
 915 subflow_default_af_ops(struct sock *sk)
 916 {
 917 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 918         if (sk->sk_family == AF_INET6)
 919                 return &subflow_v6_specific;
 920 #endif
 921         return &subflow_specific;
 922 }
 923
 924 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 925 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
 926 {
 927         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 928         struct inet_connection_sock *icsk = inet_csk(sk);
 929         struct inet_connection_sock_af_ops *target;
 930
 931         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
 932
 933         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
 934                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
 935
 936         if (likely(icsk->icsk_af_ops == target))
 937                 return;
 938
 939         subflow->icsk_af_ops = icsk->icsk_af_ops;
 940         icsk->icsk_af_ops = target;
 941 }
 942 #endif
 943
 944 static void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
 945                                 struct sockaddr_storage *addr)
 946 {
 947         memset(addr, 0, sizeof(*addr));
 948         addr->ss_family = info->family;
 949         if (addr->ss_family == AF_INET) {
 950                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
 951
 952                 in_addr->sin_addr = info->addr;
 953                 in_addr->sin_port = info->port;
 954         }
 955 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 956         else if (addr->ss_family == AF_INET6) {
 957                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
 958
 959                 in6_addr->sin6_addr = info->addr6;
 960                 in6_addr->sin6_port = info->port;
 961         }
 962 #endif
 963 }
 964
 965 int __mptcp_subflow_connect(struct sock *sk, int ifindex,
 966                             const struct mptcp_addr_info *loc,
 967                             const struct mptcp_addr_info *remote)
 968 {
 969         struct mptcp_sock *msk = mptcp_sk(sk);
 970         struct mptcp_subflow_context *subflow;
 971         struct sockaddr_storage addr;
 972         int local_id = loc->id;
 973         struct socket *sf;
 974         struct sock *ssk;
 975         u32 remote_token;
 976         int addrlen;
 977         int err;
 978
 979         if (!mptcp_is_fully_established(sk))
 980                 return -ENOTCONN;
 981
 982         err = mptcp_subflow_create_socket(sk, &sf);
 983         if (err)
 984                 return err;
 985
 986         ssk = sf->sk;
 987         subflow = mptcp_subflow_ctx(ssk);
 988         do {
 989                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
 990         } while (!subflow->local_nonce);
 991
 992         if (!local_id) {
 993                 err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
 994                 if (err < 0)
 995                         goto failed;
 996
 997                 local_id = err;
 998         }
 999
1000         subflow->remote_key = msk->remote_key;
1001         subflow->local_key = msk->local_key;
1002         subflow->token = msk->token;
1003         mptcp_info2sockaddr(loc, &addr);
1004
1005         addrlen = sizeof(struct sockaddr_in);
1006 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1007         if (loc->family == AF_INET6)
1008                 addrlen = sizeof(struct sockaddr_in6);
1009 #endif
1010         ssk->sk_bound_dev_if = ifindex;
1011         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1012         if (err)
1013                 goto failed;
1014
1015         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1016         pr_debug("msk=%p remote_token=%u local_id=%d", msk, remote_token,
1017                  local_id);
1018         subflow->remote_token = remote_token;
1019         subflow->local_id = local_id;
1020         subflow->request_join = 1;
1021         subflow->request_bkup = 1;
1022         mptcp_info2sockaddr(remote, &addr);
1023
1024         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1025         if (err && err != -EINPROGRESS)
1026                 goto failed;
1027
1028         spin_lock_bh(&msk->join_list_lock);
1029         list_add_tail(&subflow->node, &msk->join_list);
1030         spin_unlock_bh(&msk->join_list_lock);
1031
1032         return err;
1033
1034 failed:
1035         sock_release(sf);
1036         return err;
1037 }
1038
1039 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1040 {
1041         struct mptcp_subflow_context *subflow;
1042         struct net *net = sock_net(sk);
1043         struct socket *sf;
1044         int err;
1045
1046         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1047                                &sf);
1048         if (err)
1049                 return err;
1050
1051         lock_sock(sf->sk);
1052
1053         /* kernel sockets do not by default acquire net ref, but TCP timer
1054          * needs it.
1055          */
1056         sf->sk->sk_net_refcnt = 1;
1057         get_net(net);
1058 #ifdef CONFIG_PROC_FS
1059         this_cpu_add(*net->core.sock_inuse, 1);
1060 #endif
1061         err = tcp_set_ulp(sf->sk, "mptcp");
1062         release_sock(sf->sk);
1063
1064         if (err) {
1065                 sock_release(sf);
1066                 return err;
1067         }
1068
1069         /* the newly created socket really belongs to the owning MPTCP master
1070          * socket, even if for additional subflows the allocation is performed
1071          * by a kernel workqueue. Adjust inode references, so that the
1072          * procfs/diag interaces really show this one belonging to the correct
1073          * user.
1074          */
1075         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1076         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1077         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1078
1079         subflow = mptcp_subflow_ctx(sf->sk);
1080         pr_debug("subflow=%p", subflow);
1081
1082         *new_sock = sf;
1083         sock_hold(sk);
1084         subflow->conn = sk;
1085
1086         return 0;
1087 }
1088
1089 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1090                                                         gfp_t priority)
1091 {
1092         struct inet_connection_sock *icsk = inet_csk(sk);
1093         struct mptcp_subflow_context *ctx;
1094
1095         ctx = kzalloc(sizeof(*ctx), priority);
1096         if (!ctx)
1097                 return NULL;
1098
1099         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1100         INIT_LIST_HEAD(&ctx->node);
1101
1102         pr_debug("subflow=%p", ctx);
1103
1104         ctx->tcp_sock = sk;
1105
1106         return ctx;
1107 }
1108
1109 static void __subflow_state_change(struct sock *sk)
1110 {
1111         struct socket_wq *wq;
1112
1113         rcu_read_lock();
1114         wq = rcu_dereference(sk->sk_wq);
1115         if (skwq_has_sleeper(wq))
1116                 wake_up_interruptible_all(&wq->wait);
1117         rcu_read_unlock();
1118 }
1119
1120 static bool subflow_is_done(const struct sock *sk)
1121 {
1122         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1123 }
1124
1125 static void subflow_state_change(struct sock *sk)
1126 {
1127         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1128         struct sock *parent = subflow->conn;
1129
1130         __subflow_state_change(sk);
1131
1132         if (subflow_simultaneous_connect(sk)) {
1133                 mptcp_do_fallback(sk);
1134                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1135                 pr_fallback(mptcp_sk(parent));
1136                 subflow->conn_finished = 1;
1137                 if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
1138                         inet_sk_state_store(parent, TCP_ESTABLISHED);
1139                         parent->sk_state_change(parent);
1140                 }
1141         }
1142
1143         /* as recvmsg() does not acquire the subflow socket for ssk selection
1144          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1145          * the data available machinery here.
1146          */
1147         if (mptcp_subflow_data_available(sk))
1148                 mptcp_data_ready(parent, sk);
1149
1150         if (!(parent->sk_shutdown & RCV_SHUTDOWN) &&
1151             !subflow->rx_eof && subflow_is_done(sk)) {
1152                 subflow->rx_eof = 1;
1153                 mptcp_subflow_eof(parent);
1154         }
1155 }
1156
1157 static int subflow_ulp_init(struct sock *sk)
1158 {
1159         struct inet_connection_sock *icsk = inet_csk(sk);
1160         struct mptcp_subflow_context *ctx;
1161         struct tcp_sock *tp = tcp_sk(sk);
1162         int err = 0;
1163
1164         /* disallow attaching ULP to a socket unless it has been
1165          * created with sock_create_kern()
1166          */
1167         if (!sk->sk_kern_sock) {
1168                 err = -EOPNOTSUPP;
1169                 goto out;
1170         }
1171
1172         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1173         if (!ctx) {
1174                 err = -ENOMEM;
1175                 goto out;
1176         }
1177
1178         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1179
1180         tp->is_mptcp = 1;
1181         ctx->icsk_af_ops = icsk->icsk_af_ops;
1182         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1183         ctx->tcp_data_ready = sk->sk_data_ready;
1184         ctx->tcp_state_change = sk->sk_state_change;
1185         ctx->tcp_write_space = sk->sk_write_space;
1186         sk->sk_data_ready = subflow_data_ready;
1187         sk->sk_write_space = subflow_write_space;
1188         sk->sk_state_change = subflow_state_change;
1189 out:
1190         return err;
1191 }
1192
1193 static void subflow_ulp_release(struct sock *sk)
1194 {
1195         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(sk);
1196
1197         if (!ctx)
1198                 return;
1199
1200         if (ctx->conn)
1201                 sock_put(ctx->conn);
1202
1203         kfree_rcu(ctx, rcu);
1204 }
1205
1206 static void subflow_ulp_clone(const struct request_sock *req,
1207                               struct sock *newsk,
1208                               const gfp_t priority)
1209 {
1210         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1211         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1212         struct mptcp_subflow_context *new_ctx;
1213
1214         if (!tcp_rsk(req)->is_mptcp ||
1215             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1216                 subflow_ulp_fallback(newsk, old_ctx);
1217                 return;
1218         }
1219
1220         new_ctx = subflow_create_ctx(newsk, priority);
1221         if (!new_ctx) {
1222                 subflow_ulp_fallback(newsk, old_ctx);
1223                 return;
1224         }
1225
1226         new_ctx->conn_finished = 1;
1227         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1228         new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1229         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1230         new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1231         new_ctx->rel_write_seq = 1;
1232         new_ctx->tcp_sock = newsk;
1233
1234         if (subflow_req->mp_capable) {
1235                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1236                  * is fully established only after we receive the remote key
1237                  */
1238                 new_ctx->mp_capable = 1;
1239                 new_ctx->local_key = subflow_req->local_key;
1240                 new_ctx->token = subflow_req->token;
1241                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1242                 new_ctx->idsn = subflow_req->idsn;
1243         } else if (subflow_req->mp_join) {
1244                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1245                 new_ctx->mp_join = 1;
1246                 new_ctx->fully_established = 1;
1247                 new_ctx->backup = subflow_req->backup;
1248                 new_ctx->local_id = subflow_req->local_id;
1249                 new_ctx->token = subflow_req->token;
1250                 new_ctx->thmac = subflow_req->thmac;
1251         }
1252 }
1253
1254 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1255         .name           = "mptcp",
1256         .owner          = THIS_MODULE,
1257         .init           = subflow_ulp_init,
1258         .release        = subflow_ulp_release,
1259         .clone          = subflow_ulp_clone,
1260 };
1261
1262 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1263 {
1264         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1265         subflow_ops->slab_name = "request_sock_subflow";
1266
1267         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1268                                               subflow_ops->obj_size, 0,
1269                                               SLAB_ACCOUNT |
1270                                               SLAB_TYPESAFE_BY_RCU,
1271                                               NULL);
1272         if (!subflow_ops->slab)
1273                 return -ENOMEM;
1274
1275         subflow_ops->destructor = subflow_req_destructor;
1276
1277         return 0;
1278 }
1279
1280 void __init mptcp_subflow_init(void)
1281 {
1282         subflow_request_sock_ops = tcp_request_sock_ops;
1283         if (subflow_ops_init(&subflow_request_sock_ops) != 0)
1284                 panic("MPTCP: failed to init subflow request sock ops\n");
1285
1286         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1287         subflow_request_sock_ipv4_ops.init_req = subflow_v4_init_req;
1288
1289         subflow_specific = ipv4_specific;
1290         subflow_specific.conn_request = subflow_v4_conn_request;
1291         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1292         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1293
1294 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1295         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1296         subflow_request_sock_ipv6_ops.init_req = subflow_v6_init_req;
1297
1298         subflow_v6_specific = ipv6_specific;
1299         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1300         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1301         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1302
1303         subflow_v6m_specific = subflow_v6_specific;
1304         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1305         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1306         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1307         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1308         subflow_v6m_specific.net_frag_header_len = 0;
1309 #endif
1310
1311         mptcp_diag_subflow_init(&subflow_ulp_ops);
1312
1313         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1314                 panic("MPTCP: failed to register subflows to ULP\n");
1315 }