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 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
  57 {
  58         return mptcp_is_fully_established((void *)msk) &&
  59                READ_ONCE(msk->pm.accept_subflow);
  60 }
  61
  62 /* validate received token and create truncated hmac and nonce for SYN-ACK */
  63 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req,
  64                                                      const struct sk_buff *skb)
  65 {
  66         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  67         u8 hmac[SHA256_DIGEST_SIZE];
  68         struct mptcp_sock *msk;
  69         int local_id;
  70
  71         msk = mptcp_token_get_sock(subflow_req->token);
  72         if (!msk) {
  73                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
  74                 return NULL;
  75         }
  76
  77         local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
  78         if (local_id < 0) {
  79                 sock_put((struct sock *)msk);
  80                 return NULL;
  81         }
  82         subflow_req->local_id = local_id;
  83
  84         get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
  85
  86         subflow_generate_hmac(msk->local_key, msk->remote_key,
  87                               subflow_req->local_nonce,
  88                               subflow_req->remote_nonce, hmac);
  89
  90         subflow_req->thmac = get_unaligned_be64(hmac);
  91         return msk;
  92 }
  93
  94 static int __subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
  95 {
  96         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  97
  98         subflow_req->mp_capable = 0;
  99         subflow_req->mp_join = 0;
 100         subflow_req->msk = NULL;
 101         mptcp_token_init_request(req);
 102
 103 #ifdef CONFIG_TCP_MD5SIG
 104         /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
 105          * TCP option space.
 106          */
 107         if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
 108                 return -EINVAL;
 109 #endif
 110
 111         return 0;
 112 }
 113
 114 static void subflow_init_req(struct request_sock *req,
 115                              const struct sock *sk_listener,
 116                              struct sk_buff *skb)
 117 {
 118         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 119         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 120         struct mptcp_options_received mp_opt;
 121         int ret;
 122
 123         pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
 124
 125         ret = __subflow_init_req(req, sk_listener);
 126         if (ret)
 127                 return;
 128
 129         mptcp_get_options(skb, &mp_opt);
 130
 131         if (mp_opt.mp_capable) {
 132                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
 133
 134                 if (mp_opt.mp_join)
 135                         return;
 136         } else if (mp_opt.mp_join) {
 137                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
 138         }
 139
 140         if (mp_opt.mp_capable && listener->request_mptcp) {
 141                 int err, retries = 4;
 142
 143                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 144 again:
 145                 do {
 146                         get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
 147                 } while (subflow_req->local_key == 0);
 148
 149                 if (unlikely(req->syncookie)) {
 150                         mptcp_crypto_key_sha(subflow_req->local_key,
 151                                              &subflow_req->token,
 152                                              &subflow_req->idsn);
 153                         if (mptcp_token_exists(subflow_req->token)) {
 154                                 if (retries-- > 0)
 155                                         goto again;
 156                         } else {
 157                                 subflow_req->mp_capable = 1;
 158                         }
 159                         return;
 160                 }
 161
 162                 err = mptcp_token_new_request(req);
 163                 if (err == 0)
 164                         subflow_req->mp_capable = 1;
 165                 else if (retries-- > 0)
 166                         goto again;
 167
 168         } else if (mp_opt.mp_join && listener->request_mptcp) {
 169                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 170                 subflow_req->mp_join = 1;
 171                 subflow_req->backup = mp_opt.backup;
 172                 subflow_req->remote_id = mp_opt.join_id;
 173                 subflow_req->token = mp_opt.token;
 174                 subflow_req->remote_nonce = mp_opt.nonce;
 175                 subflow_req->msk = subflow_token_join_request(req, skb);
 176
 177                 if (unlikely(req->syncookie) && subflow_req->msk) {
 178                         if (mptcp_can_accept_new_subflow(subflow_req->msk))
 179                                 subflow_init_req_cookie_join_save(subflow_req, skb);
 180                 }
 181
 182                 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
 183                          subflow_req->remote_nonce, subflow_req->msk);
 184         }
 185 }
 186
 187 int mptcp_subflow_init_cookie_req(struct request_sock *req,
 188                                   const struct sock *sk_listener,
 189                                   struct sk_buff *skb)
 190 {
 191         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 192         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 193         struct mptcp_options_received mp_opt;
 194         int err;
 195
 196         err = __subflow_init_req(req, sk_listener);
 197         if (err)
 198                 return err;
 199
 200         mptcp_get_options(skb, &mp_opt);
 201
 202         if (mp_opt.mp_capable && mp_opt.mp_join)
 203                 return -EINVAL;
 204
 205         if (mp_opt.mp_capable && listener->request_mptcp) {
 206                 if (mp_opt.sndr_key == 0)
 207                         return -EINVAL;
 208
 209                 subflow_req->local_key = mp_opt.rcvr_key;
 210                 err = mptcp_token_new_request(req);
 211                 if (err)
 212                         return err;
 213
 214                 subflow_req->mp_capable = 1;
 215                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 216         } else if (mp_opt.mp_join && listener->request_mptcp) {
 217                 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
 218                         return -EINVAL;
 219
 220                 if (mptcp_can_accept_new_subflow(subflow_req->msk))
 221                         subflow_req->mp_join = 1;
 222
 223                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 224         }
 225
 226         return 0;
 227 }
 228 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
 229
 230 static void subflow_v4_init_req(struct request_sock *req,
 231                                 const struct sock *sk_listener,
 232                                 struct sk_buff *skb)
 233 {
 234         tcp_rsk(req)->is_mptcp = 1;
 235
 236         tcp_request_sock_ipv4_ops.init_req(req, sk_listener, skb);
 237
 238         subflow_init_req(req, sk_listener, skb);
 239 }
 240
 241 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 242 static void subflow_v6_init_req(struct request_sock *req,
 243                                 const struct sock *sk_listener,
 244                                 struct sk_buff *skb)
 245 {
 246         tcp_rsk(req)->is_mptcp = 1;
 247
 248         tcp_request_sock_ipv6_ops.init_req(req, sk_listener, skb);
 249
 250         subflow_init_req(req, sk_listener, skb);
 251 }
 252 #endif
 253
 254 /* validate received truncated hmac and create hmac for third ACK */
 255 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
 256 {
 257         u8 hmac[SHA256_DIGEST_SIZE];
 258         u64 thmac;
 259
 260         subflow_generate_hmac(subflow->remote_key, subflow->local_key,
 261                               subflow->remote_nonce, subflow->local_nonce,
 262                               hmac);
 263
 264         thmac = get_unaligned_be64(hmac);
 265         pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
 266                  subflow, subflow->token,
 267                  (unsigned long long)thmac,
 268                  (unsigned long long)subflow->thmac);
 269
 270         return thmac == subflow->thmac;
 271 }
 272
 273 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 274 {
 275         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 276         struct mptcp_options_received mp_opt;
 277         struct sock *parent = subflow->conn;
 278
 279         subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 280
 281         if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
 282                 inet_sk_state_store(parent, TCP_ESTABLISHED);
 283                 parent->sk_state_change(parent);
 284         }
 285
 286         /* be sure no special action on any packet other than syn-ack */
 287         if (subflow->conn_finished)
 288                 return;
 289
 290         subflow->rel_write_seq = 1;
 291         subflow->conn_finished = 1;
 292         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 293         pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
 294
 295         mptcp_get_options(skb, &mp_opt);
 296         if (subflow->request_mptcp) {
 297                 if (!mp_opt.mp_capable) {
 298                         MPTCP_INC_STATS(sock_net(sk),
 299                                         MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
 300                         mptcp_do_fallback(sk);
 301                         pr_fallback(mptcp_sk(subflow->conn));
 302                         goto fallback;
 303                 }
 304
 305                 subflow->mp_capable = 1;
 306                 subflow->can_ack = 1;
 307                 subflow->remote_key = mp_opt.sndr_key;
 308                 pr_debug("subflow=%p, remote_key=%llu", subflow,
 309                          subflow->remote_key);
 310                 mptcp_finish_connect(sk);
 311         } else if (subflow->request_join) {
 312                 u8 hmac[SHA256_DIGEST_SIZE];
 313
 314                 if (!mp_opt.mp_join)
 315                         goto do_reset;
 316
 317                 subflow->thmac = mp_opt.thmac;
 318                 subflow->remote_nonce = mp_opt.nonce;
 319                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
 320                          subflow->thmac, subflow->remote_nonce);
 321
 322                 if (!subflow_thmac_valid(subflow)) {
 323                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 324                         goto do_reset;
 325                 }
 326
 327                 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 328                                       subflow->local_nonce,
 329                                       subflow->remote_nonce,
 330                                       hmac);
 331                 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
 332
 333                 if (!mptcp_finish_join(sk))
 334                         goto do_reset;
 335
 336                 subflow->mp_join = 1;
 337                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 338         } else if (mptcp_check_fallback(sk)) {
 339 fallback:
 340                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
 341         }
 342         return;
 343
 344 do_reset:
 345         tcp_send_active_reset(sk, GFP_ATOMIC);
 346         tcp_done(sk);
 347 }
 348
 349 struct request_sock_ops mptcp_subflow_request_sock_ops;
 350 EXPORT_SYMBOL_GPL(mptcp_subflow_request_sock_ops);
 351 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
 352
 353 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 354 {
 355         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 356
 357         pr_debug("subflow=%p", subflow);
 358
 359         /* Never answer to SYNs sent to broadcast or multicast */
 360         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 361                 goto drop;
 362
 363         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 364                                 &subflow_request_sock_ipv4_ops,
 365                                 sk, skb);
 366 drop:
 367         tcp_listendrop(sk);
 368         return 0;
 369 }
 370
 371 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 372 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
 373 static struct inet_connection_sock_af_ops subflow_v6_specific;
 374 static struct inet_connection_sock_af_ops subflow_v6m_specific;
 375
 376 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 377 {
 378         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 379
 380         pr_debug("subflow=%p", subflow);
 381
 382         if (skb->protocol == htons(ETH_P_IP))
 383                 return subflow_v4_conn_request(sk, skb);
 384
 385         if (!ipv6_unicast_destination(skb))
 386                 goto drop;
 387
 388         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 389                                 &subflow_request_sock_ipv6_ops, sk, skb);
 390
 391 drop:
 392         tcp_listendrop(sk);
 393         return 0; /* don't send reset */
 394 }
 395 #endif
 396
 397 /* validate hmac received in third ACK */
 398 static bool subflow_hmac_valid(const struct request_sock *req,
 399                                const struct mptcp_options_received *mp_opt)
 400 {
 401         const struct mptcp_subflow_request_sock *subflow_req;
 402         u8 hmac[SHA256_DIGEST_SIZE];
 403         struct mptcp_sock *msk;
 404
 405         subflow_req = mptcp_subflow_rsk(req);
 406         msk = subflow_req->msk;
 407         if (!msk)
 408                 return false;
 409
 410         subflow_generate_hmac(msk->remote_key, msk->local_key,
 411                               subflow_req->remote_nonce,
 412                               subflow_req->local_nonce, hmac);
 413
 414         return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
 415 }
 416
 417 static void mptcp_sock_destruct(struct sock *sk)
 418 {
 419         /* if new mptcp socket isn't accepted, it is free'd
 420          * from the tcp listener sockets request queue, linked
 421          * from req->sk.  The tcp socket is released.
 422          * This calls the ULP release function which will
 423          * also remove the mptcp socket, via
 424          * sock_put(ctx->conn).
 425          *
 426          * Problem is that the mptcp socket will not be in
 427          * SYN_RECV state and doesn't have SOCK_DEAD flag.
 428          * Both result in warnings from inet_sock_destruct.
 429          */
 430
 431         if (sk->sk_state == TCP_SYN_RECV) {
 432                 sk->sk_state = TCP_CLOSE;
 433                 WARN_ON_ONCE(sk->sk_socket);
 434                 sock_orphan(sk);
 435         }
 436
 437         mptcp_token_destroy(mptcp_sk(sk));
 438         inet_sock_destruct(sk);
 439 }
 440
 441 static void mptcp_force_close(struct sock *sk)
 442 {
 443         inet_sk_state_store(sk, TCP_CLOSE);
 444         sk_common_release(sk);
 445 }
 446
 447 static void subflow_ulp_fallback(struct sock *sk,
 448                                  struct mptcp_subflow_context *old_ctx)
 449 {
 450         struct inet_connection_sock *icsk = inet_csk(sk);
 451
 452         mptcp_subflow_tcp_fallback(sk, old_ctx);
 453         icsk->icsk_ulp_ops = NULL;
 454         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 455         tcp_sk(sk)->is_mptcp = 0;
 456 }
 457
 458 static void subflow_drop_ctx(struct sock *ssk)
 459 {
 460         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 461
 462         if (!ctx)
 463                 return;
 464
 465         subflow_ulp_fallback(ssk, ctx);
 466         if (ctx->conn)
 467                 sock_put(ctx->conn);
 468
 469         kfree_rcu(ctx, rcu);
 470 }
 471
 472 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 473                                      struct mptcp_options_received *mp_opt)
 474 {
 475         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 476
 477         subflow->remote_key = mp_opt->sndr_key;
 478         subflow->fully_established = 1;
 479         subflow->can_ack = 1;
 480         WRITE_ONCE(msk->fully_established, true);
 481 }
 482
 483 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 484                                           struct sk_buff *skb,
 485                                           struct request_sock *req,
 486                                           struct dst_entry *dst,
 487                                           struct request_sock *req_unhash,
 488                                           bool *own_req)
 489 {
 490         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 491         struct mptcp_subflow_request_sock *subflow_req;
 492         struct mptcp_options_received mp_opt;
 493         bool fallback, fallback_is_fatal;
 494         struct sock *new_msk = NULL;
 495         struct sock *child;
 496
 497         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 498
 499         /* After child creation we must look for 'mp_capable' even when options
 500          * are not parsed
 501          */
 502         mp_opt.mp_capable = 0;
 503
 504         /* hopefully temporary handling for MP_JOIN+syncookie */
 505         subflow_req = mptcp_subflow_rsk(req);
 506         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 507         fallback = !tcp_rsk(req)->is_mptcp;
 508         if (fallback)
 509                 goto create_child;
 510
 511         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 512         if (subflow_req->mp_capable) {
 513                 if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
 514                         /* here we can receive and accept an in-window,
 515                          * out-of-order pkt, which will not carry the MP_CAPABLE
 516                          * opt even on mptcp enabled paths
 517                          */
 518                         goto create_msk;
 519                 }
 520
 521                 mptcp_get_options(skb, &mp_opt);
 522                 if (!mp_opt.mp_capable) {
 523                         fallback = true;
 524                         goto create_child;
 525                 }
 526
 527 create_msk:
 528                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 529                 if (!new_msk)
 530                         fallback = true;
 531         } else if (subflow_req->mp_join) {
 532                 mptcp_get_options(skb, &mp_opt);
 533                 if (!mp_opt.mp_join ||
 534                     !mptcp_can_accept_new_subflow(subflow_req->msk) ||
 535                     !subflow_hmac_valid(req, &mp_opt)) {
 536                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 537                         fallback = true;
 538                 }
 539         }
 540
 541 create_child:
 542         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 543                                                      req_unhash, own_req);
 544
 545         if (child && *own_req) {
 546                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 547
 548                 tcp_rsk(req)->drop_req = false;
 549
 550                 /* we need to fallback on ctx allocation failure and on pre-reqs
 551                  * checking above. In the latter scenario we additionally need
 552                  * to reset the context to non MPTCP status.
 553                  */
 554                 if (!ctx || fallback) {
 555                         if (fallback_is_fatal)
 556                                 goto dispose_child;
 557
 558                         subflow_drop_ctx(child);
 559                         goto out;
 560                 }
 561
 562                 if (ctx->mp_capable) {
 563                         /* this can't race with mptcp_close(), as the msk is
 564                          * not yet exposted to user-space
 565                          */
 566                         inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
 567
 568                         /* new mpc subflow takes ownership of the newly
 569                          * created mptcp socket
 570                          */
 571                         new_msk->sk_destruct = mptcp_sock_destruct;
 572                         mptcp_pm_new_connection(mptcp_sk(new_msk), 1);
 573                         mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
 574                         ctx->conn = new_msk;
 575                         new_msk = NULL;
 576
 577                         /* with OoO packets we can reach here without ingress
 578                          * mpc option
 579                          */
 580                         if (mp_opt.mp_capable)
 581                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 582                 } else if (ctx->mp_join) {
 583                         struct mptcp_sock *owner;
 584
 585                         owner = subflow_req->msk;
 586                         if (!owner)
 587                                 goto dispose_child;
 588
 589                         /* move the msk reference ownership to the subflow */
 590                         subflow_req->msk = NULL;
 591                         ctx->conn = (struct sock *)owner;
 592                         if (!mptcp_finish_join(child))
 593                                 goto dispose_child;
 594
 595                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 596                         tcp_rsk(req)->drop_req = true;
 597                 }
 598         }
 599
 600 out:
 601         /* dispose of the left over mptcp master, if any */
 602         if (unlikely(new_msk))
 603                 mptcp_force_close(new_msk);
 604
 605         /* check for expected invariant - should never trigger, just help
 606          * catching eariler subtle bugs
 607          */
 608         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 609                      (!mptcp_subflow_ctx(child) ||
 610                       !mptcp_subflow_ctx(child)->conn));
 611         return child;
 612
 613 dispose_child:
 614         subflow_drop_ctx(child);
 615         tcp_rsk(req)->drop_req = true;
 616         inet_csk_prepare_for_destroy_sock(child);
 617         tcp_done(child);
 618         req->rsk_ops->send_reset(sk, skb);
 619
 620         /* The last child reference will be released by the caller */
 621         return child;
 622 }
 623
 624 static struct inet_connection_sock_af_ops subflow_specific;
 625
 626 enum mapping_status {
 627         MAPPING_OK,
 628         MAPPING_INVALID,
 629         MAPPING_EMPTY,
 630         MAPPING_DATA_FIN,
 631         MAPPING_DUMMY
 632 };
 633
 634 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
 635 {
 636         if ((u32)seq == (u32)old_seq)
 637                 return old_seq;
 638
 639         /* Assume map covers data not mapped yet. */
 640         return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
 641 }
 642
 643 static void warn_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 644 {
 645         WARN_ONCE(1, "Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 646                   ssn, subflow->map_subflow_seq, subflow->map_data_len);
 647 }
 648
 649 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 650 {
 651         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 652         unsigned int skb_consumed;
 653
 654         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 655         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 656                 return true;
 657
 658         return skb->len - skb_consumed <= subflow->map_data_len -
 659                                           mptcp_subflow_get_map_offset(subflow);
 660 }
 661
 662 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 663 {
 664         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 665         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 666
 667         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 668                 /* Mapping covers data later in the subflow stream,
 669                  * currently unsupported.
 670                  */
 671                 warn_bad_map(subflow, ssn);
 672                 return false;
 673         }
 674         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 675                                   subflow->map_data_len))) {
 676                 /* Mapping does covers past subflow data, invalid */
 677                 warn_bad_map(subflow, ssn + skb->len);
 678                 return false;
 679         }
 680         return true;
 681 }
 682
 683 static enum mapping_status get_mapping_status(struct sock *ssk,
 684                                               struct mptcp_sock *msk)
 685 {
 686         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 687         struct mptcp_ext *mpext;
 688         struct sk_buff *skb;
 689         u16 data_len;
 690         u64 map_seq;
 691
 692         skb = skb_peek(&ssk->sk_receive_queue);
 693         if (!skb)
 694                 return MAPPING_EMPTY;
 695
 696         if (mptcp_check_fallback(ssk))
 697                 return MAPPING_DUMMY;
 698
 699         mpext = mptcp_get_ext(skb);
 700         if (!mpext || !mpext->use_map) {
 701                 if (!subflow->map_valid && !skb->len) {
 702                         /* the TCP stack deliver 0 len FIN pkt to the receive
 703                          * queue, that is the only 0len pkts ever expected here,
 704                          * and we can admit no mapping only for 0 len pkts
 705                          */
 706                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 707                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 708                                           TCP_SKB_CB(skb)->seq,
 709                                           TCP_SKB_CB(skb)->end_seq,
 710                                           TCP_SKB_CB(skb)->tcp_flags);
 711                         sk_eat_skb(ssk, skb);
 712                         return MAPPING_EMPTY;
 713                 }
 714
 715                 if (!subflow->map_valid)
 716                         return MAPPING_INVALID;
 717
 718                 goto validate_seq;
 719         }
 720
 721         pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
 722                  mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
 723                  mpext->data_len, mpext->data_fin);
 724
 725         data_len = mpext->data_len;
 726         if (data_len == 0) {
 727                 pr_err("Infinite mapping not handled");
 728                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 729                 return MAPPING_INVALID;
 730         }
 731
 732         if (mpext->data_fin == 1) {
 733                 if (data_len == 1) {
 734                         mptcp_update_rcv_data_fin(msk, mpext->data_seq);
 735                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
 736                         if (subflow->map_valid) {
 737                                 /* A DATA_FIN might arrive in a DSS
 738                                  * option before the previous mapping
 739                                  * has been fully consumed. Continue
 740                                  * handling the existing mapping.
 741                                  */
 742                                 skb_ext_del(skb, SKB_EXT_MPTCP);
 743                                 return MAPPING_OK;
 744                         } else {
 745                                 return MAPPING_DATA_FIN;
 746                         }
 747                 } else {
 748                         mptcp_update_rcv_data_fin(msk, mpext->data_seq + data_len);
 749                         pr_debug("DATA_FIN with mapping seq=%llu", mpext->data_seq + data_len);
 750                 }
 751
 752                 /* Adjust for DATA_FIN using 1 byte of sequence space */
 753                 data_len--;
 754         }
 755
 756         if (!mpext->dsn64) {
 757                 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
 758                                      mpext->data_seq);
 759                 subflow->use_64bit_ack = 0;
 760                 pr_debug("expanded seq=%llu", subflow->map_seq);
 761         } else {
 762                 map_seq = mpext->data_seq;
 763                 subflow->use_64bit_ack = 1;
 764         }
 765
 766         if (subflow->map_valid) {
 767                 /* Allow replacing only with an identical map */
 768                 if (subflow->map_seq == map_seq &&
 769                     subflow->map_subflow_seq == mpext->subflow_seq &&
 770                     subflow->map_data_len == data_len) {
 771                         skb_ext_del(skb, SKB_EXT_MPTCP);
 772                         return MAPPING_OK;
 773                 }
 774
 775                 /* If this skb data are fully covered by the current mapping,
 776                  * the new map would need caching, which is not supported
 777                  */
 778                 if (skb_is_fully_mapped(ssk, skb)) {
 779                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
 780                         return MAPPING_INVALID;
 781                 }
 782
 783                 /* will validate the next map after consuming the current one */
 784                 return MAPPING_OK;
 785         }
 786
 787         subflow->map_seq = map_seq;
 788         subflow->map_subflow_seq = mpext->subflow_seq;
 789         subflow->map_data_len = data_len;
 790         subflow->map_valid = 1;
 791         subflow->mpc_map = mpext->mpc_map;
 792         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
 793                  subflow->map_seq, subflow->map_subflow_seq,
 794                  subflow->map_data_len);
 795
 796 validate_seq:
 797         /* we revalidate valid mapping on new skb, because we must ensure
 798          * the current skb is completely covered by the available mapping
 799          */
 800         if (!validate_mapping(ssk, skb))
 801                 return MAPPING_INVALID;
 802
 803         skb_ext_del(skb, SKB_EXT_MPTCP);
 804         return MAPPING_OK;
 805 }
 806
 807 static int subflow_read_actor(read_descriptor_t *desc,
 808                               struct sk_buff *skb,
 809                               unsigned int offset, size_t len)
 810 {
 811         size_t copy_len = min(desc->count, len);
 812
 813         desc->count -= copy_len;
 814
 815         pr_debug("flushed %zu bytes, %zu left", copy_len, desc->count);
 816         return copy_len;
 817 }
 818
 819 static bool subflow_check_data_avail(struct sock *ssk)
 820 {
 821         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 822         enum mapping_status status;
 823         struct mptcp_sock *msk;
 824         struct sk_buff *skb;
 825
 826         pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
 827                  subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
 828         if (subflow->data_avail)
 829                 return true;
 830
 831         msk = mptcp_sk(subflow->conn);
 832         for (;;) {
 833                 u32 map_remaining;
 834                 size_t delta;
 835                 u64 ack_seq;
 836                 u64 old_ack;
 837
 838                 status = get_mapping_status(ssk, msk);
 839                 pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
 840                 if (status == MAPPING_INVALID) {
 841                         ssk->sk_err = EBADMSG;
 842                         goto fatal;
 843                 }
 844                 if (status == MAPPING_DUMMY) {
 845                         __mptcp_do_fallback(msk);
 846                         skb = skb_peek(&ssk->sk_receive_queue);
 847                         subflow->map_valid = 1;
 848                         subflow->map_seq = READ_ONCE(msk->ack_seq);
 849                         subflow->map_data_len = skb->len;
 850                         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq -
 851                                                    subflow->ssn_offset;
 852                         return true;
 853                 }
 854
 855                 if (status != MAPPING_OK)
 856                         return false;
 857
 858                 skb = skb_peek(&ssk->sk_receive_queue);
 859                 if (WARN_ON_ONCE(!skb))
 860                         return false;
 861
 862                 /* if msk lacks the remote key, this subflow must provide an
 863                  * MP_CAPABLE-based mapping
 864                  */
 865                 if (unlikely(!READ_ONCE(msk->can_ack))) {
 866                         if (!subflow->mpc_map) {
 867                                 ssk->sk_err = EBADMSG;
 868                                 goto fatal;
 869                         }
 870                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
 871                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
 872                         WRITE_ONCE(msk->can_ack, true);
 873                 }
 874
 875                 old_ack = READ_ONCE(msk->ack_seq);
 876                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
 877                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
 878                          ack_seq);
 879                 if (ack_seq == old_ack)
 880                         break;
 881
 882                 /* only accept in-sequence mapping. Old values are spurious
 883                  * retransmission; we can hit "future" values on active backup
 884                  * subflow switch, we relay on retransmissions to get
 885                  * in-sequence data.
 886                  * Cuncurrent subflows support will require subflow data
 887                  * reordering
 888                  */
 889                 map_remaining = subflow->map_data_len -
 890                                 mptcp_subflow_get_map_offset(subflow);
 891                 if (before64(ack_seq, old_ack))
 892                         delta = min_t(size_t, old_ack - ack_seq, map_remaining);
 893                 else
 894                         delta = min_t(size_t, ack_seq - old_ack, map_remaining);
 895
 896                 /* discard mapped data */
 897                 pr_debug("discarding %zu bytes, current map len=%d", delta,
 898                          map_remaining);
 899                 if (delta) {
 900                         read_descriptor_t desc = {
 901                                 .count = delta,
 902                         };
 903                         int ret;
 904
 905                         ret = tcp_read_sock(ssk, &desc, subflow_read_actor);
 906                         if (ret < 0) {
 907                                 ssk->sk_err = -ret;
 908                                 goto fatal;
 909                         }
 910                         if (ret < delta)
 911                                 return false;
 912                         if (delta == map_remaining)
 913                                 subflow->map_valid = 0;
 914                 }
 915         }
 916         return true;
 917
 918 fatal:
 919         /* fatal protocol error, close the socket */
 920         /* This barrier is coupled with smp_rmb() in tcp_poll() */
 921         smp_wmb();
 922         ssk->sk_error_report(ssk);
 923         tcp_set_state(ssk, TCP_CLOSE);
 924         tcp_send_active_reset(ssk, GFP_ATOMIC);
 925         return false;
 926 }
 927
 928 bool mptcp_subflow_data_available(struct sock *sk)
 929 {
 930         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 931         struct sk_buff *skb;
 932
 933         /* check if current mapping is still valid */
 934         if (subflow->map_valid &&
 935             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
 936                 subflow->map_valid = 0;
 937                 subflow->data_avail = 0;
 938
 939                 pr_debug("Done with mapping: seq=%u data_len=%u",
 940                          subflow->map_subflow_seq,
 941                          subflow->map_data_len);
 942         }
 943
 944         if (!subflow_check_data_avail(sk)) {
 945                 subflow->data_avail = 0;
 946                 return false;
 947         }
 948
 949         skb = skb_peek(&sk->sk_receive_queue);
 950         subflow->data_avail = skb &&
 951                        before(tcp_sk(sk)->copied_seq, TCP_SKB_CB(skb)->end_seq);
 952         return subflow->data_avail;
 953 }
 954
 955 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
 956  * not the ssk one.
 957  *
 958  * In mptcp, rwin is about the mptcp-level connection data.
 959  *
 960  * Data that is still on the ssk rx queue can thus be ignored,
 961  * as far as mptcp peer is concerened that data is still inflight.
 962  * DSS ACK is updated when skb is moved to the mptcp rx queue.
 963  */
 964 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
 965 {
 966         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 967         const struct sock *sk = subflow->conn;
 968
 969         *space = tcp_space(sk);
 970         *full_space = tcp_full_space(sk);
 971 }
 972
 973 static void subflow_data_ready(struct sock *sk)
 974 {
 975         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 976         u16 state = 1 << inet_sk_state_load(sk);
 977         struct sock *parent = subflow->conn;
 978         struct mptcp_sock *msk;
 979
 980         msk = mptcp_sk(parent);
 981         if (state & TCPF_LISTEN) {
 982                 set_bit(MPTCP_DATA_READY, &msk->flags);
 983                 parent->sk_data_ready(parent);
 984                 return;
 985         }
 986
 987         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
 988                      !subflow->mp_join && !(state & TCPF_CLOSE));
 989
 990         if (mptcp_subflow_data_available(sk))
 991                 mptcp_data_ready(parent, sk);
 992 }
 993
 994 static void subflow_write_space(struct sock *sk)
 995 {
 996         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 997         struct sock *parent = subflow->conn;
 998
 999         sk_stream_write_space(sk);
1000         if (sk_stream_is_writeable(sk)) {
1001                 set_bit(MPTCP_SEND_SPACE, &mptcp_sk(parent)->flags);
1002                 smp_mb__after_atomic();
1003                 /* set SEND_SPACE before sk_stream_write_space clears NOSPACE */
1004                 sk_stream_write_space(parent);
1005         }
1006 }
1007
1008 static struct inet_connection_sock_af_ops *
1009 subflow_default_af_ops(struct sock *sk)
1010 {
1011 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1012         if (sk->sk_family == AF_INET6)
1013                 return &subflow_v6_specific;
1014 #endif
1015         return &subflow_specific;
1016 }
1017
1018 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1019 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1020 {
1021         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1022         struct inet_connection_sock *icsk = inet_csk(sk);
1023         struct inet_connection_sock_af_ops *target;
1024
1025         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1026
1027         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1028                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1029
1030         if (likely(icsk->icsk_af_ops == target))
1031                 return;
1032
1033         subflow->icsk_af_ops = icsk->icsk_af_ops;
1034         icsk->icsk_af_ops = target;
1035 }
1036 #endif
1037
1038 static void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1039                                 struct sockaddr_storage *addr)
1040 {
1041         memset(addr, 0, sizeof(*addr));
1042         addr->ss_family = info->family;
1043         if (addr->ss_family == AF_INET) {
1044                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1045
1046                 in_addr->sin_addr = info->addr;
1047                 in_addr->sin_port = info->port;
1048         }
1049 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1050         else if (addr->ss_family == AF_INET6) {
1051                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1052
1053                 in6_addr->sin6_addr = info->addr6;
1054                 in6_addr->sin6_port = info->port;
1055         }
1056 #endif
1057 }
1058
1059 int __mptcp_subflow_connect(struct sock *sk, int ifindex,
1060                             const struct mptcp_addr_info *loc,
1061                             const struct mptcp_addr_info *remote)
1062 {
1063         struct mptcp_sock *msk = mptcp_sk(sk);
1064         struct mptcp_subflow_context *subflow;
1065         struct sockaddr_storage addr;
1066         int local_id = loc->id;
1067         struct socket *sf;
1068         struct sock *ssk;
1069         u32 remote_token;
1070         int addrlen;
1071         int err;
1072
1073         if (!mptcp_is_fully_established(sk))
1074                 return -ENOTCONN;
1075
1076         err = mptcp_subflow_create_socket(sk, &sf);
1077         if (err)
1078                 return err;
1079
1080         ssk = sf->sk;
1081         subflow = mptcp_subflow_ctx(ssk);
1082         do {
1083                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1084         } while (!subflow->local_nonce);
1085
1086         if (!local_id) {
1087                 err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
1088                 if (err < 0)
1089                         goto failed;
1090
1091                 local_id = err;
1092         }
1093
1094         subflow->remote_key = msk->remote_key;
1095         subflow->local_key = msk->local_key;
1096         subflow->token = msk->token;
1097         mptcp_info2sockaddr(loc, &addr);
1098
1099         addrlen = sizeof(struct sockaddr_in);
1100 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1101         if (loc->family == AF_INET6)
1102                 addrlen = sizeof(struct sockaddr_in6);
1103 #endif
1104         ssk->sk_bound_dev_if = ifindex;
1105         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1106         if (err)
1107                 goto failed;
1108
1109         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1110         pr_debug("msk=%p remote_token=%u local_id=%d", msk, remote_token,
1111                  local_id);
1112         subflow->remote_token = remote_token;
1113         subflow->local_id = local_id;
1114         subflow->request_join = 1;
1115         subflow->request_bkup = 1;
1116         mptcp_info2sockaddr(remote, &addr);
1117
1118         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1119         if (err && err != -EINPROGRESS)
1120                 goto failed;
1121
1122         spin_lock_bh(&msk->join_list_lock);
1123         list_add_tail(&subflow->node, &msk->join_list);
1124         spin_unlock_bh(&msk->join_list_lock);
1125
1126         return err;
1127
1128 failed:
1129         sock_release(sf);
1130         return err;
1131 }
1132
1133 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1134 {
1135         struct mptcp_subflow_context *subflow;
1136         struct net *net = sock_net(sk);
1137         struct socket *sf;
1138         int err;
1139
1140         /* un-accepted server sockets can reach here - on bad configuration
1141          * bail early to avoid greater trouble later
1142          */
1143         if (unlikely(!sk->sk_socket))
1144                 return -EINVAL;
1145
1146         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1147                                &sf);
1148         if (err)
1149                 return err;
1150
1151         lock_sock(sf->sk);
1152
1153         /* kernel sockets do not by default acquire net ref, but TCP timer
1154          * needs it.
1155          */
1156         sf->sk->sk_net_refcnt = 1;
1157         get_net(net);
1158 #ifdef CONFIG_PROC_FS
1159         this_cpu_add(*net->core.sock_inuse, 1);
1160 #endif
1161         err = tcp_set_ulp(sf->sk, "mptcp");
1162         release_sock(sf->sk);
1163
1164         if (err) {
1165                 sock_release(sf);
1166                 return err;
1167         }
1168
1169         /* the newly created socket really belongs to the owning MPTCP master
1170          * socket, even if for additional subflows the allocation is performed
1171          * by a kernel workqueue. Adjust inode references, so that the
1172          * procfs/diag interaces really show this one belonging to the correct
1173          * user.
1174          */
1175         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1176         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1177         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1178
1179         subflow = mptcp_subflow_ctx(sf->sk);
1180         pr_debug("subflow=%p", subflow);
1181
1182         *new_sock = sf;
1183         sock_hold(sk);
1184         subflow->conn = sk;
1185
1186         return 0;
1187 }
1188
1189 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1190                                                         gfp_t priority)
1191 {
1192         struct inet_connection_sock *icsk = inet_csk(sk);
1193         struct mptcp_subflow_context *ctx;
1194
1195         ctx = kzalloc(sizeof(*ctx), priority);
1196         if (!ctx)
1197                 return NULL;
1198
1199         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1200         INIT_LIST_HEAD(&ctx->node);
1201
1202         pr_debug("subflow=%p", ctx);
1203
1204         ctx->tcp_sock = sk;
1205
1206         return ctx;
1207 }
1208
1209 static void __subflow_state_change(struct sock *sk)
1210 {
1211         struct socket_wq *wq;
1212
1213         rcu_read_lock();
1214         wq = rcu_dereference(sk->sk_wq);
1215         if (skwq_has_sleeper(wq))
1216                 wake_up_interruptible_all(&wq->wait);
1217         rcu_read_unlock();
1218 }
1219
1220 static bool subflow_is_done(const struct sock *sk)
1221 {
1222         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1223 }
1224
1225 static void subflow_state_change(struct sock *sk)
1226 {
1227         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1228         struct sock *parent = subflow->conn;
1229
1230         __subflow_state_change(sk);
1231
1232         if (subflow_simultaneous_connect(sk)) {
1233                 mptcp_do_fallback(sk);
1234                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1235                 pr_fallback(mptcp_sk(parent));
1236                 subflow->conn_finished = 1;
1237                 if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
1238                         inet_sk_state_store(parent, TCP_ESTABLISHED);
1239                         parent->sk_state_change(parent);
1240                 }
1241         }
1242
1243         /* as recvmsg() does not acquire the subflow socket for ssk selection
1244          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1245          * the data available machinery here.
1246          */
1247         if (mptcp_subflow_data_available(sk))
1248                 mptcp_data_ready(parent, sk);
1249
1250         if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1251             !(parent->sk_shutdown & RCV_SHUTDOWN) &&
1252             !subflow->rx_eof && subflow_is_done(sk)) {
1253                 subflow->rx_eof = 1;
1254                 mptcp_subflow_eof(parent);
1255         }
1256 }
1257
1258 static int subflow_ulp_init(struct sock *sk)
1259 {
1260         struct inet_connection_sock *icsk = inet_csk(sk);
1261         struct mptcp_subflow_context *ctx;
1262         struct tcp_sock *tp = tcp_sk(sk);
1263         int err = 0;
1264
1265         /* disallow attaching ULP to a socket unless it has been
1266          * created with sock_create_kern()
1267          */
1268         if (!sk->sk_kern_sock) {
1269                 err = -EOPNOTSUPP;
1270                 goto out;
1271         }
1272
1273         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1274         if (!ctx) {
1275                 err = -ENOMEM;
1276                 goto out;
1277         }
1278
1279         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1280
1281         tp->is_mptcp = 1;
1282         ctx->icsk_af_ops = icsk->icsk_af_ops;
1283         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1284         ctx->tcp_data_ready = sk->sk_data_ready;
1285         ctx->tcp_state_change = sk->sk_state_change;
1286         ctx->tcp_write_space = sk->sk_write_space;
1287         sk->sk_data_ready = subflow_data_ready;
1288         sk->sk_write_space = subflow_write_space;
1289         sk->sk_state_change = subflow_state_change;
1290 out:
1291         return err;
1292 }
1293
1294 static void subflow_ulp_release(struct sock *sk)
1295 {
1296         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(sk);
1297
1298         if (!ctx)
1299                 return;
1300
1301         if (ctx->conn)
1302                 sock_put(ctx->conn);
1303
1304         kfree_rcu(ctx, rcu);
1305 }
1306
1307 static void subflow_ulp_clone(const struct request_sock *req,
1308                               struct sock *newsk,
1309                               const gfp_t priority)
1310 {
1311         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1312         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1313         struct mptcp_subflow_context *new_ctx;
1314
1315         if (!tcp_rsk(req)->is_mptcp ||
1316             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1317                 subflow_ulp_fallback(newsk, old_ctx);
1318                 return;
1319         }
1320
1321         new_ctx = subflow_create_ctx(newsk, priority);
1322         if (!new_ctx) {
1323                 subflow_ulp_fallback(newsk, old_ctx);
1324                 return;
1325         }
1326
1327         new_ctx->conn_finished = 1;
1328         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1329         new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1330         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1331         new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1332         new_ctx->rel_write_seq = 1;
1333         new_ctx->tcp_sock = newsk;
1334
1335         if (subflow_req->mp_capable) {
1336                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1337                  * is fully established only after we receive the remote key
1338                  */
1339                 new_ctx->mp_capable = 1;
1340                 new_ctx->local_key = subflow_req->local_key;
1341                 new_ctx->token = subflow_req->token;
1342                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1343                 new_ctx->idsn = subflow_req->idsn;
1344         } else if (subflow_req->mp_join) {
1345                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1346                 new_ctx->mp_join = 1;
1347                 new_ctx->fully_established = 1;
1348                 new_ctx->backup = subflow_req->backup;
1349                 new_ctx->local_id = subflow_req->local_id;
1350                 new_ctx->token = subflow_req->token;
1351                 new_ctx->thmac = subflow_req->thmac;
1352         }
1353 }
1354
1355 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1356         .name           = "mptcp",
1357         .owner          = THIS_MODULE,
1358         .init           = subflow_ulp_init,
1359         .release        = subflow_ulp_release,
1360         .clone          = subflow_ulp_clone,
1361 };
1362
1363 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1364 {
1365         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1366         subflow_ops->slab_name = "request_sock_subflow";
1367
1368         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1369                                               subflow_ops->obj_size, 0,
1370                                               SLAB_ACCOUNT |
1371                                               SLAB_TYPESAFE_BY_RCU,
1372                                               NULL);
1373         if (!subflow_ops->slab)
1374                 return -ENOMEM;
1375
1376         subflow_ops->destructor = subflow_req_destructor;
1377
1378         return 0;
1379 }
1380
1381 void __init mptcp_subflow_init(void)
1382 {
1383         mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1384         if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1385                 panic("MPTCP: failed to init subflow request sock ops\n");
1386
1387         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1388         subflow_request_sock_ipv4_ops.init_req = subflow_v4_init_req;
1389
1390         subflow_specific = ipv4_specific;
1391         subflow_specific.conn_request = subflow_v4_conn_request;
1392         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1393         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1394
1395 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1396         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1397         subflow_request_sock_ipv6_ops.init_req = subflow_v6_init_req;
1398
1399         subflow_v6_specific = ipv6_specific;
1400         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1401         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1402         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1403
1404         subflow_v6m_specific = subflow_v6_specific;
1405         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1406         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1407         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1408         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1409         subflow_v6m_specific.net_frag_header_len = 0;
1410 #endif
1411
1412         mptcp_diag_subflow_init(&subflow_ulp_ops);
1413
1414         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1415                 panic("MPTCP: failed to register subflows to ULP\n");
1416 }