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 <uapi/linux/mptcp.h>
  24 #include "protocol.h"
  25 #include "mib.h"
  26
  27 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
  28                                   enum linux_mptcp_mib_field field)
  29 {
  30         MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
  31 }
  32
  33 static void subflow_req_destructor(struct request_sock *req)
  34 {
  35         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  36
  37         pr_debug("subflow_req=%p", subflow_req);
  38
  39         if (subflow_req->msk)
  40                 sock_put((struct sock *)subflow_req->msk);
  41
  42         mptcp_token_destroy_request(req);
  43         tcp_request_sock_ops.destructor(req);
  44 }
  45
  46 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
  47                                   void *hmac)
  48 {
  49         u8 msg[8];
  50
  51         put_unaligned_be32(nonce1, &msg[0]);
  52         put_unaligned_be32(nonce2, &msg[4]);
  53
  54         mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
  55 }
  56
  57 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
  58 {
  59         return mptcp_is_fully_established((void *)msk) &&
  60                READ_ONCE(msk->pm.accept_subflow);
  61 }
  62
  63 /* validate received token and create truncated hmac and nonce for SYN-ACK */
  64 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req,
  65                                                      const struct sk_buff *skb)
  66 {
  67         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  68         u8 hmac[SHA256_DIGEST_SIZE];
  69         struct mptcp_sock *msk;
  70         int local_id;
  71
  72         msk = mptcp_token_get_sock(subflow_req->token);
  73         if (!msk) {
  74                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
  75                 return NULL;
  76         }
  77
  78         local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
  79         if (local_id < 0) {
  80                 sock_put((struct sock *)msk);
  81                 return NULL;
  82         }
  83         subflow_req->local_id = local_id;
  84
  85         get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
  86
  87         subflow_generate_hmac(msk->local_key, msk->remote_key,
  88                               subflow_req->local_nonce,
  89                               subflow_req->remote_nonce, hmac);
  90
  91         subflow_req->thmac = get_unaligned_be64(hmac);
  92         return msk;
  93 }
  94
  95 static int __subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
  96 {
  97         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  98
  99         subflow_req->mp_capable = 0;
 100         subflow_req->mp_join = 0;
 101         subflow_req->msk = NULL;
 102         mptcp_token_init_request(req);
 103
 104 #ifdef CONFIG_TCP_MD5SIG
 105         /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
 106          * TCP option space.
 107          */
 108         if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
 109                 return -EINVAL;
 110 #endif
 111
 112         return 0;
 113 }
 114
 115 static void subflow_init_req(struct request_sock *req,
 116                              const struct sock *sk_listener,
 117                              struct sk_buff *skb)
 118 {
 119         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 120         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 121         struct mptcp_options_received mp_opt;
 122         int ret;
 123
 124         pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
 125
 126         ret = __subflow_init_req(req, sk_listener);
 127         if (ret)
 128                 return;
 129
 130         mptcp_get_options(skb, &mp_opt);
 131
 132         if (mp_opt.mp_capable) {
 133                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
 134
 135                 if (mp_opt.mp_join)
 136                         return;
 137         } else if (mp_opt.mp_join) {
 138                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
 139         }
 140
 141         if (mp_opt.mp_capable && listener->request_mptcp) {
 142                 int err, retries = 4;
 143
 144                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 145 again:
 146                 do {
 147                         get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
 148                 } while (subflow_req->local_key == 0);
 149
 150                 if (unlikely(req->syncookie)) {
 151                         mptcp_crypto_key_sha(subflow_req->local_key,
 152                                              &subflow_req->token,
 153                                              &subflow_req->idsn);
 154                         if (mptcp_token_exists(subflow_req->token)) {
 155                                 if (retries-- > 0)
 156                                         goto again;
 157                         } else {
 158                                 subflow_req->mp_capable = 1;
 159                         }
 160                         return;
 161                 }
 162
 163                 err = mptcp_token_new_request(req);
 164                 if (err == 0)
 165                         subflow_req->mp_capable = 1;
 166                 else if (retries-- > 0)
 167                         goto again;
 168
 169         } else if (mp_opt.mp_join && listener->request_mptcp) {
 170                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 171                 subflow_req->mp_join = 1;
 172                 subflow_req->backup = mp_opt.backup;
 173                 subflow_req->remote_id = mp_opt.join_id;
 174                 subflow_req->token = mp_opt.token;
 175                 subflow_req->remote_nonce = mp_opt.nonce;
 176                 subflow_req->msk = subflow_token_join_request(req, skb);
 177
 178                 if (unlikely(req->syncookie) && subflow_req->msk) {
 179                         if (mptcp_can_accept_new_subflow(subflow_req->msk))
 180                                 subflow_init_req_cookie_join_save(subflow_req, skb);
 181                 }
 182
 183                 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
 184                          subflow_req->remote_nonce, subflow_req->msk);
 185         }
 186 }
 187
 188 int mptcp_subflow_init_cookie_req(struct request_sock *req,
 189                                   const struct sock *sk_listener,
 190                                   struct sk_buff *skb)
 191 {
 192         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 193         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 194         struct mptcp_options_received mp_opt;
 195         int err;
 196
 197         err = __subflow_init_req(req, sk_listener);
 198         if (err)
 199                 return err;
 200
 201         mptcp_get_options(skb, &mp_opt);
 202
 203         if (mp_opt.mp_capable && mp_opt.mp_join)
 204                 return -EINVAL;
 205
 206         if (mp_opt.mp_capable && listener->request_mptcp) {
 207                 if (mp_opt.sndr_key == 0)
 208                         return -EINVAL;
 209
 210                 subflow_req->local_key = mp_opt.rcvr_key;
 211                 err = mptcp_token_new_request(req);
 212                 if (err)
 213                         return err;
 214
 215                 subflow_req->mp_capable = 1;
 216                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 217         } else if (mp_opt.mp_join && listener->request_mptcp) {
 218                 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
 219                         return -EINVAL;
 220
 221                 if (mptcp_can_accept_new_subflow(subflow_req->msk))
 222                         subflow_req->mp_join = 1;
 223
 224                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 225         }
 226
 227         return 0;
 228 }
 229 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
 230
 231 static void subflow_v4_init_req(struct request_sock *req,
 232                                 const struct sock *sk_listener,
 233                                 struct sk_buff *skb)
 234 {
 235         tcp_rsk(req)->is_mptcp = 1;
 236
 237         tcp_request_sock_ipv4_ops.init_req(req, sk_listener, skb);
 238
 239         subflow_init_req(req, sk_listener, skb);
 240 }
 241
 242 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 243 static void subflow_v6_init_req(struct request_sock *req,
 244                                 const struct sock *sk_listener,
 245                                 struct sk_buff *skb)
 246 {
 247         tcp_rsk(req)->is_mptcp = 1;
 248
 249         tcp_request_sock_ipv6_ops.init_req(req, sk_listener, skb);
 250
 251         subflow_init_req(req, sk_listener, skb);
 252 }
 253 #endif
 254
 255 /* validate received truncated hmac and create hmac for third ACK */
 256 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
 257 {
 258         u8 hmac[SHA256_DIGEST_SIZE];
 259         u64 thmac;
 260
 261         subflow_generate_hmac(subflow->remote_key, subflow->local_key,
 262                               subflow->remote_nonce, subflow->local_nonce,
 263                               hmac);
 264
 265         thmac = get_unaligned_be64(hmac);
 266         pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
 267                  subflow, subflow->token,
 268                  (unsigned long long)thmac,
 269                  (unsigned long long)subflow->thmac);
 270
 271         return thmac == subflow->thmac;
 272 }
 273
 274 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 275 {
 276         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 277         struct mptcp_options_received mp_opt;
 278         struct sock *parent = subflow->conn;
 279
 280         subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 281
 282         if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
 283                 inet_sk_state_store(parent, TCP_ESTABLISHED);
 284                 parent->sk_state_change(parent);
 285         }
 286
 287         /* be sure no special action on any packet other than syn-ack */
 288         if (subflow->conn_finished)
 289                 return;
 290
 291         subflow->rel_write_seq = 1;
 292         subflow->conn_finished = 1;
 293         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 294         pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
 295
 296         mptcp_get_options(skb, &mp_opt);
 297         if (subflow->request_mptcp) {
 298                 if (!mp_opt.mp_capable) {
 299                         MPTCP_INC_STATS(sock_net(sk),
 300                                         MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
 301                         mptcp_do_fallback(sk);
 302                         pr_fallback(mptcp_sk(subflow->conn));
 303                         goto fallback;
 304                 }
 305
 306                 subflow->mp_capable = 1;
 307                 subflow->can_ack = 1;
 308                 subflow->remote_key = mp_opt.sndr_key;
 309                 pr_debug("subflow=%p, remote_key=%llu", subflow,
 310                          subflow->remote_key);
 311                 mptcp_finish_connect(sk);
 312         } else if (subflow->request_join) {
 313                 u8 hmac[SHA256_DIGEST_SIZE];
 314
 315                 if (!mp_opt.mp_join)
 316                         goto do_reset;
 317
 318                 subflow->thmac = mp_opt.thmac;
 319                 subflow->remote_nonce = mp_opt.nonce;
 320                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
 321                          subflow->thmac, subflow->remote_nonce);
 322
 323                 if (!subflow_thmac_valid(subflow)) {
 324                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 325                         goto do_reset;
 326                 }
 327
 328                 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 329                                       subflow->local_nonce,
 330                                       subflow->remote_nonce,
 331                                       hmac);
 332                 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
 333
 334                 if (!mptcp_finish_join(sk))
 335                         goto do_reset;
 336
 337                 subflow->mp_join = 1;
 338                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 339         } else if (mptcp_check_fallback(sk)) {
 340 fallback:
 341                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
 342         }
 343         return;
 344
 345 do_reset:
 346         tcp_send_active_reset(sk, GFP_ATOMIC);
 347         tcp_done(sk);
 348 }
 349
 350 struct request_sock_ops mptcp_subflow_request_sock_ops;
 351 EXPORT_SYMBOL_GPL(mptcp_subflow_request_sock_ops);
 352 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
 353
 354 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 355 {
 356         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 357
 358         pr_debug("subflow=%p", subflow);
 359
 360         /* Never answer to SYNs sent to broadcast or multicast */
 361         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 362                 goto drop;
 363
 364         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 365                                 &subflow_request_sock_ipv4_ops,
 366                                 sk, skb);
 367 drop:
 368         tcp_listendrop(sk);
 369         return 0;
 370 }
 371
 372 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 373 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
 374 static struct inet_connection_sock_af_ops subflow_v6_specific;
 375 static struct inet_connection_sock_af_ops subflow_v6m_specific;
 376
 377 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 378 {
 379         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 380
 381         pr_debug("subflow=%p", subflow);
 382
 383         if (skb->protocol == htons(ETH_P_IP))
 384                 return subflow_v4_conn_request(sk, skb);
 385
 386         if (!ipv6_unicast_destination(skb))
 387                 goto drop;
 388
 389         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 390                                 &subflow_request_sock_ipv6_ops, sk, skb);
 391
 392 drop:
 393         tcp_listendrop(sk);
 394         return 0; /* don't send reset */
 395 }
 396 #endif
 397
 398 /* validate hmac received in third ACK */
 399 static bool subflow_hmac_valid(const struct request_sock *req,
 400                                const struct mptcp_options_received *mp_opt)
 401 {
 402         const struct mptcp_subflow_request_sock *subflow_req;
 403         u8 hmac[SHA256_DIGEST_SIZE];
 404         struct mptcp_sock *msk;
 405
 406         subflow_req = mptcp_subflow_rsk(req);
 407         msk = subflow_req->msk;
 408         if (!msk)
 409                 return false;
 410
 411         subflow_generate_hmac(msk->remote_key, msk->local_key,
 412                               subflow_req->remote_nonce,
 413                               subflow_req->local_nonce, hmac);
 414
 415         return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
 416 }
 417
 418 static void mptcp_sock_destruct(struct sock *sk)
 419 {
 420         /* if new mptcp socket isn't accepted, it is free'd
 421          * from the tcp listener sockets request queue, linked
 422          * from req->sk.  The tcp socket is released.
 423          * This calls the ULP release function which will
 424          * also remove the mptcp socket, via
 425          * sock_put(ctx->conn).
 426          *
 427          * Problem is that the mptcp socket will be in
 428          * ESTABLISHED state and will not have the SOCK_DEAD flag.
 429          * Both result in warnings from inet_sock_destruct.
 430          */
 431
 432         if (sk->sk_state == TCP_ESTABLISHED) {
 433                 sk->sk_state = TCP_CLOSE;
 434                 WARN_ON_ONCE(sk->sk_socket);
 435                 sock_orphan(sk);
 436         }
 437
 438         mptcp_destroy_common(mptcp_sk(sk));
 439         inet_sock_destruct(sk);
 440 }
 441
 442 static void mptcp_force_close(struct sock *sk)
 443 {
 444         inet_sk_state_store(sk, TCP_CLOSE);
 445         sk_common_release(sk);
 446 }
 447
 448 static void subflow_ulp_fallback(struct sock *sk,
 449                                  struct mptcp_subflow_context *old_ctx)
 450 {
 451         struct inet_connection_sock *icsk = inet_csk(sk);
 452
 453         mptcp_subflow_tcp_fallback(sk, old_ctx);
 454         icsk->icsk_ulp_ops = NULL;
 455         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 456         tcp_sk(sk)->is_mptcp = 0;
 457 }
 458
 459 static void subflow_drop_ctx(struct sock *ssk)
 460 {
 461         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 462
 463         if (!ctx)
 464                 return;
 465
 466         subflow_ulp_fallback(ssk, ctx);
 467         if (ctx->conn)
 468                 sock_put(ctx->conn);
 469
 470         kfree_rcu(ctx, rcu);
 471 }
 472
 473 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 474                                      struct mptcp_options_received *mp_opt)
 475 {
 476         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 477
 478         subflow->remote_key = mp_opt->sndr_key;
 479         subflow->fully_established = 1;
 480         subflow->can_ack = 1;
 481         WRITE_ONCE(msk->fully_established, true);
 482 }
 483
 484 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 485                                           struct sk_buff *skb,
 486                                           struct request_sock *req,
 487                                           struct dst_entry *dst,
 488                                           struct request_sock *req_unhash,
 489                                           bool *own_req)
 490 {
 491         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 492         struct mptcp_subflow_request_sock *subflow_req;
 493         struct mptcp_options_received mp_opt;
 494         bool fallback, fallback_is_fatal;
 495         struct sock *new_msk = NULL;
 496         struct sock *child;
 497
 498         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 499
 500         /* After child creation we must look for 'mp_capable' even when options
 501          * are not parsed
 502          */
 503         mp_opt.mp_capable = 0;
 504
 505         /* hopefully temporary handling for MP_JOIN+syncookie */
 506         subflow_req = mptcp_subflow_rsk(req);
 507         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 508         fallback = !tcp_rsk(req)->is_mptcp;
 509         if (fallback)
 510                 goto create_child;
 511
 512         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 513         if (subflow_req->mp_capable) {
 514                 if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
 515                         /* here we can receive and accept an in-window,
 516                          * out-of-order pkt, which will not carry the MP_CAPABLE
 517                          * opt even on mptcp enabled paths
 518                          */
 519                         goto create_msk;
 520                 }
 521
 522                 mptcp_get_options(skb, &mp_opt);
 523                 if (!mp_opt.mp_capable) {
 524                         fallback = true;
 525                         goto create_child;
 526                 }
 527
 528 create_msk:
 529                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 530                 if (!new_msk)
 531                         fallback = true;
 532         } else if (subflow_req->mp_join) {
 533                 mptcp_get_options(skb, &mp_opt);
 534                 if (!mp_opt.mp_join ||
 535                     !mptcp_can_accept_new_subflow(subflow_req->msk) ||
 536                     !subflow_hmac_valid(req, &mp_opt)) {
 537                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 538                         fallback = true;
 539                 }
 540         }
 541
 542 create_child:
 543         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 544                                                      req_unhash, own_req);
 545
 546         if (child && *own_req) {
 547                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 548
 549                 tcp_rsk(req)->drop_req = false;
 550
 551                 /* we need to fallback on ctx allocation failure and on pre-reqs
 552                  * checking above. In the latter scenario we additionally need
 553                  * to reset the context to non MPTCP status.
 554                  */
 555                 if (!ctx || fallback) {
 556                         if (fallback_is_fatal)
 557                                 goto dispose_child;
 558
 559                         subflow_drop_ctx(child);
 560                         goto out;
 561                 }
 562
 563                 if (ctx->mp_capable) {
 564                         /* this can't race with mptcp_close(), as the msk is
 565                          * not yet exposted to user-space
 566                          */
 567                         inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
 568
 569                         /* new mpc subflow takes ownership of the newly
 570                          * created mptcp socket
 571                          */
 572                         new_msk->sk_destruct = mptcp_sock_destruct;
 573                         mptcp_pm_new_connection(mptcp_sk(new_msk), 1);
 574                         mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
 575                         ctx->conn = new_msk;
 576                         new_msk = NULL;
 577
 578                         /* with OoO packets we can reach here without ingress
 579                          * mpc option
 580                          */
 581                         if (mp_opt.mp_capable)
 582                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 583                 } else if (ctx->mp_join) {
 584                         struct mptcp_sock *owner;
 585
 586                         owner = subflow_req->msk;
 587                         if (!owner)
 588                                 goto dispose_child;
 589
 590                         /* move the msk reference ownership to the subflow */
 591                         subflow_req->msk = NULL;
 592                         ctx->conn = (struct sock *)owner;
 593                         if (!mptcp_finish_join(child))
 594                                 goto dispose_child;
 595
 596                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 597                         tcp_rsk(req)->drop_req = true;
 598                 }
 599         }
 600
 601 out:
 602         /* dispose of the left over mptcp master, if any */
 603         if (unlikely(new_msk))
 604                 mptcp_force_close(new_msk);
 605
 606         /* check for expected invariant - should never trigger, just help
 607          * catching eariler subtle bugs
 608          */
 609         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 610                      (!mptcp_subflow_ctx(child) ||
 611                       !mptcp_subflow_ctx(child)->conn));
 612         return child;
 613
 614 dispose_child:
 615         subflow_drop_ctx(child);
 616         tcp_rsk(req)->drop_req = true;
 617         inet_csk_prepare_for_destroy_sock(child);
 618         tcp_done(child);
 619         req->rsk_ops->send_reset(sk, skb);
 620
 621         /* The last child reference will be released by the caller */
 622         return child;
 623 }
 624
 625 static struct inet_connection_sock_af_ops subflow_specific;
 626
 627 enum mapping_status {
 628         MAPPING_OK,
 629         MAPPING_INVALID,
 630         MAPPING_EMPTY,
 631         MAPPING_DATA_FIN,
 632         MAPPING_DUMMY
 633 };
 634
 635 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
 636 {
 637         if ((u32)seq == (u32)old_seq)
 638                 return old_seq;
 639
 640         /* Assume map covers data not mapped yet. */
 641         return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
 642 }
 643
 644 static void warn_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 645 {
 646         WARN_ONCE(1, "Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 647                   ssn, subflow->map_subflow_seq, subflow->map_data_len);
 648 }
 649
 650 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 651 {
 652         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 653         unsigned int skb_consumed;
 654
 655         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 656         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 657                 return true;
 658
 659         return skb->len - skb_consumed <= subflow->map_data_len -
 660                                           mptcp_subflow_get_map_offset(subflow);
 661 }
 662
 663 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 664 {
 665         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 666         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 667
 668         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 669                 /* Mapping covers data later in the subflow stream,
 670                  * currently unsupported.
 671                  */
 672                 warn_bad_map(subflow, ssn);
 673                 return false;
 674         }
 675         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 676                                   subflow->map_data_len))) {
 677                 /* Mapping does covers past subflow data, invalid */
 678                 warn_bad_map(subflow, ssn + skb->len);
 679                 return false;
 680         }
 681         return true;
 682 }
 683
 684 static enum mapping_status get_mapping_status(struct sock *ssk,
 685                                               struct mptcp_sock *msk)
 686 {
 687         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 688         struct mptcp_ext *mpext;
 689         struct sk_buff *skb;
 690         u16 data_len;
 691         u64 map_seq;
 692
 693         skb = skb_peek(&ssk->sk_receive_queue);
 694         if (!skb)
 695                 return MAPPING_EMPTY;
 696
 697         if (mptcp_check_fallback(ssk))
 698                 return MAPPING_DUMMY;
 699
 700         mpext = mptcp_get_ext(skb);
 701         if (!mpext || !mpext->use_map) {
 702                 if (!subflow->map_valid && !skb->len) {
 703                         /* the TCP stack deliver 0 len FIN pkt to the receive
 704                          * queue, that is the only 0len pkts ever expected here,
 705                          * and we can admit no mapping only for 0 len pkts
 706                          */
 707                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 708                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 709                                           TCP_SKB_CB(skb)->seq,
 710                                           TCP_SKB_CB(skb)->end_seq,
 711                                           TCP_SKB_CB(skb)->tcp_flags);
 712                         sk_eat_skb(ssk, skb);
 713                         return MAPPING_EMPTY;
 714                 }
 715
 716                 if (!subflow->map_valid)
 717                         return MAPPING_INVALID;
 718
 719                 goto validate_seq;
 720         }
 721
 722         pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
 723                  mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
 724                  mpext->data_len, mpext->data_fin);
 725
 726         data_len = mpext->data_len;
 727         if (data_len == 0) {
 728                 pr_err("Infinite mapping not handled");
 729                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 730                 return MAPPING_INVALID;
 731         }
 732
 733         if (mpext->data_fin == 1) {
 734                 if (data_len == 1) {
 735                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
 736                                                                  mpext->dsn64);
 737                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
 738                         if (subflow->map_valid) {
 739                                 /* A DATA_FIN might arrive in a DSS
 740                                  * option before the previous mapping
 741                                  * has been fully consumed. Continue
 742                                  * handling the existing mapping.
 743                                  */
 744                                 skb_ext_del(skb, SKB_EXT_MPTCP);
 745                                 return MAPPING_OK;
 746                         } else {
 747                                 if (updated && schedule_work(&msk->work))
 748                                         sock_hold((struct sock *)msk);
 749
 750                                 return MAPPING_DATA_FIN;
 751                         }
 752                 } else {
 753                         u64 data_fin_seq = mpext->data_seq + data_len - 1;
 754
 755                         /* If mpext->data_seq is a 32-bit value, data_fin_seq
 756                          * must also be limited to 32 bits.
 757                          */
 758                         if (!mpext->dsn64)
 759                                 data_fin_seq &= GENMASK_ULL(31, 0);
 760
 761                         mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
 762                         pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
 763                                  data_fin_seq, mpext->dsn64);
 764                 }
 765
 766                 /* Adjust for DATA_FIN using 1 byte of sequence space */
 767                 data_len--;
 768         }
 769
 770         if (!mpext->dsn64) {
 771                 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
 772                                      mpext->data_seq);
 773                 subflow->use_64bit_ack = 0;
 774                 pr_debug("expanded seq=%llu", subflow->map_seq);
 775         } else {
 776                 map_seq = mpext->data_seq;
 777                 subflow->use_64bit_ack = 1;
 778         }
 779
 780         if (subflow->map_valid) {
 781                 /* Allow replacing only with an identical map */
 782                 if (subflow->map_seq == map_seq &&
 783                     subflow->map_subflow_seq == mpext->subflow_seq &&
 784                     subflow->map_data_len == data_len) {
 785                         skb_ext_del(skb, SKB_EXT_MPTCP);
 786                         return MAPPING_OK;
 787                 }
 788
 789                 /* If this skb data are fully covered by the current mapping,
 790                  * the new map would need caching, which is not supported
 791                  */
 792                 if (skb_is_fully_mapped(ssk, skb)) {
 793                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
 794                         return MAPPING_INVALID;
 795                 }
 796
 797                 /* will validate the next map after consuming the current one */
 798                 return MAPPING_OK;
 799         }
 800
 801         subflow->map_seq = map_seq;
 802         subflow->map_subflow_seq = mpext->subflow_seq;
 803         subflow->map_data_len = data_len;
 804         subflow->map_valid = 1;
 805         subflow->mpc_map = mpext->mpc_map;
 806         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
 807                  subflow->map_seq, subflow->map_subflow_seq,
 808                  subflow->map_data_len);
 809
 810 validate_seq:
 811         /* we revalidate valid mapping on new skb, because we must ensure
 812          * the current skb is completely covered by the available mapping
 813          */
 814         if (!validate_mapping(ssk, skb))
 815                 return MAPPING_INVALID;
 816
 817         skb_ext_del(skb, SKB_EXT_MPTCP);
 818         return MAPPING_OK;
 819 }
 820
 821 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
 822                                        u64 limit)
 823 {
 824         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 825         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
 826         u32 incr;
 827
 828         incr = limit >= skb->len ? skb->len + fin : limit;
 829
 830         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
 831                  subflow->map_subflow_seq);
 832         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
 833         tcp_sk(ssk)->copied_seq += incr;
 834         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
 835                 sk_eat_skb(ssk, skb);
 836         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
 837                 subflow->map_valid = 0;
 838         if (incr)
 839                 tcp_cleanup_rbuf(ssk, incr);
 840 }
 841
 842 static bool subflow_check_data_avail(struct sock *ssk)
 843 {
 844         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 845         enum mapping_status status;
 846         struct mptcp_sock *msk;
 847         struct sk_buff *skb;
 848
 849         pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
 850                  subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
 851         if (!skb_peek(&ssk->sk_receive_queue))
 852                 subflow->data_avail = 0;
 853         if (subflow->data_avail)
 854                 return true;
 855
 856         msk = mptcp_sk(subflow->conn);
 857         for (;;) {
 858                 u64 ack_seq;
 859                 u64 old_ack;
 860
 861                 status = get_mapping_status(ssk, msk);
 862                 pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
 863                 if (status == MAPPING_INVALID) {
 864                         ssk->sk_err = EBADMSG;
 865                         goto fatal;
 866                 }
 867                 if (status == MAPPING_DUMMY) {
 868                         __mptcp_do_fallback(msk);
 869                         skb = skb_peek(&ssk->sk_receive_queue);
 870                         subflow->map_valid = 1;
 871                         subflow->map_seq = READ_ONCE(msk->ack_seq);
 872                         subflow->map_data_len = skb->len;
 873                         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq -
 874                                                    subflow->ssn_offset;
 875                         subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
 876                         return true;
 877                 }
 878
 879                 if (status != MAPPING_OK)
 880                         return false;
 881
 882                 skb = skb_peek(&ssk->sk_receive_queue);
 883                 if (WARN_ON_ONCE(!skb))
 884                         return false;
 885
 886                 /* if msk lacks the remote key, this subflow must provide an
 887                  * MP_CAPABLE-based mapping
 888                  */
 889                 if (unlikely(!READ_ONCE(msk->can_ack))) {
 890                         if (!subflow->mpc_map) {
 891                                 ssk->sk_err = EBADMSG;
 892                                 goto fatal;
 893                         }
 894                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
 895                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
 896                         WRITE_ONCE(msk->can_ack, true);
 897                 }
 898
 899                 old_ack = READ_ONCE(msk->ack_seq);
 900                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
 901                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
 902                          ack_seq);
 903                 if (ack_seq == old_ack) {
 904                         subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
 905                         break;
 906                 } else if (after64(ack_seq, old_ack)) {
 907                         subflow->data_avail = MPTCP_SUBFLOW_OOO_DATA;
 908                         break;
 909                 }
 910
 911                 /* only accept in-sequence mapping. Old values are spurious
 912                  * retransmission
 913                  */
 914                 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
 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         subflow->data_avail = 0;
 926         return false;
 927 }
 928
 929 bool mptcp_subflow_data_available(struct sock *sk)
 930 {
 931         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 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         return subflow_check_data_avail(sk);
 945 }
 946
 947 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
 948  * not the ssk one.
 949  *
 950  * In mptcp, rwin is about the mptcp-level connection data.
 951  *
 952  * Data that is still on the ssk rx queue can thus be ignored,
 953  * as far as mptcp peer is concerened that data is still inflight.
 954  * DSS ACK is updated when skb is moved to the mptcp rx queue.
 955  */
 956 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
 957 {
 958         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 959         const struct sock *sk = subflow->conn;
 960
 961         *space = tcp_space(sk);
 962         *full_space = tcp_full_space(sk);
 963 }
 964
 965 static void subflow_data_ready(struct sock *sk)
 966 {
 967         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 968         u16 state = 1 << inet_sk_state_load(sk);
 969         struct sock *parent = subflow->conn;
 970         struct mptcp_sock *msk;
 971
 972         msk = mptcp_sk(parent);
 973         if (state & TCPF_LISTEN) {
 974                 set_bit(MPTCP_DATA_READY, &msk->flags);
 975                 parent->sk_data_ready(parent);
 976                 return;
 977         }
 978
 979         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
 980                      !subflow->mp_join && !(state & TCPF_CLOSE));
 981
 982         if (mptcp_subflow_data_available(sk))
 983                 mptcp_data_ready(parent, sk);
 984 }
 985
 986 static void subflow_write_space(struct sock *sk)
 987 {
 988         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 989         struct sock *parent = subflow->conn;
 990
 991         if (!sk_stream_is_writeable(sk))
 992                 return;
 993
 994         if (sk_stream_is_writeable(parent)) {
 995                 set_bit(MPTCP_SEND_SPACE, &mptcp_sk(parent)->flags);
 996                 smp_mb__after_atomic();
 997                 /* set SEND_SPACE before sk_stream_write_space clears NOSPACE */
 998                 sk_stream_write_space(parent);
 999         }
1000 }
1001
1002 static struct inet_connection_sock_af_ops *
1003 subflow_default_af_ops(struct sock *sk)
1004 {
1005 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1006         if (sk->sk_family == AF_INET6)
1007                 return &subflow_v6_specific;
1008 #endif
1009         return &subflow_specific;
1010 }
1011
1012 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1013 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1014 {
1015         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1016         struct inet_connection_sock *icsk = inet_csk(sk);
1017         struct inet_connection_sock_af_ops *target;
1018
1019         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1020
1021         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1022                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1023
1024         if (likely(icsk->icsk_af_ops == target))
1025                 return;
1026
1027         subflow->icsk_af_ops = icsk->icsk_af_ops;
1028         icsk->icsk_af_ops = target;
1029 }
1030 #endif
1031
1032 static void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1033                                 struct sockaddr_storage *addr)
1034 {
1035         memset(addr, 0, sizeof(*addr));
1036         addr->ss_family = info->family;
1037         if (addr->ss_family == AF_INET) {
1038                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1039
1040                 in_addr->sin_addr = info->addr;
1041                 in_addr->sin_port = info->port;
1042         }
1043 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1044         else if (addr->ss_family == AF_INET6) {
1045                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1046
1047                 in6_addr->sin6_addr = info->addr6;
1048                 in6_addr->sin6_port = info->port;
1049         }
1050 #endif
1051 }
1052
1053 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1054                             const struct mptcp_addr_info *remote)
1055 {
1056         struct mptcp_sock *msk = mptcp_sk(sk);
1057         struct mptcp_subflow_context *subflow;
1058         struct sockaddr_storage addr;
1059         int remote_id = remote->id;
1060         int local_id = loc->id;
1061         struct socket *sf;
1062         struct sock *ssk;
1063         u32 remote_token;
1064         int addrlen;
1065         int err;
1066
1067         if (!mptcp_is_fully_established(sk))
1068                 return -ENOTCONN;
1069
1070         err = mptcp_subflow_create_socket(sk, &sf);
1071         if (err)
1072                 return err;
1073
1074         ssk = sf->sk;
1075         subflow = mptcp_subflow_ctx(ssk);
1076         do {
1077                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1078         } while (!subflow->local_nonce);
1079
1080         if (!local_id) {
1081                 err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
1082                 if (err < 0)
1083                         goto failed;
1084
1085                 local_id = err;
1086         }
1087
1088         subflow->remote_key = msk->remote_key;
1089         subflow->local_key = msk->local_key;
1090         subflow->token = msk->token;
1091         mptcp_info2sockaddr(loc, &addr);
1092
1093         addrlen = sizeof(struct sockaddr_in);
1094 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1095         if (loc->family == AF_INET6)
1096                 addrlen = sizeof(struct sockaddr_in6);
1097 #endif
1098         ssk->sk_bound_dev_if = loc->ifindex;
1099         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1100         if (err)
1101                 goto failed;
1102
1103         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1104         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1105                  remote_token, local_id, remote_id);
1106         subflow->remote_token = remote_token;
1107         subflow->local_id = local_id;
1108         subflow->remote_id = remote_id;
1109         subflow->request_join = 1;
1110         subflow->request_bkup = !!(loc->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1111         mptcp_info2sockaddr(remote, &addr);
1112
1113         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1114         if (err && err != -EINPROGRESS)
1115                 goto failed;
1116
1117         spin_lock_bh(&msk->join_list_lock);
1118         list_add_tail(&subflow->node, &msk->join_list);
1119         spin_unlock_bh(&msk->join_list_lock);
1120
1121         return err;
1122
1123 failed:
1124         sock_release(sf);
1125         return err;
1126 }
1127
1128 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1129 {
1130         struct mptcp_subflow_context *subflow;
1131         struct net *net = sock_net(sk);
1132         struct socket *sf;
1133         int err;
1134
1135         /* un-accepted server sockets can reach here - on bad configuration
1136          * bail early to avoid greater trouble later
1137          */
1138         if (unlikely(!sk->sk_socket))
1139                 return -EINVAL;
1140
1141         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1142                                &sf);
1143         if (err)
1144                 return err;
1145
1146         lock_sock(sf->sk);
1147
1148         /* kernel sockets do not by default acquire net ref, but TCP timer
1149          * needs it.
1150          */
1151         sf->sk->sk_net_refcnt = 1;
1152         get_net(net);
1153 #ifdef CONFIG_PROC_FS
1154         this_cpu_add(*net->core.sock_inuse, 1);
1155 #endif
1156         err = tcp_set_ulp(sf->sk, "mptcp");
1157         release_sock(sf->sk);
1158
1159         if (err) {
1160                 sock_release(sf);
1161                 return err;
1162         }
1163
1164         /* the newly created socket really belongs to the owning MPTCP master
1165          * socket, even if for additional subflows the allocation is performed
1166          * by a kernel workqueue. Adjust inode references, so that the
1167          * procfs/diag interaces really show this one belonging to the correct
1168          * user.
1169          */
1170         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1171         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1172         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1173
1174         subflow = mptcp_subflow_ctx(sf->sk);
1175         pr_debug("subflow=%p", subflow);
1176
1177         *new_sock = sf;
1178         sock_hold(sk);
1179         subflow->conn = sk;
1180
1181         return 0;
1182 }
1183
1184 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1185                                                         gfp_t priority)
1186 {
1187         struct inet_connection_sock *icsk = inet_csk(sk);
1188         struct mptcp_subflow_context *ctx;
1189
1190         ctx = kzalloc(sizeof(*ctx), priority);
1191         if (!ctx)
1192                 return NULL;
1193
1194         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1195         INIT_LIST_HEAD(&ctx->node);
1196
1197         pr_debug("subflow=%p", ctx);
1198
1199         ctx->tcp_sock = sk;
1200
1201         return ctx;
1202 }
1203
1204 static void __subflow_state_change(struct sock *sk)
1205 {
1206         struct socket_wq *wq;
1207
1208         rcu_read_lock();
1209         wq = rcu_dereference(sk->sk_wq);
1210         if (skwq_has_sleeper(wq))
1211                 wake_up_interruptible_all(&wq->wait);
1212         rcu_read_unlock();
1213 }
1214
1215 static bool subflow_is_done(const struct sock *sk)
1216 {
1217         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1218 }
1219
1220 static void subflow_state_change(struct sock *sk)
1221 {
1222         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1223         struct sock *parent = subflow->conn;
1224
1225         __subflow_state_change(sk);
1226
1227         if (subflow_simultaneous_connect(sk)) {
1228                 mptcp_do_fallback(sk);
1229                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1230                 pr_fallback(mptcp_sk(parent));
1231                 subflow->conn_finished = 1;
1232                 if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
1233                         inet_sk_state_store(parent, TCP_ESTABLISHED);
1234                         parent->sk_state_change(parent);
1235                 }
1236         }
1237
1238         /* as recvmsg() does not acquire the subflow socket for ssk selection
1239          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1240          * the data available machinery here.
1241          */
1242         if (mptcp_subflow_data_available(sk))
1243                 mptcp_data_ready(parent, sk);
1244
1245         if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1246             !(parent->sk_shutdown & RCV_SHUTDOWN) &&
1247             !subflow->rx_eof && subflow_is_done(sk)) {
1248                 subflow->rx_eof = 1;
1249                 mptcp_subflow_eof(parent);
1250         }
1251 }
1252
1253 static int subflow_ulp_init(struct sock *sk)
1254 {
1255         struct inet_connection_sock *icsk = inet_csk(sk);
1256         struct mptcp_subflow_context *ctx;
1257         struct tcp_sock *tp = tcp_sk(sk);
1258         int err = 0;
1259
1260         /* disallow attaching ULP to a socket unless it has been
1261          * created with sock_create_kern()
1262          */
1263         if (!sk->sk_kern_sock) {
1264                 err = -EOPNOTSUPP;
1265                 goto out;
1266         }
1267
1268         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1269         if (!ctx) {
1270                 err = -ENOMEM;
1271                 goto out;
1272         }
1273
1274         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1275
1276         tp->is_mptcp = 1;
1277         ctx->icsk_af_ops = icsk->icsk_af_ops;
1278         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1279         ctx->tcp_data_ready = sk->sk_data_ready;
1280         ctx->tcp_state_change = sk->sk_state_change;
1281         ctx->tcp_write_space = sk->sk_write_space;
1282         sk->sk_data_ready = subflow_data_ready;
1283         sk->sk_write_space = subflow_write_space;
1284         sk->sk_state_change = subflow_state_change;
1285 out:
1286         return err;
1287 }
1288
1289 static void subflow_ulp_release(struct sock *sk)
1290 {
1291         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(sk);
1292
1293         if (!ctx)
1294                 return;
1295
1296         if (ctx->conn)
1297                 sock_put(ctx->conn);
1298
1299         kfree_rcu(ctx, rcu);
1300 }
1301
1302 static void subflow_ulp_clone(const struct request_sock *req,
1303                               struct sock *newsk,
1304                               const gfp_t priority)
1305 {
1306         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1307         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1308         struct mptcp_subflow_context *new_ctx;
1309
1310         if (!tcp_rsk(req)->is_mptcp ||
1311             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1312                 subflow_ulp_fallback(newsk, old_ctx);
1313                 return;
1314         }
1315
1316         new_ctx = subflow_create_ctx(newsk, priority);
1317         if (!new_ctx) {
1318                 subflow_ulp_fallback(newsk, old_ctx);
1319                 return;
1320         }
1321
1322         new_ctx->conn_finished = 1;
1323         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1324         new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1325         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1326         new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1327         new_ctx->rel_write_seq = 1;
1328         new_ctx->tcp_sock = newsk;
1329
1330         if (subflow_req->mp_capable) {
1331                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1332                  * is fully established only after we receive the remote key
1333                  */
1334                 new_ctx->mp_capable = 1;
1335                 new_ctx->local_key = subflow_req->local_key;
1336                 new_ctx->token = subflow_req->token;
1337                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1338                 new_ctx->idsn = subflow_req->idsn;
1339         } else if (subflow_req->mp_join) {
1340                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1341                 new_ctx->mp_join = 1;
1342                 new_ctx->fully_established = 1;
1343                 new_ctx->backup = subflow_req->backup;
1344                 new_ctx->local_id = subflow_req->local_id;
1345                 new_ctx->remote_id = subflow_req->remote_id;
1346                 new_ctx->token = subflow_req->token;
1347                 new_ctx->thmac = subflow_req->thmac;
1348         }
1349 }
1350
1351 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1352         .name           = "mptcp",
1353         .owner          = THIS_MODULE,
1354         .init           = subflow_ulp_init,
1355         .release        = subflow_ulp_release,
1356         .clone          = subflow_ulp_clone,
1357 };
1358
1359 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1360 {
1361         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1362         subflow_ops->slab_name = "request_sock_subflow";
1363
1364         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1365                                               subflow_ops->obj_size, 0,
1366                                               SLAB_ACCOUNT |
1367                                               SLAB_TYPESAFE_BY_RCU,
1368                                               NULL);
1369         if (!subflow_ops->slab)
1370                 return -ENOMEM;
1371
1372         subflow_ops->destructor = subflow_req_destructor;
1373
1374         return 0;
1375 }
1376
1377 void __init mptcp_subflow_init(void)
1378 {
1379         mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1380         if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1381                 panic("MPTCP: failed to init subflow request sock ops\n");
1382
1383         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1384         subflow_request_sock_ipv4_ops.init_req = subflow_v4_init_req;
1385
1386         subflow_specific = ipv4_specific;
1387         subflow_specific.conn_request = subflow_v4_conn_request;
1388         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1389         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1390
1391 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1392         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1393         subflow_request_sock_ipv6_ops.init_req = subflow_v6_init_req;
1394
1395         subflow_v6_specific = ipv6_specific;
1396         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1397         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1398         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1399
1400         subflow_v6m_specific = subflow_v6_specific;
1401         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1402         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1403         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1404         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1405         subflow_v6m_specific.net_frag_header_len = 0;
1406 #endif
1407
1408         mptcp_diag_subflow_init(&subflow_ulp_ops);
1409
1410         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1411                 panic("MPTCP: failed to register subflows to ULP\n");
1412 }