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         skb_rbtree_purge(&mptcp_sk(sk)->out_of_order_queue);
 439         mptcp_token_destroy(mptcp_sk(sk));
 440         inet_sock_destruct(sk);
 441 }
 442
 443 static void mptcp_force_close(struct sock *sk)
 444 {
 445         inet_sk_state_store(sk, TCP_CLOSE);
 446         sk_common_release(sk);
 447 }
 448
 449 static void subflow_ulp_fallback(struct sock *sk,
 450                                  struct mptcp_subflow_context *old_ctx)
 451 {
 452         struct inet_connection_sock *icsk = inet_csk(sk);
 453
 454         mptcp_subflow_tcp_fallback(sk, old_ctx);
 455         icsk->icsk_ulp_ops = NULL;
 456         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 457         tcp_sk(sk)->is_mptcp = 0;
 458 }
 459
 460 static void subflow_drop_ctx(struct sock *ssk)
 461 {
 462         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 463
 464         if (!ctx)
 465                 return;
 466
 467         subflow_ulp_fallback(ssk, ctx);
 468         if (ctx->conn)
 469                 sock_put(ctx->conn);
 470
 471         kfree_rcu(ctx, rcu);
 472 }
 473
 474 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 475                                      struct mptcp_options_received *mp_opt)
 476 {
 477         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 478
 479         subflow->remote_key = mp_opt->sndr_key;
 480         subflow->fully_established = 1;
 481         subflow->can_ack = 1;
 482         WRITE_ONCE(msk->fully_established, true);
 483 }
 484
 485 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 486                                           struct sk_buff *skb,
 487                                           struct request_sock *req,
 488                                           struct dst_entry *dst,
 489                                           struct request_sock *req_unhash,
 490                                           bool *own_req)
 491 {
 492         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 493         struct mptcp_subflow_request_sock *subflow_req;
 494         struct mptcp_options_received mp_opt;
 495         bool fallback, fallback_is_fatal;
 496         struct sock *new_msk = NULL;
 497         struct sock *child;
 498
 499         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 500
 501         /* After child creation we must look for 'mp_capable' even when options
 502          * are not parsed
 503          */
 504         mp_opt.mp_capable = 0;
 505
 506         /* hopefully temporary handling for MP_JOIN+syncookie */
 507         subflow_req = mptcp_subflow_rsk(req);
 508         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 509         fallback = !tcp_rsk(req)->is_mptcp;
 510         if (fallback)
 511                 goto create_child;
 512
 513         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 514         if (subflow_req->mp_capable) {
 515                 if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
 516                         /* here we can receive and accept an in-window,
 517                          * out-of-order pkt, which will not carry the MP_CAPABLE
 518                          * opt even on mptcp enabled paths
 519                          */
 520                         goto create_msk;
 521                 }
 522
 523                 mptcp_get_options(skb, &mp_opt);
 524                 if (!mp_opt.mp_capable) {
 525                         fallback = true;
 526                         goto create_child;
 527                 }
 528
 529 create_msk:
 530                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 531                 if (!new_msk)
 532                         fallback = true;
 533         } else if (subflow_req->mp_join) {
 534                 mptcp_get_options(skb, &mp_opt);
 535                 if (!mp_opt.mp_join ||
 536                     !mptcp_can_accept_new_subflow(subflow_req->msk) ||
 537                     !subflow_hmac_valid(req, &mp_opt)) {
 538                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 539                         fallback = true;
 540                 }
 541         }
 542
 543 create_child:
 544         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 545                                                      req_unhash, own_req);
 546
 547         if (child && *own_req) {
 548                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 549
 550                 tcp_rsk(req)->drop_req = false;
 551
 552                 /* we need to fallback on ctx allocation failure and on pre-reqs
 553                  * checking above. In the latter scenario we additionally need
 554                  * to reset the context to non MPTCP status.
 555                  */
 556                 if (!ctx || fallback) {
 557                         if (fallback_is_fatal)
 558                                 goto dispose_child;
 559
 560                         subflow_drop_ctx(child);
 561                         goto out;
 562                 }
 563
 564                 if (ctx->mp_capable) {
 565                         /* this can't race with mptcp_close(), as the msk is
 566                          * not yet exposted to user-space
 567                          */
 568                         inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
 569
 570                         /* new mpc subflow takes ownership of the newly
 571                          * created mptcp socket
 572                          */
 573                         new_msk->sk_destruct = mptcp_sock_destruct;
 574                         mptcp_pm_new_connection(mptcp_sk(new_msk), 1);
 575                         mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
 576                         ctx->conn = new_msk;
 577                         new_msk = NULL;
 578
 579                         /* with OoO packets we can reach here without ingress
 580                          * mpc option
 581                          */
 582                         if (mp_opt.mp_capable)
 583                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 584                 } else if (ctx->mp_join) {
 585                         struct mptcp_sock *owner;
 586
 587                         owner = subflow_req->msk;
 588                         if (!owner)
 589                                 goto dispose_child;
 590
 591                         /* move the msk reference ownership to the subflow */
 592                         subflow_req->msk = NULL;
 593                         ctx->conn = (struct sock *)owner;
 594                         if (!mptcp_finish_join(child))
 595                                 goto dispose_child;
 596
 597                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 598                         tcp_rsk(req)->drop_req = true;
 599                 }
 600         }
 601
 602 out:
 603         /* dispose of the left over mptcp master, if any */
 604         if (unlikely(new_msk))
 605                 mptcp_force_close(new_msk);
 606
 607         /* check for expected invariant - should never trigger, just help
 608          * catching eariler subtle bugs
 609          */
 610         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 611                      (!mptcp_subflow_ctx(child) ||
 612                       !mptcp_subflow_ctx(child)->conn));
 613         return child;
 614
 615 dispose_child:
 616         subflow_drop_ctx(child);
 617         tcp_rsk(req)->drop_req = true;
 618         inet_csk_prepare_for_destroy_sock(child);
 619         tcp_done(child);
 620         req->rsk_ops->send_reset(sk, skb);
 621
 622         /* The last child reference will be released by the caller */
 623         return child;
 624 }
 625
 626 static struct inet_connection_sock_af_ops subflow_specific;
 627
 628 enum mapping_status {
 629         MAPPING_OK,
 630         MAPPING_INVALID,
 631         MAPPING_EMPTY,
 632         MAPPING_DATA_FIN,
 633         MAPPING_DUMMY
 634 };
 635
 636 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
 637 {
 638         if ((u32)seq == (u32)old_seq)
 639                 return old_seq;
 640
 641         /* Assume map covers data not mapped yet. */
 642         return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
 643 }
 644
 645 static void warn_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 646 {
 647         WARN_ONCE(1, "Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 648                   ssn, subflow->map_subflow_seq, subflow->map_data_len);
 649 }
 650
 651 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 652 {
 653         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 654         unsigned int skb_consumed;
 655
 656         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 657         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 658                 return true;
 659
 660         return skb->len - skb_consumed <= subflow->map_data_len -
 661                                           mptcp_subflow_get_map_offset(subflow);
 662 }
 663
 664 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 665 {
 666         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 667         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 668
 669         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 670                 /* Mapping covers data later in the subflow stream,
 671                  * currently unsupported.
 672                  */
 673                 warn_bad_map(subflow, ssn);
 674                 return false;
 675         }
 676         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 677                                   subflow->map_data_len))) {
 678                 /* Mapping does covers past subflow data, invalid */
 679                 warn_bad_map(subflow, ssn + skb->len);
 680                 return false;
 681         }
 682         return true;
 683 }
 684
 685 static enum mapping_status get_mapping_status(struct sock *ssk,
 686                                               struct mptcp_sock *msk)
 687 {
 688         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 689         struct mptcp_ext *mpext;
 690         struct sk_buff *skb;
 691         u16 data_len;
 692         u64 map_seq;
 693
 694         skb = skb_peek(&ssk->sk_receive_queue);
 695         if (!skb)
 696                 return MAPPING_EMPTY;
 697
 698         if (mptcp_check_fallback(ssk))
 699                 return MAPPING_DUMMY;
 700
 701         mpext = mptcp_get_ext(skb);
 702         if (!mpext || !mpext->use_map) {
 703                 if (!subflow->map_valid && !skb->len) {
 704                         /* the TCP stack deliver 0 len FIN pkt to the receive
 705                          * queue, that is the only 0len pkts ever expected here,
 706                          * and we can admit no mapping only for 0 len pkts
 707                          */
 708                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 709                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 710                                           TCP_SKB_CB(skb)->seq,
 711                                           TCP_SKB_CB(skb)->end_seq,
 712                                           TCP_SKB_CB(skb)->tcp_flags);
 713                         sk_eat_skb(ssk, skb);
 714                         return MAPPING_EMPTY;
 715                 }
 716
 717                 if (!subflow->map_valid)
 718                         return MAPPING_INVALID;
 719
 720                 goto validate_seq;
 721         }
 722
 723         pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
 724                  mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
 725                  mpext->data_len, mpext->data_fin);
 726
 727         data_len = mpext->data_len;
 728         if (data_len == 0) {
 729                 pr_err("Infinite mapping not handled");
 730                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 731                 return MAPPING_INVALID;
 732         }
 733
 734         if (mpext->data_fin == 1) {
 735                 if (data_len == 1) {
 736                         mptcp_update_rcv_data_fin(msk, mpext->data_seq);
 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                                 return MAPPING_DATA_FIN;
 748                         }
 749                 } else {
 750                         mptcp_update_rcv_data_fin(msk, mpext->data_seq + data_len);
 751                         pr_debug("DATA_FIN with mapping seq=%llu", mpext->data_seq + data_len);
 752                 }
 753
 754                 /* Adjust for DATA_FIN using 1 byte of sequence space */
 755                 data_len--;
 756         }
 757
 758         if (!mpext->dsn64) {
 759                 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
 760                                      mpext->data_seq);
 761                 subflow->use_64bit_ack = 0;
 762                 pr_debug("expanded seq=%llu", subflow->map_seq);
 763         } else {
 764                 map_seq = mpext->data_seq;
 765                 subflow->use_64bit_ack = 1;
 766         }
 767
 768         if (subflow->map_valid) {
 769                 /* Allow replacing only with an identical map */
 770                 if (subflow->map_seq == map_seq &&
 771                     subflow->map_subflow_seq == mpext->subflow_seq &&
 772                     subflow->map_data_len == data_len) {
 773                         skb_ext_del(skb, SKB_EXT_MPTCP);
 774                         return MAPPING_OK;
 775                 }
 776
 777                 /* If this skb data are fully covered by the current mapping,
 778                  * the new map would need caching, which is not supported
 779                  */
 780                 if (skb_is_fully_mapped(ssk, skb)) {
 781                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
 782                         return MAPPING_INVALID;
 783                 }
 784
 785                 /* will validate the next map after consuming the current one */
 786                 return MAPPING_OK;
 787         }
 788
 789         subflow->map_seq = map_seq;
 790         subflow->map_subflow_seq = mpext->subflow_seq;
 791         subflow->map_data_len = data_len;
 792         subflow->map_valid = 1;
 793         subflow->mpc_map = mpext->mpc_map;
 794         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
 795                  subflow->map_seq, subflow->map_subflow_seq,
 796                  subflow->map_data_len);
 797
 798 validate_seq:
 799         /* we revalidate valid mapping on new skb, because we must ensure
 800          * the current skb is completely covered by the available mapping
 801          */
 802         if (!validate_mapping(ssk, skb))
 803                 return MAPPING_INVALID;
 804
 805         skb_ext_del(skb, SKB_EXT_MPTCP);
 806         return MAPPING_OK;
 807 }
 808
 809 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
 810                                        unsigned int limit)
 811 {
 812         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 813         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
 814         u32 incr;
 815
 816         incr = limit >= skb->len ? skb->len + fin : limit;
 817
 818         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
 819                  subflow->map_subflow_seq);
 820         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
 821         tcp_sk(ssk)->copied_seq += incr;
 822         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
 823                 sk_eat_skb(ssk, skb);
 824         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
 825                 subflow->map_valid = 0;
 826 }
 827
 828 static bool subflow_check_data_avail(struct sock *ssk)
 829 {
 830         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 831         enum mapping_status status;
 832         struct mptcp_sock *msk;
 833         struct sk_buff *skb;
 834
 835         pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
 836                  subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
 837         if (!skb_peek(&ssk->sk_receive_queue))
 838                 subflow->data_avail = 0;
 839         if (subflow->data_avail)
 840                 return true;
 841
 842         msk = mptcp_sk(subflow->conn);
 843         for (;;) {
 844                 u64 ack_seq;
 845                 u64 old_ack;
 846
 847                 status = get_mapping_status(ssk, msk);
 848                 pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
 849                 if (status == MAPPING_INVALID) {
 850                         ssk->sk_err = EBADMSG;
 851                         goto fatal;
 852                 }
 853                 if (status == MAPPING_DUMMY) {
 854                         __mptcp_do_fallback(msk);
 855                         skb = skb_peek(&ssk->sk_receive_queue);
 856                         subflow->map_valid = 1;
 857                         subflow->map_seq = READ_ONCE(msk->ack_seq);
 858                         subflow->map_data_len = skb->len;
 859                         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq -
 860                                                    subflow->ssn_offset;
 861                         subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
 862                         return true;
 863                 }
 864
 865                 if (status != MAPPING_OK)
 866                         return false;
 867
 868                 skb = skb_peek(&ssk->sk_receive_queue);
 869                 if (WARN_ON_ONCE(!skb))
 870                         return false;
 871
 872                 /* if msk lacks the remote key, this subflow must provide an
 873                  * MP_CAPABLE-based mapping
 874                  */
 875                 if (unlikely(!READ_ONCE(msk->can_ack))) {
 876                         if (!subflow->mpc_map) {
 877                                 ssk->sk_err = EBADMSG;
 878                                 goto fatal;
 879                         }
 880                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
 881                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
 882                         WRITE_ONCE(msk->can_ack, true);
 883                 }
 884
 885                 old_ack = READ_ONCE(msk->ack_seq);
 886                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
 887                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
 888                          ack_seq);
 889                 if (ack_seq == old_ack) {
 890                         subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
 891                         break;
 892                 } else if (after64(ack_seq, old_ack)) {
 893                         subflow->data_avail = MPTCP_SUBFLOW_OOO_DATA;
 894                         break;
 895                 }
 896
 897                 /* only accept in-sequence mapping. Old values are spurious
 898                  * retransmission
 899                  */
 900                 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
 901         }
 902         return true;
 903
 904 fatal:
 905         /* fatal protocol error, close the socket */
 906         /* This barrier is coupled with smp_rmb() in tcp_poll() */
 907         smp_wmb();
 908         ssk->sk_error_report(ssk);
 909         tcp_set_state(ssk, TCP_CLOSE);
 910         tcp_send_active_reset(ssk, GFP_ATOMIC);
 911         subflow->data_avail = 0;
 912         return false;
 913 }
 914
 915 bool mptcp_subflow_data_available(struct sock *sk)
 916 {
 917         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 918
 919         /* check if current mapping is still valid */
 920         if (subflow->map_valid &&
 921             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
 922                 subflow->map_valid = 0;
 923                 subflow->data_avail = 0;
 924
 925                 pr_debug("Done with mapping: seq=%u data_len=%u",
 926                          subflow->map_subflow_seq,
 927                          subflow->map_data_len);
 928         }
 929
 930         return subflow_check_data_avail(sk);
 931 }
 932
 933 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
 934  * not the ssk one.
 935  *
 936  * In mptcp, rwin is about the mptcp-level connection data.
 937  *
 938  * Data that is still on the ssk rx queue can thus be ignored,
 939  * as far as mptcp peer is concerened that data is still inflight.
 940  * DSS ACK is updated when skb is moved to the mptcp rx queue.
 941  */
 942 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
 943 {
 944         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 945         const struct sock *sk = subflow->conn;
 946
 947         *space = tcp_space(sk);
 948         *full_space = tcp_full_space(sk);
 949 }
 950
 951 static void subflow_data_ready(struct sock *sk)
 952 {
 953         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 954         u16 state = 1 << inet_sk_state_load(sk);
 955         struct sock *parent = subflow->conn;
 956         struct mptcp_sock *msk;
 957
 958         msk = mptcp_sk(parent);
 959         if (state & TCPF_LISTEN) {
 960                 set_bit(MPTCP_DATA_READY, &msk->flags);
 961                 parent->sk_data_ready(parent);
 962                 return;
 963         }
 964
 965         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
 966                      !subflow->mp_join && !(state & TCPF_CLOSE));
 967
 968         if (mptcp_subflow_data_available(sk))
 969                 mptcp_data_ready(parent, sk);
 970 }
 971
 972 static void subflow_write_space(struct sock *sk)
 973 {
 974         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 975         struct sock *parent = subflow->conn;
 976
 977         if (!sk_stream_is_writeable(sk))
 978                 return;
 979
 980         if (sk_stream_is_writeable(parent)) {
 981                 set_bit(MPTCP_SEND_SPACE, &mptcp_sk(parent)->flags);
 982                 smp_mb__after_atomic();
 983                 /* set SEND_SPACE before sk_stream_write_space clears NOSPACE */
 984                 sk_stream_write_space(parent);
 985         }
 986 }
 987
 988 static struct inet_connection_sock_af_ops *
 989 subflow_default_af_ops(struct sock *sk)
 990 {
 991 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 992         if (sk->sk_family == AF_INET6)
 993                 return &subflow_v6_specific;
 994 #endif
 995         return &subflow_specific;
 996 }
 997
 998 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 999 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1000 {
1001         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1002         struct inet_connection_sock *icsk = inet_csk(sk);
1003         struct inet_connection_sock_af_ops *target;
1004
1005         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1006
1007         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1008                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1009
1010         if (likely(icsk->icsk_af_ops == target))
1011                 return;
1012
1013         subflow->icsk_af_ops = icsk->icsk_af_ops;
1014         icsk->icsk_af_ops = target;
1015 }
1016 #endif
1017
1018 static void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1019                                 struct sockaddr_storage *addr)
1020 {
1021         memset(addr, 0, sizeof(*addr));
1022         addr->ss_family = info->family;
1023         if (addr->ss_family == AF_INET) {
1024                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1025
1026                 in_addr->sin_addr = info->addr;
1027                 in_addr->sin_port = info->port;
1028         }
1029 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1030         else if (addr->ss_family == AF_INET6) {
1031                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1032
1033                 in6_addr->sin6_addr = info->addr6;
1034                 in6_addr->sin6_port = info->port;
1035         }
1036 #endif
1037 }
1038
1039 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1040                             const struct mptcp_addr_info *remote)
1041 {
1042         struct mptcp_sock *msk = mptcp_sk(sk);
1043         struct mptcp_subflow_context *subflow;
1044         struct sockaddr_storage addr;
1045         int local_id = loc->id;
1046         struct socket *sf;
1047         struct sock *ssk;
1048         u32 remote_token;
1049         int addrlen;
1050         int err;
1051
1052         if (!mptcp_is_fully_established(sk))
1053                 return -ENOTCONN;
1054
1055         err = mptcp_subflow_create_socket(sk, &sf);
1056         if (err)
1057                 return err;
1058
1059         ssk = sf->sk;
1060         subflow = mptcp_subflow_ctx(ssk);
1061         do {
1062                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1063         } while (!subflow->local_nonce);
1064
1065         if (!local_id) {
1066                 err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
1067                 if (err < 0)
1068                         goto failed;
1069
1070                 local_id = err;
1071         }
1072
1073         subflow->remote_key = msk->remote_key;
1074         subflow->local_key = msk->local_key;
1075         subflow->token = msk->token;
1076         mptcp_info2sockaddr(loc, &addr);
1077
1078         addrlen = sizeof(struct sockaddr_in);
1079 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1080         if (loc->family == AF_INET6)
1081                 addrlen = sizeof(struct sockaddr_in6);
1082 #endif
1083         ssk->sk_bound_dev_if = loc->ifindex;
1084         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1085         if (err)
1086                 goto failed;
1087
1088         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1089         pr_debug("msk=%p remote_token=%u local_id=%d", msk, remote_token,
1090                  local_id);
1091         subflow->remote_token = remote_token;
1092         subflow->local_id = local_id;
1093         subflow->request_join = 1;
1094         subflow->request_bkup = !!(loc->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1095         mptcp_info2sockaddr(remote, &addr);
1096
1097         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1098         if (err && err != -EINPROGRESS)
1099                 goto failed;
1100
1101         spin_lock_bh(&msk->join_list_lock);
1102         list_add_tail(&subflow->node, &msk->join_list);
1103         spin_unlock_bh(&msk->join_list_lock);
1104
1105         return err;
1106
1107 failed:
1108         sock_release(sf);
1109         return err;
1110 }
1111
1112 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1113 {
1114         struct mptcp_subflow_context *subflow;
1115         struct net *net = sock_net(sk);
1116         struct socket *sf;
1117         int err;
1118
1119         /* un-accepted server sockets can reach here - on bad configuration
1120          * bail early to avoid greater trouble later
1121          */
1122         if (unlikely(!sk->sk_socket))
1123                 return -EINVAL;
1124
1125         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1126                                &sf);
1127         if (err)
1128                 return err;
1129
1130         lock_sock(sf->sk);
1131
1132         /* kernel sockets do not by default acquire net ref, but TCP timer
1133          * needs it.
1134          */
1135         sf->sk->sk_net_refcnt = 1;
1136         get_net(net);
1137 #ifdef CONFIG_PROC_FS
1138         this_cpu_add(*net->core.sock_inuse, 1);
1139 #endif
1140         err = tcp_set_ulp(sf->sk, "mptcp");
1141         release_sock(sf->sk);
1142
1143         if (err) {
1144                 sock_release(sf);
1145                 return err;
1146         }
1147
1148         /* the newly created socket really belongs to the owning MPTCP master
1149          * socket, even if for additional subflows the allocation is performed
1150          * by a kernel workqueue. Adjust inode references, so that the
1151          * procfs/diag interaces really show this one belonging to the correct
1152          * user.
1153          */
1154         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1155         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1156         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1157
1158         subflow = mptcp_subflow_ctx(sf->sk);
1159         pr_debug("subflow=%p", subflow);
1160
1161         *new_sock = sf;
1162         sock_hold(sk);
1163         subflow->conn = sk;
1164
1165         return 0;
1166 }
1167
1168 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1169                                                         gfp_t priority)
1170 {
1171         struct inet_connection_sock *icsk = inet_csk(sk);
1172         struct mptcp_subflow_context *ctx;
1173
1174         ctx = kzalloc(sizeof(*ctx), priority);
1175         if (!ctx)
1176                 return NULL;
1177
1178         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1179         INIT_LIST_HEAD(&ctx->node);
1180
1181         pr_debug("subflow=%p", ctx);
1182
1183         ctx->tcp_sock = sk;
1184
1185         return ctx;
1186 }
1187
1188 static void __subflow_state_change(struct sock *sk)
1189 {
1190         struct socket_wq *wq;
1191
1192         rcu_read_lock();
1193         wq = rcu_dereference(sk->sk_wq);
1194         if (skwq_has_sleeper(wq))
1195                 wake_up_interruptible_all(&wq->wait);
1196         rcu_read_unlock();
1197 }
1198
1199 static bool subflow_is_done(const struct sock *sk)
1200 {
1201         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1202 }
1203
1204 static void subflow_state_change(struct sock *sk)
1205 {
1206         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1207         struct sock *parent = subflow->conn;
1208
1209         __subflow_state_change(sk);
1210
1211         if (subflow_simultaneous_connect(sk)) {
1212                 mptcp_do_fallback(sk);
1213                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1214                 pr_fallback(mptcp_sk(parent));
1215                 subflow->conn_finished = 1;
1216                 if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
1217                         inet_sk_state_store(parent, TCP_ESTABLISHED);
1218                         parent->sk_state_change(parent);
1219                 }
1220         }
1221
1222         /* as recvmsg() does not acquire the subflow socket for ssk selection
1223          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1224          * the data available machinery here.
1225          */
1226         if (mptcp_subflow_data_available(sk))
1227                 mptcp_data_ready(parent, sk);
1228
1229         if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1230             !(parent->sk_shutdown & RCV_SHUTDOWN) &&
1231             !subflow->rx_eof && subflow_is_done(sk)) {
1232                 subflow->rx_eof = 1;
1233                 mptcp_subflow_eof(parent);
1234         }
1235 }
1236
1237 static int subflow_ulp_init(struct sock *sk)
1238 {
1239         struct inet_connection_sock *icsk = inet_csk(sk);
1240         struct mptcp_subflow_context *ctx;
1241         struct tcp_sock *tp = tcp_sk(sk);
1242         int err = 0;
1243
1244         /* disallow attaching ULP to a socket unless it has been
1245          * created with sock_create_kern()
1246          */
1247         if (!sk->sk_kern_sock) {
1248                 err = -EOPNOTSUPP;
1249                 goto out;
1250         }
1251
1252         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1253         if (!ctx) {
1254                 err = -ENOMEM;
1255                 goto out;
1256         }
1257
1258         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1259
1260         tp->is_mptcp = 1;
1261         ctx->icsk_af_ops = icsk->icsk_af_ops;
1262         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1263         ctx->tcp_data_ready = sk->sk_data_ready;
1264         ctx->tcp_state_change = sk->sk_state_change;
1265         ctx->tcp_write_space = sk->sk_write_space;
1266         sk->sk_data_ready = subflow_data_ready;
1267         sk->sk_write_space = subflow_write_space;
1268         sk->sk_state_change = subflow_state_change;
1269 out:
1270         return err;
1271 }
1272
1273 static void subflow_ulp_release(struct sock *sk)
1274 {
1275         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(sk);
1276
1277         if (!ctx)
1278                 return;
1279
1280         if (ctx->conn)
1281                 sock_put(ctx->conn);
1282
1283         kfree_rcu(ctx, rcu);
1284 }
1285
1286 static void subflow_ulp_clone(const struct request_sock *req,
1287                               struct sock *newsk,
1288                               const gfp_t priority)
1289 {
1290         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1291         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1292         struct mptcp_subflow_context *new_ctx;
1293
1294         if (!tcp_rsk(req)->is_mptcp ||
1295             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1296                 subflow_ulp_fallback(newsk, old_ctx);
1297                 return;
1298         }
1299
1300         new_ctx = subflow_create_ctx(newsk, priority);
1301         if (!new_ctx) {
1302                 subflow_ulp_fallback(newsk, old_ctx);
1303                 return;
1304         }
1305
1306         new_ctx->conn_finished = 1;
1307         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1308         new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1309         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1310         new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1311         new_ctx->rel_write_seq = 1;
1312         new_ctx->tcp_sock = newsk;
1313
1314         if (subflow_req->mp_capable) {
1315                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1316                  * is fully established only after we receive the remote key
1317                  */
1318                 new_ctx->mp_capable = 1;
1319                 new_ctx->local_key = subflow_req->local_key;
1320                 new_ctx->token = subflow_req->token;
1321                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1322                 new_ctx->idsn = subflow_req->idsn;
1323         } else if (subflow_req->mp_join) {
1324                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1325                 new_ctx->mp_join = 1;
1326                 new_ctx->fully_established = 1;
1327                 new_ctx->backup = subflow_req->backup;
1328                 new_ctx->local_id = subflow_req->local_id;
1329                 new_ctx->token = subflow_req->token;
1330                 new_ctx->thmac = subflow_req->thmac;
1331         }
1332 }
1333
1334 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1335         .name           = "mptcp",
1336         .owner          = THIS_MODULE,
1337         .init           = subflow_ulp_init,
1338         .release        = subflow_ulp_release,
1339         .clone          = subflow_ulp_clone,
1340 };
1341
1342 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1343 {
1344         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1345         subflow_ops->slab_name = "request_sock_subflow";
1346
1347         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1348                                               subflow_ops->obj_size, 0,
1349                                               SLAB_ACCOUNT |
1350                                               SLAB_TYPESAFE_BY_RCU,
1351                                               NULL);
1352         if (!subflow_ops->slab)
1353                 return -ENOMEM;
1354
1355         subflow_ops->destructor = subflow_req_destructor;
1356
1357         return 0;
1358 }
1359
1360 void __init mptcp_subflow_init(void)
1361 {
1362         mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1363         if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1364                 panic("MPTCP: failed to init subflow request sock ops\n");
1365
1366         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1367         subflow_request_sock_ipv4_ops.init_req = subflow_v4_init_req;
1368
1369         subflow_specific = ipv4_specific;
1370         subflow_specific.conn_request = subflow_v4_conn_request;
1371         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1372         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1373
1374 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1375         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1376         subflow_request_sock_ipv6_ops.init_req = subflow_v6_init_req;
1377
1378         subflow_v6_specific = ipv6_specific;
1379         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1380         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1381         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1382
1383         subflow_v6m_specific = subflow_v6_specific;
1384         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1385         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1386         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1387         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1388         subflow_v6m_specific.net_frag_header_len = 0;
1389 #endif
1390
1391         mptcp_diag_subflow_init(&subflow_ulp_ops);
1392
1393         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1394                 panic("MPTCP: failed to register subflows to ULP\n");
1395 }