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 void mptcp_subflow_reset(struct sock *ssk)
 275 {
 276         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 277         struct sock *sk = subflow->conn;
 278
 279         tcp_set_state(ssk, TCP_CLOSE);
 280         tcp_send_active_reset(ssk, GFP_ATOMIC);
 281         tcp_done(ssk);
 282         if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags) &&
 283             schedule_work(&mptcp_sk(sk)->work))
 284                 sock_hold(sk);
 285 }
 286
 287 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 288 {
 289         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 290         struct mptcp_options_received mp_opt;
 291         struct sock *parent = subflow->conn;
 292
 293         subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 294
 295         if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
 296                 inet_sk_state_store(parent, TCP_ESTABLISHED);
 297                 parent->sk_state_change(parent);
 298         }
 299
 300         /* be sure no special action on any packet other than syn-ack */
 301         if (subflow->conn_finished)
 302                 return;
 303
 304         subflow->rel_write_seq = 1;
 305         subflow->conn_finished = 1;
 306         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 307         pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
 308
 309         mptcp_get_options(skb, &mp_opt);
 310         if (subflow->request_mptcp) {
 311                 if (!mp_opt.mp_capable) {
 312                         MPTCP_INC_STATS(sock_net(sk),
 313                                         MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
 314                         mptcp_do_fallback(sk);
 315                         pr_fallback(mptcp_sk(subflow->conn));
 316                         goto fallback;
 317                 }
 318
 319                 subflow->mp_capable = 1;
 320                 subflow->can_ack = 1;
 321                 subflow->remote_key = mp_opt.sndr_key;
 322                 pr_debug("subflow=%p, remote_key=%llu", subflow,
 323                          subflow->remote_key);
 324                 mptcp_finish_connect(sk);
 325         } else if (subflow->request_join) {
 326                 u8 hmac[SHA256_DIGEST_SIZE];
 327
 328                 if (!mp_opt.mp_join)
 329                         goto do_reset;
 330
 331                 subflow->thmac = mp_opt.thmac;
 332                 subflow->remote_nonce = mp_opt.nonce;
 333                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
 334                          subflow->thmac, subflow->remote_nonce);
 335
 336                 if (!subflow_thmac_valid(subflow)) {
 337                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 338                         goto do_reset;
 339                 }
 340
 341                 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 342                                       subflow->local_nonce,
 343                                       subflow->remote_nonce,
 344                                       hmac);
 345                 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
 346
 347                 if (!mptcp_finish_join(sk))
 348                         goto do_reset;
 349
 350                 subflow->mp_join = 1;
 351                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 352         } else if (mptcp_check_fallback(sk)) {
 353 fallback:
 354                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
 355         }
 356         return;
 357
 358 do_reset:
 359         mptcp_subflow_reset(sk);
 360 }
 361
 362 struct request_sock_ops mptcp_subflow_request_sock_ops;
 363 EXPORT_SYMBOL_GPL(mptcp_subflow_request_sock_ops);
 364 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
 365
 366 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 367 {
 368         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 369
 370         pr_debug("subflow=%p", subflow);
 371
 372         /* Never answer to SYNs sent to broadcast or multicast */
 373         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 374                 goto drop;
 375
 376         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 377                                 &subflow_request_sock_ipv4_ops,
 378                                 sk, skb);
 379 drop:
 380         tcp_listendrop(sk);
 381         return 0;
 382 }
 383
 384 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 385 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
 386 static struct inet_connection_sock_af_ops subflow_v6_specific;
 387 static struct inet_connection_sock_af_ops subflow_v6m_specific;
 388
 389 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 390 {
 391         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 392
 393         pr_debug("subflow=%p", subflow);
 394
 395         if (skb->protocol == htons(ETH_P_IP))
 396                 return subflow_v4_conn_request(sk, skb);
 397
 398         if (!ipv6_unicast_destination(skb))
 399                 goto drop;
 400
 401         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 402                                 &subflow_request_sock_ipv6_ops, sk, skb);
 403
 404 drop:
 405         tcp_listendrop(sk);
 406         return 0; /* don't send reset */
 407 }
 408 #endif
 409
 410 /* validate hmac received in third ACK */
 411 static bool subflow_hmac_valid(const struct request_sock *req,
 412                                const struct mptcp_options_received *mp_opt)
 413 {
 414         const struct mptcp_subflow_request_sock *subflow_req;
 415         u8 hmac[SHA256_DIGEST_SIZE];
 416         struct mptcp_sock *msk;
 417
 418         subflow_req = mptcp_subflow_rsk(req);
 419         msk = subflow_req->msk;
 420         if (!msk)
 421                 return false;
 422
 423         subflow_generate_hmac(msk->remote_key, msk->local_key,
 424                               subflow_req->remote_nonce,
 425                               subflow_req->local_nonce, hmac);
 426
 427         return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
 428 }
 429
 430 static void mptcp_sock_destruct(struct sock *sk)
 431 {
 432         /* if new mptcp socket isn't accepted, it is free'd
 433          * from the tcp listener sockets request queue, linked
 434          * from req->sk.  The tcp socket is released.
 435          * This calls the ULP release function which will
 436          * also remove the mptcp socket, via
 437          * sock_put(ctx->conn).
 438          *
 439          * Problem is that the mptcp socket will be in
 440          * ESTABLISHED state and will not have the SOCK_DEAD flag.
 441          * Both result in warnings from inet_sock_destruct.
 442          */
 443
 444         if (sk->sk_state == TCP_ESTABLISHED) {
 445                 sk->sk_state = TCP_CLOSE;
 446                 WARN_ON_ONCE(sk->sk_socket);
 447                 sock_orphan(sk);
 448         }
 449
 450         mptcp_destroy_common(mptcp_sk(sk));
 451         inet_sock_destruct(sk);
 452 }
 453
 454 static void mptcp_force_close(struct sock *sk)
 455 {
 456         inet_sk_state_store(sk, TCP_CLOSE);
 457         sk_common_release(sk);
 458 }
 459
 460 static void subflow_ulp_fallback(struct sock *sk,
 461                                  struct mptcp_subflow_context *old_ctx)
 462 {
 463         struct inet_connection_sock *icsk = inet_csk(sk);
 464
 465         mptcp_subflow_tcp_fallback(sk, old_ctx);
 466         icsk->icsk_ulp_ops = NULL;
 467         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 468         tcp_sk(sk)->is_mptcp = 0;
 469 }
 470
 471 static void subflow_drop_ctx(struct sock *ssk)
 472 {
 473         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 474
 475         if (!ctx)
 476                 return;
 477
 478         subflow_ulp_fallback(ssk, ctx);
 479         if (ctx->conn)
 480                 sock_put(ctx->conn);
 481
 482         kfree_rcu(ctx, rcu);
 483 }
 484
 485 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 486                                      struct mptcp_options_received *mp_opt)
 487 {
 488         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 489
 490         subflow->remote_key = mp_opt->sndr_key;
 491         subflow->fully_established = 1;
 492         subflow->can_ack = 1;
 493         WRITE_ONCE(msk->fully_established, true);
 494 }
 495
 496 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 497                                           struct sk_buff *skb,
 498                                           struct request_sock *req,
 499                                           struct dst_entry *dst,
 500                                           struct request_sock *req_unhash,
 501                                           bool *own_req)
 502 {
 503         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 504         struct mptcp_subflow_request_sock *subflow_req;
 505         struct mptcp_options_received mp_opt;
 506         bool fallback, fallback_is_fatal;
 507         struct sock *new_msk = NULL;
 508         struct sock *child;
 509
 510         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 511
 512         /* After child creation we must look for 'mp_capable' even when options
 513          * are not parsed
 514          */
 515         mp_opt.mp_capable = 0;
 516
 517         /* hopefully temporary handling for MP_JOIN+syncookie */
 518         subflow_req = mptcp_subflow_rsk(req);
 519         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 520         fallback = !tcp_rsk(req)->is_mptcp;
 521         if (fallback)
 522                 goto create_child;
 523
 524         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 525         if (subflow_req->mp_capable) {
 526                 if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
 527                         /* here we can receive and accept an in-window,
 528                          * out-of-order pkt, which will not carry the MP_CAPABLE
 529                          * opt even on mptcp enabled paths
 530                          */
 531                         goto create_msk;
 532                 }
 533
 534                 mptcp_get_options(skb, &mp_opt);
 535                 if (!mp_opt.mp_capable) {
 536                         fallback = true;
 537                         goto create_child;
 538                 }
 539
 540 create_msk:
 541                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 542                 if (!new_msk)
 543                         fallback = true;
 544         } else if (subflow_req->mp_join) {
 545                 mptcp_get_options(skb, &mp_opt);
 546                 if (!mp_opt.mp_join || !subflow_hmac_valid(req, &mp_opt) ||
 547                     !mptcp_can_accept_new_subflow(subflow_req->msk)) {
 548                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 549                         fallback = true;
 550                 }
 551         }
 552
 553 create_child:
 554         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 555                                                      req_unhash, own_req);
 556
 557         if (child && *own_req) {
 558                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 559
 560                 tcp_rsk(req)->drop_req = false;
 561
 562                 /* we need to fallback on ctx allocation failure and on pre-reqs
 563                  * checking above. In the latter scenario we additionally need
 564                  * to reset the context to non MPTCP status.
 565                  */
 566                 if (!ctx || fallback) {
 567                         if (fallback_is_fatal)
 568                                 goto dispose_child;
 569
 570                         subflow_drop_ctx(child);
 571                         goto out;
 572                 }
 573
 574                 if (ctx->mp_capable) {
 575                         /* this can't race with mptcp_close(), as the msk is
 576                          * not yet exposted to user-space
 577                          */
 578                         inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
 579
 580                         /* link the newly created socket to the msk */
 581                         mptcp_add_pending_subflow(mptcp_sk(new_msk), ctx);
 582                         WRITE_ONCE(mptcp_sk(new_msk)->first, child);
 583
 584                         /* new mpc subflow takes ownership of the newly
 585                          * created mptcp socket
 586                          */
 587                         new_msk->sk_destruct = mptcp_sock_destruct;
 588                         mptcp_pm_new_connection(mptcp_sk(new_msk), 1);
 589                         mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
 590                         ctx->conn = new_msk;
 591                         new_msk = NULL;
 592
 593                         /* with OoO packets we can reach here without ingress
 594                          * mpc option
 595                          */
 596                         if (mp_opt.mp_capable)
 597                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 598                 } else if (ctx->mp_join) {
 599                         struct mptcp_sock *owner;
 600
 601                         owner = subflow_req->msk;
 602                         if (!owner)
 603                                 goto dispose_child;
 604
 605                         /* move the msk reference ownership to the subflow */
 606                         subflow_req->msk = NULL;
 607                         ctx->conn = (struct sock *)owner;
 608                         if (!mptcp_finish_join(child))
 609                                 goto dispose_child;
 610
 611                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 612                         tcp_rsk(req)->drop_req = true;
 613                 }
 614         }
 615
 616 out:
 617         /* dispose of the left over mptcp master, if any */
 618         if (unlikely(new_msk))
 619                 mptcp_force_close(new_msk);
 620
 621         /* check for expected invariant - should never trigger, just help
 622          * catching eariler subtle bugs
 623          */
 624         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 625                      (!mptcp_subflow_ctx(child) ||
 626                       !mptcp_subflow_ctx(child)->conn));
 627         return child;
 628
 629 dispose_child:
 630         subflow_drop_ctx(child);
 631         tcp_rsk(req)->drop_req = true;
 632         inet_csk_prepare_for_destroy_sock(child);
 633         tcp_done(child);
 634         req->rsk_ops->send_reset(sk, skb);
 635
 636         /* The last child reference will be released by the caller */
 637         return child;
 638 }
 639
 640 static struct inet_connection_sock_af_ops subflow_specific;
 641
 642 enum mapping_status {
 643         MAPPING_OK,
 644         MAPPING_INVALID,
 645         MAPPING_EMPTY,
 646         MAPPING_DATA_FIN,
 647         MAPPING_DUMMY
 648 };
 649
 650 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
 651 {
 652         if ((u32)seq == (u32)old_seq)
 653                 return old_seq;
 654
 655         /* Assume map covers data not mapped yet. */
 656         return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
 657 }
 658
 659 static void warn_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 660 {
 661         WARN_ONCE(1, "Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 662                   ssn, subflow->map_subflow_seq, subflow->map_data_len);
 663 }
 664
 665 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 666 {
 667         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 668         unsigned int skb_consumed;
 669
 670         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 671         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 672                 return true;
 673
 674         return skb->len - skb_consumed <= subflow->map_data_len -
 675                                           mptcp_subflow_get_map_offset(subflow);
 676 }
 677
 678 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 679 {
 680         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 681         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 682
 683         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 684                 /* Mapping covers data later in the subflow stream,
 685                  * currently unsupported.
 686                  */
 687                 warn_bad_map(subflow, ssn);
 688                 return false;
 689         }
 690         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 691                                   subflow->map_data_len))) {
 692                 /* Mapping does covers past subflow data, invalid */
 693                 warn_bad_map(subflow, ssn + skb->len);
 694                 return false;
 695         }
 696         return true;
 697 }
 698
 699 static enum mapping_status get_mapping_status(struct sock *ssk,
 700                                               struct mptcp_sock *msk)
 701 {
 702         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 703         struct mptcp_ext *mpext;
 704         struct sk_buff *skb;
 705         u16 data_len;
 706         u64 map_seq;
 707
 708         skb = skb_peek(&ssk->sk_receive_queue);
 709         if (!skb)
 710                 return MAPPING_EMPTY;
 711
 712         if (mptcp_check_fallback(ssk))
 713                 return MAPPING_DUMMY;
 714
 715         mpext = mptcp_get_ext(skb);
 716         if (!mpext || !mpext->use_map) {
 717                 if (!subflow->map_valid && !skb->len) {
 718                         /* the TCP stack deliver 0 len FIN pkt to the receive
 719                          * queue, that is the only 0len pkts ever expected here,
 720                          * and we can admit no mapping only for 0 len pkts
 721                          */
 722                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 723                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 724                                           TCP_SKB_CB(skb)->seq,
 725                                           TCP_SKB_CB(skb)->end_seq,
 726                                           TCP_SKB_CB(skb)->tcp_flags);
 727                         sk_eat_skb(ssk, skb);
 728                         return MAPPING_EMPTY;
 729                 }
 730
 731                 if (!subflow->map_valid)
 732                         return MAPPING_INVALID;
 733
 734                 goto validate_seq;
 735         }
 736
 737         pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
 738                  mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
 739                  mpext->data_len, mpext->data_fin);
 740
 741         data_len = mpext->data_len;
 742         if (data_len == 0) {
 743                 pr_err("Infinite mapping not handled");
 744                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 745                 return MAPPING_INVALID;
 746         }
 747
 748         if (mpext->data_fin == 1) {
 749                 if (data_len == 1) {
 750                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
 751                                                                  mpext->dsn64);
 752                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
 753                         if (subflow->map_valid) {
 754                                 /* A DATA_FIN might arrive in a DSS
 755                                  * option before the previous mapping
 756                                  * has been fully consumed. Continue
 757                                  * handling the existing mapping.
 758                                  */
 759                                 skb_ext_del(skb, SKB_EXT_MPTCP);
 760                                 return MAPPING_OK;
 761                         } else {
 762                                 if (updated && schedule_work(&msk->work))
 763                                         sock_hold((struct sock *)msk);
 764
 765                                 return MAPPING_DATA_FIN;
 766                         }
 767                 } else {
 768                         u64 data_fin_seq = mpext->data_seq + data_len - 1;
 769
 770                         /* If mpext->data_seq is a 32-bit value, data_fin_seq
 771                          * must also be limited to 32 bits.
 772                          */
 773                         if (!mpext->dsn64)
 774                                 data_fin_seq &= GENMASK_ULL(31, 0);
 775
 776                         mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
 777                         pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
 778                                  data_fin_seq, mpext->dsn64);
 779                 }
 780
 781                 /* Adjust for DATA_FIN using 1 byte of sequence space */
 782                 data_len--;
 783         }
 784
 785         if (!mpext->dsn64) {
 786                 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
 787                                      mpext->data_seq);
 788                 pr_debug("expanded seq=%llu", subflow->map_seq);
 789         } else {
 790                 map_seq = mpext->data_seq;
 791         }
 792         WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
 793
 794         if (subflow->map_valid) {
 795                 /* Allow replacing only with an identical map */
 796                 if (subflow->map_seq == map_seq &&
 797                     subflow->map_subflow_seq == mpext->subflow_seq &&
 798                     subflow->map_data_len == data_len) {
 799                         skb_ext_del(skb, SKB_EXT_MPTCP);
 800                         return MAPPING_OK;
 801                 }
 802
 803                 /* If this skb data are fully covered by the current mapping,
 804                  * the new map would need caching, which is not supported
 805                  */
 806                 if (skb_is_fully_mapped(ssk, skb)) {
 807                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
 808                         return MAPPING_INVALID;
 809                 }
 810
 811                 /* will validate the next map after consuming the current one */
 812                 return MAPPING_OK;
 813         }
 814
 815         subflow->map_seq = map_seq;
 816         subflow->map_subflow_seq = mpext->subflow_seq;
 817         subflow->map_data_len = data_len;
 818         subflow->map_valid = 1;
 819         subflow->mpc_map = mpext->mpc_map;
 820         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
 821                  subflow->map_seq, subflow->map_subflow_seq,
 822                  subflow->map_data_len);
 823
 824 validate_seq:
 825         /* we revalidate valid mapping on new skb, because we must ensure
 826          * the current skb is completely covered by the available mapping
 827          */
 828         if (!validate_mapping(ssk, skb))
 829                 return MAPPING_INVALID;
 830
 831         skb_ext_del(skb, SKB_EXT_MPTCP);
 832         return MAPPING_OK;
 833 }
 834
 835 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
 836                                        u64 limit)
 837 {
 838         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 839         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
 840         u32 incr;
 841
 842         incr = limit >= skb->len ? skb->len + fin : limit;
 843
 844         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
 845                  subflow->map_subflow_seq);
 846         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
 847         tcp_sk(ssk)->copied_seq += incr;
 848         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
 849                 sk_eat_skb(ssk, skb);
 850         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
 851                 subflow->map_valid = 0;
 852 }
 853
 854 static bool subflow_check_data_avail(struct sock *ssk)
 855 {
 856         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 857         enum mapping_status status;
 858         struct mptcp_sock *msk;
 859         struct sk_buff *skb;
 860
 861         pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
 862                  subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
 863         if (!skb_peek(&ssk->sk_receive_queue))
 864                 subflow->data_avail = 0;
 865         if (subflow->data_avail)
 866                 return true;
 867
 868         msk = mptcp_sk(subflow->conn);
 869         for (;;) {
 870                 u64 ack_seq;
 871                 u64 old_ack;
 872
 873                 status = get_mapping_status(ssk, msk);
 874                 pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
 875                 if (status == MAPPING_INVALID) {
 876                         ssk->sk_err = EBADMSG;
 877                         goto fatal;
 878                 }
 879                 if (status == MAPPING_DUMMY) {
 880                         __mptcp_do_fallback(msk);
 881                         skb = skb_peek(&ssk->sk_receive_queue);
 882                         subflow->map_valid = 1;
 883                         subflow->map_seq = READ_ONCE(msk->ack_seq);
 884                         subflow->map_data_len = skb->len;
 885                         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq -
 886                                                    subflow->ssn_offset;
 887                         subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
 888                         return true;
 889                 }
 890
 891                 if (status != MAPPING_OK)
 892                         return false;
 893
 894                 skb = skb_peek(&ssk->sk_receive_queue);
 895                 if (WARN_ON_ONCE(!skb))
 896                         return false;
 897
 898                 /* if msk lacks the remote key, this subflow must provide an
 899                  * MP_CAPABLE-based mapping
 900                  */
 901                 if (unlikely(!READ_ONCE(msk->can_ack))) {
 902                         if (!subflow->mpc_map) {
 903                                 ssk->sk_err = EBADMSG;
 904                                 goto fatal;
 905                         }
 906                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
 907                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
 908                         WRITE_ONCE(msk->can_ack, true);
 909                 }
 910
 911                 old_ack = READ_ONCE(msk->ack_seq);
 912                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
 913                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
 914                          ack_seq);
 915                 if (ack_seq == old_ack) {
 916                         subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
 917                         break;
 918                 } else if (after64(ack_seq, old_ack)) {
 919                         subflow->data_avail = MPTCP_SUBFLOW_OOO_DATA;
 920                         break;
 921                 }
 922
 923                 /* only accept in-sequence mapping. Old values are spurious
 924                  * retransmission
 925                  */
 926                 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
 927         }
 928         return true;
 929
 930 fatal:
 931         /* fatal protocol error, close the socket */
 932         /* This barrier is coupled with smp_rmb() in tcp_poll() */
 933         smp_wmb();
 934         ssk->sk_error_report(ssk);
 935         tcp_set_state(ssk, TCP_CLOSE);
 936         tcp_send_active_reset(ssk, GFP_ATOMIC);
 937         subflow->data_avail = 0;
 938         return false;
 939 }
 940
 941 bool mptcp_subflow_data_available(struct sock *sk)
 942 {
 943         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 944
 945         /* check if current mapping is still valid */
 946         if (subflow->map_valid &&
 947             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
 948                 subflow->map_valid = 0;
 949                 subflow->data_avail = 0;
 950
 951                 pr_debug("Done with mapping: seq=%u data_len=%u",
 952                          subflow->map_subflow_seq,
 953                          subflow->map_data_len);
 954         }
 955
 956         return subflow_check_data_avail(sk);
 957 }
 958
 959 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
 960  * not the ssk one.
 961  *
 962  * In mptcp, rwin is about the mptcp-level connection data.
 963  *
 964  * Data that is still on the ssk rx queue can thus be ignored,
 965  * as far as mptcp peer is concerened that data is still inflight.
 966  * DSS ACK is updated when skb is moved to the mptcp rx queue.
 967  */
 968 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
 969 {
 970         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 971         const struct sock *sk = subflow->conn;
 972
 973         *space = __mptcp_space(sk);
 974         *full_space = tcp_full_space(sk);
 975 }
 976
 977 static void subflow_data_ready(struct sock *sk)
 978 {
 979         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 980         u16 state = 1 << inet_sk_state_load(sk);
 981         struct sock *parent = subflow->conn;
 982         struct mptcp_sock *msk;
 983
 984         msk = mptcp_sk(parent);
 985         if (state & TCPF_LISTEN) {
 986                 set_bit(MPTCP_DATA_READY, &msk->flags);
 987                 parent->sk_data_ready(parent);
 988                 return;
 989         }
 990
 991         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
 992                      !subflow->mp_join && !(state & TCPF_CLOSE));
 993
 994         if (mptcp_subflow_data_available(sk))
 995                 mptcp_data_ready(parent, sk);
 996 }
 997
 998 static void subflow_write_space(struct sock *ssk)
 999 {
1000         /* we take action in __mptcp_clean_una() */
1001 }
1002
1003 static struct inet_connection_sock_af_ops *
1004 subflow_default_af_ops(struct sock *sk)
1005 {
1006 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1007         if (sk->sk_family == AF_INET6)
1008                 return &subflow_v6_specific;
1009 #endif
1010         return &subflow_specific;
1011 }
1012
1013 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1014 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1015 {
1016         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1017         struct inet_connection_sock *icsk = inet_csk(sk);
1018         struct inet_connection_sock_af_ops *target;
1019
1020         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1021
1022         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1023                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1024
1025         if (likely(icsk->icsk_af_ops == target))
1026                 return;
1027
1028         subflow->icsk_af_ops = icsk->icsk_af_ops;
1029         icsk->icsk_af_ops = target;
1030 }
1031 #endif
1032
1033 static void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1034                                 struct sockaddr_storage *addr)
1035 {
1036         memset(addr, 0, sizeof(*addr));
1037         addr->ss_family = info->family;
1038         if (addr->ss_family == AF_INET) {
1039                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1040
1041                 in_addr->sin_addr = info->addr;
1042                 in_addr->sin_port = info->port;
1043         }
1044 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1045         else if (addr->ss_family == AF_INET6) {
1046                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1047
1048                 in6_addr->sin6_addr = info->addr6;
1049                 in6_addr->sin6_port = info->port;
1050         }
1051 #endif
1052 }
1053
1054 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1055                             const struct mptcp_addr_info *remote)
1056 {
1057         struct mptcp_sock *msk = mptcp_sk(sk);
1058         struct mptcp_subflow_context *subflow;
1059         struct sockaddr_storage addr;
1060         int remote_id = remote->id;
1061         int local_id = loc->id;
1062         struct socket *sf;
1063         struct sock *ssk;
1064         u32 remote_token;
1065         int addrlen;
1066         int err;
1067
1068         if (!mptcp_is_fully_established(sk))
1069                 return -ENOTCONN;
1070
1071         err = mptcp_subflow_create_socket(sk, &sf);
1072         if (err)
1073                 return err;
1074
1075         ssk = sf->sk;
1076         subflow = mptcp_subflow_ctx(ssk);
1077         do {
1078                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1079         } while (!subflow->local_nonce);
1080
1081         if (!local_id) {
1082                 err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
1083                 if (err < 0)
1084                         goto failed;
1085
1086                 local_id = err;
1087         }
1088
1089         subflow->remote_key = msk->remote_key;
1090         subflow->local_key = msk->local_key;
1091         subflow->token = msk->token;
1092         mptcp_info2sockaddr(loc, &addr);
1093
1094         addrlen = sizeof(struct sockaddr_in);
1095 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1096         if (loc->family == AF_INET6)
1097                 addrlen = sizeof(struct sockaddr_in6);
1098 #endif
1099         ssk->sk_bound_dev_if = loc->ifindex;
1100         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1101         if (err)
1102                 goto failed;
1103
1104         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1105         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1106                  remote_token, local_id, remote_id);
1107         subflow->remote_token = remote_token;
1108         subflow->local_id = local_id;
1109         subflow->remote_id = remote_id;
1110         subflow->request_join = 1;
1111         subflow->request_bkup = !!(loc->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1112         mptcp_info2sockaddr(remote, &addr);
1113
1114         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1115         if (err && err != -EINPROGRESS)
1116                 goto failed;
1117
1118         mptcp_add_pending_subflow(msk, subflow);
1119         return err;
1120
1121 failed:
1122         subflow->disposable = 1;
1123         sock_release(sf);
1124         return err;
1125 }
1126
1127 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1128 {
1129         struct mptcp_subflow_context *subflow;
1130         struct net *net = sock_net(sk);
1131         struct socket *sf;
1132         int err;
1133
1134         /* un-accepted server sockets can reach here - on bad configuration
1135          * bail early to avoid greater trouble later
1136          */
1137         if (unlikely(!sk->sk_socket))
1138                 return -EINVAL;
1139
1140         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1141                                &sf);
1142         if (err)
1143                 return err;
1144
1145         lock_sock(sf->sk);
1146
1147         /* kernel sockets do not by default acquire net ref, but TCP timer
1148          * needs it.
1149          */
1150         sf->sk->sk_net_refcnt = 1;
1151         get_net(net);
1152 #ifdef CONFIG_PROC_FS
1153         this_cpu_add(*net->core.sock_inuse, 1);
1154 #endif
1155         err = tcp_set_ulp(sf->sk, "mptcp");
1156         release_sock(sf->sk);
1157
1158         if (err) {
1159                 sock_release(sf);
1160                 return err;
1161         }
1162
1163         /* the newly created socket really belongs to the owning MPTCP master
1164          * socket, even if for additional subflows the allocation is performed
1165          * by a kernel workqueue. Adjust inode references, so that the
1166          * procfs/diag interaces really show this one belonging to the correct
1167          * user.
1168          */
1169         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1170         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1171         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1172
1173         subflow = mptcp_subflow_ctx(sf->sk);
1174         pr_debug("subflow=%p", subflow);
1175
1176         *new_sock = sf;
1177         sock_hold(sk);
1178         subflow->conn = sk;
1179
1180         return 0;
1181 }
1182
1183 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1184                                                         gfp_t priority)
1185 {
1186         struct inet_connection_sock *icsk = inet_csk(sk);
1187         struct mptcp_subflow_context *ctx;
1188
1189         ctx = kzalloc(sizeof(*ctx), priority);
1190         if (!ctx)
1191                 return NULL;
1192
1193         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1194         INIT_LIST_HEAD(&ctx->node);
1195
1196         pr_debug("subflow=%p", ctx);
1197
1198         ctx->tcp_sock = sk;
1199
1200         return ctx;
1201 }
1202
1203 static void __subflow_state_change(struct sock *sk)
1204 {
1205         struct socket_wq *wq;
1206
1207         rcu_read_lock();
1208         wq = rcu_dereference(sk->sk_wq);
1209         if (skwq_has_sleeper(wq))
1210                 wake_up_interruptible_all(&wq->wait);
1211         rcu_read_unlock();
1212 }
1213
1214 static bool subflow_is_done(const struct sock *sk)
1215 {
1216         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1217 }
1218
1219 static void subflow_state_change(struct sock *sk)
1220 {
1221         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1222         struct sock *parent = subflow->conn;
1223
1224         __subflow_state_change(sk);
1225
1226         if (subflow_simultaneous_connect(sk)) {
1227                 mptcp_do_fallback(sk);
1228                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1229                 pr_fallback(mptcp_sk(parent));
1230                 subflow->conn_finished = 1;
1231                 if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
1232                         inet_sk_state_store(parent, TCP_ESTABLISHED);
1233                         parent->sk_state_change(parent);
1234                 }
1235         }
1236
1237         /* as recvmsg() does not acquire the subflow socket for ssk selection
1238          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1239          * the data available machinery here.
1240          */
1241         if (mptcp_subflow_data_available(sk))
1242                 mptcp_data_ready(parent, sk);
1243
1244         if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1245             !subflow->rx_eof && subflow_is_done(sk)) {
1246                 subflow->rx_eof = 1;
1247                 mptcp_subflow_eof(parent);
1248         }
1249 }
1250
1251 static int subflow_ulp_init(struct sock *sk)
1252 {
1253         struct inet_connection_sock *icsk = inet_csk(sk);
1254         struct mptcp_subflow_context *ctx;
1255         struct tcp_sock *tp = tcp_sk(sk);
1256         int err = 0;
1257
1258         /* disallow attaching ULP to a socket unless it has been
1259          * created with sock_create_kern()
1260          */
1261         if (!sk->sk_kern_sock) {
1262                 err = -EOPNOTSUPP;
1263                 goto out;
1264         }
1265
1266         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1267         if (!ctx) {
1268                 err = -ENOMEM;
1269                 goto out;
1270         }
1271
1272         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1273
1274         tp->is_mptcp = 1;
1275         ctx->icsk_af_ops = icsk->icsk_af_ops;
1276         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1277         ctx->tcp_data_ready = sk->sk_data_ready;
1278         ctx->tcp_state_change = sk->sk_state_change;
1279         ctx->tcp_write_space = sk->sk_write_space;
1280         sk->sk_data_ready = subflow_data_ready;
1281         sk->sk_write_space = subflow_write_space;
1282         sk->sk_state_change = subflow_state_change;
1283 out:
1284         return err;
1285 }
1286
1287 static void subflow_ulp_release(struct sock *ssk)
1288 {
1289         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1290         bool release = true;
1291         struct sock *sk;
1292
1293         if (!ctx)
1294                 return;
1295
1296         sk = ctx->conn;
1297         if (sk) {
1298                 /* if the msk has been orphaned, keep the ctx
1299                  * alive, will be freed by mptcp_done()
1300                  */
1301                 release = ctx->disposable;
1302                 sock_put(sk);
1303         }
1304
1305         if (release)
1306                 kfree_rcu(ctx, rcu);
1307 }
1308
1309 static void subflow_ulp_clone(const struct request_sock *req,
1310                               struct sock *newsk,
1311                               const gfp_t priority)
1312 {
1313         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1314         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1315         struct mptcp_subflow_context *new_ctx;
1316
1317         if (!tcp_rsk(req)->is_mptcp ||
1318             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1319                 subflow_ulp_fallback(newsk, old_ctx);
1320                 return;
1321         }
1322
1323         new_ctx = subflow_create_ctx(newsk, priority);
1324         if (!new_ctx) {
1325                 subflow_ulp_fallback(newsk, old_ctx);
1326                 return;
1327         }
1328
1329         new_ctx->conn_finished = 1;
1330         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1331         new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1332         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1333         new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1334         new_ctx->rel_write_seq = 1;
1335         new_ctx->tcp_sock = newsk;
1336
1337         if (subflow_req->mp_capable) {
1338                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1339                  * is fully established only after we receive the remote key
1340                  */
1341                 new_ctx->mp_capable = 1;
1342                 new_ctx->local_key = subflow_req->local_key;
1343                 new_ctx->token = subflow_req->token;
1344                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1345                 new_ctx->idsn = subflow_req->idsn;
1346         } else if (subflow_req->mp_join) {
1347                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1348                 new_ctx->mp_join = 1;
1349                 new_ctx->fully_established = 1;
1350                 new_ctx->backup = subflow_req->backup;
1351                 new_ctx->local_id = subflow_req->local_id;
1352                 new_ctx->remote_id = subflow_req->remote_id;
1353                 new_ctx->token = subflow_req->token;
1354                 new_ctx->thmac = subflow_req->thmac;
1355         }
1356 }
1357
1358 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1359         .name           = "mptcp",
1360         .owner          = THIS_MODULE,
1361         .init           = subflow_ulp_init,
1362         .release        = subflow_ulp_release,
1363         .clone          = subflow_ulp_clone,
1364 };
1365
1366 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1367 {
1368         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1369         subflow_ops->slab_name = "request_sock_subflow";
1370
1371         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1372                                               subflow_ops->obj_size, 0,
1373                                               SLAB_ACCOUNT |
1374                                               SLAB_TYPESAFE_BY_RCU,
1375                                               NULL);
1376         if (!subflow_ops->slab)
1377                 return -ENOMEM;
1378
1379         subflow_ops->destructor = subflow_req_destructor;
1380
1381         return 0;
1382 }
1383
1384 void __init mptcp_subflow_init(void)
1385 {
1386         mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1387         if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1388                 panic("MPTCP: failed to init subflow request sock ops\n");
1389
1390         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1391         subflow_request_sock_ipv4_ops.init_req = subflow_v4_init_req;
1392
1393         subflow_specific = ipv4_specific;
1394         subflow_specific.conn_request = subflow_v4_conn_request;
1395         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1396         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1397
1398 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1399         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1400         subflow_request_sock_ipv6_ops.init_req = subflow_v6_init_req;
1401
1402         subflow_v6_specific = ipv6_specific;
1403         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1404         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1405         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1406
1407         subflow_v6m_specific = subflow_v6_specific;
1408         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1409         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1410         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1411         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1412         subflow_v6m_specific.net_frag_header_len = 0;
1413 #endif
1414
1415         mptcp_diag_subflow_init(&subflow_ulp_ops);
1416
1417         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1418                 panic("MPTCP: failed to register subflows to ULP\n");
1419 }