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/sha2.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 #include <net/transp_v6.h>
  22 #endif
  23 #include <net/mptcp.h>
  24 #include <uapi/linux/mptcp.h>
  25 #include "protocol.h"
  26 #include "mib.h"
  27
  28 static void mptcp_subflow_ops_undo_override(struct sock *ssk);
  29
  30 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
  31                                   enum linux_mptcp_mib_field field)
  32 {
  33         MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
  34 }
  35
  36 static void subflow_req_destructor(struct request_sock *req)
  37 {
  38         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  39
  40         pr_debug("subflow_req=%p", subflow_req);
  41
  42         if (subflow_req->msk)
  43                 sock_put((struct sock *)subflow_req->msk);
  44
  45         mptcp_token_destroy_request(req);
  46         tcp_request_sock_ops.destructor(req);
  47 }
  48
  49 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
  50                                   void *hmac)
  51 {
  52         u8 msg[8];
  53
  54         put_unaligned_be32(nonce1, &msg[0]);
  55         put_unaligned_be32(nonce2, &msg[4]);
  56
  57         mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
  58 }
  59
  60 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
  61 {
  62         return mptcp_is_fully_established((void *)msk) &&
  63                READ_ONCE(msk->pm.accept_subflow);
  64 }
  65
  66 /* validate received token and create truncated hmac and nonce for SYN-ACK */
  67 static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req)
  68 {
  69         struct mptcp_sock *msk = subflow_req->msk;
  70         u8 hmac[SHA256_DIGEST_SIZE];
  71
  72         get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
  73
  74         subflow_generate_hmac(msk->local_key, msk->remote_key,
  75                               subflow_req->local_nonce,
  76                               subflow_req->remote_nonce, hmac);
  77
  78         subflow_req->thmac = get_unaligned_be64(hmac);
  79 }
  80
  81 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req)
  82 {
  83         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  84         struct mptcp_sock *msk;
  85         int local_id;
  86
  87         msk = mptcp_token_get_sock(subflow_req->token);
  88         if (!msk) {
  89                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
  90                 return NULL;
  91         }
  92
  93         local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
  94         if (local_id < 0) {
  95                 sock_put((struct sock *)msk);
  96                 return NULL;
  97         }
  98         subflow_req->local_id = local_id;
  99
 100         return msk;
 101 }
 102
 103 static int __subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
 104 {
 105         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 106
 107         subflow_req->mp_capable = 0;
 108         subflow_req->mp_join = 0;
 109         subflow_req->msk = NULL;
 110         mptcp_token_init_request(req);
 111
 112 #ifdef CONFIG_TCP_MD5SIG
 113         /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
 114          * TCP option space.
 115          */
 116         if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
 117                 return -EINVAL;
 118 #endif
 119
 120         return 0;
 121 }
 122
 123 static bool subflow_use_different_sport(struct mptcp_sock *msk, const struct sock *sk)
 124 {
 125         return inet_sk(sk)->inet_sport != inet_sk((struct sock *)msk)->inet_sport;
 126 }
 127
 128 /* Init mptcp request socket.
 129  *
 130  * Returns an error code if a JOIN has failed and a TCP reset
 131  * should be sent.
 132  */
 133 static int subflow_init_req(struct request_sock *req,
 134                             const struct sock *sk_listener,
 135                             struct sk_buff *skb)
 136 {
 137         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 138         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 139         struct mptcp_options_received mp_opt;
 140         int ret;
 141
 142         pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
 143
 144         ret = __subflow_init_req(req, sk_listener);
 145         if (ret)
 146                 return 0;
 147
 148         mptcp_get_options(skb, &mp_opt);
 149
 150         if (mp_opt.mp_capable) {
 151                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
 152
 153                 if (mp_opt.mp_join)
 154                         return 0;
 155         } else if (mp_opt.mp_join) {
 156                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
 157         }
 158
 159         if (mp_opt.mp_capable && listener->request_mptcp) {
 160                 int err, retries = 4;
 161
 162                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 163 again:
 164                 do {
 165                         get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
 166                 } while (subflow_req->local_key == 0);
 167
 168                 if (unlikely(req->syncookie)) {
 169                         mptcp_crypto_key_sha(subflow_req->local_key,
 170                                              &subflow_req->token,
 171                                              &subflow_req->idsn);
 172                         if (mptcp_token_exists(subflow_req->token)) {
 173                                 if (retries-- > 0)
 174                                         goto again;
 175                         } else {
 176                                 subflow_req->mp_capable = 1;
 177                         }
 178                         return 0;
 179                 }
 180
 181                 err = mptcp_token_new_request(req);
 182                 if (err == 0)
 183                         subflow_req->mp_capable = 1;
 184                 else if (retries-- > 0)
 185                         goto again;
 186
 187         } else if (mp_opt.mp_join && listener->request_mptcp) {
 188                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 189                 subflow_req->mp_join = 1;
 190                 subflow_req->backup = mp_opt.backup;
 191                 subflow_req->remote_id = mp_opt.join_id;
 192                 subflow_req->token = mp_opt.token;
 193                 subflow_req->remote_nonce = mp_opt.nonce;
 194                 subflow_req->msk = subflow_token_join_request(req);
 195
 196                 /* Can't fall back to TCP in this case. */
 197                 if (!subflow_req->msk)
 198                         return -EPERM;
 199
 200                 if (subflow_use_different_sport(subflow_req->msk, sk_listener)) {
 201                         pr_debug("syn inet_sport=%d %d",
 202                                  ntohs(inet_sk(sk_listener)->inet_sport),
 203                                  ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport));
 204                         if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) {
 205                                 sock_put((struct sock *)subflow_req->msk);
 206                                 mptcp_token_destroy_request(req);
 207                                 tcp_request_sock_ops.destructor(req);
 208                                 subflow_req->msk = NULL;
 209                                 subflow_req->mp_join = 0;
 210                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX);
 211                                 return -EPERM;
 212                         }
 213                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX);
 214                 }
 215
 216                 subflow_req_create_thmac(subflow_req);
 217
 218                 if (unlikely(req->syncookie)) {
 219                         if (mptcp_can_accept_new_subflow(subflow_req->msk))
 220                                 subflow_init_req_cookie_join_save(subflow_req, skb);
 221                 }
 222
 223                 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
 224                          subflow_req->remote_nonce, subflow_req->msk);
 225         }
 226
 227         return 0;
 228 }
 229
 230 int mptcp_subflow_init_cookie_req(struct request_sock *req,
 231                                   const struct sock *sk_listener,
 232                                   struct sk_buff *skb)
 233 {
 234         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 235         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 236         struct mptcp_options_received mp_opt;
 237         int err;
 238
 239         err = __subflow_init_req(req, sk_listener);
 240         if (err)
 241                 return err;
 242
 243         mptcp_get_options(skb, &mp_opt);
 244
 245         if (mp_opt.mp_capable && mp_opt.mp_join)
 246                 return -EINVAL;
 247
 248         if (mp_opt.mp_capable && listener->request_mptcp) {
 249                 if (mp_opt.sndr_key == 0)
 250                         return -EINVAL;
 251
 252                 subflow_req->local_key = mp_opt.rcvr_key;
 253                 err = mptcp_token_new_request(req);
 254                 if (err)
 255                         return err;
 256
 257                 subflow_req->mp_capable = 1;
 258                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 259         } else if (mp_opt.mp_join && listener->request_mptcp) {
 260                 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
 261                         return -EINVAL;
 262
 263                 if (mptcp_can_accept_new_subflow(subflow_req->msk))
 264                         subflow_req->mp_join = 1;
 265
 266                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 267         }
 268
 269         return 0;
 270 }
 271 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
 272
 273 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
 274                                               struct sk_buff *skb,
 275                                               struct flowi *fl,
 276                                               struct request_sock *req)
 277 {
 278         struct dst_entry *dst;
 279         int err;
 280
 281         tcp_rsk(req)->is_mptcp = 1;
 282
 283         dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req);
 284         if (!dst)
 285                 return NULL;
 286
 287         err = subflow_init_req(req, sk, skb);
 288         if (err == 0)
 289                 return dst;
 290
 291         dst_release(dst);
 292         if (!req->syncookie)
 293                 tcp_request_sock_ops.send_reset(sk, skb);
 294         return NULL;
 295 }
 296
 297 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 298 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
 299                                               struct sk_buff *skb,
 300                                               struct flowi *fl,
 301                                               struct request_sock *req)
 302 {
 303         struct dst_entry *dst;
 304         int err;
 305
 306         tcp_rsk(req)->is_mptcp = 1;
 307
 308         dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req);
 309         if (!dst)
 310                 return NULL;
 311
 312         err = subflow_init_req(req, sk, skb);
 313         if (err == 0)
 314                 return dst;
 315
 316         dst_release(dst);
 317         if (!req->syncookie)
 318                 tcp6_request_sock_ops.send_reset(sk, skb);
 319         return NULL;
 320 }
 321 #endif
 322
 323 /* validate received truncated hmac and create hmac for third ACK */
 324 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
 325 {
 326         u8 hmac[SHA256_DIGEST_SIZE];
 327         u64 thmac;
 328
 329         subflow_generate_hmac(subflow->remote_key, subflow->local_key,
 330                               subflow->remote_nonce, subflow->local_nonce,
 331                               hmac);
 332
 333         thmac = get_unaligned_be64(hmac);
 334         pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
 335                  subflow, subflow->token,
 336                  (unsigned long long)thmac,
 337                  (unsigned long long)subflow->thmac);
 338
 339         return thmac == subflow->thmac;
 340 }
 341
 342 void mptcp_subflow_reset(struct sock *ssk)
 343 {
 344         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 345         struct sock *sk = subflow->conn;
 346
 347         /* must hold: tcp_done() could drop last reference on parent */
 348         sock_hold(sk);
 349
 350         tcp_set_state(ssk, TCP_CLOSE);
 351         tcp_send_active_reset(ssk, GFP_ATOMIC);
 352         tcp_done(ssk);
 353         if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags) &&
 354             schedule_work(&mptcp_sk(sk)->work))
 355                 return; /* worker will put sk for us */
 356
 357         sock_put(sk);
 358 }
 359
 360 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
 361 {
 362         return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
 363 }
 364
 365 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 366 {
 367         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 368         struct mptcp_options_received mp_opt;
 369         struct sock *parent = subflow->conn;
 370
 371         subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 372
 373         if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
 374                 inet_sk_state_store(parent, TCP_ESTABLISHED);
 375                 parent->sk_state_change(parent);
 376         }
 377
 378         /* be sure no special action on any packet other than syn-ack */
 379         if (subflow->conn_finished)
 380                 return;
 381
 382         mptcp_propagate_sndbuf(parent, sk);
 383         subflow->rel_write_seq = 1;
 384         subflow->conn_finished = 1;
 385         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 386         pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
 387
 388         mptcp_get_options(skb, &mp_opt);
 389         if (subflow->request_mptcp) {
 390                 if (!mp_opt.mp_capable) {
 391                         MPTCP_INC_STATS(sock_net(sk),
 392                                         MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
 393                         mptcp_do_fallback(sk);
 394                         pr_fallback(mptcp_sk(subflow->conn));
 395                         goto fallback;
 396                 }
 397
 398                 subflow->mp_capable = 1;
 399                 subflow->can_ack = 1;
 400                 subflow->remote_key = mp_opt.sndr_key;
 401                 pr_debug("subflow=%p, remote_key=%llu", subflow,
 402                          subflow->remote_key);
 403                 mptcp_finish_connect(sk);
 404         } else if (subflow->request_join) {
 405                 u8 hmac[SHA256_DIGEST_SIZE];
 406
 407                 if (!mp_opt.mp_join)
 408                         goto do_reset;
 409
 410                 subflow->thmac = mp_opt.thmac;
 411                 subflow->remote_nonce = mp_opt.nonce;
 412                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
 413                          subflow->thmac, subflow->remote_nonce);
 414
 415                 if (!subflow_thmac_valid(subflow)) {
 416                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 417                         goto do_reset;
 418                 }
 419
 420                 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 421                                       subflow->local_nonce,
 422                                       subflow->remote_nonce,
 423                                       hmac);
 424                 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
 425
 426                 if (!mptcp_finish_join(sk))
 427                         goto do_reset;
 428
 429                 subflow->mp_join = 1;
 430                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 431
 432                 if (subflow_use_different_dport(mptcp_sk(parent), sk)) {
 433                         pr_debug("synack inet_dport=%d %d",
 434                                  ntohs(inet_sk(sk)->inet_dport),
 435                                  ntohs(inet_sk(parent)->inet_dport));
 436                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
 437                 }
 438         } else if (mptcp_check_fallback(sk)) {
 439 fallback:
 440                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
 441         }
 442         return;
 443
 444 do_reset:
 445         mptcp_subflow_reset(sk);
 446 }
 447
 448 struct request_sock_ops mptcp_subflow_request_sock_ops;
 449 EXPORT_SYMBOL_GPL(mptcp_subflow_request_sock_ops);
 450 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
 451
 452 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 453 {
 454         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 455
 456         pr_debug("subflow=%p", subflow);
 457
 458         /* Never answer to SYNs sent to broadcast or multicast */
 459         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 460                 goto drop;
 461
 462         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 463                                 &subflow_request_sock_ipv4_ops,
 464                                 sk, skb);
 465 drop:
 466         tcp_listendrop(sk);
 467         return 0;
 468 }
 469
 470 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 471 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
 472 static struct inet_connection_sock_af_ops subflow_v6_specific;
 473 static struct inet_connection_sock_af_ops subflow_v6m_specific;
 474 static struct proto tcpv6_prot_override;
 475
 476 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 477 {
 478         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 479
 480         pr_debug("subflow=%p", subflow);
 481
 482         if (skb->protocol == htons(ETH_P_IP))
 483                 return subflow_v4_conn_request(sk, skb);
 484
 485         if (!ipv6_unicast_destination(skb))
 486                 goto drop;
 487
 488         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 489                                 &subflow_request_sock_ipv6_ops, sk, skb);
 490
 491 drop:
 492         tcp_listendrop(sk);
 493         return 0; /* don't send reset */
 494 }
 495 #endif
 496
 497 /* validate hmac received in third ACK */
 498 static bool subflow_hmac_valid(const struct request_sock *req,
 499                                const struct mptcp_options_received *mp_opt)
 500 {
 501         const struct mptcp_subflow_request_sock *subflow_req;
 502         u8 hmac[SHA256_DIGEST_SIZE];
 503         struct mptcp_sock *msk;
 504
 505         subflow_req = mptcp_subflow_rsk(req);
 506         msk = subflow_req->msk;
 507         if (!msk)
 508                 return false;
 509
 510         subflow_generate_hmac(msk->remote_key, msk->local_key,
 511                               subflow_req->remote_nonce,
 512                               subflow_req->local_nonce, hmac);
 513
 514         return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
 515 }
 516
 517 static void mptcp_sock_destruct(struct sock *sk)
 518 {
 519         /* if new mptcp socket isn't accepted, it is free'd
 520          * from the tcp listener sockets request queue, linked
 521          * from req->sk.  The tcp socket is released.
 522          * This calls the ULP release function which will
 523          * also remove the mptcp socket, via
 524          * sock_put(ctx->conn).
 525          *
 526          * Problem is that the mptcp socket will be in
 527          * ESTABLISHED state and will not have the SOCK_DEAD flag.
 528          * Both result in warnings from inet_sock_destruct.
 529          */
 530
 531         if (sk->sk_state == TCP_ESTABLISHED) {
 532                 sk->sk_state = TCP_CLOSE;
 533                 WARN_ON_ONCE(sk->sk_socket);
 534                 sock_orphan(sk);
 535         }
 536
 537         mptcp_destroy_common(mptcp_sk(sk));
 538         inet_sock_destruct(sk);
 539 }
 540
 541 static void mptcp_force_close(struct sock *sk)
 542 {
 543         inet_sk_state_store(sk, TCP_CLOSE);
 544         sk_common_release(sk);
 545 }
 546
 547 static void subflow_ulp_fallback(struct sock *sk,
 548                                  struct mptcp_subflow_context *old_ctx)
 549 {
 550         struct inet_connection_sock *icsk = inet_csk(sk);
 551
 552         mptcp_subflow_tcp_fallback(sk, old_ctx);
 553         icsk->icsk_ulp_ops = NULL;
 554         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 555         tcp_sk(sk)->is_mptcp = 0;
 556
 557         mptcp_subflow_ops_undo_override(sk);
 558 }
 559
 560 static void subflow_drop_ctx(struct sock *ssk)
 561 {
 562         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 563
 564         if (!ctx)
 565                 return;
 566
 567         subflow_ulp_fallback(ssk, ctx);
 568         if (ctx->conn)
 569                 sock_put(ctx->conn);
 570
 571         kfree_rcu(ctx, rcu);
 572 }
 573
 574 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 575                                      struct mptcp_options_received *mp_opt)
 576 {
 577         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 578
 579         subflow->remote_key = mp_opt->sndr_key;
 580         subflow->fully_established = 1;
 581         subflow->can_ack = 1;
 582         WRITE_ONCE(msk->fully_established, true);
 583 }
 584
 585 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 586                                           struct sk_buff *skb,
 587                                           struct request_sock *req,
 588                                           struct dst_entry *dst,
 589                                           struct request_sock *req_unhash,
 590                                           bool *own_req)
 591 {
 592         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 593         struct mptcp_subflow_request_sock *subflow_req;
 594         struct mptcp_options_received mp_opt;
 595         bool fallback, fallback_is_fatal;
 596         struct sock *new_msk = NULL;
 597         struct sock *child;
 598
 599         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 600
 601         /* After child creation we must look for 'mp_capable' even when options
 602          * are not parsed
 603          */
 604         mp_opt.mp_capable = 0;
 605
 606         /* hopefully temporary handling for MP_JOIN+syncookie */
 607         subflow_req = mptcp_subflow_rsk(req);
 608         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 609         fallback = !tcp_rsk(req)->is_mptcp;
 610         if (fallback)
 611                 goto create_child;
 612
 613         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 614         if (subflow_req->mp_capable) {
 615                 if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
 616                         /* here we can receive and accept an in-window,
 617                          * out-of-order pkt, which will not carry the MP_CAPABLE
 618                          * opt even on mptcp enabled paths
 619                          */
 620                         goto create_msk;
 621                 }
 622
 623                 mptcp_get_options(skb, &mp_opt);
 624                 if (!mp_opt.mp_capable) {
 625                         fallback = true;
 626                         goto create_child;
 627                 }
 628
 629 create_msk:
 630                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 631                 if (!new_msk)
 632                         fallback = true;
 633         } else if (subflow_req->mp_join) {
 634                 mptcp_get_options(skb, &mp_opt);
 635                 if (!mp_opt.mp_join || !subflow_hmac_valid(req, &mp_opt) ||
 636                     !mptcp_can_accept_new_subflow(subflow_req->msk)) {
 637                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 638                         fallback = true;
 639                 }
 640         }
 641
 642 create_child:
 643         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 644                                                      req_unhash, own_req);
 645
 646         if (child && *own_req) {
 647                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 648
 649                 tcp_rsk(req)->drop_req = false;
 650
 651                 /* we need to fallback on ctx allocation failure and on pre-reqs
 652                  * checking above. In the latter scenario we additionally need
 653                  * to reset the context to non MPTCP status.
 654                  */
 655                 if (!ctx || fallback) {
 656                         if (fallback_is_fatal)
 657                                 goto dispose_child;
 658
 659                         subflow_drop_ctx(child);
 660                         goto out;
 661                 }
 662
 663                 if (ctx->mp_capable) {
 664                         /* this can't race with mptcp_close(), as the msk is
 665                          * not yet exposted to user-space
 666                          */
 667                         inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
 668
 669                         /* record the newly created socket as the first msk
 670                          * subflow, but don't link it yet into conn_list
 671                          */
 672                         WRITE_ONCE(mptcp_sk(new_msk)->first, child);
 673
 674                         /* new mpc subflow takes ownership of the newly
 675                          * created mptcp socket
 676                          */
 677                         new_msk->sk_destruct = mptcp_sock_destruct;
 678                         mptcp_pm_new_connection(mptcp_sk(new_msk), 1);
 679                         mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
 680                         ctx->conn = new_msk;
 681                         new_msk = NULL;
 682
 683                         /* with OoO packets we can reach here without ingress
 684                          * mpc option
 685                          */
 686                         if (mp_opt.mp_capable)
 687                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 688                 } else if (ctx->mp_join) {
 689                         struct mptcp_sock *owner;
 690
 691                         owner = subflow_req->msk;
 692                         if (!owner)
 693                                 goto dispose_child;
 694
 695                         /* move the msk reference ownership to the subflow */
 696                         subflow_req->msk = NULL;
 697                         ctx->conn = (struct sock *)owner;
 698                         if (!mptcp_finish_join(child))
 699                                 goto dispose_child;
 700
 701                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 702                         tcp_rsk(req)->drop_req = true;
 703
 704                         if (subflow_use_different_sport(owner, sk)) {
 705                                 pr_debug("ack inet_sport=%d %d",
 706                                          ntohs(inet_sk(sk)->inet_sport),
 707                                          ntohs(inet_sk((struct sock *)owner)->inet_sport));
 708                                 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
 709                                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
 710                                         goto out;
 711                                 }
 712                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
 713                         }
 714                 }
 715         }
 716
 717 out:
 718         /* dispose of the left over mptcp master, if any */
 719         if (unlikely(new_msk))
 720                 mptcp_force_close(new_msk);
 721
 722         /* check for expected invariant - should never trigger, just help
 723          * catching eariler subtle bugs
 724          */
 725         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 726                      (!mptcp_subflow_ctx(child) ||
 727                       !mptcp_subflow_ctx(child)->conn));
 728         return child;
 729
 730 dispose_child:
 731         subflow_drop_ctx(child);
 732         tcp_rsk(req)->drop_req = true;
 733         inet_csk_prepare_for_destroy_sock(child);
 734         tcp_done(child);
 735         req->rsk_ops->send_reset(sk, skb);
 736
 737         /* The last child reference will be released by the caller */
 738         return child;
 739 }
 740
 741 static struct inet_connection_sock_af_ops subflow_specific;
 742 static struct proto tcp_prot_override;
 743
 744 enum mapping_status {
 745         MAPPING_OK,
 746         MAPPING_INVALID,
 747         MAPPING_EMPTY,
 748         MAPPING_DATA_FIN,
 749         MAPPING_DUMMY
 750 };
 751
 752 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
 753 {
 754         if ((u32)seq == (u32)old_seq)
 755                 return old_seq;
 756
 757         /* Assume map covers data not mapped yet. */
 758         return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
 759 }
 760
 761 static void warn_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 762 {
 763         WARN_ONCE(1, "Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 764                   ssn, subflow->map_subflow_seq, subflow->map_data_len);
 765 }
 766
 767 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 768 {
 769         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 770         unsigned int skb_consumed;
 771
 772         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 773         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 774                 return true;
 775
 776         return skb->len - skb_consumed <= subflow->map_data_len -
 777                                           mptcp_subflow_get_map_offset(subflow);
 778 }
 779
 780 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 781 {
 782         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 783         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 784
 785         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 786                 /* Mapping covers data later in the subflow stream,
 787                  * currently unsupported.
 788                  */
 789                 warn_bad_map(subflow, ssn);
 790                 return false;
 791         }
 792         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 793                                   subflow->map_data_len))) {
 794                 /* Mapping does covers past subflow data, invalid */
 795                 warn_bad_map(subflow, ssn + skb->len);
 796                 return false;
 797         }
 798         return true;
 799 }
 800
 801 static enum mapping_status get_mapping_status(struct sock *ssk,
 802                                               struct mptcp_sock *msk)
 803 {
 804         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 805         struct mptcp_ext *mpext;
 806         struct sk_buff *skb;
 807         u16 data_len;
 808         u64 map_seq;
 809
 810         skb = skb_peek(&ssk->sk_receive_queue);
 811         if (!skb)
 812                 return MAPPING_EMPTY;
 813
 814         if (mptcp_check_fallback(ssk))
 815                 return MAPPING_DUMMY;
 816
 817         mpext = mptcp_get_ext(skb);
 818         if (!mpext || !mpext->use_map) {
 819                 if (!subflow->map_valid && !skb->len) {
 820                         /* the TCP stack deliver 0 len FIN pkt to the receive
 821                          * queue, that is the only 0len pkts ever expected here,
 822                          * and we can admit no mapping only for 0 len pkts
 823                          */
 824                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 825                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 826                                           TCP_SKB_CB(skb)->seq,
 827                                           TCP_SKB_CB(skb)->end_seq,
 828                                           TCP_SKB_CB(skb)->tcp_flags);
 829                         sk_eat_skb(ssk, skb);
 830                         return MAPPING_EMPTY;
 831                 }
 832
 833                 if (!subflow->map_valid)
 834                         return MAPPING_INVALID;
 835
 836                 goto validate_seq;
 837         }
 838
 839         pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
 840                  mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
 841                  mpext->data_len, mpext->data_fin);
 842
 843         data_len = mpext->data_len;
 844         if (data_len == 0) {
 845                 pr_err("Infinite mapping not handled");
 846                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 847                 return MAPPING_INVALID;
 848         }
 849
 850         if (mpext->data_fin == 1) {
 851                 if (data_len == 1) {
 852                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
 853                                                                  mpext->dsn64);
 854                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
 855                         if (subflow->map_valid) {
 856                                 /* A DATA_FIN might arrive in a DSS
 857                                  * option before the previous mapping
 858                                  * has been fully consumed. Continue
 859                                  * handling the existing mapping.
 860                                  */
 861                                 skb_ext_del(skb, SKB_EXT_MPTCP);
 862                                 return MAPPING_OK;
 863                         } else {
 864                                 if (updated && schedule_work(&msk->work))
 865                                         sock_hold((struct sock *)msk);
 866
 867                                 return MAPPING_DATA_FIN;
 868                         }
 869                 } else {
 870                         u64 data_fin_seq = mpext->data_seq + data_len - 1;
 871
 872                         /* If mpext->data_seq is a 32-bit value, data_fin_seq
 873                          * must also be limited to 32 bits.
 874                          */
 875                         if (!mpext->dsn64)
 876                                 data_fin_seq &= GENMASK_ULL(31, 0);
 877
 878                         mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
 879                         pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
 880                                  data_fin_seq, mpext->dsn64);
 881                 }
 882
 883                 /* Adjust for DATA_FIN using 1 byte of sequence space */
 884                 data_len--;
 885         }
 886
 887         if (!mpext->dsn64) {
 888                 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
 889                                      mpext->data_seq);
 890                 pr_debug("expanded seq=%llu", subflow->map_seq);
 891         } else {
 892                 map_seq = mpext->data_seq;
 893         }
 894         WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
 895
 896         if (subflow->map_valid) {
 897                 /* Allow replacing only with an identical map */
 898                 if (subflow->map_seq == map_seq &&
 899                     subflow->map_subflow_seq == mpext->subflow_seq &&
 900                     subflow->map_data_len == data_len) {
 901                         skb_ext_del(skb, SKB_EXT_MPTCP);
 902                         return MAPPING_OK;
 903                 }
 904
 905                 /* If this skb data are fully covered by the current mapping,
 906                  * the new map would need caching, which is not supported
 907                  */
 908                 if (skb_is_fully_mapped(ssk, skb)) {
 909                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
 910                         return MAPPING_INVALID;
 911                 }
 912
 913                 /* will validate the next map after consuming the current one */
 914                 return MAPPING_OK;
 915         }
 916
 917         subflow->map_seq = map_seq;
 918         subflow->map_subflow_seq = mpext->subflow_seq;
 919         subflow->map_data_len = data_len;
 920         subflow->map_valid = 1;
 921         subflow->mpc_map = mpext->mpc_map;
 922         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
 923                  subflow->map_seq, subflow->map_subflow_seq,
 924                  subflow->map_data_len);
 925
 926 validate_seq:
 927         /* we revalidate valid mapping on new skb, because we must ensure
 928          * the current skb is completely covered by the available mapping
 929          */
 930         if (!validate_mapping(ssk, skb))
 931                 return MAPPING_INVALID;
 932
 933         skb_ext_del(skb, SKB_EXT_MPTCP);
 934         return MAPPING_OK;
 935 }
 936
 937 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
 938                                        u64 limit)
 939 {
 940         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 941         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
 942         u32 incr;
 943
 944         incr = limit >= skb->len ? skb->len + fin : limit;
 945
 946         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
 947                  subflow->map_subflow_seq);
 948         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
 949         tcp_sk(ssk)->copied_seq += incr;
 950         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
 951                 sk_eat_skb(ssk, skb);
 952         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
 953                 subflow->map_valid = 0;
 954 }
 955
 956 /* sched mptcp worker to remove the subflow if no more data is pending */
 957 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
 958 {
 959         struct sock *sk = (struct sock *)msk;
 960
 961         if (likely(ssk->sk_state != TCP_CLOSE))
 962                 return;
 963
 964         if (skb_queue_empty(&ssk->sk_receive_queue) &&
 965             !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags)) {
 966                 sock_hold(sk);
 967                 if (!schedule_work(&msk->work))
 968                         sock_put(sk);
 969         }
 970 }
 971
 972 static bool subflow_check_data_avail(struct sock *ssk)
 973 {
 974         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 975         enum mapping_status status;
 976         struct mptcp_sock *msk;
 977         struct sk_buff *skb;
 978
 979         pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
 980                  subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
 981         if (!skb_peek(&ssk->sk_receive_queue))
 982                 subflow->data_avail = 0;
 983         if (subflow->data_avail)
 984                 return true;
 985
 986         msk = mptcp_sk(subflow->conn);
 987         for (;;) {
 988                 u64 ack_seq;
 989                 u64 old_ack;
 990
 991                 status = get_mapping_status(ssk, msk);
 992                 pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
 993                 if (status == MAPPING_INVALID) {
 994                         ssk->sk_err = EBADMSG;
 995                         goto fatal;
 996                 }
 997                 if (status == MAPPING_DUMMY) {
 998                         __mptcp_do_fallback(msk);
 999                         skb = skb_peek(&ssk->sk_receive_queue);
1000                         subflow->map_valid = 1;
1001                         subflow->map_seq = READ_ONCE(msk->ack_seq);
1002                         subflow->map_data_len = skb->len;
1003                         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq -
1004                                                    subflow->ssn_offset;
1005                         subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
1006                         return true;
1007                 }
1008
1009                 if (status != MAPPING_OK)
1010                         goto no_data;
1011
1012                 skb = skb_peek(&ssk->sk_receive_queue);
1013                 if (WARN_ON_ONCE(!skb))
1014                         goto no_data;
1015
1016                 /* if msk lacks the remote key, this subflow must provide an
1017                  * MP_CAPABLE-based mapping
1018                  */
1019                 if (unlikely(!READ_ONCE(msk->can_ack))) {
1020                         if (!subflow->mpc_map) {
1021                                 ssk->sk_err = EBADMSG;
1022                                 goto fatal;
1023                         }
1024                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
1025                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
1026                         WRITE_ONCE(msk->can_ack, true);
1027                 }
1028
1029                 old_ack = READ_ONCE(msk->ack_seq);
1030                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1031                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1032                          ack_seq);
1033                 if (ack_seq == old_ack) {
1034                         subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
1035                         break;
1036                 } else if (after64(ack_seq, old_ack)) {
1037                         subflow->data_avail = MPTCP_SUBFLOW_OOO_DATA;
1038                         break;
1039                 }
1040
1041                 /* only accept in-sequence mapping. Old values are spurious
1042                  * retransmission
1043                  */
1044                 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1045         }
1046         return true;
1047
1048 no_data:
1049         subflow_sched_work_if_closed(msk, ssk);
1050         return false;
1051 fatal:
1052         /* fatal protocol error, close the socket */
1053         /* This barrier is coupled with smp_rmb() in tcp_poll() */
1054         smp_wmb();
1055         ssk->sk_error_report(ssk);
1056         tcp_set_state(ssk, TCP_CLOSE);
1057         tcp_send_active_reset(ssk, GFP_ATOMIC);
1058         subflow->data_avail = 0;
1059         return false;
1060 }
1061
1062 bool mptcp_subflow_data_available(struct sock *sk)
1063 {
1064         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1065
1066         /* check if current mapping is still valid */
1067         if (subflow->map_valid &&
1068             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1069                 subflow->map_valid = 0;
1070                 subflow->data_avail = 0;
1071
1072                 pr_debug("Done with mapping: seq=%u data_len=%u",
1073                          subflow->map_subflow_seq,
1074                          subflow->map_data_len);
1075         }
1076
1077         return subflow_check_data_avail(sk);
1078 }
1079
1080 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1081  * not the ssk one.
1082  *
1083  * In mptcp, rwin is about the mptcp-level connection data.
1084  *
1085  * Data that is still on the ssk rx queue can thus be ignored,
1086  * as far as mptcp peer is concerened that data is still inflight.
1087  * DSS ACK is updated when skb is moved to the mptcp rx queue.
1088  */
1089 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1090 {
1091         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1092         const struct sock *sk = subflow->conn;
1093
1094         *space = __mptcp_space(sk);
1095         *full_space = tcp_full_space(sk);
1096 }
1097
1098 static void subflow_data_ready(struct sock *sk)
1099 {
1100         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1101         u16 state = 1 << inet_sk_state_load(sk);
1102         struct sock *parent = subflow->conn;
1103         struct mptcp_sock *msk;
1104
1105         msk = mptcp_sk(parent);
1106         if (state & TCPF_LISTEN) {
1107                 set_bit(MPTCP_DATA_READY, &msk->flags);
1108                 parent->sk_data_ready(parent);
1109                 return;
1110         }
1111
1112         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1113                      !subflow->mp_join && !(state & TCPF_CLOSE));
1114
1115         if (mptcp_subflow_data_available(sk))
1116                 mptcp_data_ready(parent, sk);
1117 }
1118
1119 static void subflow_write_space(struct sock *ssk)
1120 {
1121         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1122
1123         mptcp_propagate_sndbuf(sk, ssk);
1124         mptcp_write_space(sk);
1125 }
1126
1127 static struct inet_connection_sock_af_ops *
1128 subflow_default_af_ops(struct sock *sk)
1129 {
1130 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1131         if (sk->sk_family == AF_INET6)
1132                 return &subflow_v6_specific;
1133 #endif
1134         return &subflow_specific;
1135 }
1136
1137 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1138 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1139 {
1140         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1141         struct inet_connection_sock *icsk = inet_csk(sk);
1142         struct inet_connection_sock_af_ops *target;
1143
1144         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1145
1146         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1147                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1148
1149         if (likely(icsk->icsk_af_ops == target))
1150                 return;
1151
1152         subflow->icsk_af_ops = icsk->icsk_af_ops;
1153         icsk->icsk_af_ops = target;
1154 }
1155 #endif
1156
1157 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1158                          struct sockaddr_storage *addr,
1159                          unsigned short family)
1160 {
1161         memset(addr, 0, sizeof(*addr));
1162         addr->ss_family = family;
1163         if (addr->ss_family == AF_INET) {
1164                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1165
1166                 if (info->family == AF_INET)
1167                         in_addr->sin_addr = info->addr;
1168 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1169                 else if (ipv6_addr_v4mapped(&info->addr6))
1170                         in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1171 #endif
1172                 in_addr->sin_port = info->port;
1173         }
1174 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1175         else if (addr->ss_family == AF_INET6) {
1176                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1177
1178                 if (info->family == AF_INET)
1179                         ipv6_addr_set_v4mapped(info->addr.s_addr,
1180                                                &in6_addr->sin6_addr);
1181                 else
1182                         in6_addr->sin6_addr = info->addr6;
1183                 in6_addr->sin6_port = info->port;
1184         }
1185 #endif
1186 }
1187
1188 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1189                             const struct mptcp_addr_info *remote)
1190 {
1191         struct mptcp_sock *msk = mptcp_sk(sk);
1192         struct mptcp_subflow_context *subflow;
1193         struct sockaddr_storage addr;
1194         int remote_id = remote->id;
1195         int local_id = loc->id;
1196         struct socket *sf;
1197         struct sock *ssk;
1198         u32 remote_token;
1199         int addrlen;
1200         int err;
1201
1202         if (!mptcp_is_fully_established(sk))
1203                 return -ENOTCONN;
1204
1205         err = mptcp_subflow_create_socket(sk, &sf);
1206         if (err)
1207                 return err;
1208
1209         ssk = sf->sk;
1210         subflow = mptcp_subflow_ctx(ssk);
1211         do {
1212                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1213         } while (!subflow->local_nonce);
1214
1215         if (!local_id) {
1216                 err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
1217                 if (err < 0)
1218                         goto failed;
1219
1220                 local_id = err;
1221         }
1222
1223         subflow->remote_key = msk->remote_key;
1224         subflow->local_key = msk->local_key;
1225         subflow->token = msk->token;
1226         mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1227
1228         addrlen = sizeof(struct sockaddr_in);
1229 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1230         if (addr.ss_family == AF_INET6)
1231                 addrlen = sizeof(struct sockaddr_in6);
1232 #endif
1233         ssk->sk_bound_dev_if = loc->ifindex;
1234         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1235         if (err)
1236                 goto failed;
1237
1238         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1239         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1240                  remote_token, local_id, remote_id);
1241         subflow->remote_token = remote_token;
1242         subflow->local_id = local_id;
1243         subflow->remote_id = remote_id;
1244         subflow->request_join = 1;
1245         subflow->request_bkup = !!(loc->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1246         mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1247
1248         mptcp_add_pending_subflow(msk, subflow);
1249         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1250         if (err && err != -EINPROGRESS)
1251                 goto failed_unlink;
1252
1253         /* discard the subflow socket */
1254         mptcp_sock_graft(ssk, sk->sk_socket);
1255         iput(SOCK_INODE(sf));
1256         return err;
1257
1258 failed_unlink:
1259         spin_lock_bh(&msk->join_list_lock);
1260         list_del(&subflow->node);
1261         spin_unlock_bh(&msk->join_list_lock);
1262
1263 failed:
1264         subflow->disposable = 1;
1265         sock_release(sf);
1266         return err;
1267 }
1268
1269 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1270 {
1271 #ifdef CONFIG_SOCK_CGROUP_DATA
1272         struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1273                                 *child_skcd = &child->sk_cgrp_data;
1274
1275         /* only the additional subflows created by kworkers have to be modified */
1276         if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1277             cgroup_id(sock_cgroup_ptr(child_skcd))) {
1278 #ifdef CONFIG_MEMCG
1279                 struct mem_cgroup *memcg = parent->sk_memcg;
1280
1281                 mem_cgroup_sk_free(child);
1282                 if (memcg && css_tryget(&memcg->css))
1283                         child->sk_memcg = memcg;
1284 #endif /* CONFIG_MEMCG */
1285
1286                 cgroup_sk_free(child_skcd);
1287                 *child_skcd = *parent_skcd;
1288                 cgroup_sk_clone(child_skcd);
1289         }
1290 #endif /* CONFIG_SOCK_CGROUP_DATA */
1291 }
1292
1293 static void mptcp_subflow_ops_override(struct sock *ssk)
1294 {
1295 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1296         if (ssk->sk_prot == &tcpv6_prot)
1297                 ssk->sk_prot = &tcpv6_prot_override;
1298         else
1299 #endif
1300                 ssk->sk_prot = &tcp_prot_override;
1301 }
1302
1303 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1304 {
1305 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1306         if (ssk->sk_prot == &tcpv6_prot_override)
1307                 ssk->sk_prot = &tcpv6_prot;
1308         else
1309 #endif
1310                 ssk->sk_prot = &tcp_prot;
1311 }
1312 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1313 {
1314         struct mptcp_subflow_context *subflow;
1315         struct net *net = sock_net(sk);
1316         struct socket *sf;
1317         int err;
1318
1319         /* un-accepted server sockets can reach here - on bad configuration
1320          * bail early to avoid greater trouble later
1321          */
1322         if (unlikely(!sk->sk_socket))
1323                 return -EINVAL;
1324
1325         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1326                                &sf);
1327         if (err)
1328                 return err;
1329
1330         lock_sock(sf->sk);
1331
1332         /* the newly created socket has to be in the same cgroup as its parent */
1333         mptcp_attach_cgroup(sk, sf->sk);
1334
1335         /* kernel sockets do not by default acquire net ref, but TCP timer
1336          * needs it.
1337          */
1338         sf->sk->sk_net_refcnt = 1;
1339         get_net(net);
1340 #ifdef CONFIG_PROC_FS
1341         this_cpu_add(*net->core.sock_inuse, 1);
1342 #endif
1343         err = tcp_set_ulp(sf->sk, "mptcp");
1344         release_sock(sf->sk);
1345
1346         if (err) {
1347                 sock_release(sf);
1348                 return err;
1349         }
1350
1351         /* the newly created socket really belongs to the owning MPTCP master
1352          * socket, even if for additional subflows the allocation is performed
1353          * by a kernel workqueue. Adjust inode references, so that the
1354          * procfs/diag interaces really show this one belonging to the correct
1355          * user.
1356          */
1357         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1358         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1359         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1360
1361         subflow = mptcp_subflow_ctx(sf->sk);
1362         pr_debug("subflow=%p", subflow);
1363
1364         *new_sock = sf;
1365         sock_hold(sk);
1366         subflow->conn = sk;
1367         mptcp_subflow_ops_override(sf->sk);
1368
1369         return 0;
1370 }
1371
1372 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1373                                                         gfp_t priority)
1374 {
1375         struct inet_connection_sock *icsk = inet_csk(sk);
1376         struct mptcp_subflow_context *ctx;
1377
1378         ctx = kzalloc(sizeof(*ctx), priority);
1379         if (!ctx)
1380                 return NULL;
1381
1382         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1383         INIT_LIST_HEAD(&ctx->node);
1384         INIT_LIST_HEAD(&ctx->delegated_node);
1385
1386         pr_debug("subflow=%p", ctx);
1387
1388         ctx->tcp_sock = sk;
1389
1390         return ctx;
1391 }
1392
1393 static void __subflow_state_change(struct sock *sk)
1394 {
1395         struct socket_wq *wq;
1396
1397         rcu_read_lock();
1398         wq = rcu_dereference(sk->sk_wq);
1399         if (skwq_has_sleeper(wq))
1400                 wake_up_interruptible_all(&wq->wait);
1401         rcu_read_unlock();
1402 }
1403
1404 static bool subflow_is_done(const struct sock *sk)
1405 {
1406         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1407 }
1408
1409 static void subflow_state_change(struct sock *sk)
1410 {
1411         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1412         struct sock *parent = subflow->conn;
1413
1414         __subflow_state_change(sk);
1415
1416         if (subflow_simultaneous_connect(sk)) {
1417                 mptcp_propagate_sndbuf(parent, sk);
1418                 mptcp_do_fallback(sk);
1419                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1420                 pr_fallback(mptcp_sk(parent));
1421                 subflow->conn_finished = 1;
1422                 if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
1423                         inet_sk_state_store(parent, TCP_ESTABLISHED);
1424                         parent->sk_state_change(parent);
1425                 }
1426         }
1427
1428         /* as recvmsg() does not acquire the subflow socket for ssk selection
1429          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1430          * the data available machinery here.
1431          */
1432         if (mptcp_subflow_data_available(sk))
1433                 mptcp_data_ready(parent, sk);
1434
1435         subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1436
1437         if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1438             !subflow->rx_eof && subflow_is_done(sk)) {
1439                 subflow->rx_eof = 1;
1440                 mptcp_subflow_eof(parent);
1441         }
1442 }
1443
1444 static int subflow_ulp_init(struct sock *sk)
1445 {
1446         struct inet_connection_sock *icsk = inet_csk(sk);
1447         struct mptcp_subflow_context *ctx;
1448         struct tcp_sock *tp = tcp_sk(sk);
1449         int err = 0;
1450
1451         /* disallow attaching ULP to a socket unless it has been
1452          * created with sock_create_kern()
1453          */
1454         if (!sk->sk_kern_sock) {
1455                 err = -EOPNOTSUPP;
1456                 goto out;
1457         }
1458
1459         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1460         if (!ctx) {
1461                 err = -ENOMEM;
1462                 goto out;
1463         }
1464
1465         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1466
1467         tp->is_mptcp = 1;
1468         ctx->icsk_af_ops = icsk->icsk_af_ops;
1469         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1470         ctx->tcp_data_ready = sk->sk_data_ready;
1471         ctx->tcp_state_change = sk->sk_state_change;
1472         ctx->tcp_write_space = sk->sk_write_space;
1473         sk->sk_data_ready = subflow_data_ready;
1474         sk->sk_write_space = subflow_write_space;
1475         sk->sk_state_change = subflow_state_change;
1476 out:
1477         return err;
1478 }
1479
1480 static void subflow_ulp_release(struct sock *ssk)
1481 {
1482         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1483         bool release = true;
1484         struct sock *sk;
1485
1486         if (!ctx)
1487                 return;
1488
1489         sk = ctx->conn;
1490         if (sk) {
1491                 /* if the msk has been orphaned, keep the ctx
1492                  * alive, will be freed by __mptcp_close_ssk(),
1493                  * when the subflow is still unaccepted
1494                  */
1495                 release = ctx->disposable || list_empty(&ctx->node);
1496                 sock_put(sk);
1497         }
1498
1499         mptcp_subflow_ops_undo_override(ssk);
1500         if (release)
1501                 kfree_rcu(ctx, rcu);
1502 }
1503
1504 static void subflow_ulp_clone(const struct request_sock *req,
1505                               struct sock *newsk,
1506                               const gfp_t priority)
1507 {
1508         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1509         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1510         struct mptcp_subflow_context *new_ctx;
1511
1512         if (!tcp_rsk(req)->is_mptcp ||
1513             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1514                 subflow_ulp_fallback(newsk, old_ctx);
1515                 return;
1516         }
1517
1518         new_ctx = subflow_create_ctx(newsk, priority);
1519         if (!new_ctx) {
1520                 subflow_ulp_fallback(newsk, old_ctx);
1521                 return;
1522         }
1523
1524         new_ctx->conn_finished = 1;
1525         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1526         new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1527         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1528         new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1529         new_ctx->rel_write_seq = 1;
1530         new_ctx->tcp_sock = newsk;
1531
1532         if (subflow_req->mp_capable) {
1533                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1534                  * is fully established only after we receive the remote key
1535                  */
1536                 new_ctx->mp_capable = 1;
1537                 new_ctx->local_key = subflow_req->local_key;
1538                 new_ctx->token = subflow_req->token;
1539                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1540                 new_ctx->idsn = subflow_req->idsn;
1541         } else if (subflow_req->mp_join) {
1542                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1543                 new_ctx->mp_join = 1;
1544                 new_ctx->fully_established = 1;
1545                 new_ctx->backup = subflow_req->backup;
1546                 new_ctx->local_id = subflow_req->local_id;
1547                 new_ctx->remote_id = subflow_req->remote_id;
1548                 new_ctx->token = subflow_req->token;
1549                 new_ctx->thmac = subflow_req->thmac;
1550         }
1551 }
1552
1553 static void tcp_release_cb_override(struct sock *ssk)
1554 {
1555         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1556
1557         if (mptcp_subflow_has_delegated_action(subflow))
1558                 mptcp_subflow_process_delegated(ssk);
1559
1560         tcp_release_cb(ssk);
1561 }
1562
1563 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1564         .name           = "mptcp",
1565         .owner          = THIS_MODULE,
1566         .init           = subflow_ulp_init,
1567         .release        = subflow_ulp_release,
1568         .clone          = subflow_ulp_clone,
1569 };
1570
1571 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1572 {
1573         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1574         subflow_ops->slab_name = "request_sock_subflow";
1575
1576         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1577                                               subflow_ops->obj_size, 0,
1578                                               SLAB_ACCOUNT |
1579                                               SLAB_TYPESAFE_BY_RCU,
1580                                               NULL);
1581         if (!subflow_ops->slab)
1582                 return -ENOMEM;
1583
1584         subflow_ops->destructor = subflow_req_destructor;
1585
1586         return 0;
1587 }
1588
1589 void __init mptcp_subflow_init(void)
1590 {
1591         mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1592         if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1593                 panic("MPTCP: failed to init subflow request sock ops\n");
1594
1595         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1596         subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
1597
1598         subflow_specific = ipv4_specific;
1599         subflow_specific.conn_request = subflow_v4_conn_request;
1600         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1601         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1602
1603         tcp_prot_override = tcp_prot;
1604         tcp_prot_override.release_cb = tcp_release_cb_override;
1605
1606 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1607         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1608         subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
1609
1610         subflow_v6_specific = ipv6_specific;
1611         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1612         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1613         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1614
1615         subflow_v6m_specific = subflow_v6_specific;
1616         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1617         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1618         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1619         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1620         subflow_v6m_specific.net_frag_header_len = 0;
1621
1622         tcpv6_prot_override = tcpv6_prot;
1623         tcpv6_prot_override.release_cb = tcp_release_cb_override;
1624 #endif
1625
1626         mptcp_diag_subflow_init(&subflow_ulp_ops);
1627
1628         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1629                 panic("MPTCP: failed to register subflows to ULP\n");
1630 }