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 void 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
 113 static bool subflow_use_different_sport(struct mptcp_sock *msk, const struct sock *sk)
 114 {
 115         return inet_sk(sk)->inet_sport != inet_sk((struct sock *)msk)->inet_sport;
 116 }
 117
 118 /* Init mptcp request socket.
 119  *
 120  * Returns an error code if a JOIN has failed and a TCP reset
 121  * should be sent.
 122  */
 123 static int subflow_check_req(struct request_sock *req,
 124                              const struct sock *sk_listener,
 125                              struct sk_buff *skb)
 126 {
 127         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 128         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 129         struct mptcp_options_received mp_opt;
 130
 131         pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
 132
 133 #ifdef CONFIG_TCP_MD5SIG
 134         /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
 135          * TCP option space.
 136          */
 137         if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
 138                 return -EINVAL;
 139 #endif
 140
 141         mptcp_get_options(skb, &mp_opt);
 142
 143         if (mp_opt.mp_capable) {
 144                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
 145
 146                 if (mp_opt.mp_join)
 147                         return 0;
 148         } else if (mp_opt.mp_join) {
 149                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
 150         }
 151
 152         if (mp_opt.mp_capable && listener->request_mptcp) {
 153                 int err, retries = 4;
 154
 155                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 156 again:
 157                 do {
 158                         get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
 159                 } while (subflow_req->local_key == 0);
 160
 161                 if (unlikely(req->syncookie)) {
 162                         mptcp_crypto_key_sha(subflow_req->local_key,
 163                                              &subflow_req->token,
 164                                              &subflow_req->idsn);
 165                         if (mptcp_token_exists(subflow_req->token)) {
 166                                 if (retries-- > 0)
 167                                         goto again;
 168                         } else {
 169                                 subflow_req->mp_capable = 1;
 170                         }
 171                         return 0;
 172                 }
 173
 174                 err = mptcp_token_new_request(req);
 175                 if (err == 0)
 176                         subflow_req->mp_capable = 1;
 177                 else if (retries-- > 0)
 178                         goto again;
 179
 180         } else if (mp_opt.mp_join && listener->request_mptcp) {
 181                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 182                 subflow_req->mp_join = 1;
 183                 subflow_req->backup = mp_opt.backup;
 184                 subflow_req->remote_id = mp_opt.join_id;
 185                 subflow_req->token = mp_opt.token;
 186                 subflow_req->remote_nonce = mp_opt.nonce;
 187                 subflow_req->msk = subflow_token_join_request(req);
 188
 189                 /* Can't fall back to TCP in this case. */
 190                 if (!subflow_req->msk)
 191                         return -EPERM;
 192
 193                 if (subflow_use_different_sport(subflow_req->msk, sk_listener)) {
 194                         pr_debug("syn inet_sport=%d %d",
 195                                  ntohs(inet_sk(sk_listener)->inet_sport),
 196                                  ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport));
 197                         if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) {
 198                                 sock_put((struct sock *)subflow_req->msk);
 199                                 mptcp_token_destroy_request(req);
 200                                 tcp_request_sock_ops.destructor(req);
 201                                 subflow_req->msk = NULL;
 202                                 subflow_req->mp_join = 0;
 203                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX);
 204                                 return -EPERM;
 205                         }
 206                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX);
 207                 }
 208
 209                 subflow_req_create_thmac(subflow_req);
 210
 211                 if (unlikely(req->syncookie)) {
 212                         if (mptcp_can_accept_new_subflow(subflow_req->msk))
 213                                 subflow_init_req_cookie_join_save(subflow_req, skb);
 214                 }
 215
 216                 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
 217                          subflow_req->remote_nonce, subflow_req->msk);
 218         }
 219
 220         return 0;
 221 }
 222
 223 int mptcp_subflow_init_cookie_req(struct request_sock *req,
 224                                   const struct sock *sk_listener,
 225                                   struct sk_buff *skb)
 226 {
 227         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 228         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 229         struct mptcp_options_received mp_opt;
 230         int err;
 231
 232         subflow_init_req(req, sk_listener);
 233         mptcp_get_options(skb, &mp_opt);
 234
 235         if (mp_opt.mp_capable && mp_opt.mp_join)
 236                 return -EINVAL;
 237
 238         if (mp_opt.mp_capable && listener->request_mptcp) {
 239                 if (mp_opt.sndr_key == 0)
 240                         return -EINVAL;
 241
 242                 subflow_req->local_key = mp_opt.rcvr_key;
 243                 err = mptcp_token_new_request(req);
 244                 if (err)
 245                         return err;
 246
 247                 subflow_req->mp_capable = 1;
 248                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 249         } else if (mp_opt.mp_join && listener->request_mptcp) {
 250                 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
 251                         return -EINVAL;
 252
 253                 if (mptcp_can_accept_new_subflow(subflow_req->msk))
 254                         subflow_req->mp_join = 1;
 255
 256                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 257         }
 258
 259         return 0;
 260 }
 261 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
 262
 263 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
 264                                               struct sk_buff *skb,
 265                                               struct flowi *fl,
 266                                               struct request_sock *req)
 267 {
 268         struct dst_entry *dst;
 269         int err;
 270
 271         tcp_rsk(req)->is_mptcp = 1;
 272         subflow_init_req(req, sk);
 273
 274         dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req);
 275         if (!dst)
 276                 return NULL;
 277
 278         err = subflow_check_req(req, sk, skb);
 279         if (err == 0)
 280                 return dst;
 281
 282         dst_release(dst);
 283         if (!req->syncookie)
 284                 tcp_request_sock_ops.send_reset(sk, skb);
 285         return NULL;
 286 }
 287
 288 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 289 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
 290                                               struct sk_buff *skb,
 291                                               struct flowi *fl,
 292                                               struct request_sock *req)
 293 {
 294         struct dst_entry *dst;
 295         int err;
 296
 297         tcp_rsk(req)->is_mptcp = 1;
 298         subflow_init_req(req, sk);
 299
 300         dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req);
 301         if (!dst)
 302                 return NULL;
 303
 304         err = subflow_check_req(req, sk, skb);
 305         if (err == 0)
 306                 return dst;
 307
 308         dst_release(dst);
 309         if (!req->syncookie)
 310                 tcp6_request_sock_ops.send_reset(sk, skb);
 311         return NULL;
 312 }
 313 #endif
 314
 315 /* validate received truncated hmac and create hmac for third ACK */
 316 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
 317 {
 318         u8 hmac[SHA256_DIGEST_SIZE];
 319         u64 thmac;
 320
 321         subflow_generate_hmac(subflow->remote_key, subflow->local_key,
 322                               subflow->remote_nonce, subflow->local_nonce,
 323                               hmac);
 324
 325         thmac = get_unaligned_be64(hmac);
 326         pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
 327                  subflow, subflow->token,
 328                  (unsigned long long)thmac,
 329                  (unsigned long long)subflow->thmac);
 330
 331         return thmac == subflow->thmac;
 332 }
 333
 334 void mptcp_subflow_reset(struct sock *ssk)
 335 {
 336         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 337         struct sock *sk = subflow->conn;
 338
 339         /* must hold: tcp_done() could drop last reference on parent */
 340         sock_hold(sk);
 341
 342         tcp_set_state(ssk, TCP_CLOSE);
 343         tcp_send_active_reset(ssk, GFP_ATOMIC);
 344         tcp_done(ssk);
 345         if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags) &&
 346             schedule_work(&mptcp_sk(sk)->work))
 347                 return; /* worker will put sk for us */
 348
 349         sock_put(sk);
 350 }
 351
 352 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
 353 {
 354         return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
 355 }
 356
 357 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 358 {
 359         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 360         struct mptcp_options_received mp_opt;
 361         struct sock *parent = subflow->conn;
 362
 363         subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 364
 365         if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
 366                 inet_sk_state_store(parent, TCP_ESTABLISHED);
 367                 parent->sk_state_change(parent);
 368         }
 369
 370         /* be sure no special action on any packet other than syn-ack */
 371         if (subflow->conn_finished)
 372                 return;
 373
 374         mptcp_propagate_sndbuf(parent, sk);
 375         subflow->rel_write_seq = 1;
 376         subflow->conn_finished = 1;
 377         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 378         pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
 379
 380         mptcp_get_options(skb, &mp_opt);
 381         if (subflow->request_mptcp) {
 382                 if (!mp_opt.mp_capable) {
 383                         MPTCP_INC_STATS(sock_net(sk),
 384                                         MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
 385                         mptcp_do_fallback(sk);
 386                         pr_fallback(mptcp_sk(subflow->conn));
 387                         goto fallback;
 388                 }
 389
 390                 subflow->mp_capable = 1;
 391                 subflow->can_ack = 1;
 392                 subflow->remote_key = mp_opt.sndr_key;
 393                 pr_debug("subflow=%p, remote_key=%llu", subflow,
 394                          subflow->remote_key);
 395                 mptcp_finish_connect(sk);
 396         } else if (subflow->request_join) {
 397                 u8 hmac[SHA256_DIGEST_SIZE];
 398
 399                 if (!mp_opt.mp_join)
 400                         goto do_reset;
 401
 402                 subflow->thmac = mp_opt.thmac;
 403                 subflow->remote_nonce = mp_opt.nonce;
 404                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
 405                          subflow->thmac, subflow->remote_nonce);
 406
 407                 if (!subflow_thmac_valid(subflow)) {
 408                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 409                         goto do_reset;
 410                 }
 411
 412                 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 413                                       subflow->local_nonce,
 414                                       subflow->remote_nonce,
 415                                       hmac);
 416                 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
 417
 418                 if (!mptcp_finish_join(sk))
 419                         goto do_reset;
 420
 421                 subflow->mp_join = 1;
 422                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 423
 424                 if (subflow_use_different_dport(mptcp_sk(parent), sk)) {
 425                         pr_debug("synack inet_dport=%d %d",
 426                                  ntohs(inet_sk(sk)->inet_dport),
 427                                  ntohs(inet_sk(parent)->inet_dport));
 428                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
 429                 }
 430         } else if (mptcp_check_fallback(sk)) {
 431 fallback:
 432                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
 433         }
 434         return;
 435
 436 do_reset:
 437         mptcp_subflow_reset(sk);
 438 }
 439
 440 struct request_sock_ops mptcp_subflow_request_sock_ops;
 441 EXPORT_SYMBOL_GPL(mptcp_subflow_request_sock_ops);
 442 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
 443
 444 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 445 {
 446         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 447
 448         pr_debug("subflow=%p", subflow);
 449
 450         /* Never answer to SYNs sent to broadcast or multicast */
 451         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 452                 goto drop;
 453
 454         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 455                                 &subflow_request_sock_ipv4_ops,
 456                                 sk, skb);
 457 drop:
 458         tcp_listendrop(sk);
 459         return 0;
 460 }
 461
 462 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 463 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
 464 static struct inet_connection_sock_af_ops subflow_v6_specific;
 465 static struct inet_connection_sock_af_ops subflow_v6m_specific;
 466 static struct proto tcpv6_prot_override;
 467
 468 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 469 {
 470         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 471
 472         pr_debug("subflow=%p", subflow);
 473
 474         if (skb->protocol == htons(ETH_P_IP))
 475                 return subflow_v4_conn_request(sk, skb);
 476
 477         if (!ipv6_unicast_destination(skb))
 478                 goto drop;
 479
 480         if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
 481                 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
 482                 return 0;
 483         }
 484
 485         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 486                                 &subflow_request_sock_ipv6_ops, sk, skb);
 487
 488 drop:
 489         tcp_listendrop(sk);
 490         return 0; /* don't send reset */
 491 }
 492 #endif
 493
 494 /* validate hmac received in third ACK */
 495 static bool subflow_hmac_valid(const struct request_sock *req,
 496                                const struct mptcp_options_received *mp_opt)
 497 {
 498         const struct mptcp_subflow_request_sock *subflow_req;
 499         u8 hmac[SHA256_DIGEST_SIZE];
 500         struct mptcp_sock *msk;
 501
 502         subflow_req = mptcp_subflow_rsk(req);
 503         msk = subflow_req->msk;
 504         if (!msk)
 505                 return false;
 506
 507         subflow_generate_hmac(msk->remote_key, msk->local_key,
 508                               subflow_req->remote_nonce,
 509                               subflow_req->local_nonce, hmac);
 510
 511         return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
 512 }
 513
 514 static void mptcp_sock_destruct(struct sock *sk)
 515 {
 516         /* if new mptcp socket isn't accepted, it is free'd
 517          * from the tcp listener sockets request queue, linked
 518          * from req->sk.  The tcp socket is released.
 519          * This calls the ULP release function which will
 520          * also remove the mptcp socket, via
 521          * sock_put(ctx->conn).
 522          *
 523          * Problem is that the mptcp socket will be in
 524          * ESTABLISHED state and will not have the SOCK_DEAD flag.
 525          * Both result in warnings from inet_sock_destruct.
 526          */
 527
 528         if (sk->sk_state == TCP_ESTABLISHED) {
 529                 sk->sk_state = TCP_CLOSE;
 530                 WARN_ON_ONCE(sk->sk_socket);
 531                 sock_orphan(sk);
 532         }
 533
 534         mptcp_destroy_common(mptcp_sk(sk));
 535         inet_sock_destruct(sk);
 536 }
 537
 538 static void mptcp_force_close(struct sock *sk)
 539 {
 540         inet_sk_state_store(sk, TCP_CLOSE);
 541         sk_common_release(sk);
 542 }
 543
 544 static void subflow_ulp_fallback(struct sock *sk,
 545                                  struct mptcp_subflow_context *old_ctx)
 546 {
 547         struct inet_connection_sock *icsk = inet_csk(sk);
 548
 549         mptcp_subflow_tcp_fallback(sk, old_ctx);
 550         icsk->icsk_ulp_ops = NULL;
 551         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 552         tcp_sk(sk)->is_mptcp = 0;
 553
 554         mptcp_subflow_ops_undo_override(sk);
 555 }
 556
 557 static void subflow_drop_ctx(struct sock *ssk)
 558 {
 559         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 560
 561         if (!ctx)
 562                 return;
 563
 564         subflow_ulp_fallback(ssk, ctx);
 565         if (ctx->conn)
 566                 sock_put(ctx->conn);
 567
 568         kfree_rcu(ctx, rcu);
 569 }
 570
 571 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 572                                      struct mptcp_options_received *mp_opt)
 573 {
 574         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 575
 576         subflow->remote_key = mp_opt->sndr_key;
 577         subflow->fully_established = 1;
 578         subflow->can_ack = 1;
 579         WRITE_ONCE(msk->fully_established, true);
 580 }
 581
 582 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 583                                           struct sk_buff *skb,
 584                                           struct request_sock *req,
 585                                           struct dst_entry *dst,
 586                                           struct request_sock *req_unhash,
 587                                           bool *own_req)
 588 {
 589         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 590         struct mptcp_subflow_request_sock *subflow_req;
 591         struct mptcp_options_received mp_opt;
 592         bool fallback, fallback_is_fatal;
 593         struct sock *new_msk = NULL;
 594         struct sock *child;
 595
 596         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 597
 598         /* After child creation we must look for 'mp_capable' even when options
 599          * are not parsed
 600          */
 601         mp_opt.mp_capable = 0;
 602
 603         /* hopefully temporary handling for MP_JOIN+syncookie */
 604         subflow_req = mptcp_subflow_rsk(req);
 605         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 606         fallback = !tcp_rsk(req)->is_mptcp;
 607         if (fallback)
 608                 goto create_child;
 609
 610         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 611         if (subflow_req->mp_capable) {
 612                 if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
 613                         /* here we can receive and accept an in-window,
 614                          * out-of-order pkt, which will not carry the MP_CAPABLE
 615                          * opt even on mptcp enabled paths
 616                          */
 617                         goto create_msk;
 618                 }
 619
 620                 mptcp_get_options(skb, &mp_opt);
 621                 if (!mp_opt.mp_capable) {
 622                         fallback = true;
 623                         goto create_child;
 624                 }
 625
 626 create_msk:
 627                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 628                 if (!new_msk)
 629                         fallback = true;
 630         } else if (subflow_req->mp_join) {
 631                 mptcp_get_options(skb, &mp_opt);
 632                 if (!mp_opt.mp_join || !subflow_hmac_valid(req, &mp_opt) ||
 633                     !mptcp_can_accept_new_subflow(subflow_req->msk)) {
 634                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 635                         fallback = true;
 636                 }
 637         }
 638
 639 create_child:
 640         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 641                                                      req_unhash, own_req);
 642
 643         if (child && *own_req) {
 644                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 645
 646                 tcp_rsk(req)->drop_req = false;
 647
 648                 /* we need to fallback on ctx allocation failure and on pre-reqs
 649                  * checking above. In the latter scenario we additionally need
 650                  * to reset the context to non MPTCP status.
 651                  */
 652                 if (!ctx || fallback) {
 653                         if (fallback_is_fatal)
 654                                 goto dispose_child;
 655
 656                         subflow_drop_ctx(child);
 657                         goto out;
 658                 }
 659
 660                 if (ctx->mp_capable) {
 661                         /* this can't race with mptcp_close(), as the msk is
 662                          * not yet exposted to user-space
 663                          */
 664                         inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
 665
 666                         /* record the newly created socket as the first msk
 667                          * subflow, but don't link it yet into conn_list
 668                          */
 669                         WRITE_ONCE(mptcp_sk(new_msk)->first, child);
 670
 671                         /* new mpc subflow takes ownership of the newly
 672                          * created mptcp socket
 673                          */
 674                         new_msk->sk_destruct = mptcp_sock_destruct;
 675                         mptcp_pm_new_connection(mptcp_sk(new_msk), child, 1);
 676                         mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
 677                         ctx->conn = new_msk;
 678                         new_msk = NULL;
 679
 680                         /* with OoO packets we can reach here without ingress
 681                          * mpc option
 682                          */
 683                         if (mp_opt.mp_capable)
 684                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 685                 } else if (ctx->mp_join) {
 686                         struct mptcp_sock *owner;
 687
 688                         owner = subflow_req->msk;
 689                         if (!owner)
 690                                 goto dispose_child;
 691
 692                         /* move the msk reference ownership to the subflow */
 693                         subflow_req->msk = NULL;
 694                         ctx->conn = (struct sock *)owner;
 695
 696                         if (subflow_use_different_sport(owner, sk)) {
 697                                 pr_debug("ack inet_sport=%d %d",
 698                                          ntohs(inet_sk(sk)->inet_sport),
 699                                          ntohs(inet_sk((struct sock *)owner)->inet_sport));
 700                                 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
 701                                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
 702                                         goto dispose_child;
 703                                 }
 704                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
 705                         }
 706
 707                         if (!mptcp_finish_join(child))
 708                                 goto dispose_child;
 709
 710                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 711                         tcp_rsk(req)->drop_req = true;
 712                 }
 713         }
 714
 715 out:
 716         /* dispose of the left over mptcp master, if any */
 717         if (unlikely(new_msk))
 718                 mptcp_force_close(new_msk);
 719
 720         /* check for expected invariant - should never trigger, just help
 721          * catching eariler subtle bugs
 722          */
 723         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 724                      (!mptcp_subflow_ctx(child) ||
 725                       !mptcp_subflow_ctx(child)->conn));
 726         return child;
 727
 728 dispose_child:
 729         subflow_drop_ctx(child);
 730         tcp_rsk(req)->drop_req = true;
 731         inet_csk_prepare_for_destroy_sock(child);
 732         tcp_done(child);
 733         req->rsk_ops->send_reset(sk, skb);
 734
 735         /* The last child reference will be released by the caller */
 736         return child;
 737 }
 738
 739 static struct inet_connection_sock_af_ops subflow_specific;
 740 static struct proto tcp_prot_override;
 741
 742 enum mapping_status {
 743         MAPPING_OK,
 744         MAPPING_INVALID,
 745         MAPPING_EMPTY,
 746         MAPPING_DATA_FIN,
 747         MAPPING_DUMMY
 748 };
 749
 750 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
 751 {
 752         if ((u32)seq == (u32)old_seq)
 753                 return old_seq;
 754
 755         /* Assume map covers data not mapped yet. */
 756         return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
 757 }
 758
 759 static void warn_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 760 {
 761         WARN_ONCE(1, "Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 762                   ssn, subflow->map_subflow_seq, subflow->map_data_len);
 763 }
 764
 765 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 766 {
 767         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 768         unsigned int skb_consumed;
 769
 770         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 771         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 772                 return true;
 773
 774         return skb->len - skb_consumed <= subflow->map_data_len -
 775                                           mptcp_subflow_get_map_offset(subflow);
 776 }
 777
 778 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 779 {
 780         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 781         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 782
 783         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 784                 /* Mapping covers data later in the subflow stream,
 785                  * currently unsupported.
 786                  */
 787                 warn_bad_map(subflow, ssn);
 788                 return false;
 789         }
 790         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 791                                   subflow->map_data_len))) {
 792                 /* Mapping does covers past subflow data, invalid */
 793                 warn_bad_map(subflow, ssn + skb->len);
 794                 return false;
 795         }
 796         return true;
 797 }
 798
 799 static enum mapping_status get_mapping_status(struct sock *ssk,
 800                                               struct mptcp_sock *msk)
 801 {
 802         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 803         struct mptcp_ext *mpext;
 804         struct sk_buff *skb;
 805         u16 data_len;
 806         u64 map_seq;
 807
 808         skb = skb_peek(&ssk->sk_receive_queue);
 809         if (!skb)
 810                 return MAPPING_EMPTY;
 811
 812         if (mptcp_check_fallback(ssk))
 813                 return MAPPING_DUMMY;
 814
 815         mpext = mptcp_get_ext(skb);
 816         if (!mpext || !mpext->use_map) {
 817                 if (!subflow->map_valid && !skb->len) {
 818                         /* the TCP stack deliver 0 len FIN pkt to the receive
 819                          * queue, that is the only 0len pkts ever expected here,
 820                          * and we can admit no mapping only for 0 len pkts
 821                          */
 822                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 823                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 824                                           TCP_SKB_CB(skb)->seq,
 825                                           TCP_SKB_CB(skb)->end_seq,
 826                                           TCP_SKB_CB(skb)->tcp_flags);
 827                         sk_eat_skb(ssk, skb);
 828                         return MAPPING_EMPTY;
 829                 }
 830
 831                 if (!subflow->map_valid)
 832                         return MAPPING_INVALID;
 833
 834                 goto validate_seq;
 835         }
 836
 837         pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
 838                  mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
 839                  mpext->data_len, mpext->data_fin);
 840
 841         data_len = mpext->data_len;
 842         if (data_len == 0) {
 843                 pr_err("Infinite mapping not handled");
 844                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 845                 return MAPPING_INVALID;
 846         }
 847
 848         if (mpext->data_fin == 1) {
 849                 if (data_len == 1) {
 850                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
 851                                                                  mpext->dsn64);
 852                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
 853                         if (subflow->map_valid) {
 854                                 /* A DATA_FIN might arrive in a DSS
 855                                  * option before the previous mapping
 856                                  * has been fully consumed. Continue
 857                                  * handling the existing mapping.
 858                                  */
 859                                 skb_ext_del(skb, SKB_EXT_MPTCP);
 860                                 return MAPPING_OK;
 861                         } else {
 862                                 if (updated && schedule_work(&msk->work))
 863                                         sock_hold((struct sock *)msk);
 864
 865                                 return MAPPING_DATA_FIN;
 866                         }
 867                 } else {
 868                         u64 data_fin_seq = mpext->data_seq + data_len - 1;
 869
 870                         /* If mpext->data_seq is a 32-bit value, data_fin_seq
 871                          * must also be limited to 32 bits.
 872                          */
 873                         if (!mpext->dsn64)
 874                                 data_fin_seq &= GENMASK_ULL(31, 0);
 875
 876                         mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
 877                         pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
 878                                  data_fin_seq, mpext->dsn64);
 879                 }
 880
 881                 /* Adjust for DATA_FIN using 1 byte of sequence space */
 882                 data_len--;
 883         }
 884
 885         if (!mpext->dsn64) {
 886                 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
 887                                      mpext->data_seq);
 888                 pr_debug("expanded seq=%llu", subflow->map_seq);
 889         } else {
 890                 map_seq = mpext->data_seq;
 891         }
 892         WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
 893
 894         if (subflow->map_valid) {
 895                 /* Allow replacing only with an identical map */
 896                 if (subflow->map_seq == map_seq &&
 897                     subflow->map_subflow_seq == mpext->subflow_seq &&
 898                     subflow->map_data_len == data_len) {
 899                         skb_ext_del(skb, SKB_EXT_MPTCP);
 900                         return MAPPING_OK;
 901                 }
 902
 903                 /* If this skb data are fully covered by the current mapping,
 904                  * the new map would need caching, which is not supported
 905                  */
 906                 if (skb_is_fully_mapped(ssk, skb)) {
 907                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
 908                         return MAPPING_INVALID;
 909                 }
 910
 911                 /* will validate the next map after consuming the current one */
 912                 return MAPPING_OK;
 913         }
 914
 915         subflow->map_seq = map_seq;
 916         subflow->map_subflow_seq = mpext->subflow_seq;
 917         subflow->map_data_len = data_len;
 918         subflow->map_valid = 1;
 919         subflow->mpc_map = mpext->mpc_map;
 920         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
 921                  subflow->map_seq, subflow->map_subflow_seq,
 922                  subflow->map_data_len);
 923
 924 validate_seq:
 925         /* we revalidate valid mapping on new skb, because we must ensure
 926          * the current skb is completely covered by the available mapping
 927          */
 928         if (!validate_mapping(ssk, skb))
 929                 return MAPPING_INVALID;
 930
 931         skb_ext_del(skb, SKB_EXT_MPTCP);
 932         return MAPPING_OK;
 933 }
 934
 935 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
 936                                        u64 limit)
 937 {
 938         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 939         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
 940         u32 incr;
 941
 942         incr = limit >= skb->len ? skb->len + fin : limit;
 943
 944         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
 945                  subflow->map_subflow_seq);
 946         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
 947         tcp_sk(ssk)->copied_seq += incr;
 948         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
 949                 sk_eat_skb(ssk, skb);
 950         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
 951                 subflow->map_valid = 0;
 952 }
 953
 954 /* sched mptcp worker to remove the subflow if no more data is pending */
 955 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
 956 {
 957         struct sock *sk = (struct sock *)msk;
 958
 959         if (likely(ssk->sk_state != TCP_CLOSE))
 960                 return;
 961
 962         if (skb_queue_empty(&ssk->sk_receive_queue) &&
 963             !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags)) {
 964                 sock_hold(sk);
 965                 if (!schedule_work(&msk->work))
 966                         sock_put(sk);
 967         }
 968 }
 969
 970 static bool subflow_check_data_avail(struct sock *ssk)
 971 {
 972         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 973         enum mapping_status status;
 974         struct mptcp_sock *msk;
 975         struct sk_buff *skb;
 976
 977         pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
 978                  subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
 979         if (!skb_peek(&ssk->sk_receive_queue))
 980                 subflow->data_avail = 0;
 981         if (subflow->data_avail)
 982                 return true;
 983
 984         msk = mptcp_sk(subflow->conn);
 985         for (;;) {
 986                 u64 ack_seq;
 987                 u64 old_ack;
 988
 989                 status = get_mapping_status(ssk, msk);
 990                 pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
 991                 if (status == MAPPING_INVALID) {
 992                         ssk->sk_err = EBADMSG;
 993                         goto fatal;
 994                 }
 995                 if (status == MAPPING_DUMMY) {
 996                         __mptcp_do_fallback(msk);
 997                         skb = skb_peek(&ssk->sk_receive_queue);
 998                         subflow->map_valid = 1;
 999                         subflow->map_seq = READ_ONCE(msk->ack_seq);
1000                         subflow->map_data_len = skb->len;
1001                         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq -
1002                                                    subflow->ssn_offset;
1003                         subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
1004                         return true;
1005                 }
1006
1007                 if (status != MAPPING_OK)
1008                         goto no_data;
1009
1010                 skb = skb_peek(&ssk->sk_receive_queue);
1011                 if (WARN_ON_ONCE(!skb))
1012                         goto no_data;
1013
1014                 /* if msk lacks the remote key, this subflow must provide an
1015                  * MP_CAPABLE-based mapping
1016                  */
1017                 if (unlikely(!READ_ONCE(msk->can_ack))) {
1018                         if (!subflow->mpc_map) {
1019                                 ssk->sk_err = EBADMSG;
1020                                 goto fatal;
1021                         }
1022                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
1023                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
1024                         WRITE_ONCE(msk->can_ack, true);
1025                 }
1026
1027                 old_ack = READ_ONCE(msk->ack_seq);
1028                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1029                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1030                          ack_seq);
1031                 if (ack_seq == old_ack) {
1032                         subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
1033                         break;
1034                 } else if (after64(ack_seq, old_ack)) {
1035                         subflow->data_avail = MPTCP_SUBFLOW_OOO_DATA;
1036                         break;
1037                 }
1038
1039                 /* only accept in-sequence mapping. Old values are spurious
1040                  * retransmission
1041                  */
1042                 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1043         }
1044         return true;
1045
1046 no_data:
1047         subflow_sched_work_if_closed(msk, ssk);
1048         return false;
1049 fatal:
1050         /* fatal protocol error, close the socket */
1051         /* This barrier is coupled with smp_rmb() in tcp_poll() */
1052         smp_wmb();
1053         ssk->sk_error_report(ssk);
1054         tcp_set_state(ssk, TCP_CLOSE);
1055         tcp_send_active_reset(ssk, GFP_ATOMIC);
1056         subflow->data_avail = 0;
1057         return false;
1058 }
1059
1060 bool mptcp_subflow_data_available(struct sock *sk)
1061 {
1062         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1063
1064         /* check if current mapping is still valid */
1065         if (subflow->map_valid &&
1066             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1067                 subflow->map_valid = 0;
1068                 subflow->data_avail = 0;
1069
1070                 pr_debug("Done with mapping: seq=%u data_len=%u",
1071                          subflow->map_subflow_seq,
1072                          subflow->map_data_len);
1073         }
1074
1075         return subflow_check_data_avail(sk);
1076 }
1077
1078 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1079  * not the ssk one.
1080  *
1081  * In mptcp, rwin is about the mptcp-level connection data.
1082  *
1083  * Data that is still on the ssk rx queue can thus be ignored,
1084  * as far as mptcp peer is concerened that data is still inflight.
1085  * DSS ACK is updated when skb is moved to the mptcp rx queue.
1086  */
1087 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1088 {
1089         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1090         const struct sock *sk = subflow->conn;
1091
1092         *space = __mptcp_space(sk);
1093         *full_space = tcp_full_space(sk);
1094 }
1095
1096 static void subflow_data_ready(struct sock *sk)
1097 {
1098         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1099         u16 state = 1 << inet_sk_state_load(sk);
1100         struct sock *parent = subflow->conn;
1101         struct mptcp_sock *msk;
1102
1103         msk = mptcp_sk(parent);
1104         if (state & TCPF_LISTEN) {
1105                 /* MPJ subflow are removed from accept queue before reaching here,
1106                  * avoid stray wakeups
1107                  */
1108                 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1109                         return;
1110
1111                 set_bit(MPTCP_DATA_READY, &msk->flags);
1112                 parent->sk_data_ready(parent);
1113                 return;
1114         }
1115
1116         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1117                      !subflow->mp_join && !(state & TCPF_CLOSE));
1118
1119         if (mptcp_subflow_data_available(sk))
1120                 mptcp_data_ready(parent, sk);
1121 }
1122
1123 static void subflow_write_space(struct sock *ssk)
1124 {
1125         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1126
1127         mptcp_propagate_sndbuf(sk, ssk);
1128         mptcp_write_space(sk);
1129 }
1130
1131 void __mptcp_error_report(struct sock *sk)
1132 {
1133         struct mptcp_subflow_context *subflow;
1134         struct mptcp_sock *msk = mptcp_sk(sk);
1135
1136         mptcp_for_each_subflow(msk, subflow) {
1137                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1138                 int err = sock_error(ssk);
1139
1140                 if (!err)
1141                         continue;
1142
1143                 /* only propagate errors on fallen-back sockets or
1144                  * on MPC connect
1145                  */
1146                 if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(msk))
1147                         continue;
1148
1149                 inet_sk_state_store(sk, inet_sk_state_load(ssk));
1150                 sk->sk_err = -err;
1151
1152                 /* This barrier is coupled with smp_rmb() in mptcp_poll() */
1153                 smp_wmb();
1154                 sk->sk_error_report(sk);
1155                 break;
1156         }
1157 }
1158
1159 static void subflow_error_report(struct sock *ssk)
1160 {
1161         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1162
1163         mptcp_data_lock(sk);
1164         if (!sock_owned_by_user(sk))
1165                 __mptcp_error_report(sk);
1166         else
1167                 set_bit(MPTCP_ERROR_REPORT,  &mptcp_sk(sk)->flags);
1168         mptcp_data_unlock(sk);
1169 }
1170
1171 static struct inet_connection_sock_af_ops *
1172 subflow_default_af_ops(struct sock *sk)
1173 {
1174 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1175         if (sk->sk_family == AF_INET6)
1176                 return &subflow_v6_specific;
1177 #endif
1178         return &subflow_specific;
1179 }
1180
1181 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1182 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1183 {
1184         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1185         struct inet_connection_sock *icsk = inet_csk(sk);
1186         struct inet_connection_sock_af_ops *target;
1187
1188         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1189
1190         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1191                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1192
1193         if (likely(icsk->icsk_af_ops == target))
1194                 return;
1195
1196         subflow->icsk_af_ops = icsk->icsk_af_ops;
1197         icsk->icsk_af_ops = target;
1198 }
1199 #endif
1200
1201 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1202                          struct sockaddr_storage *addr,
1203                          unsigned short family)
1204 {
1205         memset(addr, 0, sizeof(*addr));
1206         addr->ss_family = family;
1207         if (addr->ss_family == AF_INET) {
1208                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1209
1210                 if (info->family == AF_INET)
1211                         in_addr->sin_addr = info->addr;
1212 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1213                 else if (ipv6_addr_v4mapped(&info->addr6))
1214                         in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1215 #endif
1216                 in_addr->sin_port = info->port;
1217         }
1218 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1219         else if (addr->ss_family == AF_INET6) {
1220                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1221
1222                 if (info->family == AF_INET)
1223                         ipv6_addr_set_v4mapped(info->addr.s_addr,
1224                                                &in6_addr->sin6_addr);
1225                 else
1226                         in6_addr->sin6_addr = info->addr6;
1227                 in6_addr->sin6_port = info->port;
1228         }
1229 #endif
1230 }
1231
1232 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1233                             const struct mptcp_addr_info *remote)
1234 {
1235         struct mptcp_sock *msk = mptcp_sk(sk);
1236         struct mptcp_subflow_context *subflow;
1237         struct sockaddr_storage addr;
1238         int remote_id = remote->id;
1239         int local_id = loc->id;
1240         struct socket *sf;
1241         struct sock *ssk;
1242         u32 remote_token;
1243         int addrlen;
1244         int err;
1245
1246         if (!mptcp_is_fully_established(sk))
1247                 return -ENOTCONN;
1248
1249         err = mptcp_subflow_create_socket(sk, &sf);
1250         if (err)
1251                 return err;
1252
1253         ssk = sf->sk;
1254         subflow = mptcp_subflow_ctx(ssk);
1255         do {
1256                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1257         } while (!subflow->local_nonce);
1258
1259         if (!local_id) {
1260                 err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
1261                 if (err < 0)
1262                         goto failed;
1263
1264                 local_id = err;
1265         }
1266
1267         subflow->remote_key = msk->remote_key;
1268         subflow->local_key = msk->local_key;
1269         subflow->token = msk->token;
1270         mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1271
1272         addrlen = sizeof(struct sockaddr_in);
1273 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1274         if (addr.ss_family == AF_INET6)
1275                 addrlen = sizeof(struct sockaddr_in6);
1276 #endif
1277         ssk->sk_bound_dev_if = loc->ifindex;
1278         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1279         if (err)
1280                 goto failed;
1281
1282         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1283         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1284                  remote_token, local_id, remote_id);
1285         subflow->remote_token = remote_token;
1286         subflow->local_id = local_id;
1287         subflow->remote_id = remote_id;
1288         subflow->request_join = 1;
1289         subflow->request_bkup = !!(loc->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1290         mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1291
1292         mptcp_add_pending_subflow(msk, subflow);
1293         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1294         if (err && err != -EINPROGRESS)
1295                 goto failed_unlink;
1296
1297         /* discard the subflow socket */
1298         mptcp_sock_graft(ssk, sk->sk_socket);
1299         iput(SOCK_INODE(sf));
1300         return err;
1301
1302 failed_unlink:
1303         spin_lock_bh(&msk->join_list_lock);
1304         list_del(&subflow->node);
1305         spin_unlock_bh(&msk->join_list_lock);
1306         sock_put(mptcp_subflow_tcp_sock(subflow));
1307
1308 failed:
1309         subflow->disposable = 1;
1310         sock_release(sf);
1311         return err;
1312 }
1313
1314 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1315 {
1316 #ifdef CONFIG_SOCK_CGROUP_DATA
1317         struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1318                                 *child_skcd = &child->sk_cgrp_data;
1319
1320         /* only the additional subflows created by kworkers have to be modified */
1321         if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1322             cgroup_id(sock_cgroup_ptr(child_skcd))) {
1323 #ifdef CONFIG_MEMCG
1324                 struct mem_cgroup *memcg = parent->sk_memcg;
1325
1326                 mem_cgroup_sk_free(child);
1327                 if (memcg && css_tryget(&memcg->css))
1328                         child->sk_memcg = memcg;
1329 #endif /* CONFIG_MEMCG */
1330
1331                 cgroup_sk_free(child_skcd);
1332                 *child_skcd = *parent_skcd;
1333                 cgroup_sk_clone(child_skcd);
1334         }
1335 #endif /* CONFIG_SOCK_CGROUP_DATA */
1336 }
1337
1338 static void mptcp_subflow_ops_override(struct sock *ssk)
1339 {
1340 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1341         if (ssk->sk_prot == &tcpv6_prot)
1342                 ssk->sk_prot = &tcpv6_prot_override;
1343         else
1344 #endif
1345                 ssk->sk_prot = &tcp_prot_override;
1346 }
1347
1348 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1349 {
1350 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1351         if (ssk->sk_prot == &tcpv6_prot_override)
1352                 ssk->sk_prot = &tcpv6_prot;
1353         else
1354 #endif
1355                 ssk->sk_prot = &tcp_prot;
1356 }
1357 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1358 {
1359         struct mptcp_subflow_context *subflow;
1360         struct net *net = sock_net(sk);
1361         struct socket *sf;
1362         int err;
1363
1364         /* un-accepted server sockets can reach here - on bad configuration
1365          * bail early to avoid greater trouble later
1366          */
1367         if (unlikely(!sk->sk_socket))
1368                 return -EINVAL;
1369
1370         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1371                                &sf);
1372         if (err)
1373                 return err;
1374
1375         lock_sock(sf->sk);
1376
1377         /* the newly created socket has to be in the same cgroup as its parent */
1378         mptcp_attach_cgroup(sk, sf->sk);
1379
1380         /* kernel sockets do not by default acquire net ref, but TCP timer
1381          * needs it.
1382          */
1383         sf->sk->sk_net_refcnt = 1;
1384         get_net(net);
1385 #ifdef CONFIG_PROC_FS
1386         this_cpu_add(*net->core.sock_inuse, 1);
1387 #endif
1388         err = tcp_set_ulp(sf->sk, "mptcp");
1389         release_sock(sf->sk);
1390
1391         if (err) {
1392                 sock_release(sf);
1393                 return err;
1394         }
1395
1396         /* the newly created socket really belongs to the owning MPTCP master
1397          * socket, even if for additional subflows the allocation is performed
1398          * by a kernel workqueue. Adjust inode references, so that the
1399          * procfs/diag interaces really show this one belonging to the correct
1400          * user.
1401          */
1402         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1403         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1404         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1405
1406         subflow = mptcp_subflow_ctx(sf->sk);
1407         pr_debug("subflow=%p", subflow);
1408
1409         *new_sock = sf;
1410         sock_hold(sk);
1411         subflow->conn = sk;
1412         mptcp_subflow_ops_override(sf->sk);
1413
1414         return 0;
1415 }
1416
1417 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1418                                                         gfp_t priority)
1419 {
1420         struct inet_connection_sock *icsk = inet_csk(sk);
1421         struct mptcp_subflow_context *ctx;
1422
1423         ctx = kzalloc(sizeof(*ctx), priority);
1424         if (!ctx)
1425                 return NULL;
1426
1427         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1428         INIT_LIST_HEAD(&ctx->node);
1429         INIT_LIST_HEAD(&ctx->delegated_node);
1430
1431         pr_debug("subflow=%p", ctx);
1432
1433         ctx->tcp_sock = sk;
1434
1435         return ctx;
1436 }
1437
1438 static void __subflow_state_change(struct sock *sk)
1439 {
1440         struct socket_wq *wq;
1441
1442         rcu_read_lock();
1443         wq = rcu_dereference(sk->sk_wq);
1444         if (skwq_has_sleeper(wq))
1445                 wake_up_interruptible_all(&wq->wait);
1446         rcu_read_unlock();
1447 }
1448
1449 static bool subflow_is_done(const struct sock *sk)
1450 {
1451         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1452 }
1453
1454 static void subflow_state_change(struct sock *sk)
1455 {
1456         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1457         struct sock *parent = subflow->conn;
1458
1459         __subflow_state_change(sk);
1460
1461         if (subflow_simultaneous_connect(sk)) {
1462                 mptcp_propagate_sndbuf(parent, sk);
1463                 mptcp_do_fallback(sk);
1464                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1465                 pr_fallback(mptcp_sk(parent));
1466                 subflow->conn_finished = 1;
1467                 if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
1468                         inet_sk_state_store(parent, TCP_ESTABLISHED);
1469                         parent->sk_state_change(parent);
1470                 }
1471         }
1472
1473         /* as recvmsg() does not acquire the subflow socket for ssk selection
1474          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1475          * the data available machinery here.
1476          */
1477         if (mptcp_subflow_data_available(sk))
1478                 mptcp_data_ready(parent, sk);
1479
1480         subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1481
1482         if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1483             !subflow->rx_eof && subflow_is_done(sk)) {
1484                 subflow->rx_eof = 1;
1485                 mptcp_subflow_eof(parent);
1486         }
1487 }
1488
1489 static int subflow_ulp_init(struct sock *sk)
1490 {
1491         struct inet_connection_sock *icsk = inet_csk(sk);
1492         struct mptcp_subflow_context *ctx;
1493         struct tcp_sock *tp = tcp_sk(sk);
1494         int err = 0;
1495
1496         /* disallow attaching ULP to a socket unless it has been
1497          * created with sock_create_kern()
1498          */
1499         if (!sk->sk_kern_sock) {
1500                 err = -EOPNOTSUPP;
1501                 goto out;
1502         }
1503
1504         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1505         if (!ctx) {
1506                 err = -ENOMEM;
1507                 goto out;
1508         }
1509
1510         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1511
1512         tp->is_mptcp = 1;
1513         ctx->icsk_af_ops = icsk->icsk_af_ops;
1514         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1515         ctx->tcp_data_ready = sk->sk_data_ready;
1516         ctx->tcp_state_change = sk->sk_state_change;
1517         ctx->tcp_write_space = sk->sk_write_space;
1518         ctx->tcp_error_report = sk->sk_error_report;
1519         sk->sk_data_ready = subflow_data_ready;
1520         sk->sk_write_space = subflow_write_space;
1521         sk->sk_state_change = subflow_state_change;
1522         sk->sk_error_report = subflow_error_report;
1523 out:
1524         return err;
1525 }
1526
1527 static void subflow_ulp_release(struct sock *ssk)
1528 {
1529         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1530         bool release = true;
1531         struct sock *sk;
1532
1533         if (!ctx)
1534                 return;
1535
1536         sk = ctx->conn;
1537         if (sk) {
1538                 /* if the msk has been orphaned, keep the ctx
1539                  * alive, will be freed by __mptcp_close_ssk(),
1540                  * when the subflow is still unaccepted
1541                  */
1542                 release = ctx->disposable || list_empty(&ctx->node);
1543                 sock_put(sk);
1544         }
1545
1546         mptcp_subflow_ops_undo_override(ssk);
1547         if (release)
1548                 kfree_rcu(ctx, rcu);
1549 }
1550
1551 static void subflow_ulp_clone(const struct request_sock *req,
1552                               struct sock *newsk,
1553                               const gfp_t priority)
1554 {
1555         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1556         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1557         struct mptcp_subflow_context *new_ctx;
1558
1559         if (!tcp_rsk(req)->is_mptcp ||
1560             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1561                 subflow_ulp_fallback(newsk, old_ctx);
1562                 return;
1563         }
1564
1565         new_ctx = subflow_create_ctx(newsk, priority);
1566         if (!new_ctx) {
1567                 subflow_ulp_fallback(newsk, old_ctx);
1568                 return;
1569         }
1570
1571         new_ctx->conn_finished = 1;
1572         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1573         new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1574         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1575         new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1576         new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1577         new_ctx->rel_write_seq = 1;
1578         new_ctx->tcp_sock = newsk;
1579
1580         if (subflow_req->mp_capable) {
1581                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1582                  * is fully established only after we receive the remote key
1583                  */
1584                 new_ctx->mp_capable = 1;
1585                 new_ctx->local_key = subflow_req->local_key;
1586                 new_ctx->token = subflow_req->token;
1587                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1588                 new_ctx->idsn = subflow_req->idsn;
1589         } else if (subflow_req->mp_join) {
1590                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1591                 new_ctx->mp_join = 1;
1592                 new_ctx->fully_established = 1;
1593                 new_ctx->backup = subflow_req->backup;
1594                 new_ctx->local_id = subflow_req->local_id;
1595                 new_ctx->remote_id = subflow_req->remote_id;
1596                 new_ctx->token = subflow_req->token;
1597                 new_ctx->thmac = subflow_req->thmac;
1598         }
1599 }
1600
1601 static void tcp_release_cb_override(struct sock *ssk)
1602 {
1603         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1604
1605         if (mptcp_subflow_has_delegated_action(subflow))
1606                 mptcp_subflow_process_delegated(ssk);
1607
1608         tcp_release_cb(ssk);
1609 }
1610
1611 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1612         .name           = "mptcp",
1613         .owner          = THIS_MODULE,
1614         .init           = subflow_ulp_init,
1615         .release        = subflow_ulp_release,
1616         .clone          = subflow_ulp_clone,
1617 };
1618
1619 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1620 {
1621         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1622         subflow_ops->slab_name = "request_sock_subflow";
1623
1624         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1625                                               subflow_ops->obj_size, 0,
1626                                               SLAB_ACCOUNT |
1627                                               SLAB_TYPESAFE_BY_RCU,
1628                                               NULL);
1629         if (!subflow_ops->slab)
1630                 return -ENOMEM;
1631
1632         subflow_ops->destructor = subflow_req_destructor;
1633
1634         return 0;
1635 }
1636
1637 void __init mptcp_subflow_init(void)
1638 {
1639         mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1640         if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1641                 panic("MPTCP: failed to init subflow request sock ops\n");
1642
1643         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1644         subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
1645
1646         subflow_specific = ipv4_specific;
1647         subflow_specific.conn_request = subflow_v4_conn_request;
1648         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1649         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1650
1651         tcp_prot_override = tcp_prot;
1652         tcp_prot_override.release_cb = tcp_release_cb_override;
1653
1654 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1655         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1656         subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
1657
1658         subflow_v6_specific = ipv6_specific;
1659         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1660         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1661         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1662
1663         subflow_v6m_specific = subflow_v6_specific;
1664         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1665         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1666         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1667         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1668         subflow_v6m_specific.net_frag_header_len = 0;
1669
1670         tcpv6_prot_override = tcpv6_prot;
1671         tcpv6_prot_override.release_cb = tcp_release_cb_override;
1672 #endif
1673
1674         mptcp_diag_subflow_init(&subflow_ulp_ops);
1675
1676         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1677                 panic("MPTCP: failed to register subflows to ULP\n");
1678 }