net/mptcp/subflow.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Multipath TCP
   3  *
   4  * Copyright (c) 2017 - 2019, Intel Corporation.
   5  */
   6
   7 #define pr_fmt(fmt) "MPTCP: " fmt
   8
   9 #include <linux/kernel.h>
  10 #include <linux/module.h>
  11 #include <linux/netdevice.h>
  12 #include <crypto/algapi.h>
  13 #include <crypto/sha.h>
  14 #include <net/sock.h>
  15 #include <net/inet_common.h>
  16 #include <net/inet_hashtables.h>
  17 #include <net/protocol.h>
  18 #include <net/tcp.h>
  19 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
  20 #include <net/ip6_route.h>
  21 #endif
  22 #include <net/mptcp.h>
  23 #include <uapi/linux/mptcp.h>
  24 #include "protocol.h"
  25 #include "mib.h"
  26
  27 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
  28                                   enum linux_mptcp_mib_field field)
  29 {
  30         MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
  31 }
  32
  33 static void subflow_req_destructor(struct request_sock *req)
  34 {
  35         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  36
  37         pr_debug("subflow_req=%p", subflow_req);
  38
  39         if (subflow_req->msk)
  40                 sock_put((struct sock *)subflow_req->msk);
  41
  42         mptcp_token_destroy_request(req);
  43         tcp_request_sock_ops.destructor(req);
  44 }
  45
  46 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
  47                                   void *hmac)
  48 {
  49         u8 msg[8];
  50
  51         put_unaligned_be32(nonce1, &msg[0]);
  52         put_unaligned_be32(nonce2, &msg[4]);
  53
  54         mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
  55 }
  56
  57 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
  58 {
  59         return mptcp_is_fully_established((void *)msk) &&
  60                READ_ONCE(msk->pm.accept_subflow);
  61 }
  62
  63 /* validate received token and create truncated hmac and nonce for SYN-ACK */
  64 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req,
  65                                                      const struct sk_buff *skb)
  66 {
  67         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  68         u8 hmac[SHA256_DIGEST_SIZE];
  69         struct mptcp_sock *msk;
  70         int local_id;
  71
  72         msk = mptcp_token_get_sock(sock_net(req_to_sk(req)), subflow_req->token);
  73         if (!msk) {
  74                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
  75                 return NULL;
  76         }
  77
  78         local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
  79         if (local_id < 0) {
  80                 sock_put((struct sock *)msk);
  81                 return NULL;
  82         }
  83         subflow_req->local_id = local_id;
  84
  85         get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
  86
  87         subflow_generate_hmac(msk->local_key, msk->remote_key,
  88                               subflow_req->local_nonce,
  89                               subflow_req->remote_nonce, hmac);
  90
  91         subflow_req->thmac = get_unaligned_be64(hmac);
  92         return msk;
  93 }
  94
  95 static int __subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
  96 {
  97         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  98
  99         subflow_req->mp_capable = 0;
 100         subflow_req->mp_join = 0;
 101         subflow_req->msk = NULL;
 102         mptcp_token_init_request(req);
 103
 104 #ifdef CONFIG_TCP_MD5SIG
 105         /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
 106          * TCP option space.
 107          */
 108         if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
 109                 return -EINVAL;
 110 #endif
 111
 112         return 0;
 113 }
 114
 115 static void subflow_init_req(struct request_sock *req,
 116                              const struct sock *sk_listener,
 117                              struct sk_buff *skb)
 118 {
 119         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 120         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 121         struct mptcp_options_received mp_opt;
 122         int ret;
 123
 124         pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
 125
 126         ret = __subflow_init_req(req, sk_listener);
 127         if (ret)
 128                 return;
 129
 130         mptcp_get_options(skb, &mp_opt);
 131
 132         if (mp_opt.mp_capable) {
 133                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
 134
 135                 if (mp_opt.mp_join)
 136                         return;
 137         } else if (mp_opt.mp_join) {
 138                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
 139         }
 140
 141         if (mp_opt.mp_capable && listener->request_mptcp) {
 142                 int err, retries = 4;
 143
 144                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 145 again:
 146                 do {
 147                         get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
 148                 } while (subflow_req->local_key == 0);
 149
 150                 if (unlikely(req->syncookie)) {
 151                         mptcp_crypto_key_sha(subflow_req->local_key,
 152                                              &subflow_req->token,
 153                                              &subflow_req->idsn);
 154                         if (mptcp_token_exists(subflow_req->token)) {
 155                                 if (retries-- > 0)
 156                                         goto again;
 157                         } else {
 158                                 subflow_req->mp_capable = 1;
 159                         }
 160                         return;
 161                 }
 162
 163                 err = mptcp_token_new_request(req);
 164                 if (err == 0)
 165                         subflow_req->mp_capable = 1;
 166                 else if (retries-- > 0)
 167                         goto again;
 168
 169         } else if (mp_opt.mp_join && listener->request_mptcp) {
 170                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 171                 subflow_req->mp_join = 1;
 172                 subflow_req->backup = mp_opt.backup;
 173                 subflow_req->remote_id = mp_opt.join_id;
 174                 subflow_req->token = mp_opt.token;
 175                 subflow_req->remote_nonce = mp_opt.nonce;
 176                 subflow_req->msk = subflow_token_join_request(req, skb);
 177
 178                 if (unlikely(req->syncookie) && subflow_req->msk) {
 179                         if (mptcp_can_accept_new_subflow(subflow_req->msk))
 180                                 subflow_init_req_cookie_join_save(subflow_req, skb);
 181                 }
 182
 183                 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
 184                          subflow_req->remote_nonce, subflow_req->msk);
 185         }
 186 }
 187
 188 int mptcp_subflow_init_cookie_req(struct request_sock *req,
 189                                   const struct sock *sk_listener,
 190                                   struct sk_buff *skb)
 191 {
 192         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 193         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 194         struct mptcp_options_received mp_opt;
 195         int err;
 196
 197         err = __subflow_init_req(req, sk_listener);
 198         if (err)
 199                 return err;
 200
 201         mptcp_get_options(skb, &mp_opt);
 202
 203         if (mp_opt.mp_capable && mp_opt.mp_join)
 204                 return -EINVAL;
 205
 206         if (mp_opt.mp_capable && listener->request_mptcp) {
 207                 if (mp_opt.sndr_key == 0)
 208                         return -EINVAL;
 209
 210                 subflow_req->local_key = mp_opt.rcvr_key;
 211                 err = mptcp_token_new_request(req);
 212                 if (err)
 213                         return err;
 214
 215                 subflow_req->mp_capable = 1;
 216                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 217         } else if (mp_opt.mp_join && listener->request_mptcp) {
 218                 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
 219                         return -EINVAL;
 220
 221                 if (mptcp_can_accept_new_subflow(subflow_req->msk))
 222                         subflow_req->mp_join = 1;
 223
 224                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 225         }
 226
 227         return 0;
 228 }
 229 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
 230
 231 static void subflow_v4_init_req(struct request_sock *req,
 232                                 const struct sock *sk_listener,
 233                                 struct sk_buff *skb)
 234 {
 235         tcp_rsk(req)->is_mptcp = 1;
 236
 237         tcp_request_sock_ipv4_ops.init_req(req, sk_listener, skb);
 238
 239         subflow_init_req(req, sk_listener, skb);
 240 }
 241
 242 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 243 static void subflow_v6_init_req(struct request_sock *req,
 244                                 const struct sock *sk_listener,
 245                                 struct sk_buff *skb)
 246 {
 247         tcp_rsk(req)->is_mptcp = 1;
 248
 249         tcp_request_sock_ipv6_ops.init_req(req, sk_listener, skb);
 250
 251         subflow_init_req(req, sk_listener, skb);
 252 }
 253 #endif
 254
 255 /* validate received truncated hmac and create hmac for third ACK */
 256 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
 257 {
 258         u8 hmac[SHA256_DIGEST_SIZE];
 259         u64 thmac;
 260
 261         subflow_generate_hmac(subflow->remote_key, subflow->local_key,
 262                               subflow->remote_nonce, subflow->local_nonce,
 263                               hmac);
 264
 265         thmac = get_unaligned_be64(hmac);
 266         pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
 267                  subflow, subflow->token,
 268                  (unsigned long long)thmac,
 269                  (unsigned long long)subflow->thmac);
 270
 271         return thmac == subflow->thmac;
 272 }
 273
 274 void mptcp_subflow_reset(struct sock *ssk)
 275 {
 276         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 277         struct sock *sk = subflow->conn;
 278
 279         tcp_set_state(ssk, TCP_CLOSE);
 280         tcp_send_active_reset(ssk, GFP_ATOMIC);
 281         tcp_done(ssk);
 282         if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags) &&
 283             schedule_work(&mptcp_sk(sk)->work))
 284                 sock_hold(sk);
 285 }
 286
 287 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 288 {
 289         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 290         struct mptcp_options_received mp_opt;
 291         struct sock *parent = subflow->conn;
 292
 293         subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 294
 295         if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
 296                 inet_sk_state_store(parent, TCP_ESTABLISHED);
 297                 parent->sk_state_change(parent);
 298         }
 299
 300         /* be sure no special action on any packet other than syn-ack */
 301         if (subflow->conn_finished)
 302                 return;
 303
 304         subflow->rel_write_seq = 1;
 305         subflow->conn_finished = 1;
 306         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 307         pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
 308
 309         mptcp_get_options(skb, &mp_opt);
 310         if (subflow->request_mptcp) {
 311                 if (!mp_opt.mp_capable) {
 312                         MPTCP_INC_STATS(sock_net(sk),
 313                                         MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
 314                         mptcp_do_fallback(sk);
 315                         pr_fallback(mptcp_sk(subflow->conn));
 316                         goto fallback;
 317                 }
 318
 319                 subflow->mp_capable = 1;
 320                 subflow->can_ack = 1;
 321                 subflow->remote_key = mp_opt.sndr_key;
 322                 pr_debug("subflow=%p, remote_key=%llu", subflow,
 323                          subflow->remote_key);
 324                 mptcp_finish_connect(sk);
 325         } else if (subflow->request_join) {
 326                 u8 hmac[SHA256_DIGEST_SIZE];
 327
 328                 if (!mp_opt.mp_join)
 329                         goto do_reset;
 330
 331                 subflow->thmac = mp_opt.thmac;
 332                 subflow->remote_nonce = mp_opt.nonce;
 333                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
 334                          subflow->thmac, subflow->remote_nonce);
 335
 336                 if (!subflow_thmac_valid(subflow)) {
 337                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 338                         goto do_reset;
 339                 }
 340
 341                 if (!mptcp_finish_join(sk))
 342                         goto do_reset;
 343
 344                 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 345                                       subflow->local_nonce,
 346                                       subflow->remote_nonce,
 347                                       hmac);
 348                 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
 349
 350                 subflow->mp_join = 1;
 351                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 352         } else if (mptcp_check_fallback(sk)) {
 353 fallback:
 354                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
 355         }
 356         return;
 357
 358 do_reset:
 359         mptcp_subflow_reset(sk);
 360 }
 361
 362 struct request_sock_ops mptcp_subflow_request_sock_ops;
 363 EXPORT_SYMBOL_GPL(mptcp_subflow_request_sock_ops);
 364 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
 365
 366 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 367 {
 368         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 369
 370         pr_debug("subflow=%p", subflow);
 371
 372         /* Never answer to SYNs sent to broadcast or multicast */
 373         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 374                 goto drop;
 375
 376         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 377                                 &subflow_request_sock_ipv4_ops,
 378                                 sk, skb);
 379 drop:
 380         tcp_listendrop(sk);
 381         return 0;
 382 }
 383
 384 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 385 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
 386 static struct inet_connection_sock_af_ops subflow_v6_specific;
 387 static struct inet_connection_sock_af_ops subflow_v6m_specific;
 388
 389 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 390 {
 391         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 392
 393         pr_debug("subflow=%p", subflow);
 394
 395         if (skb->protocol == htons(ETH_P_IP))
 396                 return subflow_v4_conn_request(sk, skb);
 397
 398         if (!ipv6_unicast_destination(skb))
 399                 goto drop;
 400
 401         if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
 402                 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
 403                 return 0;
 404         }
 405
 406         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 407                                 &subflow_request_sock_ipv6_ops, sk, skb);
 408
 409 drop:
 410         tcp_listendrop(sk);
 411         return 0; /* don't send reset */
 412 }
 413 #endif
 414
 415 /* validate hmac received in third ACK */
 416 static bool subflow_hmac_valid(const struct request_sock *req,
 417                                const struct mptcp_options_received *mp_opt)
 418 {
 419         const struct mptcp_subflow_request_sock *subflow_req;
 420         u8 hmac[SHA256_DIGEST_SIZE];
 421         struct mptcp_sock *msk;
 422
 423         subflow_req = mptcp_subflow_rsk(req);
 424         msk = subflow_req->msk;
 425         if (!msk)
 426                 return false;
 427
 428         subflow_generate_hmac(msk->remote_key, msk->local_key,
 429                               subflow_req->remote_nonce,
 430                               subflow_req->local_nonce, hmac);
 431
 432         return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
 433 }
 434
 435 static void mptcp_sock_destruct(struct sock *sk)
 436 {
 437         /* if new mptcp socket isn't accepted, it is free'd
 438          * from the tcp listener sockets request queue, linked
 439          * from req->sk.  The tcp socket is released.
 440          * This calls the ULP release function which will
 441          * also remove the mptcp socket, via
 442          * sock_put(ctx->conn).
 443          *
 444          * Problem is that the mptcp socket will be in
 445          * ESTABLISHED state and will not have the SOCK_DEAD flag.
 446          * Both result in warnings from inet_sock_destruct.
 447          */
 448         if ((1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
 449                 sk->sk_state = TCP_CLOSE;
 450                 WARN_ON_ONCE(sk->sk_socket);
 451                 sock_orphan(sk);
 452         }
 453
 454         mptcp_destroy_common(mptcp_sk(sk));
 455         inet_sock_destruct(sk);
 456 }
 457
 458 static void mptcp_force_close(struct sock *sk)
 459 {
 460         inet_sk_state_store(sk, TCP_CLOSE);
 461         sk_common_release(sk);
 462 }
 463
 464 static void subflow_ulp_fallback(struct sock *sk,
 465                                  struct mptcp_subflow_context *old_ctx)
 466 {
 467         struct inet_connection_sock *icsk = inet_csk(sk);
 468
 469         mptcp_subflow_tcp_fallback(sk, old_ctx);
 470         icsk->icsk_ulp_ops = NULL;
 471         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 472         tcp_sk(sk)->is_mptcp = 0;
 473 }
 474
 475 static void subflow_drop_ctx(struct sock *ssk)
 476 {
 477         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 478
 479         if (!ctx)
 480                 return;
 481
 482         subflow_ulp_fallback(ssk, ctx);
 483         if (ctx->conn)
 484                 sock_put(ctx->conn);
 485
 486         kfree_rcu(ctx, rcu);
 487 }
 488
 489 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 490                                      struct mptcp_options_received *mp_opt)
 491 {
 492         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 493
 494         subflow->remote_key = mp_opt->sndr_key;
 495         subflow->fully_established = 1;
 496         subflow->can_ack = 1;
 497         WRITE_ONCE(msk->fully_established, true);
 498 }
 499
 500 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 501                                           struct sk_buff *skb,
 502                                           struct request_sock *req,
 503                                           struct dst_entry *dst,
 504                                           struct request_sock *req_unhash,
 505                                           bool *own_req)
 506 {
 507         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 508         struct mptcp_subflow_request_sock *subflow_req;
 509         struct mptcp_options_received mp_opt;
 510         bool fallback, fallback_is_fatal;
 511         struct sock *new_msk = NULL;
 512         struct sock *child;
 513
 514         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 515
 516         /* After child creation we must look for 'mp_capable' even when options
 517          * are not parsed
 518          */
 519         mp_opt.mp_capable = 0;
 520
 521         /* hopefully temporary handling for MP_JOIN+syncookie */
 522         subflow_req = mptcp_subflow_rsk(req);
 523         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 524         fallback = !tcp_rsk(req)->is_mptcp;
 525         if (fallback)
 526                 goto create_child;
 527
 528         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 529         if (subflow_req->mp_capable) {
 530                 /* we can receive and accept an in-window, out-of-order pkt,
 531                  * which may not carry the MP_CAPABLE opt even on mptcp enabled
 532                  * paths: always try to extract the peer key, and fallback
 533                  * for packets missing it.
 534                  * Even OoO DSS packets coming legitly after dropped or
 535                  * reordered MPC will cause fallback, but we don't have other
 536                  * options.
 537                  */
 538                 mptcp_get_options(skb, &mp_opt);
 539                 if (!mp_opt.mp_capable) {
 540                         fallback = true;
 541                         goto create_child;
 542                 }
 543
 544                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 545                 if (!new_msk)
 546                         fallback = true;
 547         } else if (subflow_req->mp_join) {
 548                 mptcp_get_options(skb, &mp_opt);
 549                 if (!mp_opt.mp_join || !subflow_hmac_valid(req, &mp_opt) ||
 550                     !mptcp_can_accept_new_subflow(subflow_req->msk)) {
 551                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 552                         fallback = true;
 553                 }
 554         }
 555
 556 create_child:
 557         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 558                                                      req_unhash, own_req);
 559
 560         if (child && *own_req) {
 561                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 562
 563                 tcp_rsk(req)->drop_req = false;
 564
 565                 /* we need to fallback on ctx allocation failure and on pre-reqs
 566                  * checking above. In the latter scenario we additionally need
 567                  * to reset the context to non MPTCP status.
 568                  */
 569                 if (!ctx || fallback) {
 570                         if (fallback_is_fatal)
 571                                 goto dispose_child;
 572
 573                         subflow_drop_ctx(child);
 574                         goto out;
 575                 }
 576
 577                 if (ctx->mp_capable) {
 578                         /* this can't race with mptcp_close(), as the msk is
 579                          * not yet exposted to user-space
 580                          */
 581                         inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
 582
 583                         /* new mpc subflow takes ownership of the newly
 584                          * created mptcp socket
 585                          */
 586                         new_msk->sk_destruct = mptcp_sock_destruct;
 587                         mptcp_pm_new_connection(mptcp_sk(new_msk), 1);
 588                         mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
 589                         ctx->conn = new_msk;
 590                         new_msk = NULL;
 591
 592                         /* with OoO packets we can reach here without ingress
 593                          * mpc option
 594                          */
 595                         if (mp_opt.mp_capable)
 596                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 597                 } else if (ctx->mp_join) {
 598                         struct mptcp_sock *owner;
 599
 600                         owner = subflow_req->msk;
 601                         if (!owner)
 602                                 goto dispose_child;
 603
 604                         /* move the msk reference ownership to the subflow */
 605                         subflow_req->msk = NULL;
 606                         ctx->conn = (struct sock *)owner;
 607                         if (!mptcp_finish_join(child))
 608                                 goto dispose_child;
 609
 610                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 611                         tcp_rsk(req)->drop_req = true;
 612                 }
 613         }
 614
 615 out:
 616         /* dispose of the left over mptcp master, if any */
 617         if (unlikely(new_msk))
 618                 mptcp_force_close(new_msk);
 619
 620         /* check for expected invariant - should never trigger, just help
 621          * catching eariler subtle bugs
 622          */
 623         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 624                      (!mptcp_subflow_ctx(child) ||
 625                       !mptcp_subflow_ctx(child)->conn));
 626         return child;
 627
 628 dispose_child:
 629         subflow_drop_ctx(child);
 630         tcp_rsk(req)->drop_req = true;
 631         inet_csk_prepare_for_destroy_sock(child);
 632         tcp_done(child);
 633         req->rsk_ops->send_reset(sk, skb);
 634
 635         /* The last child reference will be released by the caller */
 636         return child;
 637 }
 638
 639 static struct inet_connection_sock_af_ops subflow_specific;
 640
 641 enum mapping_status {
 642         MAPPING_OK,
 643         MAPPING_INVALID,
 644         MAPPING_EMPTY,
 645         MAPPING_DATA_FIN,
 646         MAPPING_DUMMY
 647 };
 648
 649 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
 650 {
 651         if ((u32)seq == (u32)old_seq)
 652                 return old_seq;
 653
 654         /* Assume map covers data not mapped yet. */
 655         return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
 656 }
 657
 658 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 659 {
 660         pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 661                  ssn, subflow->map_subflow_seq, subflow->map_data_len);
 662 }
 663
 664 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 665 {
 666         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 667         unsigned int skb_consumed;
 668
 669         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 670         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 671                 return true;
 672
 673         return skb->len - skb_consumed <= subflow->map_data_len -
 674                                           mptcp_subflow_get_map_offset(subflow);
 675 }
 676
 677 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 678 {
 679         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 680         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 681
 682         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 683                 /* Mapping covers data later in the subflow stream,
 684                  * currently unsupported.
 685                  */
 686                 dbg_bad_map(subflow, ssn);
 687                 return false;
 688         }
 689         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 690                                   subflow->map_data_len))) {
 691                 /* Mapping does covers past subflow data, invalid */
 692                 dbg_bad_map(subflow, ssn);
 693                 return false;
 694         }
 695         return true;
 696 }
 697
 698 static enum mapping_status get_mapping_status(struct sock *ssk,
 699                                               struct mptcp_sock *msk)
 700 {
 701         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 702         struct mptcp_ext *mpext;
 703         struct sk_buff *skb;
 704         u16 data_len;
 705         u64 map_seq;
 706
 707         skb = skb_peek(&ssk->sk_receive_queue);
 708         if (!skb)
 709                 return MAPPING_EMPTY;
 710
 711         if (mptcp_check_fallback(ssk))
 712                 return MAPPING_DUMMY;
 713
 714         mpext = mptcp_get_ext(skb);
 715         if (!mpext || !mpext->use_map) {
 716                 if (!subflow->map_valid && !skb->len) {
 717                         /* the TCP stack deliver 0 len FIN pkt to the receive
 718                          * queue, that is the only 0len pkts ever expected here,
 719                          * and we can admit no mapping only for 0 len pkts
 720                          */
 721                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 722                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 723                                           TCP_SKB_CB(skb)->seq,
 724                                           TCP_SKB_CB(skb)->end_seq,
 725                                           TCP_SKB_CB(skb)->tcp_flags);
 726                         sk_eat_skb(ssk, skb);
 727                         return MAPPING_EMPTY;
 728                 }
 729
 730                 if (!subflow->map_valid)
 731                         return MAPPING_INVALID;
 732
 733                 goto validate_seq;
 734         }
 735
 736         pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
 737                  mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
 738                  mpext->data_len, mpext->data_fin);
 739
 740         data_len = mpext->data_len;
 741         if (data_len == 0) {
 742                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 743                 return MAPPING_INVALID;
 744         }
 745
 746         if (mpext->data_fin == 1) {
 747                 if (data_len == 1) {
 748                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
 749                                                                  mpext->dsn64);
 750                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
 751                         if (subflow->map_valid) {
 752                                 /* A DATA_FIN might arrive in a DSS
 753                                  * option before the previous mapping
 754                                  * has been fully consumed. Continue
 755                                  * handling the existing mapping.
 756                                  */
 757                                 skb_ext_del(skb, SKB_EXT_MPTCP);
 758                                 return MAPPING_OK;
 759                         } else {
 760                                 if (updated && schedule_work(&msk->work))
 761                                         sock_hold((struct sock *)msk);
 762
 763                                 return MAPPING_DATA_FIN;
 764                         }
 765                 } else {
 766                         u64 data_fin_seq = mpext->data_seq + data_len - 1;
 767
 768                         /* If mpext->data_seq is a 32-bit value, data_fin_seq
 769                          * must also be limited to 32 bits.
 770                          */
 771                         if (!mpext->dsn64)
 772                                 data_fin_seq &= GENMASK_ULL(31, 0);
 773
 774                         mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
 775                         pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
 776                                  data_fin_seq, mpext->dsn64);
 777                 }
 778
 779                 /* Adjust for DATA_FIN using 1 byte of sequence space */
 780                 data_len--;
 781         }
 782
 783         if (!mpext->dsn64) {
 784                 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
 785                                      mpext->data_seq);
 786                 pr_debug("expanded seq=%llu", subflow->map_seq);
 787         } else {
 788                 map_seq = mpext->data_seq;
 789         }
 790         WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
 791
 792         if (subflow->map_valid) {
 793                 /* Allow replacing only with an identical map */
 794                 if (subflow->map_seq == map_seq &&
 795                     subflow->map_subflow_seq == mpext->subflow_seq &&
 796                     subflow->map_data_len == data_len) {
 797                         skb_ext_del(skb, SKB_EXT_MPTCP);
 798                         return MAPPING_OK;
 799                 }
 800
 801                 /* If this skb data are fully covered by the current mapping,
 802                  * the new map would need caching, which is not supported
 803                  */
 804                 if (skb_is_fully_mapped(ssk, skb)) {
 805                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
 806                         return MAPPING_INVALID;
 807                 }
 808
 809                 /* will validate the next map after consuming the current one */
 810                 return MAPPING_OK;
 811         }
 812
 813         subflow->map_seq = map_seq;
 814         subflow->map_subflow_seq = mpext->subflow_seq;
 815         subflow->map_data_len = data_len;
 816         subflow->map_valid = 1;
 817         subflow->mpc_map = mpext->mpc_map;
 818         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
 819                  subflow->map_seq, subflow->map_subflow_seq,
 820                  subflow->map_data_len);
 821
 822 validate_seq:
 823         /* we revalidate valid mapping on new skb, because we must ensure
 824          * the current skb is completely covered by the available mapping
 825          */
 826         if (!validate_mapping(ssk, skb))
 827                 return MAPPING_INVALID;
 828
 829         skb_ext_del(skb, SKB_EXT_MPTCP);
 830         return MAPPING_OK;
 831 }
 832
 833 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
 834                                        u64 limit)
 835 {
 836         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 837         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
 838         u32 incr;
 839
 840         incr = limit >= skb->len ? skb->len + fin : limit;
 841
 842         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
 843                  subflow->map_subflow_seq);
 844         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
 845         tcp_sk(ssk)->copied_seq += incr;
 846         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
 847                 sk_eat_skb(ssk, skb);
 848         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
 849                 subflow->map_valid = 0;
 850         if (incr)
 851                 tcp_cleanup_rbuf(ssk, incr);
 852 }
 853
 854 static bool subflow_check_data_avail(struct sock *ssk)
 855 {
 856         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 857         enum mapping_status status;
 858         struct mptcp_sock *msk;
 859         struct sk_buff *skb;
 860
 861         pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
 862                  subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
 863         if (!skb_peek(&ssk->sk_receive_queue))
 864                 subflow->data_avail = 0;
 865         if (subflow->data_avail)
 866                 return true;
 867
 868         msk = mptcp_sk(subflow->conn);
 869         for (;;) {
 870                 u64 ack_seq;
 871                 u64 old_ack;
 872
 873                 status = get_mapping_status(ssk, msk);
 874                 pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
 875                 if (status == MAPPING_INVALID) {
 876                         ssk->sk_err = EBADMSG;
 877                         goto fatal;
 878                 }
 879                 if (status == MAPPING_DUMMY) {
 880                         __mptcp_do_fallback(msk);
 881                         skb = skb_peek(&ssk->sk_receive_queue);
 882                         subflow->map_valid = 1;
 883                         subflow->map_seq = READ_ONCE(msk->ack_seq);
 884                         subflow->map_data_len = skb->len;
 885                         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq -
 886                                                    subflow->ssn_offset;
 887                         subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
 888                         return true;
 889                 }
 890
 891                 if (status != MAPPING_OK)
 892                         return false;
 893
 894                 skb = skb_peek(&ssk->sk_receive_queue);
 895                 if (WARN_ON_ONCE(!skb))
 896                         return false;
 897
 898                 /* if msk lacks the remote key, this subflow must provide an
 899                  * MP_CAPABLE-based mapping
 900                  */
 901                 if (unlikely(!READ_ONCE(msk->can_ack))) {
 902                         if (!subflow->mpc_map) {
 903                                 ssk->sk_err = EBADMSG;
 904                                 goto fatal;
 905                         }
 906                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
 907                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
 908                         WRITE_ONCE(msk->can_ack, true);
 909                 }
 910
 911                 old_ack = READ_ONCE(msk->ack_seq);
 912                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
 913                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
 914                          ack_seq);
 915                 if (ack_seq == old_ack) {
 916                         subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
 917                         break;
 918                 } else if (after64(ack_seq, old_ack)) {
 919                         subflow->data_avail = MPTCP_SUBFLOW_OOO_DATA;
 920                         break;
 921                 }
 922
 923                 /* only accept in-sequence mapping. Old values are spurious
 924                  * retransmission
 925                  */
 926                 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
 927         }
 928         return true;
 929
 930 fatal:
 931         /* fatal protocol error, close the socket */
 932         /* This barrier is coupled with smp_rmb() in tcp_poll() */
 933         smp_wmb();
 934         ssk->sk_error_report(ssk);
 935         tcp_set_state(ssk, TCP_CLOSE);
 936         tcp_send_active_reset(ssk, GFP_ATOMIC);
 937         subflow->data_avail = 0;
 938         return false;
 939 }
 940
 941 bool mptcp_subflow_data_available(struct sock *sk)
 942 {
 943         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 944
 945         /* check if current mapping is still valid */
 946         if (subflow->map_valid &&
 947             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
 948                 subflow->map_valid = 0;
 949                 subflow->data_avail = 0;
 950
 951                 pr_debug("Done with mapping: seq=%u data_len=%u",
 952                          subflow->map_subflow_seq,
 953                          subflow->map_data_len);
 954         }
 955
 956         return subflow_check_data_avail(sk);
 957 }
 958
 959 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
 960  * not the ssk one.
 961  *
 962  * In mptcp, rwin is about the mptcp-level connection data.
 963  *
 964  * Data that is still on the ssk rx queue can thus be ignored,
 965  * as far as mptcp peer is concerened that data is still inflight.
 966  * DSS ACK is updated when skb is moved to the mptcp rx queue.
 967  */
 968 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
 969 {
 970         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 971         const struct sock *sk = subflow->conn;
 972
 973         *space = tcp_space(sk);
 974         *full_space = tcp_full_space(sk);
 975 }
 976
 977 static void subflow_data_ready(struct sock *sk)
 978 {
 979         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 980         u16 state = 1 << inet_sk_state_load(sk);
 981         struct sock *parent = subflow->conn;
 982         struct mptcp_sock *msk;
 983
 984         msk = mptcp_sk(parent);
 985         if (state & TCPF_LISTEN) {
 986                 /* MPJ subflow are removed from accept queue before reaching here,
 987                  * avoid stray wakeups
 988                  */
 989                 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
 990                         return;
 991
 992                 set_bit(MPTCP_DATA_READY, &msk->flags);
 993                 parent->sk_data_ready(parent);
 994                 return;
 995         }
 996
 997         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
 998                      !subflow->mp_join && !(state & TCPF_CLOSE));
 999
1000         if (mptcp_subflow_data_available(sk))
1001                 mptcp_data_ready(parent, sk);
1002 }
1003
1004 static void subflow_write_space(struct sock *sk)
1005 {
1006         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1007         struct sock *parent = subflow->conn;
1008
1009         if (!sk_stream_is_writeable(sk))
1010                 return;
1011
1012         if (sk_stream_is_writeable(parent)) {
1013                 set_bit(MPTCP_SEND_SPACE, &mptcp_sk(parent)->flags);
1014                 smp_mb__after_atomic();
1015                 /* set SEND_SPACE before sk_stream_write_space clears NOSPACE */
1016                 sk_stream_write_space(parent);
1017         }
1018 }
1019
1020 static struct inet_connection_sock_af_ops *
1021 subflow_default_af_ops(struct sock *sk)
1022 {
1023 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1024         if (sk->sk_family == AF_INET6)
1025                 return &subflow_v6_specific;
1026 #endif
1027         return &subflow_specific;
1028 }
1029
1030 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1031 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1032 {
1033         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1034         struct inet_connection_sock *icsk = inet_csk(sk);
1035         struct inet_connection_sock_af_ops *target;
1036
1037         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1038
1039         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1040                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1041
1042         if (likely(icsk->icsk_af_ops == target))
1043                 return;
1044
1045         subflow->icsk_af_ops = icsk->icsk_af_ops;
1046         icsk->icsk_af_ops = target;
1047 }
1048 #endif
1049
1050 static void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1051                                 struct sockaddr_storage *addr)
1052 {
1053         memset(addr, 0, sizeof(*addr));
1054         addr->ss_family = info->family;
1055         if (addr->ss_family == AF_INET) {
1056                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1057
1058                 in_addr->sin_addr = info->addr;
1059                 in_addr->sin_port = info->port;
1060         }
1061 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1062         else if (addr->ss_family == AF_INET6) {
1063                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1064
1065                 in6_addr->sin6_addr = info->addr6;
1066                 in6_addr->sin6_port = info->port;
1067         }
1068 #endif
1069 }
1070
1071 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1072                             const struct mptcp_addr_info *remote)
1073 {
1074         struct mptcp_sock *msk = mptcp_sk(sk);
1075         struct mptcp_subflow_context *subflow;
1076         struct sockaddr_storage addr;
1077         int remote_id = remote->id;
1078         int local_id = loc->id;
1079         struct socket *sf;
1080         struct sock *ssk;
1081         u32 remote_token;
1082         int addrlen;
1083         int err;
1084
1085         if (!mptcp_is_fully_established(sk))
1086                 return -ENOTCONN;
1087
1088         err = mptcp_subflow_create_socket(sk, &sf);
1089         if (err)
1090                 return err;
1091
1092         ssk = sf->sk;
1093         subflow = mptcp_subflow_ctx(ssk);
1094         do {
1095                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1096         } while (!subflow->local_nonce);
1097
1098         if (!local_id) {
1099                 err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
1100                 if (err < 0)
1101                         goto failed;
1102
1103                 local_id = err;
1104         }
1105
1106         subflow->remote_key = msk->remote_key;
1107         subflow->local_key = msk->local_key;
1108         subflow->token = msk->token;
1109         mptcp_info2sockaddr(loc, &addr);
1110
1111         addrlen = sizeof(struct sockaddr_in);
1112 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1113         if (loc->family == AF_INET6)
1114                 addrlen = sizeof(struct sockaddr_in6);
1115 #endif
1116         ssk->sk_bound_dev_if = loc->ifindex;
1117         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1118         if (err)
1119                 goto failed;
1120
1121         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1122         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1123                  remote_token, local_id, remote_id);
1124         subflow->remote_token = remote_token;
1125         subflow->local_id = local_id;
1126         subflow->remote_id = remote_id;
1127         subflow->request_join = 1;
1128         subflow->request_bkup = !!(loc->flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1129         mptcp_info2sockaddr(remote, &addr);
1130
1131         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1132         if (err && err != -EINPROGRESS)
1133                 goto failed;
1134
1135         spin_lock_bh(&msk->join_list_lock);
1136         list_add_tail(&subflow->node, &msk->join_list);
1137         spin_unlock_bh(&msk->join_list_lock);
1138
1139         return err;
1140
1141 failed:
1142         sock_release(sf);
1143         return err;
1144 }
1145
1146 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1147 {
1148         struct mptcp_subflow_context *subflow;
1149         struct net *net = sock_net(sk);
1150         struct socket *sf;
1151         int err;
1152
1153         /* un-accepted server sockets can reach here - on bad configuration
1154          * bail early to avoid greater trouble later
1155          */
1156         if (unlikely(!sk->sk_socket))
1157                 return -EINVAL;
1158
1159         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1160                                &sf);
1161         if (err)
1162                 return err;
1163
1164         lock_sock(sf->sk);
1165
1166         /* kernel sockets do not by default acquire net ref, but TCP timer
1167          * needs it.
1168          */
1169         sf->sk->sk_net_refcnt = 1;
1170         get_net(net);
1171 #ifdef CONFIG_PROC_FS
1172         this_cpu_add(*net->core.sock_inuse, 1);
1173 #endif
1174         err = tcp_set_ulp(sf->sk, "mptcp");
1175         release_sock(sf->sk);
1176
1177         if (err) {
1178                 sock_release(sf);
1179                 return err;
1180         }
1181
1182         /* the newly created socket really belongs to the owning MPTCP master
1183          * socket, even if for additional subflows the allocation is performed
1184          * by a kernel workqueue. Adjust inode references, so that the
1185          * procfs/diag interaces really show this one belonging to the correct
1186          * user.
1187          */
1188         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1189         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1190         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1191
1192         subflow = mptcp_subflow_ctx(sf->sk);
1193         pr_debug("subflow=%p", subflow);
1194
1195         *new_sock = sf;
1196         sock_hold(sk);
1197         subflow->conn = sk;
1198
1199         return 0;
1200 }
1201
1202 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1203                                                         gfp_t priority)
1204 {
1205         struct inet_connection_sock *icsk = inet_csk(sk);
1206         struct mptcp_subflow_context *ctx;
1207
1208         ctx = kzalloc(sizeof(*ctx), priority);
1209         if (!ctx)
1210                 return NULL;
1211
1212         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1213         INIT_LIST_HEAD(&ctx->node);
1214
1215         pr_debug("subflow=%p", ctx);
1216
1217         ctx->tcp_sock = sk;
1218
1219         return ctx;
1220 }
1221
1222 static void __subflow_state_change(struct sock *sk)
1223 {
1224         struct socket_wq *wq;
1225
1226         rcu_read_lock();
1227         wq = rcu_dereference(sk->sk_wq);
1228         if (skwq_has_sleeper(wq))
1229                 wake_up_interruptible_all(&wq->wait);
1230         rcu_read_unlock();
1231 }
1232
1233 static bool subflow_is_done(const struct sock *sk)
1234 {
1235         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1236 }
1237
1238 static void subflow_state_change(struct sock *sk)
1239 {
1240         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1241         struct sock *parent = subflow->conn;
1242
1243         __subflow_state_change(sk);
1244
1245         if (subflow_simultaneous_connect(sk)) {
1246                 mptcp_do_fallback(sk);
1247                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1248                 pr_fallback(mptcp_sk(parent));
1249                 subflow->conn_finished = 1;
1250                 if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
1251                         inet_sk_state_store(parent, TCP_ESTABLISHED);
1252                         parent->sk_state_change(parent);
1253                 }
1254         }
1255
1256         /* as recvmsg() does not acquire the subflow socket for ssk selection
1257          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1258          * the data available machinery here.
1259          */
1260         if (mptcp_subflow_data_available(sk))
1261                 mptcp_data_ready(parent, sk);
1262
1263         if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1264             !(parent->sk_shutdown & RCV_SHUTDOWN) &&
1265             !subflow->rx_eof && subflow_is_done(sk)) {
1266                 subflow->rx_eof = 1;
1267                 mptcp_subflow_eof(parent);
1268         }
1269 }
1270
1271 static int subflow_ulp_init(struct sock *sk)
1272 {
1273         struct inet_connection_sock *icsk = inet_csk(sk);
1274         struct mptcp_subflow_context *ctx;
1275         struct tcp_sock *tp = tcp_sk(sk);
1276         int err = 0;
1277
1278         /* disallow attaching ULP to a socket unless it has been
1279          * created with sock_create_kern()
1280          */
1281         if (!sk->sk_kern_sock) {
1282                 err = -EOPNOTSUPP;
1283                 goto out;
1284         }
1285
1286         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1287         if (!ctx) {
1288                 err = -ENOMEM;
1289                 goto out;
1290         }
1291
1292         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1293
1294         tp->is_mptcp = 1;
1295         ctx->icsk_af_ops = icsk->icsk_af_ops;
1296         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1297         ctx->tcp_data_ready = sk->sk_data_ready;
1298         ctx->tcp_state_change = sk->sk_state_change;
1299         ctx->tcp_write_space = sk->sk_write_space;
1300         sk->sk_data_ready = subflow_data_ready;
1301         sk->sk_write_space = subflow_write_space;
1302         sk->sk_state_change = subflow_state_change;
1303 out:
1304         return err;
1305 }
1306
1307 static void subflow_ulp_release(struct sock *sk)
1308 {
1309         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(sk);
1310
1311         if (!ctx)
1312                 return;
1313
1314         if (ctx->conn)
1315                 sock_put(ctx->conn);
1316
1317         kfree_rcu(ctx, rcu);
1318 }
1319
1320 static void subflow_ulp_clone(const struct request_sock *req,
1321                               struct sock *newsk,
1322                               const gfp_t priority)
1323 {
1324         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1325         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1326         struct mptcp_subflow_context *new_ctx;
1327
1328         if (!tcp_rsk(req)->is_mptcp ||
1329             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1330                 subflow_ulp_fallback(newsk, old_ctx);
1331                 return;
1332         }
1333
1334         new_ctx = subflow_create_ctx(newsk, priority);
1335         if (!new_ctx) {
1336                 subflow_ulp_fallback(newsk, old_ctx);
1337                 return;
1338         }
1339
1340         new_ctx->conn_finished = 1;
1341         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1342         new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1343         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1344         new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1345         new_ctx->rel_write_seq = 1;
1346         new_ctx->tcp_sock = newsk;
1347
1348         if (subflow_req->mp_capable) {
1349                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1350                  * is fully established only after we receive the remote key
1351                  */
1352                 new_ctx->mp_capable = 1;
1353                 new_ctx->local_key = subflow_req->local_key;
1354                 new_ctx->token = subflow_req->token;
1355                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1356                 new_ctx->idsn = subflow_req->idsn;
1357         } else if (subflow_req->mp_join) {
1358                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1359                 new_ctx->mp_join = 1;
1360                 new_ctx->fully_established = 1;
1361                 new_ctx->backup = subflow_req->backup;
1362                 new_ctx->local_id = subflow_req->local_id;
1363                 new_ctx->remote_id = subflow_req->remote_id;
1364                 new_ctx->token = subflow_req->token;
1365                 new_ctx->thmac = subflow_req->thmac;
1366         }
1367 }
1368
1369 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1370         .name           = "mptcp",
1371         .owner          = THIS_MODULE,
1372         .init           = subflow_ulp_init,
1373         .release        = subflow_ulp_release,
1374         .clone          = subflow_ulp_clone,
1375 };
1376
1377 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1378 {
1379         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1380         subflow_ops->slab_name = "request_sock_subflow";
1381
1382         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1383                                               subflow_ops->obj_size, 0,
1384                                               SLAB_ACCOUNT |
1385                                               SLAB_TYPESAFE_BY_RCU,
1386                                               NULL);
1387         if (!subflow_ops->slab)
1388                 return -ENOMEM;
1389
1390         subflow_ops->destructor = subflow_req_destructor;
1391
1392         return 0;
1393 }
1394
1395 void __init mptcp_subflow_init(void)
1396 {
1397         mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1398         if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1399                 panic("MPTCP: failed to init subflow request sock ops\n");
1400
1401         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1402         subflow_request_sock_ipv4_ops.init_req = subflow_v4_init_req;
1403
1404         subflow_specific = ipv4_specific;
1405         subflow_specific.conn_request = subflow_v4_conn_request;
1406         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1407         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1408
1409 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1410         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1411         subflow_request_sock_ipv6_ops.init_req = subflow_v6_init_req;
1412
1413         subflow_v6_specific = ipv6_specific;
1414         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1415         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1416         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1417
1418         subflow_v6m_specific = subflow_v6_specific;
1419         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1420         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1421         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1422         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1423         subflow_v6m_specific.net_frag_header_len = 0;
1424 #endif
1425
1426         mptcp_diag_subflow_init(&subflow_ulp_ops);
1427
1428         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1429                 panic("MPTCP: failed to register subflows to ULP\n");
1430 }