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 #include <trace/events/mptcp.h>
  29
  30 static void mptcp_subflow_ops_undo_override(struct sock *ssk);
  31
  32 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
  33                                   enum linux_mptcp_mib_field field)
  34 {
  35         MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
  36 }
  37
  38 static void subflow_req_destructor(struct request_sock *req)
  39 {
  40         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  41
  42         pr_debug("subflow_req=%p", subflow_req);
  43
  44         if (subflow_req->msk)
  45                 sock_put((struct sock *)subflow_req->msk);
  46
  47         mptcp_token_destroy_request(req);
  48         tcp_request_sock_ops.destructor(req);
  49 }
  50
  51 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
  52                                   void *hmac)
  53 {
  54         u8 msg[8];
  55
  56         put_unaligned_be32(nonce1, &msg[0]);
  57         put_unaligned_be32(nonce2, &msg[4]);
  58
  59         mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
  60 }
  61
  62 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
  63 {
  64         return mptcp_is_fully_established((void *)msk) &&
  65                READ_ONCE(msk->pm.accept_subflow);
  66 }
  67
  68 /* validate received token and create truncated hmac and nonce for SYN-ACK */
  69 static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req)
  70 {
  71         struct mptcp_sock *msk = subflow_req->msk;
  72         u8 hmac[SHA256_DIGEST_SIZE];
  73
  74         get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
  75
  76         subflow_generate_hmac(msk->local_key, msk->remote_key,
  77                               subflow_req->local_nonce,
  78                               subflow_req->remote_nonce, hmac);
  79
  80         subflow_req->thmac = get_unaligned_be64(hmac);
  81 }
  82
  83 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req)
  84 {
  85         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  86         struct mptcp_sock *msk;
  87         int local_id;
  88
  89         msk = mptcp_token_get_sock(subflow_req->token);
  90         if (!msk) {
  91                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
  92                 return NULL;
  93         }
  94
  95         local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
  96         if (local_id < 0) {
  97                 sock_put((struct sock *)msk);
  98                 return NULL;
  99         }
 100         subflow_req->local_id = local_id;
 101
 102         return msk;
 103 }
 104
 105 static void subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
 106 {
 107         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 108
 109         subflow_req->mp_capable = 0;
 110         subflow_req->mp_join = 0;
 111         subflow_req->msk = NULL;
 112         mptcp_token_init_request(req);
 113 }
 114
 115 static bool subflow_use_different_sport(struct mptcp_sock *msk, const struct sock *sk)
 116 {
 117         return inet_sk(sk)->inet_sport != inet_sk((struct sock *)msk)->inet_sport;
 118 }
 119
 120 static void subflow_add_reset_reason(struct sk_buff *skb, u8 reason)
 121 {
 122         struct mptcp_ext *mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
 123
 124         if (mpext) {
 125                 memset(mpext, 0, sizeof(*mpext));
 126                 mpext->reset_reason = reason;
 127         }
 128 }
 129
 130 /* Init mptcp request socket.
 131  *
 132  * Returns an error code if a JOIN has failed and a TCP reset
 133  * should be sent.
 134  */
 135 static int subflow_check_req(struct request_sock *req,
 136                              const struct sock *sk_listener,
 137                              struct sk_buff *skb)
 138 {
 139         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 140         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 141         struct mptcp_options_received mp_opt;
 142
 143         pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
 144
 145 #ifdef CONFIG_TCP_MD5SIG
 146         /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
 147          * TCP option space.
 148          */
 149         if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
 150                 return -EINVAL;
 151 #endif
 152
 153         mptcp_get_options(skb, &mp_opt);
 154
 155         if (mp_opt.mp_capable) {
 156                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
 157
 158                 if (mp_opt.mp_join)
 159                         return 0;
 160         } else if (mp_opt.mp_join) {
 161                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
 162         }
 163
 164         if (mp_opt.mp_capable && listener->request_mptcp) {
 165                 int err, retries = 4;
 166
 167                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 168 again:
 169                 do {
 170                         get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
 171                 } while (subflow_req->local_key == 0);
 172
 173                 if (unlikely(req->syncookie)) {
 174                         mptcp_crypto_key_sha(subflow_req->local_key,
 175                                              &subflow_req->token,
 176                                              &subflow_req->idsn);
 177                         if (mptcp_token_exists(subflow_req->token)) {
 178                                 if (retries-- > 0)
 179                                         goto again;
 180                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
 181                         } else {
 182                                 subflow_req->mp_capable = 1;
 183                         }
 184                         return 0;
 185                 }
 186
 187                 err = mptcp_token_new_request(req);
 188                 if (err == 0)
 189                         subflow_req->mp_capable = 1;
 190                 else if (retries-- > 0)
 191                         goto again;
 192                 else
 193                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
 194
 195         } else if (mp_opt.mp_join && listener->request_mptcp) {
 196                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 197                 subflow_req->mp_join = 1;
 198                 subflow_req->backup = mp_opt.backup;
 199                 subflow_req->remote_id = mp_opt.join_id;
 200                 subflow_req->token = mp_opt.token;
 201                 subflow_req->remote_nonce = mp_opt.nonce;
 202                 subflow_req->msk = subflow_token_join_request(req);
 203
 204                 /* Can't fall back to TCP in this case. */
 205                 if (!subflow_req->msk) {
 206                         subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 207                         return -EPERM;
 208                 }
 209
 210                 if (subflow_use_different_sport(subflow_req->msk, sk_listener)) {
 211                         pr_debug("syn inet_sport=%d %d",
 212                                  ntohs(inet_sk(sk_listener)->inet_sport),
 213                                  ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport));
 214                         if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) {
 215                                 sock_put((struct sock *)subflow_req->msk);
 216                                 mptcp_token_destroy_request(req);
 217                                 tcp_request_sock_ops.destructor(req);
 218                                 subflow_req->msk = NULL;
 219                                 subflow_req->mp_join = 0;
 220                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX);
 221                                 return -EPERM;
 222                         }
 223                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX);
 224                 }
 225
 226                 subflow_req_create_thmac(subflow_req);
 227
 228                 if (unlikely(req->syncookie)) {
 229                         if (mptcp_can_accept_new_subflow(subflow_req->msk))
 230                                 subflow_init_req_cookie_join_save(subflow_req, skb);
 231                 }
 232
 233                 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
 234                          subflow_req->remote_nonce, subflow_req->msk);
 235         }
 236
 237         return 0;
 238 }
 239
 240 int mptcp_subflow_init_cookie_req(struct request_sock *req,
 241                                   const struct sock *sk_listener,
 242                                   struct sk_buff *skb)
 243 {
 244         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 245         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 246         struct mptcp_options_received mp_opt;
 247         int err;
 248
 249         subflow_init_req(req, sk_listener);
 250         mptcp_get_options(skb, &mp_opt);
 251
 252         if (mp_opt.mp_capable && mp_opt.mp_join)
 253                 return -EINVAL;
 254
 255         if (mp_opt.mp_capable && listener->request_mptcp) {
 256                 if (mp_opt.sndr_key == 0)
 257                         return -EINVAL;
 258
 259                 subflow_req->local_key = mp_opt.rcvr_key;
 260                 err = mptcp_token_new_request(req);
 261                 if (err)
 262                         return err;
 263
 264                 subflow_req->mp_capable = 1;
 265                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 266         } else if (mp_opt.mp_join && listener->request_mptcp) {
 267                 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
 268                         return -EINVAL;
 269
 270                 if (mptcp_can_accept_new_subflow(subflow_req->msk))
 271                         subflow_req->mp_join = 1;
 272
 273                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 274         }
 275
 276         return 0;
 277 }
 278 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
 279
 280 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
 281                                               struct sk_buff *skb,
 282                                               struct flowi *fl,
 283                                               struct request_sock *req)
 284 {
 285         struct dst_entry *dst;
 286         int err;
 287
 288         tcp_rsk(req)->is_mptcp = 1;
 289         subflow_init_req(req, sk);
 290
 291         dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req);
 292         if (!dst)
 293                 return NULL;
 294
 295         err = subflow_check_req(req, sk, skb);
 296         if (err == 0)
 297                 return dst;
 298
 299         dst_release(dst);
 300         if (!req->syncookie)
 301                 tcp_request_sock_ops.send_reset(sk, skb);
 302         return NULL;
 303 }
 304
 305 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 306 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
 307                                               struct sk_buff *skb,
 308                                               struct flowi *fl,
 309                                               struct request_sock *req)
 310 {
 311         struct dst_entry *dst;
 312         int err;
 313
 314         tcp_rsk(req)->is_mptcp = 1;
 315         subflow_init_req(req, sk);
 316
 317         dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req);
 318         if (!dst)
 319                 return NULL;
 320
 321         err = subflow_check_req(req, sk, skb);
 322         if (err == 0)
 323                 return dst;
 324
 325         dst_release(dst);
 326         if (!req->syncookie)
 327                 tcp6_request_sock_ops.send_reset(sk, skb);
 328         return NULL;
 329 }
 330 #endif
 331
 332 /* validate received truncated hmac and create hmac for third ACK */
 333 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
 334 {
 335         u8 hmac[SHA256_DIGEST_SIZE];
 336         u64 thmac;
 337
 338         subflow_generate_hmac(subflow->remote_key, subflow->local_key,
 339                               subflow->remote_nonce, subflow->local_nonce,
 340                               hmac);
 341
 342         thmac = get_unaligned_be64(hmac);
 343         pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
 344                  subflow, subflow->token,
 345                  (unsigned long long)thmac,
 346                  (unsigned long long)subflow->thmac);
 347
 348         return thmac == subflow->thmac;
 349 }
 350
 351 void mptcp_subflow_reset(struct sock *ssk)
 352 {
 353         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 354         struct sock *sk = subflow->conn;
 355
 356         /* must hold: tcp_done() could drop last reference on parent */
 357         sock_hold(sk);
 358
 359         tcp_set_state(ssk, TCP_CLOSE);
 360         tcp_send_active_reset(ssk, GFP_ATOMIC);
 361         tcp_done(ssk);
 362         if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags) &&
 363             schedule_work(&mptcp_sk(sk)->work))
 364                 return; /* worker will put sk for us */
 365
 366         sock_put(sk);
 367 }
 368
 369 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
 370 {
 371         return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
 372 }
 373
 374 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 375 {
 376         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 377         struct mptcp_options_received mp_opt;
 378         struct sock *parent = subflow->conn;
 379
 380         subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 381
 382         if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
 383                 inet_sk_state_store(parent, TCP_ESTABLISHED);
 384                 parent->sk_state_change(parent);
 385         }
 386
 387         /* be sure no special action on any packet other than syn-ack */
 388         if (subflow->conn_finished)
 389                 return;
 390
 391         mptcp_propagate_sndbuf(parent, sk);
 392         subflow->rel_write_seq = 1;
 393         subflow->conn_finished = 1;
 394         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 395         pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
 396
 397         mptcp_get_options(skb, &mp_opt);
 398         if (subflow->request_mptcp) {
 399                 if (!mp_opt.mp_capable) {
 400                         MPTCP_INC_STATS(sock_net(sk),
 401                                         MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
 402                         mptcp_do_fallback(sk);
 403                         pr_fallback(mptcp_sk(subflow->conn));
 404                         goto fallback;
 405                 }
 406
 407                 subflow->mp_capable = 1;
 408                 subflow->can_ack = 1;
 409                 subflow->remote_key = mp_opt.sndr_key;
 410                 pr_debug("subflow=%p, remote_key=%llu", subflow,
 411                          subflow->remote_key);
 412                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
 413                 mptcp_finish_connect(sk);
 414         } else if (subflow->request_join) {
 415                 u8 hmac[SHA256_DIGEST_SIZE];
 416
 417                 if (!mp_opt.mp_join) {
 418                         subflow->reset_reason = MPTCP_RST_EMPTCP;
 419                         goto do_reset;
 420                 }
 421
 422                 subflow->thmac = mp_opt.thmac;
 423                 subflow->remote_nonce = mp_opt.nonce;
 424                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
 425                          subflow->thmac, subflow->remote_nonce);
 426
 427                 if (!subflow_thmac_valid(subflow)) {
 428                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 429                         subflow->reset_reason = MPTCP_RST_EMPTCP;
 430                         goto do_reset;
 431                 }
 432
 433                 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 434                                       subflow->local_nonce,
 435                                       subflow->remote_nonce,
 436                                       hmac);
 437                 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
 438
 439                 if (!mptcp_finish_join(sk))
 440                         goto do_reset;
 441
 442                 subflow->mp_join = 1;
 443                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 444
 445                 if (subflow_use_different_dport(mptcp_sk(parent), sk)) {
 446                         pr_debug("synack inet_dport=%d %d",
 447                                  ntohs(inet_sk(sk)->inet_dport),
 448                                  ntohs(inet_sk(parent)->inet_dport));
 449                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
 450                 }
 451         } else if (mptcp_check_fallback(sk)) {
 452 fallback:
 453                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
 454         }
 455         return;
 456
 457 do_reset:
 458         subflow->reset_transient = 0;
 459         mptcp_subflow_reset(sk);
 460 }
 461
 462 struct request_sock_ops mptcp_subflow_request_sock_ops;
 463 EXPORT_SYMBOL_GPL(mptcp_subflow_request_sock_ops);
 464 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
 465
 466 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 467 {
 468         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 469
 470         pr_debug("subflow=%p", subflow);
 471
 472         /* Never answer to SYNs sent to broadcast or multicast */
 473         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 474                 goto drop;
 475
 476         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 477                                 &subflow_request_sock_ipv4_ops,
 478                                 sk, skb);
 479 drop:
 480         tcp_listendrop(sk);
 481         return 0;
 482 }
 483
 484 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 485 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
 486 static struct inet_connection_sock_af_ops subflow_v6_specific;
 487 static struct inet_connection_sock_af_ops subflow_v6m_specific;
 488 static struct proto tcpv6_prot_override;
 489
 490 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 491 {
 492         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 493
 494         pr_debug("subflow=%p", subflow);
 495
 496         if (skb->protocol == htons(ETH_P_IP))
 497                 return subflow_v4_conn_request(sk, skb);
 498
 499         if (!ipv6_unicast_destination(skb))
 500                 goto drop;
 501
 502         if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
 503                 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
 504                 return 0;
 505         }
 506
 507         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 508                                 &subflow_request_sock_ipv6_ops, sk, skb);
 509
 510 drop:
 511         tcp_listendrop(sk);
 512         return 0; /* don't send reset */
 513 }
 514 #endif
 515
 516 /* validate hmac received in third ACK */
 517 static bool subflow_hmac_valid(const struct request_sock *req,
 518                                const struct mptcp_options_received *mp_opt)
 519 {
 520         const struct mptcp_subflow_request_sock *subflow_req;
 521         u8 hmac[SHA256_DIGEST_SIZE];
 522         struct mptcp_sock *msk;
 523
 524         subflow_req = mptcp_subflow_rsk(req);
 525         msk = subflow_req->msk;
 526         if (!msk)
 527                 return false;
 528
 529         subflow_generate_hmac(msk->remote_key, msk->local_key,
 530                               subflow_req->remote_nonce,
 531                               subflow_req->local_nonce, hmac);
 532
 533         return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
 534 }
 535
 536 static void mptcp_sock_destruct(struct sock *sk)
 537 {
 538         /* if new mptcp socket isn't accepted, it is free'd
 539          * from the tcp listener sockets request queue, linked
 540          * from req->sk.  The tcp socket is released.
 541          * This calls the ULP release function which will
 542          * also remove the mptcp socket, via
 543          * sock_put(ctx->conn).
 544          *
 545          * Problem is that the mptcp socket will be in
 546          * ESTABLISHED state and will not have the SOCK_DEAD flag.
 547          * Both result in warnings from inet_sock_destruct.
 548          */
 549         if ((1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
 550                 sk->sk_state = TCP_CLOSE;
 551                 WARN_ON_ONCE(sk->sk_socket);
 552                 sock_orphan(sk);
 553         }
 554
 555         mptcp_destroy_common(mptcp_sk(sk));
 556         inet_sock_destruct(sk);
 557 }
 558
 559 static void mptcp_force_close(struct sock *sk)
 560 {
 561         inet_sk_state_store(sk, TCP_CLOSE);
 562         sk_common_release(sk);
 563 }
 564
 565 static void subflow_ulp_fallback(struct sock *sk,
 566                                  struct mptcp_subflow_context *old_ctx)
 567 {
 568         struct inet_connection_sock *icsk = inet_csk(sk);
 569
 570         mptcp_subflow_tcp_fallback(sk, old_ctx);
 571         icsk->icsk_ulp_ops = NULL;
 572         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 573         tcp_sk(sk)->is_mptcp = 0;
 574
 575         mptcp_subflow_ops_undo_override(sk);
 576 }
 577
 578 static void subflow_drop_ctx(struct sock *ssk)
 579 {
 580         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 581
 582         if (!ctx)
 583                 return;
 584
 585         subflow_ulp_fallback(ssk, ctx);
 586         if (ctx->conn)
 587                 sock_put(ctx->conn);
 588
 589         kfree_rcu(ctx, rcu);
 590 }
 591
 592 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 593                                      struct mptcp_options_received *mp_opt)
 594 {
 595         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 596
 597         subflow->remote_key = mp_opt->sndr_key;
 598         subflow->fully_established = 1;
 599         subflow->can_ack = 1;
 600         WRITE_ONCE(msk->fully_established, true);
 601 }
 602
 603 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 604                                           struct sk_buff *skb,
 605                                           struct request_sock *req,
 606                                           struct dst_entry *dst,
 607                                           struct request_sock *req_unhash,
 608                                           bool *own_req)
 609 {
 610         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 611         struct mptcp_subflow_request_sock *subflow_req;
 612         struct mptcp_options_received mp_opt;
 613         bool fallback, fallback_is_fatal;
 614         struct sock *new_msk = NULL;
 615         struct sock *child;
 616
 617         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 618
 619         /* After child creation we must look for 'mp_capable' even when options
 620          * are not parsed
 621          */
 622         mp_opt.mp_capable = 0;
 623
 624         /* hopefully temporary handling for MP_JOIN+syncookie */
 625         subflow_req = mptcp_subflow_rsk(req);
 626         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 627         fallback = !tcp_rsk(req)->is_mptcp;
 628         if (fallback)
 629                 goto create_child;
 630
 631         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 632         if (subflow_req->mp_capable) {
 633                 if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
 634                         /* here we can receive and accept an in-window,
 635                          * out-of-order pkt, which will not carry the MP_CAPABLE
 636                          * opt even on mptcp enabled paths
 637                          */
 638                         goto create_msk;
 639                 }
 640
 641                 mptcp_get_options(skb, &mp_opt);
 642                 if (!mp_opt.mp_capable) {
 643                         fallback = true;
 644                         goto create_child;
 645                 }
 646
 647 create_msk:
 648                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 649                 if (!new_msk)
 650                         fallback = true;
 651         } else if (subflow_req->mp_join) {
 652                 mptcp_get_options(skb, &mp_opt);
 653                 if (!mp_opt.mp_join || !subflow_hmac_valid(req, &mp_opt) ||
 654                     !mptcp_can_accept_new_subflow(subflow_req->msk)) {
 655                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 656                         fallback = true;
 657                 }
 658         }
 659
 660 create_child:
 661         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 662                                                      req_unhash, own_req);
 663
 664         if (child && *own_req) {
 665                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 666
 667                 tcp_rsk(req)->drop_req = false;
 668
 669                 /* we need to fallback on ctx allocation failure and on pre-reqs
 670                  * checking above. In the latter scenario we additionally need
 671                  * to reset the context to non MPTCP status.
 672                  */
 673                 if (!ctx || fallback) {
 674                         if (fallback_is_fatal) {
 675                                 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 676                                 goto dispose_child;
 677                         }
 678
 679                         subflow_drop_ctx(child);
 680                         goto out;
 681                 }
 682
 683                 /* ssk inherits options of listener sk */
 684                 ctx->setsockopt_seq = listener->setsockopt_seq;
 685
 686                 if (ctx->mp_capable) {
 687                         /* this can't race with mptcp_close(), as the msk is
 688                          * not yet exposted to user-space
 689                          */
 690                         inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
 691
 692                         /* record the newly created socket as the first msk
 693                          * subflow, but don't link it yet into conn_list
 694                          */
 695                         WRITE_ONCE(mptcp_sk(new_msk)->first, child);
 696
 697                         /* new mpc subflow takes ownership of the newly
 698                          * created mptcp socket
 699                          */
 700                         new_msk->sk_destruct = mptcp_sock_destruct;
 701                         mptcp_sk(new_msk)->setsockopt_seq = ctx->setsockopt_seq;
 702                         mptcp_pm_new_connection(mptcp_sk(new_msk), child, 1);
 703                         mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
 704                         ctx->conn = new_msk;
 705                         new_msk = NULL;
 706
 707                         /* with OoO packets we can reach here without ingress
 708                          * mpc option
 709                          */
 710                         if (mp_opt.mp_capable)
 711                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 712                 } else if (ctx->mp_join) {
 713                         struct mptcp_sock *owner;
 714
 715                         owner = subflow_req->msk;
 716                         if (!owner) {
 717                                 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 718                                 goto dispose_child;
 719                         }
 720
 721                         /* move the msk reference ownership to the subflow */
 722                         subflow_req->msk = NULL;
 723                         ctx->conn = (struct sock *)owner;
 724
 725                         if (subflow_use_different_sport(owner, sk)) {
 726                                 pr_debug("ack inet_sport=%d %d",
 727                                          ntohs(inet_sk(sk)->inet_sport),
 728                                          ntohs(inet_sk((struct sock *)owner)->inet_sport));
 729                                 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
 730                                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
 731                                         goto dispose_child;
 732                                 }
 733                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
 734                         }
 735
 736                         if (!mptcp_finish_join(child))
 737                                 goto dispose_child;
 738
 739                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 740                         tcp_rsk(req)->drop_req = true;
 741                 }
 742         }
 743
 744 out:
 745         /* dispose of the left over mptcp master, if any */
 746         if (unlikely(new_msk))
 747                 mptcp_force_close(new_msk);
 748
 749         /* check for expected invariant - should never trigger, just help
 750          * catching eariler subtle bugs
 751          */
 752         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 753                      (!mptcp_subflow_ctx(child) ||
 754                       !mptcp_subflow_ctx(child)->conn));
 755         return child;
 756
 757 dispose_child:
 758         subflow_drop_ctx(child);
 759         tcp_rsk(req)->drop_req = true;
 760         inet_csk_prepare_for_destroy_sock(child);
 761         tcp_done(child);
 762         req->rsk_ops->send_reset(sk, skb);
 763
 764         /* The last child reference will be released by the caller */
 765         return child;
 766 }
 767
 768 static struct inet_connection_sock_af_ops subflow_specific;
 769 static struct proto tcp_prot_override;
 770
 771 enum mapping_status {
 772         MAPPING_OK,
 773         MAPPING_INVALID,
 774         MAPPING_EMPTY,
 775         MAPPING_DATA_FIN,
 776         MAPPING_DUMMY
 777 };
 778
 779 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
 780 {
 781         if ((u32)seq == (u32)old_seq)
 782                 return old_seq;
 783
 784         /* Assume map covers data not mapped yet. */
 785         return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
 786 }
 787
 788 static void warn_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 789 {
 790         WARN_ONCE(1, "Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 791                   ssn, subflow->map_subflow_seq, subflow->map_data_len);
 792 }
 793
 794 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 795 {
 796         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 797         unsigned int skb_consumed;
 798
 799         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 800         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 801                 return true;
 802
 803         return skb->len - skb_consumed <= subflow->map_data_len -
 804                                           mptcp_subflow_get_map_offset(subflow);
 805 }
 806
 807 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 808 {
 809         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 810         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 811
 812         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 813                 /* Mapping covers data later in the subflow stream,
 814                  * currently unsupported.
 815                  */
 816                 warn_bad_map(subflow, ssn);
 817                 return false;
 818         }
 819         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 820                                   subflow->map_data_len))) {
 821                 /* Mapping does covers past subflow data, invalid */
 822                 warn_bad_map(subflow, ssn + skb->len);
 823                 return false;
 824         }
 825         return true;
 826 }
 827
 828 static enum mapping_status get_mapping_status(struct sock *ssk,
 829                                               struct mptcp_sock *msk)
 830 {
 831         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 832         struct mptcp_ext *mpext;
 833         struct sk_buff *skb;
 834         u16 data_len;
 835         u64 map_seq;
 836
 837         skb = skb_peek(&ssk->sk_receive_queue);
 838         if (!skb)
 839                 return MAPPING_EMPTY;
 840
 841         if (mptcp_check_fallback(ssk))
 842                 return MAPPING_DUMMY;
 843
 844         mpext = mptcp_get_ext(skb);
 845         if (!mpext || !mpext->use_map) {
 846                 if (!subflow->map_valid && !skb->len) {
 847                         /* the TCP stack deliver 0 len FIN pkt to the receive
 848                          * queue, that is the only 0len pkts ever expected here,
 849                          * and we can admit no mapping only for 0 len pkts
 850                          */
 851                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 852                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 853                                           TCP_SKB_CB(skb)->seq,
 854                                           TCP_SKB_CB(skb)->end_seq,
 855                                           TCP_SKB_CB(skb)->tcp_flags);
 856                         sk_eat_skb(ssk, skb);
 857                         return MAPPING_EMPTY;
 858                 }
 859
 860                 if (!subflow->map_valid)
 861                         return MAPPING_INVALID;
 862
 863                 goto validate_seq;
 864         }
 865
 866         trace_get_mapping_status(mpext);
 867
 868         data_len = mpext->data_len;
 869         if (data_len == 0) {
 870                 pr_err("Infinite mapping not handled");
 871                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 872                 return MAPPING_INVALID;
 873         }
 874
 875         if (mpext->data_fin == 1) {
 876                 if (data_len == 1) {
 877                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
 878                                                                  mpext->dsn64);
 879                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
 880                         if (subflow->map_valid) {
 881                                 /* A DATA_FIN might arrive in a DSS
 882                                  * option before the previous mapping
 883                                  * has been fully consumed. Continue
 884                                  * handling the existing mapping.
 885                                  */
 886                                 skb_ext_del(skb, SKB_EXT_MPTCP);
 887                                 return MAPPING_OK;
 888                         } else {
 889                                 if (updated && schedule_work(&msk->work))
 890                                         sock_hold((struct sock *)msk);
 891
 892                                 return MAPPING_DATA_FIN;
 893                         }
 894                 } else {
 895                         u64 data_fin_seq = mpext->data_seq + data_len - 1;
 896
 897                         /* If mpext->data_seq is a 32-bit value, data_fin_seq
 898                          * must also be limited to 32 bits.
 899                          */
 900                         if (!mpext->dsn64)
 901                                 data_fin_seq &= GENMASK_ULL(31, 0);
 902
 903                         mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
 904                         pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
 905                                  data_fin_seq, mpext->dsn64);
 906                 }
 907
 908                 /* Adjust for DATA_FIN using 1 byte of sequence space */
 909                 data_len--;
 910         }
 911
 912         if (!mpext->dsn64) {
 913                 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
 914                                      mpext->data_seq);
 915                 pr_debug("expanded seq=%llu", subflow->map_seq);
 916         } else {
 917                 map_seq = mpext->data_seq;
 918         }
 919         WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
 920
 921         if (subflow->map_valid) {
 922                 /* Allow replacing only with an identical map */
 923                 if (subflow->map_seq == map_seq &&
 924                     subflow->map_subflow_seq == mpext->subflow_seq &&
 925                     subflow->map_data_len == data_len) {
 926                         skb_ext_del(skb, SKB_EXT_MPTCP);
 927                         return MAPPING_OK;
 928                 }
 929
 930                 /* If this skb data are fully covered by the current mapping,
 931                  * the new map would need caching, which is not supported
 932                  */
 933                 if (skb_is_fully_mapped(ssk, skb)) {
 934                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
 935                         return MAPPING_INVALID;
 936                 }
 937
 938                 /* will validate the next map after consuming the current one */
 939                 return MAPPING_OK;
 940         }
 941
 942         subflow->map_seq = map_seq;
 943         subflow->map_subflow_seq = mpext->subflow_seq;
 944         subflow->map_data_len = data_len;
 945         subflow->map_valid = 1;
 946         subflow->mpc_map = mpext->mpc_map;
 947         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
 948                  subflow->map_seq, subflow->map_subflow_seq,
 949                  subflow->map_data_len);
 950
 951 validate_seq:
 952         /* we revalidate valid mapping on new skb, because we must ensure
 953          * the current skb is completely covered by the available mapping
 954          */
 955         if (!validate_mapping(ssk, skb))
 956                 return MAPPING_INVALID;
 957
 958         skb_ext_del(skb, SKB_EXT_MPTCP);
 959         return MAPPING_OK;
 960 }
 961
 962 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
 963                                        u64 limit)
 964 {
 965         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 966         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
 967         u32 incr;
 968
 969         incr = limit >= skb->len ? skb->len + fin : limit;
 970
 971         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
 972                  subflow->map_subflow_seq);
 973         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
 974         tcp_sk(ssk)->copied_seq += incr;
 975         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
 976                 sk_eat_skb(ssk, skb);
 977         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
 978                 subflow->map_valid = 0;
 979 }
 980
 981 /* sched mptcp worker to remove the subflow if no more data is pending */
 982 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
 983 {
 984         struct sock *sk = (struct sock *)msk;
 985
 986         if (likely(ssk->sk_state != TCP_CLOSE))
 987                 return;
 988
 989         if (skb_queue_empty(&ssk->sk_receive_queue) &&
 990             !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags)) {
 991                 sock_hold(sk);
 992                 if (!schedule_work(&msk->work))
 993                         sock_put(sk);
 994         }
 995 }
 996
 997 static bool subflow_check_data_avail(struct sock *ssk)
 998 {
 999         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1000         enum mapping_status status;
1001         struct mptcp_sock *msk;
1002         struct sk_buff *skb;
1003
1004         if (!skb_peek(&ssk->sk_receive_queue))
1005                 subflow->data_avail = 0;
1006         if (subflow->data_avail)
1007                 return true;
1008
1009         msk = mptcp_sk(subflow->conn);
1010         for (;;) {
1011                 u64 ack_seq;
1012                 u64 old_ack;
1013
1014                 status = get_mapping_status(ssk, msk);
1015                 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1016                 if (status == MAPPING_INVALID) {
1017                         ssk->sk_err = EBADMSG;
1018                         goto fatal;
1019                 }
1020                 if (status == MAPPING_DUMMY) {
1021                         __mptcp_do_fallback(msk);
1022                         skb = skb_peek(&ssk->sk_receive_queue);
1023                         subflow->map_valid = 1;
1024                         subflow->map_seq = READ_ONCE(msk->ack_seq);
1025                         subflow->map_data_len = skb->len;
1026                         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq -
1027                                                    subflow->ssn_offset;
1028                         subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
1029                         return true;
1030                 }
1031
1032                 if (status != MAPPING_OK)
1033                         goto no_data;
1034
1035                 skb = skb_peek(&ssk->sk_receive_queue);
1036                 if (WARN_ON_ONCE(!skb))
1037                         goto no_data;
1038
1039                 /* if msk lacks the remote key, this subflow must provide an
1040                  * MP_CAPABLE-based mapping
1041                  */
1042                 if (unlikely(!READ_ONCE(msk->can_ack))) {
1043                         if (!subflow->mpc_map) {
1044                                 ssk->sk_err = EBADMSG;
1045                                 goto fatal;
1046                         }
1047                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
1048                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
1049                         WRITE_ONCE(msk->can_ack, true);
1050                 }
1051
1052                 old_ack = READ_ONCE(msk->ack_seq);
1053                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1054                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1055                          ack_seq);
1056                 if (ack_seq == old_ack) {
1057                         subflow->data_avail = MPTCP_SUBFLOW_DATA_AVAIL;
1058                         break;
1059                 } else if (after64(ack_seq, old_ack)) {
1060                         subflow->data_avail = MPTCP_SUBFLOW_OOO_DATA;
1061                         break;
1062                 }
1063
1064                 /* only accept in-sequence mapping. Old values are spurious
1065                  * retransmission
1066                  */
1067                 mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1068         }
1069         return true;
1070
1071 no_data:
1072         subflow_sched_work_if_closed(msk, ssk);
1073         return false;
1074 fatal:
1075         /* fatal protocol error, close the socket */
1076         /* This barrier is coupled with smp_rmb() in tcp_poll() */
1077         smp_wmb();
1078         ssk->sk_error_report(ssk);
1079         tcp_set_state(ssk, TCP_CLOSE);
1080         subflow->reset_transient = 0;
1081         subflow->reset_reason = MPTCP_RST_EMPTCP;
1082         tcp_send_active_reset(ssk, GFP_ATOMIC);
1083         subflow->data_avail = 0;
1084         return false;
1085 }
1086
1087 bool mptcp_subflow_data_available(struct sock *sk)
1088 {
1089         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1090
1091         /* check if current mapping is still valid */
1092         if (subflow->map_valid &&
1093             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1094                 subflow->map_valid = 0;
1095                 subflow->data_avail = 0;
1096
1097                 pr_debug("Done with mapping: seq=%u data_len=%u",
1098                          subflow->map_subflow_seq,
1099                          subflow->map_data_len);
1100         }
1101
1102         return subflow_check_data_avail(sk);
1103 }
1104
1105 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1106  * not the ssk one.
1107  *
1108  * In mptcp, rwin is about the mptcp-level connection data.
1109  *
1110  * Data that is still on the ssk rx queue can thus be ignored,
1111  * as far as mptcp peer is concerned that data is still inflight.
1112  * DSS ACK is updated when skb is moved to the mptcp rx queue.
1113  */
1114 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1115 {
1116         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1117         const struct sock *sk = subflow->conn;
1118
1119         *space = __mptcp_space(sk);
1120         *full_space = tcp_full_space(sk);
1121 }
1122
1123 static void subflow_data_ready(struct sock *sk)
1124 {
1125         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1126         u16 state = 1 << inet_sk_state_load(sk);
1127         struct sock *parent = subflow->conn;
1128         struct mptcp_sock *msk;
1129
1130         msk = mptcp_sk(parent);
1131         if (state & TCPF_LISTEN) {
1132                 /* MPJ subflow are removed from accept queue before reaching here,
1133                  * avoid stray wakeups
1134                  */
1135                 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1136                         return;
1137
1138                 set_bit(MPTCP_DATA_READY, &msk->flags);
1139                 parent->sk_data_ready(parent);
1140                 return;
1141         }
1142
1143         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1144                      !subflow->mp_join && !(state & TCPF_CLOSE));
1145
1146         if (mptcp_subflow_data_available(sk))
1147                 mptcp_data_ready(parent, sk);
1148 }
1149
1150 static void subflow_write_space(struct sock *ssk)
1151 {
1152         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1153
1154         mptcp_propagate_sndbuf(sk, ssk);
1155         mptcp_write_space(sk);
1156 }
1157
1158 void __mptcp_error_report(struct sock *sk)
1159 {
1160         struct mptcp_subflow_context *subflow;
1161         struct mptcp_sock *msk = mptcp_sk(sk);
1162
1163         mptcp_for_each_subflow(msk, subflow) {
1164                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1165                 int err = sock_error(ssk);
1166
1167                 if (!err)
1168                         continue;
1169
1170                 /* only propagate errors on fallen-back sockets or
1171                  * on MPC connect
1172                  */
1173                 if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(msk))
1174                         continue;
1175
1176                 inet_sk_state_store(sk, inet_sk_state_load(ssk));
1177                 sk->sk_err = -err;
1178
1179                 /* This barrier is coupled with smp_rmb() in mptcp_poll() */
1180                 smp_wmb();
1181                 sk->sk_error_report(sk);
1182                 break;
1183         }
1184 }
1185
1186 static void subflow_error_report(struct sock *ssk)
1187 {
1188         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1189
1190         mptcp_data_lock(sk);
1191         if (!sock_owned_by_user(sk))
1192                 __mptcp_error_report(sk);
1193         else
1194                 set_bit(MPTCP_ERROR_REPORT,  &mptcp_sk(sk)->flags);
1195         mptcp_data_unlock(sk);
1196 }
1197
1198 static struct inet_connection_sock_af_ops *
1199 subflow_default_af_ops(struct sock *sk)
1200 {
1201 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1202         if (sk->sk_family == AF_INET6)
1203                 return &subflow_v6_specific;
1204 #endif
1205         return &subflow_specific;
1206 }
1207
1208 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1209 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1210 {
1211         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1212         struct inet_connection_sock *icsk = inet_csk(sk);
1213         struct inet_connection_sock_af_ops *target;
1214
1215         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1216
1217         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1218                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1219
1220         if (likely(icsk->icsk_af_ops == target))
1221                 return;
1222
1223         subflow->icsk_af_ops = icsk->icsk_af_ops;
1224         icsk->icsk_af_ops = target;
1225 }
1226 #endif
1227
1228 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1229                          struct sockaddr_storage *addr,
1230                          unsigned short family)
1231 {
1232         memset(addr, 0, sizeof(*addr));
1233         addr->ss_family = family;
1234         if (addr->ss_family == AF_INET) {
1235                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1236
1237                 if (info->family == AF_INET)
1238                         in_addr->sin_addr = info->addr;
1239 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1240                 else if (ipv6_addr_v4mapped(&info->addr6))
1241                         in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1242 #endif
1243                 in_addr->sin_port = info->port;
1244         }
1245 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1246         else if (addr->ss_family == AF_INET6) {
1247                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1248
1249                 if (info->family == AF_INET)
1250                         ipv6_addr_set_v4mapped(info->addr.s_addr,
1251                                                &in6_addr->sin6_addr);
1252                 else
1253                         in6_addr->sin6_addr = info->addr6;
1254                 in6_addr->sin6_port = info->port;
1255         }
1256 #endif
1257 }
1258
1259 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1260                             const struct mptcp_addr_info *remote,
1261                             u8 flags, int ifindex)
1262 {
1263         struct mptcp_sock *msk = mptcp_sk(sk);
1264         struct mptcp_subflow_context *subflow;
1265         struct sockaddr_storage addr;
1266         int remote_id = remote->id;
1267         int local_id = loc->id;
1268         struct socket *sf;
1269         struct sock *ssk;
1270         u32 remote_token;
1271         int addrlen;
1272         int err;
1273
1274         if (!mptcp_is_fully_established(sk))
1275                 return -ENOTCONN;
1276
1277         err = mptcp_subflow_create_socket(sk, &sf);
1278         if (err)
1279                 return err;
1280
1281         ssk = sf->sk;
1282         subflow = mptcp_subflow_ctx(ssk);
1283         do {
1284                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1285         } while (!subflow->local_nonce);
1286
1287         if (!local_id) {
1288                 err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
1289                 if (err < 0)
1290                         goto failed;
1291
1292                 local_id = err;
1293         }
1294
1295         subflow->remote_key = msk->remote_key;
1296         subflow->local_key = msk->local_key;
1297         subflow->token = msk->token;
1298         mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1299
1300         addrlen = sizeof(struct sockaddr_in);
1301 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1302         if (addr.ss_family == AF_INET6)
1303                 addrlen = sizeof(struct sockaddr_in6);
1304 #endif
1305         ssk->sk_bound_dev_if = ifindex;
1306         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1307         if (err)
1308                 goto failed;
1309
1310         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1311         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1312                  remote_token, local_id, remote_id);
1313         subflow->remote_token = remote_token;
1314         subflow->local_id = local_id;
1315         subflow->remote_id = remote_id;
1316         subflow->request_join = 1;
1317         subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1318         mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1319
1320         mptcp_add_pending_subflow(msk, subflow);
1321         mptcp_sockopt_sync(msk, ssk);
1322         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1323         if (err && err != -EINPROGRESS)
1324                 goto failed_unlink;
1325
1326         /* discard the subflow socket */
1327         mptcp_sock_graft(ssk, sk->sk_socket);
1328         iput(SOCK_INODE(sf));
1329         return err;
1330
1331 failed_unlink:
1332         spin_lock_bh(&msk->join_list_lock);
1333         list_del(&subflow->node);
1334         spin_unlock_bh(&msk->join_list_lock);
1335         sock_put(mptcp_subflow_tcp_sock(subflow));
1336
1337 failed:
1338         subflow->disposable = 1;
1339         sock_release(sf);
1340         return err;
1341 }
1342
1343 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1344 {
1345 #ifdef CONFIG_SOCK_CGROUP_DATA
1346         struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1347                                 *child_skcd = &child->sk_cgrp_data;
1348
1349         /* only the additional subflows created by kworkers have to be modified */
1350         if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1351             cgroup_id(sock_cgroup_ptr(child_skcd))) {
1352 #ifdef CONFIG_MEMCG
1353                 struct mem_cgroup *memcg = parent->sk_memcg;
1354
1355                 mem_cgroup_sk_free(child);
1356                 if (memcg && css_tryget(&memcg->css))
1357                         child->sk_memcg = memcg;
1358 #endif /* CONFIG_MEMCG */
1359
1360                 cgroup_sk_free(child_skcd);
1361                 *child_skcd = *parent_skcd;
1362                 cgroup_sk_clone(child_skcd);
1363         }
1364 #endif /* CONFIG_SOCK_CGROUP_DATA */
1365 }
1366
1367 static void mptcp_subflow_ops_override(struct sock *ssk)
1368 {
1369 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1370         if (ssk->sk_prot == &tcpv6_prot)
1371                 ssk->sk_prot = &tcpv6_prot_override;
1372         else
1373 #endif
1374                 ssk->sk_prot = &tcp_prot_override;
1375 }
1376
1377 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1378 {
1379 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1380         if (ssk->sk_prot == &tcpv6_prot_override)
1381                 ssk->sk_prot = &tcpv6_prot;
1382         else
1383 #endif
1384                 ssk->sk_prot = &tcp_prot;
1385 }
1386 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1387 {
1388         struct mptcp_subflow_context *subflow;
1389         struct net *net = sock_net(sk);
1390         struct socket *sf;
1391         int err;
1392
1393         /* un-accepted server sockets can reach here - on bad configuration
1394          * bail early to avoid greater trouble later
1395          */
1396         if (unlikely(!sk->sk_socket))
1397                 return -EINVAL;
1398
1399         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1400                                &sf);
1401         if (err)
1402                 return err;
1403
1404         lock_sock(sf->sk);
1405
1406         /* the newly created socket has to be in the same cgroup as its parent */
1407         mptcp_attach_cgroup(sk, sf->sk);
1408
1409         /* kernel sockets do not by default acquire net ref, but TCP timer
1410          * needs it.
1411          */
1412         sf->sk->sk_net_refcnt = 1;
1413         get_net(net);
1414 #ifdef CONFIG_PROC_FS
1415         this_cpu_add(*net->core.sock_inuse, 1);
1416 #endif
1417         err = tcp_set_ulp(sf->sk, "mptcp");
1418         release_sock(sf->sk);
1419
1420         if (err) {
1421                 sock_release(sf);
1422                 return err;
1423         }
1424
1425         /* the newly created socket really belongs to the owning MPTCP master
1426          * socket, even if for additional subflows the allocation is performed
1427          * by a kernel workqueue. Adjust inode references, so that the
1428          * procfs/diag interaces really show this one belonging to the correct
1429          * user.
1430          */
1431         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1432         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1433         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1434
1435         subflow = mptcp_subflow_ctx(sf->sk);
1436         pr_debug("subflow=%p", subflow);
1437
1438         *new_sock = sf;
1439         sock_hold(sk);
1440         subflow->conn = sk;
1441         mptcp_subflow_ops_override(sf->sk);
1442
1443         return 0;
1444 }
1445
1446 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1447                                                         gfp_t priority)
1448 {
1449         struct inet_connection_sock *icsk = inet_csk(sk);
1450         struct mptcp_subflow_context *ctx;
1451
1452         ctx = kzalloc(sizeof(*ctx), priority);
1453         if (!ctx)
1454                 return NULL;
1455
1456         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1457         INIT_LIST_HEAD(&ctx->node);
1458         INIT_LIST_HEAD(&ctx->delegated_node);
1459
1460         pr_debug("subflow=%p", ctx);
1461
1462         ctx->tcp_sock = sk;
1463
1464         return ctx;
1465 }
1466
1467 static void __subflow_state_change(struct sock *sk)
1468 {
1469         struct socket_wq *wq;
1470
1471         rcu_read_lock();
1472         wq = rcu_dereference(sk->sk_wq);
1473         if (skwq_has_sleeper(wq))
1474                 wake_up_interruptible_all(&wq->wait);
1475         rcu_read_unlock();
1476 }
1477
1478 static bool subflow_is_done(const struct sock *sk)
1479 {
1480         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1481 }
1482
1483 static void subflow_state_change(struct sock *sk)
1484 {
1485         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1486         struct sock *parent = subflow->conn;
1487
1488         __subflow_state_change(sk);
1489
1490         if (subflow_simultaneous_connect(sk)) {
1491                 mptcp_propagate_sndbuf(parent, sk);
1492                 mptcp_do_fallback(sk);
1493                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1494                 pr_fallback(mptcp_sk(parent));
1495                 subflow->conn_finished = 1;
1496                 if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
1497                         inet_sk_state_store(parent, TCP_ESTABLISHED);
1498                         parent->sk_state_change(parent);
1499                 }
1500         }
1501
1502         /* as recvmsg() does not acquire the subflow socket for ssk selection
1503          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1504          * the data available machinery here.
1505          */
1506         if (mptcp_subflow_data_available(sk))
1507                 mptcp_data_ready(parent, sk);
1508
1509         subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1510
1511         if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1512             !subflow->rx_eof && subflow_is_done(sk)) {
1513                 subflow->rx_eof = 1;
1514                 mptcp_subflow_eof(parent);
1515         }
1516 }
1517
1518 static int subflow_ulp_init(struct sock *sk)
1519 {
1520         struct inet_connection_sock *icsk = inet_csk(sk);
1521         struct mptcp_subflow_context *ctx;
1522         struct tcp_sock *tp = tcp_sk(sk);
1523         int err = 0;
1524
1525         /* disallow attaching ULP to a socket unless it has been
1526          * created with sock_create_kern()
1527          */
1528         if (!sk->sk_kern_sock) {
1529                 err = -EOPNOTSUPP;
1530                 goto out;
1531         }
1532
1533         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1534         if (!ctx) {
1535                 err = -ENOMEM;
1536                 goto out;
1537         }
1538
1539         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1540
1541         tp->is_mptcp = 1;
1542         ctx->icsk_af_ops = icsk->icsk_af_ops;
1543         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1544         ctx->tcp_data_ready = sk->sk_data_ready;
1545         ctx->tcp_state_change = sk->sk_state_change;
1546         ctx->tcp_write_space = sk->sk_write_space;
1547         ctx->tcp_error_report = sk->sk_error_report;
1548         sk->sk_data_ready = subflow_data_ready;
1549         sk->sk_write_space = subflow_write_space;
1550         sk->sk_state_change = subflow_state_change;
1551         sk->sk_error_report = subflow_error_report;
1552 out:
1553         return err;
1554 }
1555
1556 static void subflow_ulp_release(struct sock *ssk)
1557 {
1558         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1559         bool release = true;
1560         struct sock *sk;
1561
1562         if (!ctx)
1563                 return;
1564
1565         sk = ctx->conn;
1566         if (sk) {
1567                 /* if the msk has been orphaned, keep the ctx
1568                  * alive, will be freed by __mptcp_close_ssk(),
1569                  * when the subflow is still unaccepted
1570                  */
1571                 release = ctx->disposable || list_empty(&ctx->node);
1572                 sock_put(sk);
1573         }
1574
1575         mptcp_subflow_ops_undo_override(ssk);
1576         if (release)
1577                 kfree_rcu(ctx, rcu);
1578 }
1579
1580 static void subflow_ulp_clone(const struct request_sock *req,
1581                               struct sock *newsk,
1582                               const gfp_t priority)
1583 {
1584         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1585         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1586         struct mptcp_subflow_context *new_ctx;
1587
1588         if (!tcp_rsk(req)->is_mptcp ||
1589             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1590                 subflow_ulp_fallback(newsk, old_ctx);
1591                 return;
1592         }
1593
1594         new_ctx = subflow_create_ctx(newsk, priority);
1595         if (!new_ctx) {
1596                 subflow_ulp_fallback(newsk, old_ctx);
1597                 return;
1598         }
1599
1600         new_ctx->conn_finished = 1;
1601         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1602         new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1603         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1604         new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1605         new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1606         new_ctx->rel_write_seq = 1;
1607         new_ctx->tcp_sock = newsk;
1608
1609         if (subflow_req->mp_capable) {
1610                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1611                  * is fully established only after we receive the remote key
1612                  */
1613                 new_ctx->mp_capable = 1;
1614                 new_ctx->local_key = subflow_req->local_key;
1615                 new_ctx->token = subflow_req->token;
1616                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1617                 new_ctx->idsn = subflow_req->idsn;
1618         } else if (subflow_req->mp_join) {
1619                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1620                 new_ctx->mp_join = 1;
1621                 new_ctx->fully_established = 1;
1622                 new_ctx->backup = subflow_req->backup;
1623                 new_ctx->local_id = subflow_req->local_id;
1624                 new_ctx->remote_id = subflow_req->remote_id;
1625                 new_ctx->token = subflow_req->token;
1626                 new_ctx->thmac = subflow_req->thmac;
1627         }
1628 }
1629
1630 static void tcp_release_cb_override(struct sock *ssk)
1631 {
1632         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1633
1634         if (mptcp_subflow_has_delegated_action(subflow))
1635                 mptcp_subflow_process_delegated(ssk);
1636
1637         tcp_release_cb(ssk);
1638 }
1639
1640 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1641         .name           = "mptcp",
1642         .owner          = THIS_MODULE,
1643         .init           = subflow_ulp_init,
1644         .release        = subflow_ulp_release,
1645         .clone          = subflow_ulp_clone,
1646 };
1647
1648 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1649 {
1650         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1651         subflow_ops->slab_name = "request_sock_subflow";
1652
1653         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1654                                               subflow_ops->obj_size, 0,
1655                                               SLAB_ACCOUNT |
1656                                               SLAB_TYPESAFE_BY_RCU,
1657                                               NULL);
1658         if (!subflow_ops->slab)
1659                 return -ENOMEM;
1660
1661         subflow_ops->destructor = subflow_req_destructor;
1662
1663         return 0;
1664 }
1665
1666 void __init mptcp_subflow_init(void)
1667 {
1668         mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1669         if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1670                 panic("MPTCP: failed to init subflow request sock ops\n");
1671
1672         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1673         subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
1674
1675         subflow_specific = ipv4_specific;
1676         subflow_specific.conn_request = subflow_v4_conn_request;
1677         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1678         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1679
1680         tcp_prot_override = tcp_prot;
1681         tcp_prot_override.release_cb = tcp_release_cb_override;
1682
1683 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1684         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1685         subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
1686
1687         subflow_v6_specific = ipv6_specific;
1688         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1689         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1690         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1691
1692         subflow_v6m_specific = subflow_v6_specific;
1693         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1694         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1695         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1696         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1697         subflow_v6m_specific.net_frag_header_len = 0;
1698
1699         tcpv6_prot_override = tcpv6_prot;
1700         tcpv6_prot_override.release_cb = tcp_release_cb_override;
1701 #endif
1702
1703         mptcp_diag_subflow_init(&subflow_ulp_ops);
1704
1705         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1706                 panic("MPTCP: failed to register subflows to ULP\n");
1707 }