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                 /* we can receive and accept an in-window, out-of-order pkt,
 634                  * which may not carry the MP_CAPABLE opt even on mptcp enabled
 635                  * paths: always try to extract the peer key, and fallback
 636                  * for packets missing it.
 637                  * Even OoO DSS packets coming legitly after dropped or
 638                  * reordered MPC will cause fallback, but we don't have other
 639                  * options.
 640                  */
 641                 mptcp_get_options(skb, &mp_opt);
 642                 if (!mp_opt.mp_capable) {
 643                         fallback = true;
 644                         goto create_child;
 645                 }
 646
 647                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 648                 if (!new_msk)
 649                         fallback = true;
 650         } else if (subflow_req->mp_join) {
 651                 mptcp_get_options(skb, &mp_opt);
 652                 if (!mp_opt.mp_join || !subflow_hmac_valid(req, &mp_opt) ||
 653                     !mptcp_can_accept_new_subflow(subflow_req->msk)) {
 654                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 655                         fallback = true;
 656                 }
 657         }
 658
 659 create_child:
 660         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 661                                                      req_unhash, own_req);
 662
 663         if (child && *own_req) {
 664                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 665
 666                 tcp_rsk(req)->drop_req = false;
 667
 668                 /* we need to fallback on ctx allocation failure and on pre-reqs
 669                  * checking above. In the latter scenario we additionally need
 670                  * to reset the context to non MPTCP status.
 671                  */
 672                 if (!ctx || fallback) {
 673                         if (fallback_is_fatal) {
 674                                 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 675                                 goto dispose_child;
 676                         }
 677
 678                         subflow_drop_ctx(child);
 679                         goto out;
 680                 }
 681
 682                 /* ssk inherits options of listener sk */
 683                 ctx->setsockopt_seq = listener->setsockopt_seq;
 684
 685                 if (ctx->mp_capable) {
 686                         /* this can't race with mptcp_close(), as the msk is
 687                          * not yet exposted to user-space
 688                          */
 689                         inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
 690
 691                         /* record the newly created socket as the first msk
 692                          * subflow, but don't link it yet into conn_list
 693                          */
 694                         WRITE_ONCE(mptcp_sk(new_msk)->first, child);
 695
 696                         /* new mpc subflow takes ownership of the newly
 697                          * created mptcp socket
 698                          */
 699                         new_msk->sk_destruct = mptcp_sock_destruct;
 700                         mptcp_sk(new_msk)->setsockopt_seq = ctx->setsockopt_seq;
 701                         mptcp_pm_new_connection(mptcp_sk(new_msk), child, 1);
 702                         mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
 703                         ctx->conn = new_msk;
 704                         new_msk = NULL;
 705
 706                         /* with OoO packets we can reach here without ingress
 707                          * mpc option
 708                          */
 709                         if (mp_opt.mp_capable)
 710                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 711                 } else if (ctx->mp_join) {
 712                         struct mptcp_sock *owner;
 713
 714                         owner = subflow_req->msk;
 715                         if (!owner) {
 716                                 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 717                                 goto dispose_child;
 718                         }
 719
 720                         /* move the msk reference ownership to the subflow */
 721                         subflow_req->msk = NULL;
 722                         ctx->conn = (struct sock *)owner;
 723
 724                         if (subflow_use_different_sport(owner, sk)) {
 725                                 pr_debug("ack inet_sport=%d %d",
 726                                          ntohs(inet_sk(sk)->inet_sport),
 727                                          ntohs(inet_sk((struct sock *)owner)->inet_sport));
 728                                 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
 729                                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
 730                                         goto dispose_child;
 731                                 }
 732                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
 733                         }
 734
 735                         if (!mptcp_finish_join(child))
 736                                 goto dispose_child;
 737
 738                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 739                         tcp_rsk(req)->drop_req = true;
 740                 }
 741         }
 742
 743 out:
 744         /* dispose of the left over mptcp master, if any */
 745         if (unlikely(new_msk))
 746                 mptcp_force_close(new_msk);
 747
 748         /* check for expected invariant - should never trigger, just help
 749          * catching eariler subtle bugs
 750          */
 751         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 752                      (!mptcp_subflow_ctx(child) ||
 753                       !mptcp_subflow_ctx(child)->conn));
 754         return child;
 755
 756 dispose_child:
 757         subflow_drop_ctx(child);
 758         tcp_rsk(req)->drop_req = true;
 759         inet_csk_prepare_for_destroy_sock(child);
 760         tcp_done(child);
 761         req->rsk_ops->send_reset(sk, skb);
 762
 763         /* The last child reference will be released by the caller */
 764         return child;
 765 }
 766
 767 static struct inet_connection_sock_af_ops subflow_specific;
 768 static struct proto tcp_prot_override;
 769
 770 enum mapping_status {
 771         MAPPING_OK,
 772         MAPPING_INVALID,
 773         MAPPING_EMPTY,
 774         MAPPING_DATA_FIN,
 775         MAPPING_DUMMY
 776 };
 777
 778 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
 779 {
 780         if ((u32)seq == (u32)old_seq)
 781                 return old_seq;
 782
 783         /* Assume map covers data not mapped yet. */
 784         return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
 785 }
 786
 787 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 788 {
 789         pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 790                  ssn, subflow->map_subflow_seq, subflow->map_data_len);
 791 }
 792
 793 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 794 {
 795         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 796         unsigned int skb_consumed;
 797
 798         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 799         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 800                 return true;
 801
 802         return skb->len - skb_consumed <= subflow->map_data_len -
 803                                           mptcp_subflow_get_map_offset(subflow);
 804 }
 805
 806 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 807 {
 808         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 809         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 810
 811         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 812                 /* Mapping covers data later in the subflow stream,
 813                  * currently unsupported.
 814                  */
 815                 dbg_bad_map(subflow, ssn);
 816                 return false;
 817         }
 818         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 819                                   subflow->map_data_len))) {
 820                 /* Mapping does covers past subflow data, invalid */
 821                 dbg_bad_map(subflow, ssn);
 822                 return false;
 823         }
 824         return true;
 825 }
 826
 827 static enum mapping_status get_mapping_status(struct sock *ssk,
 828                                               struct mptcp_sock *msk)
 829 {
 830         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 831         struct mptcp_ext *mpext;
 832         struct sk_buff *skb;
 833         u16 data_len;
 834         u64 map_seq;
 835
 836         skb = skb_peek(&ssk->sk_receive_queue);
 837         if (!skb)
 838                 return MAPPING_EMPTY;
 839
 840         if (mptcp_check_fallback(ssk))
 841                 return MAPPING_DUMMY;
 842
 843         mpext = mptcp_get_ext(skb);
 844         if (!mpext || !mpext->use_map) {
 845                 if (!subflow->map_valid && !skb->len) {
 846                         /* the TCP stack deliver 0 len FIN pkt to the receive
 847                          * queue, that is the only 0len pkts ever expected here,
 848                          * and we can admit no mapping only for 0 len pkts
 849                          */
 850                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 851                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 852                                           TCP_SKB_CB(skb)->seq,
 853                                           TCP_SKB_CB(skb)->end_seq,
 854                                           TCP_SKB_CB(skb)->tcp_flags);
 855                         sk_eat_skb(ssk, skb);
 856                         return MAPPING_EMPTY;
 857                 }
 858
 859                 if (!subflow->map_valid)
 860                         return MAPPING_INVALID;
 861
 862                 goto validate_seq;
 863         }
 864
 865         trace_get_mapping_status(mpext);
 866
 867         data_len = mpext->data_len;
 868         if (data_len == 0) {
 869                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 870                 return MAPPING_INVALID;
 871         }
 872
 873         if (mpext->data_fin == 1) {
 874                 if (data_len == 1) {
 875                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
 876                                                                  mpext->dsn64);
 877                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
 878                         if (subflow->map_valid) {
 879                                 /* A DATA_FIN might arrive in a DSS
 880                                  * option before the previous mapping
 881                                  * has been fully consumed. Continue
 882                                  * handling the existing mapping.
 883                                  */
 884                                 skb_ext_del(skb, SKB_EXT_MPTCP);
 885                                 return MAPPING_OK;
 886                         } else {
 887                                 if (updated && schedule_work(&msk->work))
 888                                         sock_hold((struct sock *)msk);
 889
 890                                 return MAPPING_DATA_FIN;
 891                         }
 892                 } else {
 893                         u64 data_fin_seq = mpext->data_seq + data_len - 1;
 894
 895                         /* If mpext->data_seq is a 32-bit value, data_fin_seq
 896                          * must also be limited to 32 bits.
 897                          */
 898                         if (!mpext->dsn64)
 899                                 data_fin_seq &= GENMASK_ULL(31, 0);
 900
 901                         mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
 902                         pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
 903                                  data_fin_seq, mpext->dsn64);
 904                 }
 905
 906                 /* Adjust for DATA_FIN using 1 byte of sequence space */
 907                 data_len--;
 908         }
 909
 910         if (!mpext->dsn64) {
 911                 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
 912                                      mpext->data_seq);
 913                 pr_debug("expanded seq=%llu", subflow->map_seq);
 914         } else {
 915                 map_seq = mpext->data_seq;
 916         }
 917         WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
 918
 919         if (subflow->map_valid) {
 920                 /* Allow replacing only with an identical map */
 921                 if (subflow->map_seq == map_seq &&
 922                     subflow->map_subflow_seq == mpext->subflow_seq &&
 923                     subflow->map_data_len == data_len) {
 924                         skb_ext_del(skb, SKB_EXT_MPTCP);
 925                         return MAPPING_OK;
 926                 }
 927
 928                 /* If this skb data are fully covered by the current mapping,
 929                  * the new map would need caching, which is not supported
 930                  */
 931                 if (skb_is_fully_mapped(ssk, skb)) {
 932                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
 933                         return MAPPING_INVALID;
 934                 }
 935
 936                 /* will validate the next map after consuming the current one */
 937                 return MAPPING_OK;
 938         }
 939
 940         subflow->map_seq = map_seq;
 941         subflow->map_subflow_seq = mpext->subflow_seq;
 942         subflow->map_data_len = data_len;
 943         subflow->map_valid = 1;
 944         subflow->mpc_map = mpext->mpc_map;
 945         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
 946                  subflow->map_seq, subflow->map_subflow_seq,
 947                  subflow->map_data_len);
 948
 949 validate_seq:
 950         /* we revalidate valid mapping on new skb, because we must ensure
 951          * the current skb is completely covered by the available mapping
 952          */
 953         if (!validate_mapping(ssk, skb))
 954                 return MAPPING_INVALID;
 955
 956         skb_ext_del(skb, SKB_EXT_MPTCP);
 957         return MAPPING_OK;
 958 }
 959
 960 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
 961                                        u64 limit)
 962 {
 963         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 964         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
 965         u32 incr;
 966
 967         incr = limit >= skb->len ? skb->len + fin : limit;
 968
 969         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
 970                  subflow->map_subflow_seq);
 971         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
 972         tcp_sk(ssk)->copied_seq += incr;
 973         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
 974                 sk_eat_skb(ssk, skb);
 975         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
 976                 subflow->map_valid = 0;
 977 }
 978
 979 /* sched mptcp worker to remove the subflow if no more data is pending */
 980 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
 981 {
 982         struct sock *sk = (struct sock *)msk;
 983
 984         if (likely(ssk->sk_state != TCP_CLOSE))
 985                 return;
 986
 987         if (skb_queue_empty(&ssk->sk_receive_queue) &&
 988             !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags)) {
 989                 sock_hold(sk);
 990                 if (!schedule_work(&msk->work))
 991                         sock_put(sk);
 992         }
 993 }
 994
 995 static bool subflow_check_data_avail(struct sock *ssk)
 996 {
 997         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 998         enum mapping_status status;
 999         struct mptcp_sock *msk;
1000         struct sk_buff *skb;
1001
1002         if (!skb_peek(&ssk->sk_receive_queue))
1003                 WRITE_ONCE(subflow->data_avail, 0);
1004         if (subflow->data_avail)
1005                 return true;
1006
1007         msk = mptcp_sk(subflow->conn);
1008         for (;;) {
1009                 u64 ack_seq;
1010                 u64 old_ack;
1011
1012                 status = get_mapping_status(ssk, msk);
1013                 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1014                 if (unlikely(status == MAPPING_INVALID))
1015                         goto fallback;
1016
1017                 if (unlikely(status == MAPPING_DUMMY))
1018                         goto fallback;
1019
1020                 if (status != MAPPING_OK)
1021                         goto no_data;
1022
1023                 skb = skb_peek(&ssk->sk_receive_queue);
1024                 if (WARN_ON_ONCE(!skb))
1025                         goto no_data;
1026
1027                 /* if msk lacks the remote key, this subflow must provide an
1028                  * MP_CAPABLE-based mapping
1029                  */
1030                 if (unlikely(!READ_ONCE(msk->can_ack))) {
1031                         if (!subflow->mpc_map)
1032                                 goto fallback;
1033                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
1034                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
1035                         WRITE_ONCE(msk->can_ack, true);
1036                 }
1037
1038                 old_ack = READ_ONCE(msk->ack_seq);
1039                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1040                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1041                          ack_seq);
1042                 if (unlikely(before64(ack_seq, old_ack))) {
1043                         mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1044                         continue;
1045                 }
1046
1047                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1048                 break;
1049         }
1050         return true;
1051
1052 no_data:
1053         subflow_sched_work_if_closed(msk, ssk);
1054         return false;
1055
1056 fallback:
1057         /* RFC 8684 section 3.7. */
1058         if (subflow->mp_join || subflow->fully_established) {
1059                 /* fatal protocol error, close the socket.
1060                  * subflow_error_report() will introduce the appropriate barriers
1061                  */
1062                 ssk->sk_err = EBADMSG;
1063                 tcp_set_state(ssk, TCP_CLOSE);
1064                 subflow->reset_transient = 0;
1065                 subflow->reset_reason = MPTCP_RST_EMPTCP;
1066                 tcp_send_active_reset(ssk, GFP_ATOMIC);
1067                 WRITE_ONCE(subflow->data_avail, 0);
1068                 return false;
1069         }
1070
1071         __mptcp_do_fallback(msk);
1072         skb = skb_peek(&ssk->sk_receive_queue);
1073         subflow->map_valid = 1;
1074         subflow->map_seq = READ_ONCE(msk->ack_seq);
1075         subflow->map_data_len = skb->len;
1076         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1077         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1078         return true;
1079 }
1080
1081 bool mptcp_subflow_data_available(struct sock *sk)
1082 {
1083         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1084
1085         /* check if current mapping is still valid */
1086         if (subflow->map_valid &&
1087             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1088                 subflow->map_valid = 0;
1089                 WRITE_ONCE(subflow->data_avail, 0);
1090
1091                 pr_debug("Done with mapping: seq=%u data_len=%u",
1092                          subflow->map_subflow_seq,
1093                          subflow->map_data_len);
1094         }
1095
1096         return subflow_check_data_avail(sk);
1097 }
1098
1099 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1100  * not the ssk one.
1101  *
1102  * In mptcp, rwin is about the mptcp-level connection data.
1103  *
1104  * Data that is still on the ssk rx queue can thus be ignored,
1105  * as far as mptcp peer is concerned that data is still inflight.
1106  * DSS ACK is updated when skb is moved to the mptcp rx queue.
1107  */
1108 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1109 {
1110         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1111         const struct sock *sk = subflow->conn;
1112
1113         *space = __mptcp_space(sk);
1114         *full_space = tcp_full_space(sk);
1115 }
1116
1117 void __mptcp_error_report(struct sock *sk)
1118 {
1119         struct mptcp_subflow_context *subflow;
1120         struct mptcp_sock *msk = mptcp_sk(sk);
1121
1122         mptcp_for_each_subflow(msk, subflow) {
1123                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1124                 int err = sock_error(ssk);
1125
1126                 if (!err)
1127                         continue;
1128
1129                 /* only propagate errors on fallen-back sockets or
1130                  * on MPC connect
1131                  */
1132                 if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(msk))
1133                         continue;
1134
1135                 inet_sk_state_store(sk, inet_sk_state_load(ssk));
1136                 sk->sk_err = -err;
1137
1138                 /* This barrier is coupled with smp_rmb() in mptcp_poll() */
1139                 smp_wmb();
1140                 sk->sk_error_report(sk);
1141                 break;
1142         }
1143 }
1144
1145 static void subflow_error_report(struct sock *ssk)
1146 {
1147         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1148
1149         mptcp_data_lock(sk);
1150         if (!sock_owned_by_user(sk))
1151                 __mptcp_error_report(sk);
1152         else
1153                 set_bit(MPTCP_ERROR_REPORT,  &mptcp_sk(sk)->flags);
1154         mptcp_data_unlock(sk);
1155 }
1156
1157 static void subflow_data_ready(struct sock *sk)
1158 {
1159         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1160         u16 state = 1 << inet_sk_state_load(sk);
1161         struct sock *parent = subflow->conn;
1162         struct mptcp_sock *msk;
1163
1164         msk = mptcp_sk(parent);
1165         if (state & TCPF_LISTEN) {
1166                 /* MPJ subflow are removed from accept queue before reaching here,
1167                  * avoid stray wakeups
1168                  */
1169                 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1170                         return;
1171
1172                 set_bit(MPTCP_DATA_READY, &msk->flags);
1173                 parent->sk_data_ready(parent);
1174                 return;
1175         }
1176
1177         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1178                      !subflow->mp_join && !(state & TCPF_CLOSE));
1179
1180         if (mptcp_subflow_data_available(sk))
1181                 mptcp_data_ready(parent, sk);
1182         else if (unlikely(sk->sk_err))
1183                 subflow_error_report(sk);
1184 }
1185
1186 static void subflow_write_space(struct sock *ssk)
1187 {
1188         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1189
1190         mptcp_propagate_sndbuf(sk, ssk);
1191         mptcp_write_space(sk);
1192 }
1193
1194 static struct inet_connection_sock_af_ops *
1195 subflow_default_af_ops(struct sock *sk)
1196 {
1197 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1198         if (sk->sk_family == AF_INET6)
1199                 return &subflow_v6_specific;
1200 #endif
1201         return &subflow_specific;
1202 }
1203
1204 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1205 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1206 {
1207         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1208         struct inet_connection_sock *icsk = inet_csk(sk);
1209         struct inet_connection_sock_af_ops *target;
1210
1211         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1212
1213         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1214                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1215
1216         if (likely(icsk->icsk_af_ops == target))
1217                 return;
1218
1219         subflow->icsk_af_ops = icsk->icsk_af_ops;
1220         icsk->icsk_af_ops = target;
1221 }
1222 #endif
1223
1224 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1225                          struct sockaddr_storage *addr,
1226                          unsigned short family)
1227 {
1228         memset(addr, 0, sizeof(*addr));
1229         addr->ss_family = family;
1230         if (addr->ss_family == AF_INET) {
1231                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1232
1233                 if (info->family == AF_INET)
1234                         in_addr->sin_addr = info->addr;
1235 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1236                 else if (ipv6_addr_v4mapped(&info->addr6))
1237                         in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1238 #endif
1239                 in_addr->sin_port = info->port;
1240         }
1241 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1242         else if (addr->ss_family == AF_INET6) {
1243                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1244
1245                 if (info->family == AF_INET)
1246                         ipv6_addr_set_v4mapped(info->addr.s_addr,
1247                                                &in6_addr->sin6_addr);
1248                 else
1249                         in6_addr->sin6_addr = info->addr6;
1250                 in6_addr->sin6_port = info->port;
1251         }
1252 #endif
1253 }
1254
1255 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1256                             const struct mptcp_addr_info *remote,
1257                             u8 flags, int ifindex)
1258 {
1259         struct mptcp_sock *msk = mptcp_sk(sk);
1260         struct mptcp_subflow_context *subflow;
1261         struct sockaddr_storage addr;
1262         int remote_id = remote->id;
1263         int local_id = loc->id;
1264         struct socket *sf;
1265         struct sock *ssk;
1266         u32 remote_token;
1267         int addrlen;
1268         int err;
1269
1270         if (!mptcp_is_fully_established(sk))
1271                 return -ENOTCONN;
1272
1273         err = mptcp_subflow_create_socket(sk, &sf);
1274         if (err)
1275                 return err;
1276
1277         ssk = sf->sk;
1278         subflow = mptcp_subflow_ctx(ssk);
1279         do {
1280                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1281         } while (!subflow->local_nonce);
1282
1283         if (!local_id) {
1284                 err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
1285                 if (err < 0)
1286                         goto failed;
1287
1288                 local_id = err;
1289         }
1290
1291         subflow->remote_key = msk->remote_key;
1292         subflow->local_key = msk->local_key;
1293         subflow->token = msk->token;
1294         mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1295
1296         addrlen = sizeof(struct sockaddr_in);
1297 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1298         if (addr.ss_family == AF_INET6)
1299                 addrlen = sizeof(struct sockaddr_in6);
1300 #endif
1301         ssk->sk_bound_dev_if = ifindex;
1302         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1303         if (err)
1304                 goto failed;
1305
1306         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1307         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1308                  remote_token, local_id, remote_id);
1309         subflow->remote_token = remote_token;
1310         subflow->local_id = local_id;
1311         subflow->remote_id = remote_id;
1312         subflow->request_join = 1;
1313         subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1314         mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1315
1316         mptcp_add_pending_subflow(msk, subflow);
1317         mptcp_sockopt_sync(msk, ssk);
1318         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1319         if (err && err != -EINPROGRESS)
1320                 goto failed_unlink;
1321
1322         /* discard the subflow socket */
1323         mptcp_sock_graft(ssk, sk->sk_socket);
1324         iput(SOCK_INODE(sf));
1325         return err;
1326
1327 failed_unlink:
1328         spin_lock_bh(&msk->join_list_lock);
1329         list_del(&subflow->node);
1330         spin_unlock_bh(&msk->join_list_lock);
1331         sock_put(mptcp_subflow_tcp_sock(subflow));
1332
1333 failed:
1334         subflow->disposable = 1;
1335         sock_release(sf);
1336         return err;
1337 }
1338
1339 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1340 {
1341 #ifdef CONFIG_SOCK_CGROUP_DATA
1342         struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1343                                 *child_skcd = &child->sk_cgrp_data;
1344
1345         /* only the additional subflows created by kworkers have to be modified */
1346         if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1347             cgroup_id(sock_cgroup_ptr(child_skcd))) {
1348 #ifdef CONFIG_MEMCG
1349                 struct mem_cgroup *memcg = parent->sk_memcg;
1350
1351                 mem_cgroup_sk_free(child);
1352                 if (memcg && css_tryget(&memcg->css))
1353                         child->sk_memcg = memcg;
1354 #endif /* CONFIG_MEMCG */
1355
1356                 cgroup_sk_free(child_skcd);
1357                 *child_skcd = *parent_skcd;
1358                 cgroup_sk_clone(child_skcd);
1359         }
1360 #endif /* CONFIG_SOCK_CGROUP_DATA */
1361 }
1362
1363 static void mptcp_subflow_ops_override(struct sock *ssk)
1364 {
1365 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1366         if (ssk->sk_prot == &tcpv6_prot)
1367                 ssk->sk_prot = &tcpv6_prot_override;
1368         else
1369 #endif
1370                 ssk->sk_prot = &tcp_prot_override;
1371 }
1372
1373 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1374 {
1375 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1376         if (ssk->sk_prot == &tcpv6_prot_override)
1377                 ssk->sk_prot = &tcpv6_prot;
1378         else
1379 #endif
1380                 ssk->sk_prot = &tcp_prot;
1381 }
1382 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1383 {
1384         struct mptcp_subflow_context *subflow;
1385         struct net *net = sock_net(sk);
1386         struct socket *sf;
1387         int err;
1388
1389         /* un-accepted server sockets can reach here - on bad configuration
1390          * bail early to avoid greater trouble later
1391          */
1392         if (unlikely(!sk->sk_socket))
1393                 return -EINVAL;
1394
1395         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1396                                &sf);
1397         if (err)
1398                 return err;
1399
1400         lock_sock(sf->sk);
1401
1402         /* the newly created socket has to be in the same cgroup as its parent */
1403         mptcp_attach_cgroup(sk, sf->sk);
1404
1405         /* kernel sockets do not by default acquire net ref, but TCP timer
1406          * needs it.
1407          */
1408         sf->sk->sk_net_refcnt = 1;
1409         get_net(net);
1410 #ifdef CONFIG_PROC_FS
1411         this_cpu_add(*net->core.sock_inuse, 1);
1412 #endif
1413         err = tcp_set_ulp(sf->sk, "mptcp");
1414         release_sock(sf->sk);
1415
1416         if (err) {
1417                 sock_release(sf);
1418                 return err;
1419         }
1420
1421         /* the newly created socket really belongs to the owning MPTCP master
1422          * socket, even if for additional subflows the allocation is performed
1423          * by a kernel workqueue. Adjust inode references, so that the
1424          * procfs/diag interaces really show this one belonging to the correct
1425          * user.
1426          */
1427         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1428         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1429         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1430
1431         subflow = mptcp_subflow_ctx(sf->sk);
1432         pr_debug("subflow=%p", subflow);
1433
1434         *new_sock = sf;
1435         sock_hold(sk);
1436         subflow->conn = sk;
1437         mptcp_subflow_ops_override(sf->sk);
1438
1439         return 0;
1440 }
1441
1442 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1443                                                         gfp_t priority)
1444 {
1445         struct inet_connection_sock *icsk = inet_csk(sk);
1446         struct mptcp_subflow_context *ctx;
1447
1448         ctx = kzalloc(sizeof(*ctx), priority);
1449         if (!ctx)
1450                 return NULL;
1451
1452         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1453         INIT_LIST_HEAD(&ctx->node);
1454         INIT_LIST_HEAD(&ctx->delegated_node);
1455
1456         pr_debug("subflow=%p", ctx);
1457
1458         ctx->tcp_sock = sk;
1459
1460         return ctx;
1461 }
1462
1463 static void __subflow_state_change(struct sock *sk)
1464 {
1465         struct socket_wq *wq;
1466
1467         rcu_read_lock();
1468         wq = rcu_dereference(sk->sk_wq);
1469         if (skwq_has_sleeper(wq))
1470                 wake_up_interruptible_all(&wq->wait);
1471         rcu_read_unlock();
1472 }
1473
1474 static bool subflow_is_done(const struct sock *sk)
1475 {
1476         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1477 }
1478
1479 static void subflow_state_change(struct sock *sk)
1480 {
1481         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1482         struct sock *parent = subflow->conn;
1483
1484         __subflow_state_change(sk);
1485
1486         if (subflow_simultaneous_connect(sk)) {
1487                 mptcp_propagate_sndbuf(parent, sk);
1488                 mptcp_do_fallback(sk);
1489                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1490                 pr_fallback(mptcp_sk(parent));
1491                 subflow->conn_finished = 1;
1492                 if (inet_sk_state_load(parent) == TCP_SYN_SENT) {
1493                         inet_sk_state_store(parent, TCP_ESTABLISHED);
1494                         parent->sk_state_change(parent);
1495                 }
1496         }
1497
1498         /* as recvmsg() does not acquire the subflow socket for ssk selection
1499          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1500          * the data available machinery here.
1501          */
1502         if (mptcp_subflow_data_available(sk))
1503                 mptcp_data_ready(parent, sk);
1504         else if (unlikely(sk->sk_err))
1505                 subflow_error_report(sk);
1506
1507         subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1508
1509         if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1510             !subflow->rx_eof && subflow_is_done(sk)) {
1511                 subflow->rx_eof = 1;
1512                 mptcp_subflow_eof(parent);
1513         }
1514 }
1515
1516 static int subflow_ulp_init(struct sock *sk)
1517 {
1518         struct inet_connection_sock *icsk = inet_csk(sk);
1519         struct mptcp_subflow_context *ctx;
1520         struct tcp_sock *tp = tcp_sk(sk);
1521         int err = 0;
1522
1523         /* disallow attaching ULP to a socket unless it has been
1524          * created with sock_create_kern()
1525          */
1526         if (!sk->sk_kern_sock) {
1527                 err = -EOPNOTSUPP;
1528                 goto out;
1529         }
1530
1531         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1532         if (!ctx) {
1533                 err = -ENOMEM;
1534                 goto out;
1535         }
1536
1537         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1538
1539         tp->is_mptcp = 1;
1540         ctx->icsk_af_ops = icsk->icsk_af_ops;
1541         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1542         ctx->tcp_data_ready = sk->sk_data_ready;
1543         ctx->tcp_state_change = sk->sk_state_change;
1544         ctx->tcp_write_space = sk->sk_write_space;
1545         ctx->tcp_error_report = sk->sk_error_report;
1546         sk->sk_data_ready = subflow_data_ready;
1547         sk->sk_write_space = subflow_write_space;
1548         sk->sk_state_change = subflow_state_change;
1549         sk->sk_error_report = subflow_error_report;
1550 out:
1551         return err;
1552 }
1553
1554 static void subflow_ulp_release(struct sock *ssk)
1555 {
1556         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1557         bool release = true;
1558         struct sock *sk;
1559
1560         if (!ctx)
1561                 return;
1562
1563         sk = ctx->conn;
1564         if (sk) {
1565                 /* if the msk has been orphaned, keep the ctx
1566                  * alive, will be freed by __mptcp_close_ssk(),
1567                  * when the subflow is still unaccepted
1568                  */
1569                 release = ctx->disposable || list_empty(&ctx->node);
1570                 sock_put(sk);
1571         }
1572
1573         mptcp_subflow_ops_undo_override(ssk);
1574         if (release)
1575                 kfree_rcu(ctx, rcu);
1576 }
1577
1578 static void subflow_ulp_clone(const struct request_sock *req,
1579                               struct sock *newsk,
1580                               const gfp_t priority)
1581 {
1582         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1583         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1584         struct mptcp_subflow_context *new_ctx;
1585
1586         if (!tcp_rsk(req)->is_mptcp ||
1587             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1588                 subflow_ulp_fallback(newsk, old_ctx);
1589                 return;
1590         }
1591
1592         new_ctx = subflow_create_ctx(newsk, priority);
1593         if (!new_ctx) {
1594                 subflow_ulp_fallback(newsk, old_ctx);
1595                 return;
1596         }
1597
1598         new_ctx->conn_finished = 1;
1599         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1600         new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1601         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1602         new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1603         new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1604         new_ctx->rel_write_seq = 1;
1605         new_ctx->tcp_sock = newsk;
1606
1607         if (subflow_req->mp_capable) {
1608                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1609                  * is fully established only after we receive the remote key
1610                  */
1611                 new_ctx->mp_capable = 1;
1612                 new_ctx->local_key = subflow_req->local_key;
1613                 new_ctx->token = subflow_req->token;
1614                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1615                 new_ctx->idsn = subflow_req->idsn;
1616         } else if (subflow_req->mp_join) {
1617                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1618                 new_ctx->mp_join = 1;
1619                 new_ctx->fully_established = 1;
1620                 new_ctx->backup = subflow_req->backup;
1621                 new_ctx->local_id = subflow_req->local_id;
1622                 new_ctx->remote_id = subflow_req->remote_id;
1623                 new_ctx->token = subflow_req->token;
1624                 new_ctx->thmac = subflow_req->thmac;
1625         }
1626 }
1627
1628 static void tcp_release_cb_override(struct sock *ssk)
1629 {
1630         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1631
1632         if (mptcp_subflow_has_delegated_action(subflow))
1633                 mptcp_subflow_process_delegated(ssk);
1634
1635         tcp_release_cb(ssk);
1636 }
1637
1638 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1639         .name           = "mptcp",
1640         .owner          = THIS_MODULE,
1641         .init           = subflow_ulp_init,
1642         .release        = subflow_ulp_release,
1643         .clone          = subflow_ulp_clone,
1644 };
1645
1646 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1647 {
1648         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1649         subflow_ops->slab_name = "request_sock_subflow";
1650
1651         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1652                                               subflow_ops->obj_size, 0,
1653                                               SLAB_ACCOUNT |
1654                                               SLAB_TYPESAFE_BY_RCU,
1655                                               NULL);
1656         if (!subflow_ops->slab)
1657                 return -ENOMEM;
1658
1659         subflow_ops->destructor = subflow_req_destructor;
1660
1661         return 0;
1662 }
1663
1664 void __init mptcp_subflow_init(void)
1665 {
1666         mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1667         if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1668                 panic("MPTCP: failed to init subflow request sock ops\n");
1669
1670         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1671         subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
1672
1673         subflow_specific = ipv4_specific;
1674         subflow_specific.conn_request = subflow_v4_conn_request;
1675         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1676         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1677
1678         tcp_prot_override = tcp_prot;
1679         tcp_prot_override.release_cb = tcp_release_cb_override;
1680
1681 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1682         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1683         subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
1684
1685         subflow_v6_specific = ipv6_specific;
1686         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1687         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1688         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1689
1690         subflow_v6m_specific = subflow_v6_specific;
1691         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1692         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1693         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1694         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1695         subflow_v6m_specific.net_frag_header_len = 0;
1696
1697         tcpv6_prot_override = tcpv6_prot;
1698         tcpv6_prot_override.release_cb = tcp_release_cb_override;
1699 #endif
1700
1701         mptcp_diag_subflow_init(&subflow_ulp_ops);
1702
1703         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1704                 panic("MPTCP: failed to register subflows to ULP\n");
1705 }