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(sk_listener, 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                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX);
 216                                 return -EPERM;
 217                         }
 218                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX);
 219                 }
 220
 221                 subflow_req_create_thmac(subflow_req);
 222
 223                 if (unlikely(req->syncookie)) {
 224                         if (mptcp_can_accept_new_subflow(subflow_req->msk))
 225                                 subflow_init_req_cookie_join_save(subflow_req, skb);
 226                         else
 227                                 return -EPERM;
 228                 }
 229
 230                 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
 231                          subflow_req->remote_nonce, subflow_req->msk);
 232         }
 233
 234         return 0;
 235 }
 236
 237 int mptcp_subflow_init_cookie_req(struct request_sock *req,
 238                                   const struct sock *sk_listener,
 239                                   struct sk_buff *skb)
 240 {
 241         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 242         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 243         struct mptcp_options_received mp_opt;
 244         int err;
 245
 246         subflow_init_req(req, sk_listener);
 247         mptcp_get_options(sk_listener, skb, &mp_opt);
 248
 249         if (mp_opt.mp_capable && mp_opt.mp_join)
 250                 return -EINVAL;
 251
 252         if (mp_opt.mp_capable && listener->request_mptcp) {
 253                 if (mp_opt.sndr_key == 0)
 254                         return -EINVAL;
 255
 256                 subflow_req->local_key = mp_opt.rcvr_key;
 257                 err = mptcp_token_new_request(req);
 258                 if (err)
 259                         return err;
 260
 261                 subflow_req->mp_capable = 1;
 262                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 263         } else if (mp_opt.mp_join && listener->request_mptcp) {
 264                 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
 265                         return -EINVAL;
 266
 267                 subflow_req->mp_join = 1;
 268                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 269         }
 270
 271         return 0;
 272 }
 273 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
 274
 275 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
 276                                               struct sk_buff *skb,
 277                                               struct flowi *fl,
 278                                               struct request_sock *req)
 279 {
 280         struct dst_entry *dst;
 281         int err;
 282
 283         tcp_rsk(req)->is_mptcp = 1;
 284         subflow_init_req(req, sk);
 285
 286         dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req);
 287         if (!dst)
 288                 return NULL;
 289
 290         err = subflow_check_req(req, sk, skb);
 291         if (err == 0)
 292                 return dst;
 293
 294         dst_release(dst);
 295         if (!req->syncookie)
 296                 tcp_request_sock_ops.send_reset(sk, skb);
 297         return NULL;
 298 }
 299
 300 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 301 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
 302                                               struct sk_buff *skb,
 303                                               struct flowi *fl,
 304                                               struct request_sock *req)
 305 {
 306         struct dst_entry *dst;
 307         int err;
 308
 309         tcp_rsk(req)->is_mptcp = 1;
 310         subflow_init_req(req, sk);
 311
 312         dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req);
 313         if (!dst)
 314                 return NULL;
 315
 316         err = subflow_check_req(req, sk, skb);
 317         if (err == 0)
 318                 return dst;
 319
 320         dst_release(dst);
 321         if (!req->syncookie)
 322                 tcp6_request_sock_ops.send_reset(sk, skb);
 323         return NULL;
 324 }
 325 #endif
 326
 327 /* validate received truncated hmac and create hmac for third ACK */
 328 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
 329 {
 330         u8 hmac[SHA256_DIGEST_SIZE];
 331         u64 thmac;
 332
 333         subflow_generate_hmac(subflow->remote_key, subflow->local_key,
 334                               subflow->remote_nonce, subflow->local_nonce,
 335                               hmac);
 336
 337         thmac = get_unaligned_be64(hmac);
 338         pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
 339                  subflow, subflow->token,
 340                  (unsigned long long)thmac,
 341                  (unsigned long long)subflow->thmac);
 342
 343         return thmac == subflow->thmac;
 344 }
 345
 346 void mptcp_subflow_reset(struct sock *ssk)
 347 {
 348         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 349         struct sock *sk = subflow->conn;
 350
 351         /* must hold: tcp_done() could drop last reference on parent */
 352         sock_hold(sk);
 353
 354         tcp_set_state(ssk, TCP_CLOSE);
 355         tcp_send_active_reset(ssk, GFP_ATOMIC);
 356         tcp_done(ssk);
 357         if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags) &&
 358             schedule_work(&mptcp_sk(sk)->work))
 359                 return; /* worker will put sk for us */
 360
 361         sock_put(sk);
 362 }
 363
 364 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
 365 {
 366         return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
 367 }
 368
 369 void __mptcp_set_connected(struct sock *sk)
 370 {
 371         if (sk->sk_state == TCP_SYN_SENT) {
 372                 inet_sk_state_store(sk, TCP_ESTABLISHED);
 373                 sk->sk_state_change(sk);
 374         }
 375 }
 376
 377 static void mptcp_set_connected(struct sock *sk)
 378 {
 379         mptcp_data_lock(sk);
 380         if (!sock_owned_by_user(sk))
 381                 __mptcp_set_connected(sk);
 382         else
 383                 set_bit(MPTCP_CONNECTED, &mptcp_sk(sk)->flags);
 384         mptcp_data_unlock(sk);
 385 }
 386
 387 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 388 {
 389         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 390         struct mptcp_options_received mp_opt;
 391         struct sock *parent = subflow->conn;
 392
 393         subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 394
 395
 396         /* be sure no special action on any packet other than syn-ack */
 397         if (subflow->conn_finished)
 398                 return;
 399
 400         mptcp_propagate_sndbuf(parent, sk);
 401         subflow->rel_write_seq = 1;
 402         subflow->conn_finished = 1;
 403         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 404         pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
 405
 406         mptcp_get_options(sk, skb, &mp_opt);
 407         if (subflow->request_mptcp) {
 408                 if (!mp_opt.mp_capable) {
 409                         MPTCP_INC_STATS(sock_net(sk),
 410                                         MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
 411                         mptcp_do_fallback(sk);
 412                         pr_fallback(mptcp_sk(subflow->conn));
 413                         goto fallback;
 414                 }
 415
 416                 subflow->mp_capable = 1;
 417                 subflow->can_ack = 1;
 418                 subflow->remote_key = mp_opt.sndr_key;
 419                 pr_debug("subflow=%p, remote_key=%llu", subflow,
 420                          subflow->remote_key);
 421                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
 422                 mptcp_finish_connect(sk);
 423                 mptcp_set_connected(parent);
 424         } else if (subflow->request_join) {
 425                 u8 hmac[SHA256_DIGEST_SIZE];
 426
 427                 if (!mp_opt.mp_join) {
 428                         subflow->reset_reason = MPTCP_RST_EMPTCP;
 429                         goto do_reset;
 430                 }
 431
 432                 subflow->thmac = mp_opt.thmac;
 433                 subflow->remote_nonce = mp_opt.nonce;
 434                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u", subflow,
 435                          subflow->thmac, subflow->remote_nonce);
 436
 437                 if (!subflow_thmac_valid(subflow)) {
 438                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 439                         subflow->reset_reason = MPTCP_RST_EMPTCP;
 440                         goto do_reset;
 441                 }
 442
 443                 if (!mptcp_finish_join(sk))
 444                         goto do_reset;
 445
 446                 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 447                                       subflow->local_nonce,
 448                                       subflow->remote_nonce,
 449                                       hmac);
 450                 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
 451
 452                 subflow->mp_join = 1;
 453                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 454
 455                 if (subflow_use_different_dport(mptcp_sk(parent), sk)) {
 456                         pr_debug("synack inet_dport=%d %d",
 457                                  ntohs(inet_sk(sk)->inet_dport),
 458                                  ntohs(inet_sk(parent)->inet_dport));
 459                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
 460                 }
 461         } else if (mptcp_check_fallback(sk)) {
 462 fallback:
 463                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
 464                 mptcp_set_connected(parent);
 465         }
 466         return;
 467
 468 do_reset:
 469         subflow->reset_transient = 0;
 470         mptcp_subflow_reset(sk);
 471 }
 472
 473 struct request_sock_ops mptcp_subflow_request_sock_ops;
 474 EXPORT_SYMBOL_GPL(mptcp_subflow_request_sock_ops);
 475 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
 476
 477 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 478 {
 479         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 480
 481         pr_debug("subflow=%p", subflow);
 482
 483         /* Never answer to SYNs sent to broadcast or multicast */
 484         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 485                 goto drop;
 486
 487         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 488                                 &subflow_request_sock_ipv4_ops,
 489                                 sk, skb);
 490 drop:
 491         tcp_listendrop(sk);
 492         return 0;
 493 }
 494
 495 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 496 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
 497 static struct inet_connection_sock_af_ops subflow_v6_specific;
 498 static struct inet_connection_sock_af_ops subflow_v6m_specific;
 499 static struct proto tcpv6_prot_override;
 500
 501 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 502 {
 503         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 504
 505         pr_debug("subflow=%p", subflow);
 506
 507         if (skb->protocol == htons(ETH_P_IP))
 508                 return subflow_v4_conn_request(sk, skb);
 509
 510         if (!ipv6_unicast_destination(skb))
 511                 goto drop;
 512
 513         if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
 514                 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
 515                 return 0;
 516         }
 517
 518         return tcp_conn_request(&mptcp_subflow_request_sock_ops,
 519                                 &subflow_request_sock_ipv6_ops, sk, skb);
 520
 521 drop:
 522         tcp_listendrop(sk);
 523         return 0; /* don't send reset */
 524 }
 525 #endif
 526
 527 /* validate hmac received in third ACK */
 528 static bool subflow_hmac_valid(const struct request_sock *req,
 529                                const struct mptcp_options_received *mp_opt)
 530 {
 531         const struct mptcp_subflow_request_sock *subflow_req;
 532         u8 hmac[SHA256_DIGEST_SIZE];
 533         struct mptcp_sock *msk;
 534
 535         subflow_req = mptcp_subflow_rsk(req);
 536         msk = subflow_req->msk;
 537         if (!msk)
 538                 return false;
 539
 540         subflow_generate_hmac(msk->remote_key, msk->local_key,
 541                               subflow_req->remote_nonce,
 542                               subflow_req->local_nonce, hmac);
 543
 544         return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
 545 }
 546
 547 static void mptcp_sock_destruct(struct sock *sk)
 548 {
 549         /* if new mptcp socket isn't accepted, it is free'd
 550          * from the tcp listener sockets request queue, linked
 551          * from req->sk.  The tcp socket is released.
 552          * This calls the ULP release function which will
 553          * also remove the mptcp socket, via
 554          * sock_put(ctx->conn).
 555          *
 556          * Problem is that the mptcp socket will be in
 557          * ESTABLISHED state and will not have the SOCK_DEAD flag.
 558          * Both result in warnings from inet_sock_destruct.
 559          */
 560         if ((1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
 561                 sk->sk_state = TCP_CLOSE;
 562                 WARN_ON_ONCE(sk->sk_socket);
 563                 sock_orphan(sk);
 564         }
 565
 566         mptcp_destroy_common(mptcp_sk(sk));
 567         inet_sock_destruct(sk);
 568 }
 569
 570 static void mptcp_force_close(struct sock *sk)
 571 {
 572         /* the msk is not yet exposed to user-space */
 573         inet_sk_state_store(sk, TCP_CLOSE);
 574         sk_common_release(sk);
 575 }
 576
 577 static void subflow_ulp_fallback(struct sock *sk,
 578                                  struct mptcp_subflow_context *old_ctx)
 579 {
 580         struct inet_connection_sock *icsk = inet_csk(sk);
 581
 582         mptcp_subflow_tcp_fallback(sk, old_ctx);
 583         icsk->icsk_ulp_ops = NULL;
 584         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 585         tcp_sk(sk)->is_mptcp = 0;
 586
 587         mptcp_subflow_ops_undo_override(sk);
 588 }
 589
 590 static void subflow_drop_ctx(struct sock *ssk)
 591 {
 592         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 593
 594         if (!ctx)
 595                 return;
 596
 597         subflow_ulp_fallback(ssk, ctx);
 598         if (ctx->conn)
 599                 sock_put(ctx->conn);
 600
 601         kfree_rcu(ctx, rcu);
 602 }
 603
 604 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 605                                      struct mptcp_options_received *mp_opt)
 606 {
 607         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 608
 609         subflow->remote_key = mp_opt->sndr_key;
 610         subflow->fully_established = 1;
 611         subflow->can_ack = 1;
 612         WRITE_ONCE(msk->fully_established, true);
 613 }
 614
 615 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 616                                           struct sk_buff *skb,
 617                                           struct request_sock *req,
 618                                           struct dst_entry *dst,
 619                                           struct request_sock *req_unhash,
 620                                           bool *own_req)
 621 {
 622         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 623         struct mptcp_subflow_request_sock *subflow_req;
 624         struct mptcp_options_received mp_opt;
 625         bool fallback, fallback_is_fatal;
 626         struct sock *new_msk = NULL;
 627         struct sock *child;
 628
 629         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 630
 631         /* After child creation we must look for 'mp_capable' even when options
 632          * are not parsed
 633          */
 634         mp_opt.mp_capable = 0;
 635
 636         /* hopefully temporary handling for MP_JOIN+syncookie */
 637         subflow_req = mptcp_subflow_rsk(req);
 638         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 639         fallback = !tcp_rsk(req)->is_mptcp;
 640         if (fallback)
 641                 goto create_child;
 642
 643         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 644         if (subflow_req->mp_capable) {
 645                 /* we can receive and accept an in-window, out-of-order pkt,
 646                  * which may not carry the MP_CAPABLE opt even on mptcp enabled
 647                  * paths: always try to extract the peer key, and fallback
 648                  * for packets missing it.
 649                  * Even OoO DSS packets coming legitly after dropped or
 650                  * reordered MPC will cause fallback, but we don't have other
 651                  * options.
 652                  */
 653                 mptcp_get_options(sk, skb, &mp_opt);
 654                 if (!mp_opt.mp_capable) {
 655                         fallback = true;
 656                         goto create_child;
 657                 }
 658
 659                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 660                 if (!new_msk)
 661                         fallback = true;
 662         } else if (subflow_req->mp_join) {
 663                 mptcp_get_options(sk, skb, &mp_opt);
 664                 if (!mp_opt.mp_join || !subflow_hmac_valid(req, &mp_opt) ||
 665                     !mptcp_can_accept_new_subflow(subflow_req->msk)) {
 666                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 667                         fallback = true;
 668                 }
 669         }
 670
 671 create_child:
 672         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 673                                                      req_unhash, own_req);
 674
 675         if (child && *own_req) {
 676                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 677
 678                 tcp_rsk(req)->drop_req = false;
 679
 680                 /* we need to fallback on ctx allocation failure and on pre-reqs
 681                  * checking above. In the latter scenario we additionally need
 682                  * to reset the context to non MPTCP status.
 683                  */
 684                 if (!ctx || fallback) {
 685                         if (fallback_is_fatal) {
 686                                 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 687                                 goto dispose_child;
 688                         }
 689
 690                         subflow_drop_ctx(child);
 691                         goto out;
 692                 }
 693
 694                 /* ssk inherits options of listener sk */
 695                 ctx->setsockopt_seq = listener->setsockopt_seq;
 696
 697                 if (ctx->mp_capable) {
 698                         /* this can't race with mptcp_close(), as the msk is
 699                          * not yet exposted to user-space
 700                          */
 701                         inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
 702
 703                         /* record the newly created socket as the first msk
 704                          * subflow, but don't link it yet into conn_list
 705                          */
 706                         WRITE_ONCE(mptcp_sk(new_msk)->first, child);
 707
 708                         /* new mpc subflow takes ownership of the newly
 709                          * created mptcp socket
 710                          */
 711                         new_msk->sk_destruct = mptcp_sock_destruct;
 712                         mptcp_sk(new_msk)->setsockopt_seq = ctx->setsockopt_seq;
 713                         mptcp_pm_new_connection(mptcp_sk(new_msk), child, 1);
 714                         mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
 715                         ctx->conn = new_msk;
 716                         new_msk = NULL;
 717
 718                         /* with OoO packets we can reach here without ingress
 719                          * mpc option
 720                          */
 721                         if (mp_opt.mp_capable)
 722                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 723                 } else if (ctx->mp_join) {
 724                         struct mptcp_sock *owner;
 725
 726                         owner = subflow_req->msk;
 727                         if (!owner) {
 728                                 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 729                                 goto dispose_child;
 730                         }
 731
 732                         /* move the msk reference ownership to the subflow */
 733                         subflow_req->msk = NULL;
 734                         ctx->conn = (struct sock *)owner;
 735
 736                         if (subflow_use_different_sport(owner, sk)) {
 737                                 pr_debug("ack inet_sport=%d %d",
 738                                          ntohs(inet_sk(sk)->inet_sport),
 739                                          ntohs(inet_sk((struct sock *)owner)->inet_sport));
 740                                 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
 741                                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
 742                                         goto dispose_child;
 743                                 }
 744                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
 745                         }
 746
 747                         if (!mptcp_finish_join(child))
 748                                 goto dispose_child;
 749
 750                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 751                         tcp_rsk(req)->drop_req = true;
 752                 }
 753         }
 754
 755 out:
 756         /* dispose of the left over mptcp master, if any */
 757         if (unlikely(new_msk))
 758                 mptcp_force_close(new_msk);
 759
 760         /* check for expected invariant - should never trigger, just help
 761          * catching eariler subtle bugs
 762          */
 763         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 764                      (!mptcp_subflow_ctx(child) ||
 765                       !mptcp_subflow_ctx(child)->conn));
 766         return child;
 767
 768 dispose_child:
 769         subflow_drop_ctx(child);
 770         tcp_rsk(req)->drop_req = true;
 771         inet_csk_prepare_for_destroy_sock(child);
 772         tcp_done(child);
 773         req->rsk_ops->send_reset(sk, skb);
 774
 775         /* The last child reference will be released by the caller */
 776         return child;
 777 }
 778
 779 static struct inet_connection_sock_af_ops subflow_specific;
 780 static struct proto tcp_prot_override;
 781
 782 enum mapping_status {
 783         MAPPING_OK,
 784         MAPPING_INVALID,
 785         MAPPING_EMPTY,
 786         MAPPING_DATA_FIN,
 787         MAPPING_DUMMY
 788 };
 789
 790 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 791 {
 792         pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 793                  ssn, subflow->map_subflow_seq, subflow->map_data_len);
 794 }
 795
 796 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 797 {
 798         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 799         unsigned int skb_consumed;
 800
 801         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 802         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 803                 return true;
 804
 805         return skb->len - skb_consumed <= subflow->map_data_len -
 806                                           mptcp_subflow_get_map_offset(subflow);
 807 }
 808
 809 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 810 {
 811         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 812         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 813
 814         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 815                 /* Mapping covers data later in the subflow stream,
 816                  * currently unsupported.
 817                  */
 818                 dbg_bad_map(subflow, ssn);
 819                 return false;
 820         }
 821         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 822                                   subflow->map_data_len))) {
 823                 /* Mapping does covers past subflow data, invalid */
 824                 dbg_bad_map(subflow, ssn);
 825                 return false;
 826         }
 827         return true;
 828 }
 829
 830 static enum mapping_status get_mapping_status(struct sock *ssk,
 831                                               struct mptcp_sock *msk)
 832 {
 833         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 834         struct mptcp_ext *mpext;
 835         struct sk_buff *skb;
 836         u16 data_len;
 837         u64 map_seq;
 838
 839         skb = skb_peek(&ssk->sk_receive_queue);
 840         if (!skb)
 841                 return MAPPING_EMPTY;
 842
 843         if (mptcp_check_fallback(ssk))
 844                 return MAPPING_DUMMY;
 845
 846         mpext = mptcp_get_ext(skb);
 847         if (!mpext || !mpext->use_map) {
 848                 if (!subflow->map_valid && !skb->len) {
 849                         /* the TCP stack deliver 0 len FIN pkt to the receive
 850                          * queue, that is the only 0len pkts ever expected here,
 851                          * and we can admit no mapping only for 0 len pkts
 852                          */
 853                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 854                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 855                                           TCP_SKB_CB(skb)->seq,
 856                                           TCP_SKB_CB(skb)->end_seq,
 857                                           TCP_SKB_CB(skb)->tcp_flags);
 858                         sk_eat_skb(ssk, skb);
 859                         return MAPPING_EMPTY;
 860                 }
 861
 862                 if (!subflow->map_valid)
 863                         return MAPPING_INVALID;
 864
 865                 goto validate_seq;
 866         }
 867
 868         trace_get_mapping_status(mpext);
 869
 870         data_len = mpext->data_len;
 871         if (data_len == 0) {
 872                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 873                 return MAPPING_INVALID;
 874         }
 875
 876         if (mpext->data_fin == 1) {
 877                 if (data_len == 1) {
 878                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
 879                                                                  mpext->dsn64);
 880                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
 881                         if (subflow->map_valid) {
 882                                 /* A DATA_FIN might arrive in a DSS
 883                                  * option before the previous mapping
 884                                  * has been fully consumed. Continue
 885                                  * handling the existing mapping.
 886                                  */
 887                                 skb_ext_del(skb, SKB_EXT_MPTCP);
 888                                 return MAPPING_OK;
 889                         } else {
 890                                 if (updated && schedule_work(&msk->work))
 891                                         sock_hold((struct sock *)msk);
 892
 893                                 return MAPPING_DATA_FIN;
 894                         }
 895                 } else {
 896                         u64 data_fin_seq = mpext->data_seq + data_len - 1;
 897
 898                         /* If mpext->data_seq is a 32-bit value, data_fin_seq
 899                          * must also be limited to 32 bits.
 900                          */
 901                         if (!mpext->dsn64)
 902                                 data_fin_seq &= GENMASK_ULL(31, 0);
 903
 904                         mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
 905                         pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
 906                                  data_fin_seq, mpext->dsn64);
 907                 }
 908
 909                 /* Adjust for DATA_FIN using 1 byte of sequence space */
 910                 data_len--;
 911         }
 912
 913         map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
 914         WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
 915
 916         if (subflow->map_valid) {
 917                 /* Allow replacing only with an identical map */
 918                 if (subflow->map_seq == map_seq &&
 919                     subflow->map_subflow_seq == mpext->subflow_seq &&
 920                     subflow->map_data_len == data_len) {
 921                         skb_ext_del(skb, SKB_EXT_MPTCP);
 922                         return MAPPING_OK;
 923                 }
 924
 925                 /* If this skb data are fully covered by the current mapping,
 926                  * the new map would need caching, which is not supported
 927                  */
 928                 if (skb_is_fully_mapped(ssk, skb)) {
 929                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
 930                         return MAPPING_INVALID;
 931                 }
 932
 933                 /* will validate the next map after consuming the current one */
 934                 return MAPPING_OK;
 935         }
 936
 937         subflow->map_seq = map_seq;
 938         subflow->map_subflow_seq = mpext->subflow_seq;
 939         subflow->map_data_len = data_len;
 940         subflow->map_valid = 1;
 941         subflow->mpc_map = mpext->mpc_map;
 942         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
 943                  subflow->map_seq, subflow->map_subflow_seq,
 944                  subflow->map_data_len);
 945
 946 validate_seq:
 947         /* we revalidate valid mapping on new skb, because we must ensure
 948          * the current skb is completely covered by the available mapping
 949          */
 950         if (!validate_mapping(ssk, skb))
 951                 return MAPPING_INVALID;
 952
 953         skb_ext_del(skb, SKB_EXT_MPTCP);
 954         return MAPPING_OK;
 955 }
 956
 957 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
 958                                        u64 limit)
 959 {
 960         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 961         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
 962         u32 incr;
 963
 964         incr = limit >= skb->len ? skb->len + fin : limit;
 965
 966         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
 967                  subflow->map_subflow_seq);
 968         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
 969         tcp_sk(ssk)->copied_seq += incr;
 970         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
 971                 sk_eat_skb(ssk, skb);
 972         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
 973                 subflow->map_valid = 0;
 974 }
 975
 976 /* sched mptcp worker to remove the subflow if no more data is pending */
 977 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
 978 {
 979         struct sock *sk = (struct sock *)msk;
 980
 981         if (likely(ssk->sk_state != TCP_CLOSE))
 982                 return;
 983
 984         if (skb_queue_empty(&ssk->sk_receive_queue) &&
 985             !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags)) {
 986                 sock_hold(sk);
 987                 if (!schedule_work(&msk->work))
 988                         sock_put(sk);
 989         }
 990 }
 991
 992 static bool subflow_check_data_avail(struct sock *ssk)
 993 {
 994         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 995         enum mapping_status status;
 996         struct mptcp_sock *msk;
 997         struct sk_buff *skb;
 998
 999         if (!skb_peek(&ssk->sk_receive_queue))
1000                 WRITE_ONCE(subflow->data_avail, 0);
1001         if (subflow->data_avail)
1002                 return true;
1003
1004         msk = mptcp_sk(subflow->conn);
1005         for (;;) {
1006                 u64 ack_seq;
1007                 u64 old_ack;
1008
1009                 status = get_mapping_status(ssk, msk);
1010                 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1011                 if (unlikely(status == MAPPING_INVALID))
1012                         goto fallback;
1013
1014                 if (unlikely(status == MAPPING_DUMMY))
1015                         goto fallback;
1016
1017                 if (status != MAPPING_OK)
1018                         goto no_data;
1019
1020                 skb = skb_peek(&ssk->sk_receive_queue);
1021                 if (WARN_ON_ONCE(!skb))
1022                         goto no_data;
1023
1024                 /* if msk lacks the remote key, this subflow must provide an
1025                  * MP_CAPABLE-based mapping
1026                  */
1027                 if (unlikely(!READ_ONCE(msk->can_ack))) {
1028                         if (!subflow->mpc_map)
1029                                 goto fallback;
1030                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
1031                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
1032                         WRITE_ONCE(msk->can_ack, true);
1033                 }
1034
1035                 old_ack = READ_ONCE(msk->ack_seq);
1036                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1037                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1038                          ack_seq);
1039                 if (unlikely(before64(ack_seq, old_ack))) {
1040                         mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1041                         continue;
1042                 }
1043
1044                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1045                 break;
1046         }
1047         return true;
1048
1049 no_data:
1050         subflow_sched_work_if_closed(msk, ssk);
1051         return false;
1052
1053 fallback:
1054         /* RFC 8684 section 3.7. */
1055         if (subflow->mp_join || subflow->fully_established) {
1056                 /* fatal protocol error, close the socket.
1057                  * subflow_error_report() will introduce the appropriate barriers
1058                  */
1059                 ssk->sk_err = EBADMSG;
1060                 tcp_set_state(ssk, TCP_CLOSE);
1061                 subflow->reset_transient = 0;
1062                 subflow->reset_reason = MPTCP_RST_EMPTCP;
1063                 tcp_send_active_reset(ssk, GFP_ATOMIC);
1064                 WRITE_ONCE(subflow->data_avail, 0);
1065                 return false;
1066         }
1067
1068         __mptcp_do_fallback(msk);
1069         skb = skb_peek(&ssk->sk_receive_queue);
1070         subflow->map_valid = 1;
1071         subflow->map_seq = READ_ONCE(msk->ack_seq);
1072         subflow->map_data_len = skb->len;
1073         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1074         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1075         return true;
1076 }
1077
1078 bool mptcp_subflow_data_available(struct sock *sk)
1079 {
1080         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1081
1082         /* check if current mapping is still valid */
1083         if (subflow->map_valid &&
1084             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1085                 subflow->map_valid = 0;
1086                 WRITE_ONCE(subflow->data_avail, 0);
1087
1088                 pr_debug("Done with mapping: seq=%u data_len=%u",
1089                          subflow->map_subflow_seq,
1090                          subflow->map_data_len);
1091         }
1092
1093         return subflow_check_data_avail(sk);
1094 }
1095
1096 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1097  * not the ssk one.
1098  *
1099  * In mptcp, rwin is about the mptcp-level connection data.
1100  *
1101  * Data that is still on the ssk rx queue can thus be ignored,
1102  * as far as mptcp peer is concerned that data is still inflight.
1103  * DSS ACK is updated when skb is moved to the mptcp rx queue.
1104  */
1105 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1106 {
1107         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1108         const struct sock *sk = subflow->conn;
1109
1110         *space = __mptcp_space(sk);
1111         *full_space = tcp_full_space(sk);
1112 }
1113
1114 void __mptcp_error_report(struct sock *sk)
1115 {
1116         struct mptcp_subflow_context *subflow;
1117         struct mptcp_sock *msk = mptcp_sk(sk);
1118
1119         mptcp_for_each_subflow(msk, subflow) {
1120                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1121                 int err = sock_error(ssk);
1122
1123                 if (!err)
1124                         continue;
1125
1126                 /* only propagate errors on fallen-back sockets or
1127                  * on MPC connect
1128                  */
1129                 if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(msk))
1130                         continue;
1131
1132                 inet_sk_state_store(sk, inet_sk_state_load(ssk));
1133                 sk->sk_err = -err;
1134
1135                 /* This barrier is coupled with smp_rmb() in mptcp_poll() */
1136                 smp_wmb();
1137                 sk->sk_error_report(sk);
1138                 break;
1139         }
1140 }
1141
1142 static void subflow_error_report(struct sock *ssk)
1143 {
1144         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1145
1146         mptcp_data_lock(sk);
1147         if (!sock_owned_by_user(sk))
1148                 __mptcp_error_report(sk);
1149         else
1150                 set_bit(MPTCP_ERROR_REPORT,  &mptcp_sk(sk)->flags);
1151         mptcp_data_unlock(sk);
1152 }
1153
1154 static void subflow_data_ready(struct sock *sk)
1155 {
1156         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1157         u16 state = 1 << inet_sk_state_load(sk);
1158         struct sock *parent = subflow->conn;
1159         struct mptcp_sock *msk;
1160
1161         msk = mptcp_sk(parent);
1162         if (state & TCPF_LISTEN) {
1163                 /* MPJ subflow are removed from accept queue before reaching here,
1164                  * avoid stray wakeups
1165                  */
1166                 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1167                         return;
1168
1169                 set_bit(MPTCP_DATA_READY, &msk->flags);
1170                 parent->sk_data_ready(parent);
1171                 return;
1172         }
1173
1174         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1175                      !subflow->mp_join && !(state & TCPF_CLOSE));
1176
1177         if (mptcp_subflow_data_available(sk))
1178                 mptcp_data_ready(parent, sk);
1179         else if (unlikely(sk->sk_err))
1180                 subflow_error_report(sk);
1181 }
1182
1183 static void subflow_write_space(struct sock *ssk)
1184 {
1185         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1186
1187         mptcp_propagate_sndbuf(sk, ssk);
1188         mptcp_write_space(sk);
1189 }
1190
1191 static struct inet_connection_sock_af_ops *
1192 subflow_default_af_ops(struct sock *sk)
1193 {
1194 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1195         if (sk->sk_family == AF_INET6)
1196                 return &subflow_v6_specific;
1197 #endif
1198         return &subflow_specific;
1199 }
1200
1201 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1202 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1203 {
1204         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1205         struct inet_connection_sock *icsk = inet_csk(sk);
1206         struct inet_connection_sock_af_ops *target;
1207
1208         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1209
1210         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1211                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1212
1213         if (likely(icsk->icsk_af_ops == target))
1214                 return;
1215
1216         subflow->icsk_af_ops = icsk->icsk_af_ops;
1217         icsk->icsk_af_ops = target;
1218 }
1219 #endif
1220
1221 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1222                          struct sockaddr_storage *addr,
1223                          unsigned short family)
1224 {
1225         memset(addr, 0, sizeof(*addr));
1226         addr->ss_family = family;
1227         if (addr->ss_family == AF_INET) {
1228                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1229
1230                 if (info->family == AF_INET)
1231                         in_addr->sin_addr = info->addr;
1232 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1233                 else if (ipv6_addr_v4mapped(&info->addr6))
1234                         in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1235 #endif
1236                 in_addr->sin_port = info->port;
1237         }
1238 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1239         else if (addr->ss_family == AF_INET6) {
1240                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1241
1242                 if (info->family == AF_INET)
1243                         ipv6_addr_set_v4mapped(info->addr.s_addr,
1244                                                &in6_addr->sin6_addr);
1245                 else
1246                         in6_addr->sin6_addr = info->addr6;
1247                 in6_addr->sin6_port = info->port;
1248         }
1249 #endif
1250 }
1251
1252 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1253                             const struct mptcp_addr_info *remote,
1254                             u8 flags, int ifindex)
1255 {
1256         struct mptcp_sock *msk = mptcp_sk(sk);
1257         struct mptcp_subflow_context *subflow;
1258         struct sockaddr_storage addr;
1259         int remote_id = remote->id;
1260         int local_id = loc->id;
1261         struct socket *sf;
1262         struct sock *ssk;
1263         u32 remote_token;
1264         int addrlen;
1265         int err;
1266
1267         if (!mptcp_is_fully_established(sk))
1268                 return -ENOTCONN;
1269
1270         err = mptcp_subflow_create_socket(sk, &sf);
1271         if (err)
1272                 return err;
1273
1274         ssk = sf->sk;
1275         subflow = mptcp_subflow_ctx(ssk);
1276         do {
1277                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1278         } while (!subflow->local_nonce);
1279
1280         if (!local_id) {
1281                 err = mptcp_pm_get_local_id(msk, (struct sock_common *)ssk);
1282                 if (err < 0)
1283                         goto failed;
1284
1285                 local_id = err;
1286         }
1287
1288         subflow->remote_key = msk->remote_key;
1289         subflow->local_key = msk->local_key;
1290         subflow->token = msk->token;
1291         mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1292
1293         addrlen = sizeof(struct sockaddr_in);
1294 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1295         if (addr.ss_family == AF_INET6)
1296                 addrlen = sizeof(struct sockaddr_in6);
1297 #endif
1298         ssk->sk_bound_dev_if = ifindex;
1299         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1300         if (err)
1301                 goto failed;
1302
1303         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1304         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1305                  remote_token, local_id, remote_id);
1306         subflow->remote_token = remote_token;
1307         subflow->local_id = local_id;
1308         subflow->remote_id = remote_id;
1309         subflow->request_join = 1;
1310         subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1311         mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1312
1313         mptcp_add_pending_subflow(msk, subflow);
1314         mptcp_sockopt_sync(msk, ssk);
1315         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1316         if (err && err != -EINPROGRESS)
1317                 goto failed_unlink;
1318
1319         /* discard the subflow socket */
1320         mptcp_sock_graft(ssk, sk->sk_socket);
1321         iput(SOCK_INODE(sf));
1322         return err;
1323
1324 failed_unlink:
1325         spin_lock_bh(&msk->join_list_lock);
1326         list_del(&subflow->node);
1327         spin_unlock_bh(&msk->join_list_lock);
1328         sock_put(mptcp_subflow_tcp_sock(subflow));
1329
1330 failed:
1331         subflow->disposable = 1;
1332         sock_release(sf);
1333         return err;
1334 }
1335
1336 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1337 {
1338 #ifdef CONFIG_SOCK_CGROUP_DATA
1339         struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1340                                 *child_skcd = &child->sk_cgrp_data;
1341
1342         /* only the additional subflows created by kworkers have to be modified */
1343         if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1344             cgroup_id(sock_cgroup_ptr(child_skcd))) {
1345 #ifdef CONFIG_MEMCG
1346                 struct mem_cgroup *memcg = parent->sk_memcg;
1347
1348                 mem_cgroup_sk_free(child);
1349                 if (memcg && css_tryget(&memcg->css))
1350                         child->sk_memcg = memcg;
1351 #endif /* CONFIG_MEMCG */
1352
1353                 cgroup_sk_free(child_skcd);
1354                 *child_skcd = *parent_skcd;
1355                 cgroup_sk_clone(child_skcd);
1356         }
1357 #endif /* CONFIG_SOCK_CGROUP_DATA */
1358 }
1359
1360 static void mptcp_subflow_ops_override(struct sock *ssk)
1361 {
1362 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1363         if (ssk->sk_prot == &tcpv6_prot)
1364                 ssk->sk_prot = &tcpv6_prot_override;
1365         else
1366 #endif
1367                 ssk->sk_prot = &tcp_prot_override;
1368 }
1369
1370 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1371 {
1372 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1373         if (ssk->sk_prot == &tcpv6_prot_override)
1374                 ssk->sk_prot = &tcpv6_prot;
1375         else
1376 #endif
1377                 ssk->sk_prot = &tcp_prot;
1378 }
1379 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1380 {
1381         struct mptcp_subflow_context *subflow;
1382         struct net *net = sock_net(sk);
1383         struct socket *sf;
1384         int err;
1385
1386         /* un-accepted server sockets can reach here - on bad configuration
1387          * bail early to avoid greater trouble later
1388          */
1389         if (unlikely(!sk->sk_socket))
1390                 return -EINVAL;
1391
1392         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1393                                &sf);
1394         if (err)
1395                 return err;
1396
1397         lock_sock(sf->sk);
1398
1399         /* the newly created socket has to be in the same cgroup as its parent */
1400         mptcp_attach_cgroup(sk, sf->sk);
1401
1402         /* kernel sockets do not by default acquire net ref, but TCP timer
1403          * needs it.
1404          */
1405         sf->sk->sk_net_refcnt = 1;
1406         get_net(net);
1407 #ifdef CONFIG_PROC_FS
1408         this_cpu_add(*net->core.sock_inuse, 1);
1409 #endif
1410         err = tcp_set_ulp(sf->sk, "mptcp");
1411         release_sock(sf->sk);
1412
1413         if (err) {
1414                 sock_release(sf);
1415                 return err;
1416         }
1417
1418         /* the newly created socket really belongs to the owning MPTCP master
1419          * socket, even if for additional subflows the allocation is performed
1420          * by a kernel workqueue. Adjust inode references, so that the
1421          * procfs/diag interaces really show this one belonging to the correct
1422          * user.
1423          */
1424         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1425         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1426         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1427
1428         subflow = mptcp_subflow_ctx(sf->sk);
1429         pr_debug("subflow=%p", subflow);
1430
1431         *new_sock = sf;
1432         sock_hold(sk);
1433         subflow->conn = sk;
1434         mptcp_subflow_ops_override(sf->sk);
1435
1436         return 0;
1437 }
1438
1439 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1440                                                         gfp_t priority)
1441 {
1442         struct inet_connection_sock *icsk = inet_csk(sk);
1443         struct mptcp_subflow_context *ctx;
1444
1445         ctx = kzalloc(sizeof(*ctx), priority);
1446         if (!ctx)
1447                 return NULL;
1448
1449         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1450         INIT_LIST_HEAD(&ctx->node);
1451         INIT_LIST_HEAD(&ctx->delegated_node);
1452
1453         pr_debug("subflow=%p", ctx);
1454
1455         ctx->tcp_sock = sk;
1456
1457         return ctx;
1458 }
1459
1460 static void __subflow_state_change(struct sock *sk)
1461 {
1462         struct socket_wq *wq;
1463
1464         rcu_read_lock();
1465         wq = rcu_dereference(sk->sk_wq);
1466         if (skwq_has_sleeper(wq))
1467                 wake_up_interruptible_all(&wq->wait);
1468         rcu_read_unlock();
1469 }
1470
1471 static bool subflow_is_done(const struct sock *sk)
1472 {
1473         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1474 }
1475
1476 static void subflow_state_change(struct sock *sk)
1477 {
1478         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1479         struct sock *parent = subflow->conn;
1480
1481         __subflow_state_change(sk);
1482
1483         if (subflow_simultaneous_connect(sk)) {
1484                 mptcp_propagate_sndbuf(parent, sk);
1485                 mptcp_do_fallback(sk);
1486                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1487                 pr_fallback(mptcp_sk(parent));
1488                 subflow->conn_finished = 1;
1489                 mptcp_set_connected(parent);
1490         }
1491
1492         /* as recvmsg() does not acquire the subflow socket for ssk selection
1493          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1494          * the data available machinery here.
1495          */
1496         if (mptcp_subflow_data_available(sk))
1497                 mptcp_data_ready(parent, sk);
1498         else if (unlikely(sk->sk_err))
1499                 subflow_error_report(sk);
1500
1501         subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1502
1503         if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1504             !subflow->rx_eof && subflow_is_done(sk)) {
1505                 subflow->rx_eof = 1;
1506                 mptcp_subflow_eof(parent);
1507         }
1508 }
1509
1510 static int subflow_ulp_init(struct sock *sk)
1511 {
1512         struct inet_connection_sock *icsk = inet_csk(sk);
1513         struct mptcp_subflow_context *ctx;
1514         struct tcp_sock *tp = tcp_sk(sk);
1515         int err = 0;
1516
1517         /* disallow attaching ULP to a socket unless it has been
1518          * created with sock_create_kern()
1519          */
1520         if (!sk->sk_kern_sock) {
1521                 err = -EOPNOTSUPP;
1522                 goto out;
1523         }
1524
1525         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1526         if (!ctx) {
1527                 err = -ENOMEM;
1528                 goto out;
1529         }
1530
1531         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1532
1533         tp->is_mptcp = 1;
1534         ctx->icsk_af_ops = icsk->icsk_af_ops;
1535         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1536         ctx->tcp_data_ready = sk->sk_data_ready;
1537         ctx->tcp_state_change = sk->sk_state_change;
1538         ctx->tcp_write_space = sk->sk_write_space;
1539         ctx->tcp_error_report = sk->sk_error_report;
1540         sk->sk_data_ready = subflow_data_ready;
1541         sk->sk_write_space = subflow_write_space;
1542         sk->sk_state_change = subflow_state_change;
1543         sk->sk_error_report = subflow_error_report;
1544 out:
1545         return err;
1546 }
1547
1548 static void subflow_ulp_release(struct sock *ssk)
1549 {
1550         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1551         bool release = true;
1552         struct sock *sk;
1553
1554         if (!ctx)
1555                 return;
1556
1557         sk = ctx->conn;
1558         if (sk) {
1559                 /* if the msk has been orphaned, keep the ctx
1560                  * alive, will be freed by __mptcp_close_ssk(),
1561                  * when the subflow is still unaccepted
1562                  */
1563                 release = ctx->disposable || list_empty(&ctx->node);
1564                 sock_put(sk);
1565         }
1566
1567         mptcp_subflow_ops_undo_override(ssk);
1568         if (release)
1569                 kfree_rcu(ctx, rcu);
1570 }
1571
1572 static void subflow_ulp_clone(const struct request_sock *req,
1573                               struct sock *newsk,
1574                               const gfp_t priority)
1575 {
1576         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1577         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1578         struct mptcp_subflow_context *new_ctx;
1579
1580         if (!tcp_rsk(req)->is_mptcp ||
1581             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1582                 subflow_ulp_fallback(newsk, old_ctx);
1583                 return;
1584         }
1585
1586         new_ctx = subflow_create_ctx(newsk, priority);
1587         if (!new_ctx) {
1588                 subflow_ulp_fallback(newsk, old_ctx);
1589                 return;
1590         }
1591
1592         new_ctx->conn_finished = 1;
1593         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1594         new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1595         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1596         new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1597         new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1598         new_ctx->rel_write_seq = 1;
1599         new_ctx->tcp_sock = newsk;
1600
1601         if (subflow_req->mp_capable) {
1602                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1603                  * is fully established only after we receive the remote key
1604                  */
1605                 new_ctx->mp_capable = 1;
1606                 new_ctx->local_key = subflow_req->local_key;
1607                 new_ctx->token = subflow_req->token;
1608                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1609                 new_ctx->idsn = subflow_req->idsn;
1610         } else if (subflow_req->mp_join) {
1611                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1612                 new_ctx->mp_join = 1;
1613                 new_ctx->fully_established = 1;
1614                 new_ctx->backup = subflow_req->backup;
1615                 new_ctx->local_id = subflow_req->local_id;
1616                 new_ctx->remote_id = subflow_req->remote_id;
1617                 new_ctx->token = subflow_req->token;
1618                 new_ctx->thmac = subflow_req->thmac;
1619         }
1620 }
1621
1622 static void tcp_release_cb_override(struct sock *ssk)
1623 {
1624         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1625
1626         if (mptcp_subflow_has_delegated_action(subflow))
1627                 mptcp_subflow_process_delegated(ssk);
1628
1629         tcp_release_cb(ssk);
1630 }
1631
1632 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1633         .name           = "mptcp",
1634         .owner          = THIS_MODULE,
1635         .init           = subflow_ulp_init,
1636         .release        = subflow_ulp_release,
1637         .clone          = subflow_ulp_clone,
1638 };
1639
1640 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1641 {
1642         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1643         subflow_ops->slab_name = "request_sock_subflow";
1644
1645         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1646                                               subflow_ops->obj_size, 0,
1647                                               SLAB_ACCOUNT |
1648                                               SLAB_TYPESAFE_BY_RCU,
1649                                               NULL);
1650         if (!subflow_ops->slab)
1651                 return -ENOMEM;
1652
1653         subflow_ops->destructor = subflow_req_destructor;
1654
1655         return 0;
1656 }
1657
1658 void __init mptcp_subflow_init(void)
1659 {
1660         mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1661         if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1662                 panic("MPTCP: failed to init subflow request sock ops\n");
1663
1664         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1665         subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
1666
1667         subflow_specific = ipv4_specific;
1668         subflow_specific.conn_request = subflow_v4_conn_request;
1669         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1670         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1671
1672         tcp_prot_override = tcp_prot;
1673         tcp_prot_override.release_cb = tcp_release_cb_override;
1674
1675 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1676         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1677         subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
1678
1679         subflow_v6_specific = ipv6_specific;
1680         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1681         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1682         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1683
1684         subflow_v6m_specific = subflow_v6_specific;
1685         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1686         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1687         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1688         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1689         subflow_v6m_specific.net_frag_header_len = 0;
1690
1691         tcpv6_prot_override = tcpv6_prot;
1692         tcpv6_prot_override.release_cb = tcp_release_cb_override;
1693 #endif
1694
1695         mptcp_diag_subflow_init(&subflow_ulp_ops);
1696
1697         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1698                 panic("MPTCP: failed to register subflows to ULP\n");
1699 }