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 <net/sock.h>
  14 #include <net/inet_common.h>
  15 #include <net/inet_hashtables.h>
  16 #include <net/protocol.h>
  17 #include <net/tcp.h>
  18 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
  19 #include <net/ip6_route.h>
  20 #endif
  21 #include <net/mptcp.h>
  22 #include "protocol.h"
  23 #include "mib.h"
  24
  25 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
  26                                   enum linux_mptcp_mib_field field)
  27 {
  28         MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
  29 }
  30
  31 static int subflow_rebuild_header(struct sock *sk)
  32 {
  33         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
  34         int local_id, err = 0;
  35
  36         if (subflow->request_mptcp && !subflow->token) {
  37                 pr_debug("subflow=%p", sk);
  38                 err = mptcp_token_new_connect(sk);
  39         } else if (subflow->request_join && !subflow->local_nonce) {
  40                 struct mptcp_sock *msk = (struct mptcp_sock *)subflow->conn;
  41
  42                 pr_debug("subflow=%p", sk);
  43
  44                 do {
  45                         get_random_bytes(&subflow->local_nonce, sizeof(u32));
  46                 } while (!subflow->local_nonce);
  47
  48                 if (subflow->local_id)
  49                         goto out;
  50
  51                 local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
  52                 if (local_id < 0)
  53                         return -EINVAL;
  54
  55                 subflow->local_id = local_id;
  56         }
  57
  58 out:
  59         if (err)
  60                 return err;
  61
  62         return subflow->icsk_af_ops->rebuild_header(sk);
  63 }
  64
  65 static void subflow_req_destructor(struct request_sock *req)
  66 {
  67         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  68
  69         pr_debug("subflow_req=%p", subflow_req);
  70
  71         if (subflow_req->mp_capable)
  72                 mptcp_token_destroy_request(subflow_req->token);
  73         tcp_request_sock_ops.destructor(req);
  74 }
  75
  76 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
  77                                   void *hmac)
  78 {
  79         u8 msg[8];
  80
  81         put_unaligned_be32(nonce1, &msg[0]);
  82         put_unaligned_be32(nonce2, &msg[4]);
  83
  84         mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
  85 }
  86
  87 /* validate received token and create truncated hmac and nonce for SYN-ACK */
  88 static bool subflow_token_join_request(struct request_sock *req,
  89                                        const struct sk_buff *skb)
  90 {
  91         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  92         u8 hmac[MPTCPOPT_HMAC_LEN];
  93         struct mptcp_sock *msk;
  94         int local_id;
  95
  96         msk = mptcp_token_get_sock(subflow_req->token);
  97         if (!msk) {
  98                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
  99                 return false;
 100         }
 101
 102         local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
 103         if (local_id < 0) {
 104                 sock_put((struct sock *)msk);
 105                 return false;
 106         }
 107         subflow_req->local_id = local_id;
 108
 109         get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
 110
 111         subflow_generate_hmac(msk->local_key, msk->remote_key,
 112                               subflow_req->local_nonce,
 113                               subflow_req->remote_nonce, hmac);
 114
 115         subflow_req->thmac = get_unaligned_be64(hmac);
 116
 117         sock_put((struct sock *)msk);
 118         return true;
 119 }
 120
 121 static void subflow_init_req(struct request_sock *req,
 122                              const struct sock *sk_listener,
 123                              struct sk_buff *skb)
 124 {
 125         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 126         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 127         struct tcp_options_received rx_opt;
 128
 129         pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
 130
 131         memset(&rx_opt.mptcp, 0, sizeof(rx_opt.mptcp));
 132         mptcp_get_options(skb, &rx_opt);
 133
 134         subflow_req->mp_capable = 0;
 135         subflow_req->mp_join = 0;
 136
 137 #ifdef CONFIG_TCP_MD5SIG
 138         /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
 139          * TCP option space.
 140          */
 141         if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
 142                 return;
 143 #endif
 144
 145         if (rx_opt.mptcp.mp_capable) {
 146                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
 147
 148                 if (rx_opt.mptcp.mp_join)
 149                         return;
 150         } else if (rx_opt.mptcp.mp_join) {
 151                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
 152         }
 153
 154         if (rx_opt.mptcp.mp_capable && listener->request_mptcp) {
 155                 int err;
 156
 157                 err = mptcp_token_new_request(req);
 158                 if (err == 0)
 159                         subflow_req->mp_capable = 1;
 160
 161                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 162         } else if (rx_opt.mptcp.mp_join && listener->request_mptcp) {
 163                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 164                 subflow_req->mp_join = 1;
 165                 subflow_req->backup = rx_opt.mptcp.backup;
 166                 subflow_req->remote_id = rx_opt.mptcp.join_id;
 167                 subflow_req->token = rx_opt.mptcp.token;
 168                 subflow_req->remote_nonce = rx_opt.mptcp.nonce;
 169                 pr_debug("token=%u, remote_nonce=%u", subflow_req->token,
 170                          subflow_req->remote_nonce);
 171                 if (!subflow_token_join_request(req, skb)) {
 172                         subflow_req->mp_join = 0;
 173                         // @@ need to trigger RST
 174                 }
 175         }
 176 }
 177
 178 static void subflow_v4_init_req(struct request_sock *req,
 179                                 const struct sock *sk_listener,
 180                                 struct sk_buff *skb)
 181 {
 182         tcp_rsk(req)->is_mptcp = 1;
 183
 184         tcp_request_sock_ipv4_ops.init_req(req, sk_listener, skb);
 185
 186         subflow_init_req(req, sk_listener, skb);
 187 }
 188
 189 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 190 static void subflow_v6_init_req(struct request_sock *req,
 191                                 const struct sock *sk_listener,
 192                                 struct sk_buff *skb)
 193 {
 194         tcp_rsk(req)->is_mptcp = 1;
 195
 196         tcp_request_sock_ipv6_ops.init_req(req, sk_listener, skb);
 197
 198         subflow_init_req(req, sk_listener, skb);
 199 }
 200 #endif
 201
 202 /* validate received truncated hmac and create hmac for third ACK */
 203 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
 204 {
 205         u8 hmac[MPTCPOPT_HMAC_LEN];
 206         u64 thmac;
 207
 208         subflow_generate_hmac(subflow->remote_key, subflow->local_key,
 209                               subflow->remote_nonce, subflow->local_nonce,
 210                               hmac);
 211
 212         thmac = get_unaligned_be64(hmac);
 213         pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
 214                  subflow, subflow->token,
 215                  (unsigned long long)thmac,
 216                  (unsigned long long)subflow->thmac);
 217
 218         return thmac == subflow->thmac;
 219 }
 220
 221 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 222 {
 223         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 224         struct sock *parent = subflow->conn;
 225
 226         subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 227
 228         if (inet_sk_state_load(parent) != TCP_ESTABLISHED) {
 229                 inet_sk_state_store(parent, TCP_ESTABLISHED);
 230                 parent->sk_state_change(parent);
 231         }
 232
 233         if (subflow->conn_finished || !tcp_sk(sk)->is_mptcp)
 234                 return;
 235
 236         if (subflow->mp_capable) {
 237                 pr_debug("subflow=%p, remote_key=%llu", mptcp_subflow_ctx(sk),
 238                          subflow->remote_key);
 239                 mptcp_finish_connect(sk);
 240                 subflow->conn_finished = 1;
 241
 242                 if (skb) {
 243                         pr_debug("synack seq=%u", TCP_SKB_CB(skb)->seq);
 244                         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 245                 }
 246         } else if (subflow->mp_join) {
 247                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u",
 248                          subflow, subflow->thmac,
 249                          subflow->remote_nonce);
 250                 if (!subflow_thmac_valid(subflow)) {
 251                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 252                         subflow->mp_join = 0;
 253                         goto do_reset;
 254                 }
 255
 256                 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 257                                       subflow->local_nonce,
 258                                       subflow->remote_nonce,
 259                                       subflow->hmac);
 260
 261                 if (skb)
 262                         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 263
 264                 if (!mptcp_finish_join(sk))
 265                         goto do_reset;
 266
 267                 subflow->conn_finished = 1;
 268                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 269         } else {
 270 do_reset:
 271                 tcp_send_active_reset(sk, GFP_ATOMIC);
 272                 tcp_done(sk);
 273         }
 274 }
 275
 276 static struct request_sock_ops subflow_request_sock_ops;
 277 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops;
 278
 279 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 280 {
 281         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 282
 283         pr_debug("subflow=%p", subflow);
 284
 285         /* Never answer to SYNs sent to broadcast or multicast */
 286         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 287                 goto drop;
 288
 289         return tcp_conn_request(&subflow_request_sock_ops,
 290                                 &subflow_request_sock_ipv4_ops,
 291                                 sk, skb);
 292 drop:
 293         tcp_listendrop(sk);
 294         return 0;
 295 }
 296
 297 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 298 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops;
 299 static struct inet_connection_sock_af_ops subflow_v6_specific;
 300 static struct inet_connection_sock_af_ops subflow_v6m_specific;
 301
 302 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 303 {
 304         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 305
 306         pr_debug("subflow=%p", subflow);
 307
 308         if (skb->protocol == htons(ETH_P_IP))
 309                 return subflow_v4_conn_request(sk, skb);
 310
 311         if (!ipv6_unicast_destination(skb))
 312                 goto drop;
 313
 314         return tcp_conn_request(&subflow_request_sock_ops,
 315                                 &subflow_request_sock_ipv6_ops, sk, skb);
 316
 317 drop:
 318         tcp_listendrop(sk);
 319         return 0; /* don't send reset */
 320 }
 321 #endif
 322
 323 /* validate hmac received in third ACK */
 324 static bool subflow_hmac_valid(const struct request_sock *req,
 325                                const struct tcp_options_received *rx_opt)
 326 {
 327         const struct mptcp_subflow_request_sock *subflow_req;
 328         u8 hmac[MPTCPOPT_HMAC_LEN];
 329         struct mptcp_sock *msk;
 330         bool ret;
 331
 332         subflow_req = mptcp_subflow_rsk(req);
 333         msk = mptcp_token_get_sock(subflow_req->token);
 334         if (!msk)
 335                 return false;
 336
 337         subflow_generate_hmac(msk->remote_key, msk->local_key,
 338                               subflow_req->remote_nonce,
 339                               subflow_req->local_nonce, hmac);
 340
 341         ret = true;
 342         if (crypto_memneq(hmac, rx_opt->mptcp.hmac, sizeof(hmac)))
 343                 ret = false;
 344
 345         sock_put((struct sock *)msk);
 346         return ret;
 347 }
 348
 349 static void mptcp_sock_destruct(struct sock *sk)
 350 {
 351         /* if new mptcp socket isn't accepted, it is free'd
 352          * from the tcp listener sockets request queue, linked
 353          * from req->sk.  The tcp socket is released.
 354          * This calls the ULP release function which will
 355          * also remove the mptcp socket, via
 356          * sock_put(ctx->conn).
 357          *
 358          * Problem is that the mptcp socket will not be in
 359          * SYN_RECV state and doesn't have SOCK_DEAD flag.
 360          * Both result in warnings from inet_sock_destruct.
 361          */
 362
 363         if (sk->sk_state == TCP_SYN_RECV) {
 364                 sk->sk_state = TCP_CLOSE;
 365                 WARN_ON_ONCE(sk->sk_socket);
 366                 sock_orphan(sk);
 367         }
 368
 369         inet_sock_destruct(sk);
 370 }
 371
 372 static void mptcp_force_close(struct sock *sk)
 373 {
 374         inet_sk_state_store(sk, TCP_CLOSE);
 375         sk_common_release(sk);
 376 }
 377
 378 static void subflow_ulp_fallback(struct sock *sk,
 379                                  struct mptcp_subflow_context *old_ctx)
 380 {
 381         struct inet_connection_sock *icsk = inet_csk(sk);
 382
 383         mptcp_subflow_tcp_fallback(sk, old_ctx);
 384         icsk->icsk_ulp_ops = NULL;
 385         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 386         tcp_sk(sk)->is_mptcp = 0;
 387 }
 388
 389 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 390                                           struct sk_buff *skb,
 391                                           struct request_sock *req,
 392                                           struct dst_entry *dst,
 393                                           struct request_sock *req_unhash,
 394                                           bool *own_req)
 395 {
 396         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 397         struct mptcp_subflow_request_sock *subflow_req;
 398         struct tcp_options_received opt_rx;
 399         bool fallback_is_fatal = false;
 400         struct sock *new_msk = NULL;
 401         bool fallback = false;
 402         struct sock *child;
 403
 404         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 405
 406         opt_rx.mptcp.mp_capable = 0;
 407         if (tcp_rsk(req)->is_mptcp == 0)
 408                 goto create_child;
 409
 410         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 411         subflow_req = mptcp_subflow_rsk(req);
 412         if (subflow_req->mp_capable) {
 413                 if (TCP_SKB_CB(skb)->seq != subflow_req->ssn_offset + 1) {
 414                         /* here we can receive and accept an in-window,
 415                          * out-of-order pkt, which will not carry the MP_CAPABLE
 416                          * opt even on mptcp enabled paths
 417                          */
 418                         goto create_msk;
 419                 }
 420
 421                 mptcp_get_options(skb, &opt_rx);
 422                 if (!opt_rx.mptcp.mp_capable) {
 423                         fallback = true;
 424                         goto create_child;
 425                 }
 426
 427 create_msk:
 428                 new_msk = mptcp_sk_clone(listener->conn, &opt_rx, req);
 429                 if (!new_msk)
 430                         fallback = true;
 431         } else if (subflow_req->mp_join) {
 432                 fallback_is_fatal = true;
 433                 opt_rx.mptcp.mp_join = 0;
 434                 mptcp_get_options(skb, &opt_rx);
 435                 if (!opt_rx.mptcp.mp_join ||
 436                     !subflow_hmac_valid(req, &opt_rx)) {
 437                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 438                         return NULL;
 439                 }
 440         }
 441
 442 create_child:
 443         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 444                                                      req_unhash, own_req);
 445
 446         if (child && *own_req) {
 447                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 448
 449                 /* we need to fallback on ctx allocation failure and on pre-reqs
 450                  * checking above. In the latter scenario we additionally need
 451                  * to reset the context to non MPTCP status.
 452                  */
 453                 if (!ctx || fallback) {
 454                         if (fallback_is_fatal)
 455                                 goto close_child;
 456
 457                         if (ctx) {
 458                                 subflow_ulp_fallback(child, ctx);
 459                                 kfree_rcu(ctx, rcu);
 460                         }
 461                         goto out;
 462                 }
 463
 464                 if (ctx->mp_capable) {
 465                         /* new mpc subflow takes ownership of the newly
 466                          * created mptcp socket
 467                          */
 468                         new_msk->sk_destruct = mptcp_sock_destruct;
 469                         mptcp_pm_new_connection(mptcp_sk(new_msk), 1);
 470                         ctx->conn = new_msk;
 471                         new_msk = NULL;
 472
 473                         /* with OoO packets we can reach here without ingress
 474                          * mpc option
 475                          */
 476                         ctx->remote_key = opt_rx.mptcp.sndr_key;
 477                         ctx->fully_established = opt_rx.mptcp.mp_capable;
 478                         ctx->can_ack = opt_rx.mptcp.mp_capable;
 479                 } else if (ctx->mp_join) {
 480                         struct mptcp_sock *owner;
 481
 482                         owner = mptcp_token_get_sock(ctx->token);
 483                         if (!owner)
 484                                 goto close_child;
 485
 486                         ctx->conn = (struct sock *)owner;
 487                         if (!mptcp_finish_join(child))
 488                                 goto close_child;
 489
 490                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 491                 }
 492         }
 493
 494 out:
 495         /* dispose of the left over mptcp master, if any */
 496         if (unlikely(new_msk))
 497                 mptcp_force_close(new_msk);
 498
 499         /* check for expected invariant - should never trigger, just help
 500          * catching eariler subtle bugs
 501          */
 502         WARN_ON_ONCE(*own_req && child && tcp_sk(child)->is_mptcp &&
 503                      (!mptcp_subflow_ctx(child) ||
 504                       !mptcp_subflow_ctx(child)->conn));
 505         return child;
 506
 507 close_child:
 508         tcp_send_active_reset(child, GFP_ATOMIC);
 509         inet_csk_prepare_forced_close(child);
 510         tcp_done(child);
 511         return NULL;
 512 }
 513
 514 static struct inet_connection_sock_af_ops subflow_specific;
 515
 516 enum mapping_status {
 517         MAPPING_OK,
 518         MAPPING_INVALID,
 519         MAPPING_EMPTY,
 520         MAPPING_DATA_FIN
 521 };
 522
 523 static u64 expand_seq(u64 old_seq, u16 old_data_len, u64 seq)
 524 {
 525         if ((u32)seq == (u32)old_seq)
 526                 return old_seq;
 527
 528         /* Assume map covers data not mapped yet. */
 529         return seq | ((old_seq + old_data_len + 1) & GENMASK_ULL(63, 32));
 530 }
 531
 532 static void warn_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 533 {
 534         WARN_ONCE(1, "Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 535                   ssn, subflow->map_subflow_seq, subflow->map_data_len);
 536 }
 537
 538 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 539 {
 540         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 541         unsigned int skb_consumed;
 542
 543         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 544         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 545                 return true;
 546
 547         return skb->len - skb_consumed <= subflow->map_data_len -
 548                                           mptcp_subflow_get_map_offset(subflow);
 549 }
 550
 551 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 552 {
 553         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 554         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 555
 556         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 557                 /* Mapping covers data later in the subflow stream,
 558                  * currently unsupported.
 559                  */
 560                 warn_bad_map(subflow, ssn);
 561                 return false;
 562         }
 563         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 564                                   subflow->map_data_len))) {
 565                 /* Mapping does covers past subflow data, invalid */
 566                 warn_bad_map(subflow, ssn + skb->len);
 567                 return false;
 568         }
 569         return true;
 570 }
 571
 572 static enum mapping_status get_mapping_status(struct sock *ssk)
 573 {
 574         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 575         struct mptcp_ext *mpext;
 576         struct sk_buff *skb;
 577         u16 data_len;
 578         u64 map_seq;
 579
 580         skb = skb_peek(&ssk->sk_receive_queue);
 581         if (!skb)
 582                 return MAPPING_EMPTY;
 583
 584         mpext = mptcp_get_ext(skb);
 585         if (!mpext || !mpext->use_map) {
 586                 if (!subflow->map_valid && !skb->len) {
 587                         /* the TCP stack deliver 0 len FIN pkt to the receive
 588                          * queue, that is the only 0len pkts ever expected here,
 589                          * and we can admit no mapping only for 0 len pkts
 590                          */
 591                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 592                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 593                                           TCP_SKB_CB(skb)->seq,
 594                                           TCP_SKB_CB(skb)->end_seq,
 595                                           TCP_SKB_CB(skb)->tcp_flags);
 596                         sk_eat_skb(ssk, skb);
 597                         return MAPPING_EMPTY;
 598                 }
 599
 600                 if (!subflow->map_valid)
 601                         return MAPPING_INVALID;
 602
 603                 goto validate_seq;
 604         }
 605
 606         pr_debug("seq=%llu is64=%d ssn=%u data_len=%u data_fin=%d",
 607                  mpext->data_seq, mpext->dsn64, mpext->subflow_seq,
 608                  mpext->data_len, mpext->data_fin);
 609
 610         data_len = mpext->data_len;
 611         if (data_len == 0) {
 612                 pr_err("Infinite mapping not handled");
 613                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
 614                 return MAPPING_INVALID;
 615         }
 616
 617         if (mpext->data_fin == 1) {
 618                 if (data_len == 1) {
 619                         pr_debug("DATA_FIN with no payload");
 620                         if (subflow->map_valid) {
 621                                 /* A DATA_FIN might arrive in a DSS
 622                                  * option before the previous mapping
 623                                  * has been fully consumed. Continue
 624                                  * handling the existing mapping.
 625                                  */
 626                                 skb_ext_del(skb, SKB_EXT_MPTCP);
 627                                 return MAPPING_OK;
 628                         } else {
 629                                 return MAPPING_DATA_FIN;
 630                         }
 631                 }
 632
 633                 /* Adjust for DATA_FIN using 1 byte of sequence space */
 634                 data_len--;
 635         }
 636
 637         if (!mpext->dsn64) {
 638                 map_seq = expand_seq(subflow->map_seq, subflow->map_data_len,
 639                                      mpext->data_seq);
 640                 pr_debug("expanded seq=%llu", subflow->map_seq);
 641         } else {
 642                 map_seq = mpext->data_seq;
 643         }
 644
 645         if (subflow->map_valid) {
 646                 /* Allow replacing only with an identical map */
 647                 if (subflow->map_seq == map_seq &&
 648                     subflow->map_subflow_seq == mpext->subflow_seq &&
 649                     subflow->map_data_len == data_len) {
 650                         skb_ext_del(skb, SKB_EXT_MPTCP);
 651                         return MAPPING_OK;
 652                 }
 653
 654                 /* If this skb data are fully covered by the current mapping,
 655                  * the new map would need caching, which is not supported
 656                  */
 657                 if (skb_is_fully_mapped(ssk, skb)) {
 658                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
 659                         return MAPPING_INVALID;
 660                 }
 661
 662                 /* will validate the next map after consuming the current one */
 663                 return MAPPING_OK;
 664         }
 665
 666         subflow->map_seq = map_seq;
 667         subflow->map_subflow_seq = mpext->subflow_seq;
 668         subflow->map_data_len = data_len;
 669         subflow->map_valid = 1;
 670         subflow->mpc_map = mpext->mpc_map;
 671         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u",
 672                  subflow->map_seq, subflow->map_subflow_seq,
 673                  subflow->map_data_len);
 674
 675 validate_seq:
 676         /* we revalidate valid mapping on new skb, because we must ensure
 677          * the current skb is completely covered by the available mapping
 678          */
 679         if (!validate_mapping(ssk, skb))
 680                 return MAPPING_INVALID;
 681
 682         skb_ext_del(skb, SKB_EXT_MPTCP);
 683         return MAPPING_OK;
 684 }
 685
 686 static int subflow_read_actor(read_descriptor_t *desc,
 687                               struct sk_buff *skb,
 688                               unsigned int offset, size_t len)
 689 {
 690         size_t copy_len = min(desc->count, len);
 691
 692         desc->count -= copy_len;
 693
 694         pr_debug("flushed %zu bytes, %zu left", copy_len, desc->count);
 695         return copy_len;
 696 }
 697
 698 static bool subflow_check_data_avail(struct sock *ssk)
 699 {
 700         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 701         enum mapping_status status;
 702         struct mptcp_sock *msk;
 703         struct sk_buff *skb;
 704
 705         pr_debug("msk=%p ssk=%p data_avail=%d skb=%p", subflow->conn, ssk,
 706                  subflow->data_avail, skb_peek(&ssk->sk_receive_queue));
 707         if (subflow->data_avail)
 708                 return true;
 709
 710         msk = mptcp_sk(subflow->conn);
 711         for (;;) {
 712                 u32 map_remaining;
 713                 size_t delta;
 714                 u64 ack_seq;
 715                 u64 old_ack;
 716
 717                 status = get_mapping_status(ssk);
 718                 pr_debug("msk=%p ssk=%p status=%d", msk, ssk, status);
 719                 if (status == MAPPING_INVALID) {
 720                         ssk->sk_err = EBADMSG;
 721                         goto fatal;
 722                 }
 723
 724                 if (status != MAPPING_OK)
 725                         return false;
 726
 727                 skb = skb_peek(&ssk->sk_receive_queue);
 728                 if (WARN_ON_ONCE(!skb))
 729                         return false;
 730
 731                 /* if msk lacks the remote key, this subflow must provide an
 732                  * MP_CAPABLE-based mapping
 733                  */
 734                 if (unlikely(!READ_ONCE(msk->can_ack))) {
 735                         if (!subflow->mpc_map) {
 736                                 ssk->sk_err = EBADMSG;
 737                                 goto fatal;
 738                         }
 739                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
 740                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
 741                         WRITE_ONCE(msk->can_ack, true);
 742                 }
 743
 744                 old_ack = READ_ONCE(msk->ack_seq);
 745                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
 746                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
 747                          ack_seq);
 748                 if (ack_seq == old_ack)
 749                         break;
 750
 751                 /* only accept in-sequence mapping. Old values are spurious
 752                  * retransmission; we can hit "future" values on active backup
 753                  * subflow switch, we relay on retransmissions to get
 754                  * in-sequence data.
 755                  * Cuncurrent subflows support will require subflow data
 756                  * reordering
 757                  */
 758                 map_remaining = subflow->map_data_len -
 759                                 mptcp_subflow_get_map_offset(subflow);
 760                 if (before64(ack_seq, old_ack))
 761                         delta = min_t(size_t, old_ack - ack_seq, map_remaining);
 762                 else
 763                         delta = min_t(size_t, ack_seq - old_ack, map_remaining);
 764
 765                 /* discard mapped data */
 766                 pr_debug("discarding %zu bytes, current map len=%d", delta,
 767                          map_remaining);
 768                 if (delta) {
 769                         read_descriptor_t desc = {
 770                                 .count = delta,
 771                         };
 772                         int ret;
 773
 774                         ret = tcp_read_sock(ssk, &desc, subflow_read_actor);
 775                         if (ret < 0) {
 776                                 ssk->sk_err = -ret;
 777                                 goto fatal;
 778                         }
 779                         if (ret < delta)
 780                                 return false;
 781                         if (delta == map_remaining)
 782                                 subflow->map_valid = 0;
 783                 }
 784         }
 785         return true;
 786
 787 fatal:
 788         /* fatal protocol error, close the socket */
 789         /* This barrier is coupled with smp_rmb() in tcp_poll() */
 790         smp_wmb();
 791         ssk->sk_error_report(ssk);
 792         tcp_set_state(ssk, TCP_CLOSE);
 793         tcp_send_active_reset(ssk, GFP_ATOMIC);
 794         return false;
 795 }
 796
 797 bool mptcp_subflow_data_available(struct sock *sk)
 798 {
 799         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 800         struct sk_buff *skb;
 801
 802         /* check if current mapping is still valid */
 803         if (subflow->map_valid &&
 804             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
 805                 subflow->map_valid = 0;
 806                 subflow->data_avail = 0;
 807
 808                 pr_debug("Done with mapping: seq=%u data_len=%u",
 809                          subflow->map_subflow_seq,
 810                          subflow->map_data_len);
 811         }
 812
 813         if (!subflow_check_data_avail(sk)) {
 814                 subflow->data_avail = 0;
 815                 return false;
 816         }
 817
 818         skb = skb_peek(&sk->sk_receive_queue);
 819         subflow->data_avail = skb &&
 820                        before(tcp_sk(sk)->copied_seq, TCP_SKB_CB(skb)->end_seq);
 821         return subflow->data_avail;
 822 }
 823
 824 static void subflow_data_ready(struct sock *sk)
 825 {
 826         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 827         struct sock *parent = subflow->conn;
 828
 829         if (!subflow->mp_capable && !subflow->mp_join) {
 830                 subflow->tcp_data_ready(sk);
 831
 832                 parent->sk_data_ready(parent);
 833                 return;
 834         }
 835
 836         if (mptcp_subflow_data_available(sk))
 837                 mptcp_data_ready(parent, sk);
 838 }
 839
 840 static void subflow_write_space(struct sock *sk)
 841 {
 842         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 843         struct sock *parent = subflow->conn;
 844
 845         sk_stream_write_space(sk);
 846         if (sk_stream_is_writeable(sk)) {
 847                 set_bit(MPTCP_SEND_SPACE, &mptcp_sk(parent)->flags);
 848                 smp_mb__after_atomic();
 849                 /* set SEND_SPACE before sk_stream_write_space clears NOSPACE */
 850                 sk_stream_write_space(parent);
 851         }
 852 }
 853
 854 static struct inet_connection_sock_af_ops *
 855 subflow_default_af_ops(struct sock *sk)
 856 {
 857 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 858         if (sk->sk_family == AF_INET6)
 859                 return &subflow_v6_specific;
 860 #endif
 861         return &subflow_specific;
 862 }
 863
 864 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 865 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
 866 {
 867         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 868         struct inet_connection_sock *icsk = inet_csk(sk);
 869         struct inet_connection_sock_af_ops *target;
 870
 871         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
 872
 873         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
 874                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
 875
 876         if (likely(icsk->icsk_af_ops == target))
 877                 return;
 878
 879         subflow->icsk_af_ops = icsk->icsk_af_ops;
 880         icsk->icsk_af_ops = target;
 881 }
 882 #endif
 883
 884 static void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
 885                                 struct sockaddr_storage *addr)
 886 {
 887         memset(addr, 0, sizeof(*addr));
 888         addr->ss_family = info->family;
 889         if (addr->ss_family == AF_INET) {
 890                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
 891
 892                 in_addr->sin_addr = info->addr;
 893                 in_addr->sin_port = info->port;
 894         }
 895 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 896         else if (addr->ss_family == AF_INET6) {
 897                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
 898
 899                 in6_addr->sin6_addr = info->addr6;
 900                 in6_addr->sin6_port = info->port;
 901         }
 902 #endif
 903 }
 904
 905 int __mptcp_subflow_connect(struct sock *sk, int ifindex,
 906                             const struct mptcp_addr_info *loc,
 907                             const struct mptcp_addr_info *remote)
 908 {
 909         struct mptcp_sock *msk = mptcp_sk(sk);
 910         struct mptcp_subflow_context *subflow;
 911         struct sockaddr_storage addr;
 912         struct socket *sf;
 913         u32 remote_token;
 914         int addrlen;
 915         int err;
 916
 917         if (sk->sk_state != TCP_ESTABLISHED)
 918                 return -ENOTCONN;
 919
 920         err = mptcp_subflow_create_socket(sk, &sf);
 921         if (err)
 922                 return err;
 923
 924         subflow = mptcp_subflow_ctx(sf->sk);
 925         subflow->remote_key = msk->remote_key;
 926         subflow->local_key = msk->local_key;
 927         subflow->token = msk->token;
 928         mptcp_info2sockaddr(loc, &addr);
 929
 930         addrlen = sizeof(struct sockaddr_in);
 931 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 932         if (loc->family == AF_INET6)
 933                 addrlen = sizeof(struct sockaddr_in6);
 934 #endif
 935         sf->sk->sk_bound_dev_if = ifindex;
 936         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
 937         if (err)
 938                 goto failed;
 939
 940         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
 941         pr_debug("msk=%p remote_token=%u", msk, remote_token);
 942         subflow->remote_token = remote_token;
 943         subflow->local_id = loc->id;
 944         subflow->request_join = 1;
 945         subflow->request_bkup = 1;
 946         mptcp_info2sockaddr(remote, &addr);
 947
 948         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
 949         if (err && err != -EINPROGRESS)
 950                 goto failed;
 951
 952         spin_lock_bh(&msk->join_list_lock);
 953         list_add_tail(&subflow->node, &msk->join_list);
 954         spin_unlock_bh(&msk->join_list_lock);
 955
 956         return err;
 957
 958 failed:
 959         sock_release(sf);
 960         return err;
 961 }
 962
 963 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
 964 {
 965         struct mptcp_subflow_context *subflow;
 966         struct net *net = sock_net(sk);
 967         struct socket *sf;
 968         int err;
 969
 970         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
 971                                &sf);
 972         if (err)
 973                 return err;
 974
 975         lock_sock(sf->sk);
 976
 977         /* kernel sockets do not by default acquire net ref, but TCP timer
 978          * needs it.
 979          */
 980         sf->sk->sk_net_refcnt = 1;
 981         get_net(net);
 982 #ifdef CONFIG_PROC_FS
 983         this_cpu_add(*net->core.sock_inuse, 1);
 984 #endif
 985         err = tcp_set_ulp(sf->sk, "mptcp");
 986         release_sock(sf->sk);
 987
 988         if (err)
 989                 return err;
 990
 991         subflow = mptcp_subflow_ctx(sf->sk);
 992         pr_debug("subflow=%p", subflow);
 993
 994         *new_sock = sf;
 995         sock_hold(sk);
 996         subflow->conn = sk;
 997
 998         return 0;
 999 }
1000
1001 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1002                                                         gfp_t priority)
1003 {
1004         struct inet_connection_sock *icsk = inet_csk(sk);
1005         struct mptcp_subflow_context *ctx;
1006
1007         ctx = kzalloc(sizeof(*ctx), priority);
1008         if (!ctx)
1009                 return NULL;
1010
1011         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1012         INIT_LIST_HEAD(&ctx->node);
1013
1014         pr_debug("subflow=%p", ctx);
1015
1016         ctx->tcp_sock = sk;
1017
1018         return ctx;
1019 }
1020
1021 static void __subflow_state_change(struct sock *sk)
1022 {
1023         struct socket_wq *wq;
1024
1025         rcu_read_lock();
1026         wq = rcu_dereference(sk->sk_wq);
1027         if (skwq_has_sleeper(wq))
1028                 wake_up_interruptible_all(&wq->wait);
1029         rcu_read_unlock();
1030 }
1031
1032 static bool subflow_is_done(const struct sock *sk)
1033 {
1034         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1035 }
1036
1037 static void subflow_state_change(struct sock *sk)
1038 {
1039         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1040         struct sock *parent = subflow->conn;
1041
1042         __subflow_state_change(sk);
1043
1044         /* as recvmsg() does not acquire the subflow socket for ssk selection
1045          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1046          * the data available machinery here.
1047          */
1048         if (subflow->mp_capable && mptcp_subflow_data_available(sk))
1049                 mptcp_data_ready(parent, sk);
1050
1051         if (!(parent->sk_shutdown & RCV_SHUTDOWN) &&
1052             !subflow->rx_eof && subflow_is_done(sk)) {
1053                 subflow->rx_eof = 1;
1054                 mptcp_subflow_eof(parent);
1055         }
1056 }
1057
1058 static int subflow_ulp_init(struct sock *sk)
1059 {
1060         struct inet_connection_sock *icsk = inet_csk(sk);
1061         struct mptcp_subflow_context *ctx;
1062         struct tcp_sock *tp = tcp_sk(sk);
1063         int err = 0;
1064
1065         /* disallow attaching ULP to a socket unless it has been
1066          * created with sock_create_kern()
1067          */
1068         if (!sk->sk_kern_sock) {
1069                 err = -EOPNOTSUPP;
1070                 goto out;
1071         }
1072
1073         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1074         if (!ctx) {
1075                 err = -ENOMEM;
1076                 goto out;
1077         }
1078
1079         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1080
1081         tp->is_mptcp = 1;
1082         ctx->icsk_af_ops = icsk->icsk_af_ops;
1083         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1084         ctx->tcp_data_ready = sk->sk_data_ready;
1085         ctx->tcp_state_change = sk->sk_state_change;
1086         ctx->tcp_write_space = sk->sk_write_space;
1087         sk->sk_data_ready = subflow_data_ready;
1088         sk->sk_write_space = subflow_write_space;
1089         sk->sk_state_change = subflow_state_change;
1090 out:
1091         return err;
1092 }
1093
1094 static void subflow_ulp_release(struct sock *sk)
1095 {
1096         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(sk);
1097
1098         if (!ctx)
1099                 return;
1100
1101         if (ctx->conn)
1102                 sock_put(ctx->conn);
1103
1104         kfree_rcu(ctx, rcu);
1105 }
1106
1107 static void subflow_ulp_clone(const struct request_sock *req,
1108                               struct sock *newsk,
1109                               const gfp_t priority)
1110 {
1111         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1112         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1113         struct mptcp_subflow_context *new_ctx;
1114
1115         if (!tcp_rsk(req)->is_mptcp ||
1116             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1117                 subflow_ulp_fallback(newsk, old_ctx);
1118                 return;
1119         }
1120
1121         new_ctx = subflow_create_ctx(newsk, priority);
1122         if (!new_ctx) {
1123                 subflow_ulp_fallback(newsk, old_ctx);
1124                 return;
1125         }
1126
1127         new_ctx->conn_finished = 1;
1128         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1129         new_ctx->tcp_data_ready = old_ctx->tcp_data_ready;
1130         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1131         new_ctx->tcp_write_space = old_ctx->tcp_write_space;
1132         new_ctx->rel_write_seq = 1;
1133         new_ctx->tcp_sock = newsk;
1134
1135         if (subflow_req->mp_capable) {
1136                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1137                  * is fully established only after we receive the remote key
1138                  */
1139                 new_ctx->mp_capable = 1;
1140                 new_ctx->local_key = subflow_req->local_key;
1141                 new_ctx->token = subflow_req->token;
1142                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1143                 new_ctx->idsn = subflow_req->idsn;
1144         } else if (subflow_req->mp_join) {
1145                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1146                 new_ctx->mp_join = 1;
1147                 new_ctx->fully_established = 1;
1148                 new_ctx->backup = subflow_req->backup;
1149                 new_ctx->local_id = subflow_req->local_id;
1150                 new_ctx->token = subflow_req->token;
1151                 new_ctx->thmac = subflow_req->thmac;
1152         }
1153 }
1154
1155 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1156         .name           = "mptcp",
1157         .owner          = THIS_MODULE,
1158         .init           = subflow_ulp_init,
1159         .release        = subflow_ulp_release,
1160         .clone          = subflow_ulp_clone,
1161 };
1162
1163 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1164 {
1165         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1166         subflow_ops->slab_name = "request_sock_subflow";
1167
1168         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1169                                               subflow_ops->obj_size, 0,
1170                                               SLAB_ACCOUNT |
1171                                               SLAB_TYPESAFE_BY_RCU,
1172                                               NULL);
1173         if (!subflow_ops->slab)
1174                 return -ENOMEM;
1175
1176         subflow_ops->destructor = subflow_req_destructor;
1177
1178         return 0;
1179 }
1180
1181 void mptcp_subflow_init(void)
1182 {
1183         subflow_request_sock_ops = tcp_request_sock_ops;
1184         if (subflow_ops_init(&subflow_request_sock_ops) != 0)
1185                 panic("MPTCP: failed to init subflow request sock ops\n");
1186
1187         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1188         subflow_request_sock_ipv4_ops.init_req = subflow_v4_init_req;
1189
1190         subflow_specific = ipv4_specific;
1191         subflow_specific.conn_request = subflow_v4_conn_request;
1192         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1193         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1194         subflow_specific.rebuild_header = subflow_rebuild_header;
1195
1196 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1197         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1198         subflow_request_sock_ipv6_ops.init_req = subflow_v6_init_req;
1199
1200         subflow_v6_specific = ipv6_specific;
1201         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1202         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1203         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1204         subflow_v6_specific.rebuild_header = subflow_rebuild_header;
1205
1206         subflow_v6m_specific = subflow_v6_specific;
1207         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1208         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1209         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1210         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1211         subflow_v6m_specific.net_frag_header_len = 0;
1212 #endif
1213
1214         mptcp_diag_subflow_init(&subflow_ulp_ops);
1215
1216         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1217                 panic("MPTCP: failed to register subflows to ULP\n");
1218 }