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