net/mptcp/subflow.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Multipath TCP
   3  *
   4  * Copyright (c) 2017 - 2019, Intel Corporation.
   5  */
   6
   7 #define pr_fmt(fmt) "MPTCP: " fmt
   8
   9 #include <linux/kernel.h>
  10 #include <linux/module.h>
  11 #include <linux/netdevice.h>
  12 #include <crypto/algapi.h>
  13 #include <crypto/sha2.h>
  14 #include <net/sock.h>
  15 #include <net/inet_common.h>
  16 #include <net/inet_hashtables.h>
  17 #include <net/protocol.h>
  18 #include <net/tcp.h>
  19 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
  20 #include <net/ip6_route.h>
  21 #include <net/transp_v6.h>
  22 #endif
  23 #include <net/mptcp.h>
  24 #include <uapi/linux/mptcp.h>
  25 #include "protocol.h"
  26 #include "mib.h"
  27
  28 #include <trace/events/mptcp.h>
  29 #include <trace/events/sock.h>
  30
  31 static void mptcp_subflow_ops_undo_override(struct sock *ssk);
  32
  33 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
  34                                   enum linux_mptcp_mib_field field)
  35 {
  36         MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
  37 }
  38
  39 static void subflow_req_destructor(struct request_sock *req)
  40 {
  41         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  42
  43         pr_debug("subflow_req=%p", subflow_req);
  44
  45         if (subflow_req->msk)
  46                 sock_put((struct sock *)subflow_req->msk);
  47
  48         mptcp_token_destroy_request(req);
  49 }
  50
  51 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
  52                                   void *hmac)
  53 {
  54         u8 msg[8];
  55
  56         put_unaligned_be32(nonce1, &msg[0]);
  57         put_unaligned_be32(nonce2, &msg[4]);
  58
  59         mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
  60 }
  61
  62 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
  63 {
  64         return mptcp_is_fully_established((void *)msk) &&
  65                 ((mptcp_pm_is_userspace(msk) &&
  66                   mptcp_userspace_pm_active(msk)) ||
  67                  READ_ONCE(msk->pm.accept_subflow));
  68 }
  69
  70 /* validate received token and create truncated hmac and nonce for SYN-ACK */
  71 static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req)
  72 {
  73         struct mptcp_sock *msk = subflow_req->msk;
  74         u8 hmac[SHA256_DIGEST_SIZE];
  75
  76         get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
  77
  78         subflow_generate_hmac(msk->local_key, msk->remote_key,
  79                               subflow_req->local_nonce,
  80                               subflow_req->remote_nonce, hmac);
  81
  82         subflow_req->thmac = get_unaligned_be64(hmac);
  83 }
  84
  85 static struct mptcp_sock *subflow_token_join_request(struct request_sock *req)
  86 {
  87         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  88         struct mptcp_sock *msk;
  89         int local_id;
  90
  91         msk = mptcp_token_get_sock(sock_net(req_to_sk(req)), subflow_req->token);
  92         if (!msk) {
  93                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINNOTOKEN);
  94                 return NULL;
  95         }
  96
  97         local_id = mptcp_pm_get_local_id(msk, (struct sock_common *)req);
  98         if (local_id < 0) {
  99                 sock_put((struct sock *)msk);
 100                 return NULL;
 101         }
 102         subflow_req->local_id = local_id;
 103
 104         return msk;
 105 }
 106
 107 static void subflow_init_req(struct request_sock *req, const struct sock *sk_listener)
 108 {
 109         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 110
 111         subflow_req->mp_capable = 0;
 112         subflow_req->mp_join = 0;
 113         subflow_req->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk_listener));
 114         subflow_req->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk_listener));
 115         subflow_req->msk = NULL;
 116         mptcp_token_init_request(req);
 117 }
 118
 119 static bool subflow_use_different_sport(struct mptcp_sock *msk, const struct sock *sk)
 120 {
 121         return inet_sk(sk)->inet_sport != inet_sk((struct sock *)msk)->inet_sport;
 122 }
 123
 124 static void subflow_add_reset_reason(struct sk_buff *skb, u8 reason)
 125 {
 126         struct mptcp_ext *mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
 127
 128         if (mpext) {
 129                 memset(mpext, 0, sizeof(*mpext));
 130                 mpext->reset_reason = reason;
 131         }
 132 }
 133
 134 /* Init mptcp request socket.
 135  *
 136  * Returns an error code if a JOIN has failed and a TCP reset
 137  * should be sent.
 138  */
 139 static int subflow_check_req(struct request_sock *req,
 140                              const struct sock *sk_listener,
 141                              struct sk_buff *skb)
 142 {
 143         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 144         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 145         struct mptcp_options_received mp_opt;
 146         bool opt_mp_capable, opt_mp_join;
 147
 148         pr_debug("subflow_req=%p, listener=%p", subflow_req, listener);
 149
 150 #ifdef CONFIG_TCP_MD5SIG
 151         /* no MPTCP if MD5SIG is enabled on this socket or we may run out of
 152          * TCP option space.
 153          */
 154         if (rcu_access_pointer(tcp_sk(sk_listener)->md5sig_info))
 155                 return -EINVAL;
 156 #endif
 157
 158         mptcp_get_options(skb, &mp_opt);
 159
 160         opt_mp_capable = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPC);
 161         opt_mp_join = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ);
 162         if (opt_mp_capable) {
 163                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
 164
 165                 if (opt_mp_join)
 166                         return 0;
 167         } else if (opt_mp_join) {
 168                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
 169         }
 170
 171         if (opt_mp_capable && listener->request_mptcp) {
 172                 int err, retries = MPTCP_TOKEN_MAX_RETRIES;
 173
 174                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 175 again:
 176                 do {
 177                         get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
 178                 } while (subflow_req->local_key == 0);
 179
 180                 if (unlikely(req->syncookie)) {
 181                         mptcp_crypto_key_sha(subflow_req->local_key,
 182                                              &subflow_req->token,
 183                                              &subflow_req->idsn);
 184                         if (mptcp_token_exists(subflow_req->token)) {
 185                                 if (retries-- > 0)
 186                                         goto again;
 187                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
 188                         } else {
 189                                 subflow_req->mp_capable = 1;
 190                         }
 191                         return 0;
 192                 }
 193
 194                 err = mptcp_token_new_request(req);
 195                 if (err == 0)
 196                         subflow_req->mp_capable = 1;
 197                 else if (retries-- > 0)
 198                         goto again;
 199                 else
 200                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
 201
 202         } else if (opt_mp_join && listener->request_mptcp) {
 203                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 204                 subflow_req->mp_join = 1;
 205                 subflow_req->backup = mp_opt.backup;
 206                 subflow_req->remote_id = mp_opt.join_id;
 207                 subflow_req->token = mp_opt.token;
 208                 subflow_req->remote_nonce = mp_opt.nonce;
 209                 subflow_req->msk = subflow_token_join_request(req);
 210
 211                 /* Can't fall back to TCP in this case. */
 212                 if (!subflow_req->msk) {
 213                         subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 214                         return -EPERM;
 215                 }
 216
 217                 if (subflow_use_different_sport(subflow_req->msk, sk_listener)) {
 218                         pr_debug("syn inet_sport=%d %d",
 219                                  ntohs(inet_sk(sk_listener)->inet_sport),
 220                                  ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport));
 221                         if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) {
 222                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX);
 223                                 return -EPERM;
 224                         }
 225                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX);
 226                 }
 227
 228                 subflow_req_create_thmac(subflow_req);
 229
 230                 if (unlikely(req->syncookie)) {
 231                         if (mptcp_can_accept_new_subflow(subflow_req->msk))
 232                                 subflow_init_req_cookie_join_save(subflow_req, skb);
 233                         else
 234                                 return -EPERM;
 235                 }
 236
 237                 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
 238                          subflow_req->remote_nonce, subflow_req->msk);
 239         }
 240
 241         return 0;
 242 }
 243
 244 int mptcp_subflow_init_cookie_req(struct request_sock *req,
 245                                   const struct sock *sk_listener,
 246                                   struct sk_buff *skb)
 247 {
 248         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 249         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 250         struct mptcp_options_received mp_opt;
 251         bool opt_mp_capable, opt_mp_join;
 252         int err;
 253
 254         subflow_init_req(req, sk_listener);
 255         mptcp_get_options(skb, &mp_opt);
 256
 257         opt_mp_capable = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPC);
 258         opt_mp_join = !!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ);
 259         if (opt_mp_capable && opt_mp_join)
 260                 return -EINVAL;
 261
 262         if (opt_mp_capable && listener->request_mptcp) {
 263                 if (mp_opt.sndr_key == 0)
 264                         return -EINVAL;
 265
 266                 subflow_req->local_key = mp_opt.rcvr_key;
 267                 err = mptcp_token_new_request(req);
 268                 if (err)
 269                         return err;
 270
 271                 subflow_req->mp_capable = 1;
 272                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 273         } else if (opt_mp_join && listener->request_mptcp) {
 274                 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
 275                         return -EINVAL;
 276
 277                 subflow_req->mp_join = 1;
 278                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 279         }
 280
 281         return 0;
 282 }
 283 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
 284
 285 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
 286                                               struct sk_buff *skb,
 287                                               struct flowi *fl,
 288                                               struct request_sock *req)
 289 {
 290         struct dst_entry *dst;
 291         int err;
 292
 293         tcp_rsk(req)->is_mptcp = 1;
 294         subflow_init_req(req, sk);
 295
 296         dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req);
 297         if (!dst)
 298                 return NULL;
 299
 300         err = subflow_check_req(req, sk, skb);
 301         if (err == 0)
 302                 return dst;
 303
 304         dst_release(dst);
 305         if (!req->syncookie)
 306                 tcp_request_sock_ops.send_reset(sk, skb);
 307         return NULL;
 308 }
 309
 310 static void subflow_prep_synack(const struct sock *sk, struct request_sock *req,
 311                                 struct tcp_fastopen_cookie *foc,
 312                                 enum tcp_synack_type synack_type)
 313 {
 314         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 315         struct inet_request_sock *ireq = inet_rsk(req);
 316
 317         /* clear tstamp_ok, as needed depending on cookie */
 318         if (foc && foc->len > -1)
 319                 ireq->tstamp_ok = 0;
 320
 321         if (synack_type == TCP_SYNACK_FASTOPEN)
 322                 mptcp_fastopen_subflow_synack_set_params(subflow, req);
 323 }
 324
 325 static int subflow_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
 326                                   struct flowi *fl,
 327                                   struct request_sock *req,
 328                                   struct tcp_fastopen_cookie *foc,
 329                                   enum tcp_synack_type synack_type,
 330                                   struct sk_buff *syn_skb)
 331 {
 332         subflow_prep_synack(sk, req, foc, synack_type);
 333
 334         return tcp_request_sock_ipv4_ops.send_synack(sk, dst, fl, req, foc,
 335                                                      synack_type, syn_skb);
 336 }
 337
 338 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 339 static int subflow_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
 340                                   struct flowi *fl,
 341                                   struct request_sock *req,
 342                                   struct tcp_fastopen_cookie *foc,
 343                                   enum tcp_synack_type synack_type,
 344                                   struct sk_buff *syn_skb)
 345 {
 346         subflow_prep_synack(sk, req, foc, synack_type);
 347
 348         return tcp_request_sock_ipv6_ops.send_synack(sk, dst, fl, req, foc,
 349                                                      synack_type, syn_skb);
 350 }
 351
 352 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
 353                                               struct sk_buff *skb,
 354                                               struct flowi *fl,
 355                                               struct request_sock *req)
 356 {
 357         struct dst_entry *dst;
 358         int err;
 359
 360         tcp_rsk(req)->is_mptcp = 1;
 361         subflow_init_req(req, sk);
 362
 363         dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req);
 364         if (!dst)
 365                 return NULL;
 366
 367         err = subflow_check_req(req, sk, skb);
 368         if (err == 0)
 369                 return dst;
 370
 371         dst_release(dst);
 372         if (!req->syncookie)
 373                 tcp6_request_sock_ops.send_reset(sk, skb);
 374         return NULL;
 375 }
 376 #endif
 377
 378 /* validate received truncated hmac and create hmac for third ACK */
 379 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
 380 {
 381         u8 hmac[SHA256_DIGEST_SIZE];
 382         u64 thmac;
 383
 384         subflow_generate_hmac(subflow->remote_key, subflow->local_key,
 385                               subflow->remote_nonce, subflow->local_nonce,
 386                               hmac);
 387
 388         thmac = get_unaligned_be64(hmac);
 389         pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
 390                  subflow, subflow->token, thmac, subflow->thmac);
 391
 392         return thmac == subflow->thmac;
 393 }
 394
 395 void mptcp_subflow_reset(struct sock *ssk)
 396 {
 397         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 398         struct sock *sk = subflow->conn;
 399
 400         /* mptcp_mp_fail_no_response() can reach here on an already closed
 401          * socket
 402          */
 403         if (ssk->sk_state == TCP_CLOSE)
 404                 return;
 405
 406         /* must hold: tcp_done() could drop last reference on parent */
 407         sock_hold(sk);
 408
 409         tcp_send_active_reset(ssk, GFP_ATOMIC);
 410         tcp_done(ssk);
 411         if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags))
 412                 mptcp_schedule_work(sk);
 413
 414         sock_put(sk);
 415 }
 416
 417 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
 418 {
 419         return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
 420 }
 421
 422 void __mptcp_set_connected(struct sock *sk)
 423 {
 424         if (sk->sk_state == TCP_SYN_SENT) {
 425                 inet_sk_state_store(sk, TCP_ESTABLISHED);
 426                 sk->sk_state_change(sk);
 427         }
 428 }
 429
 430 static void mptcp_set_connected(struct sock *sk)
 431 {
 432         mptcp_data_lock(sk);
 433         if (!sock_owned_by_user(sk))
 434                 __mptcp_set_connected(sk);
 435         else
 436                 __set_bit(MPTCP_CONNECTED, &mptcp_sk(sk)->cb_flags);
 437         mptcp_data_unlock(sk);
 438 }
 439
 440 static void subflow_set_remote_key(struct mptcp_sock *msk,
 441                                    struct mptcp_subflow_context *subflow,
 442                                    const struct mptcp_options_received *mp_opt)
 443 {
 444         /* active MPC subflow will reach here multiple times:
 445          * at subflow_finish_connect() time and at 4th ack time
 446          */
 447         if (subflow->remote_key_valid)
 448                 return;
 449
 450         subflow->remote_key_valid = 1;
 451         subflow->remote_key = mp_opt->sndr_key;
 452         mptcp_crypto_key_sha(subflow->remote_key, NULL, &subflow->iasn);
 453         subflow->iasn++;
 454
 455         WRITE_ONCE(msk->remote_key, subflow->remote_key);
 456         WRITE_ONCE(msk->ack_seq, subflow->iasn);
 457         WRITE_ONCE(msk->can_ack, true);
 458         atomic64_set(&msk->rcv_wnd_sent, subflow->iasn);
 459 }
 460
 461 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 462 {
 463         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 464         struct mptcp_options_received mp_opt;
 465         struct sock *parent = subflow->conn;
 466         struct mptcp_sock *msk;
 467
 468         subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 469
 470         /* be sure no special action on any packet other than syn-ack */
 471         if (subflow->conn_finished)
 472                 return;
 473
 474         msk = mptcp_sk(parent);
 475         mptcp_propagate_sndbuf(parent, sk);
 476         subflow->rel_write_seq = 1;
 477         subflow->conn_finished = 1;
 478         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 479         pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
 480
 481         mptcp_get_options(skb, &mp_opt);
 482         if (subflow->request_mptcp) {
 483                 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPC)) {
 484                         MPTCP_INC_STATS(sock_net(sk),
 485                                         MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
 486                         mptcp_do_fallback(sk);
 487                         pr_fallback(msk);
 488                         goto fallback;
 489                 }
 490
 491                 if (mp_opt.suboptions & OPTION_MPTCP_CSUMREQD)
 492                         WRITE_ONCE(msk->csum_enabled, true);
 493                 if (mp_opt.deny_join_id0)
 494                         WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
 495                 subflow->mp_capable = 1;
 496                 subflow_set_remote_key(msk, subflow, &mp_opt);
 497                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
 498                 mptcp_finish_connect(sk);
 499                 mptcp_set_connected(parent);
 500         } else if (subflow->request_join) {
 501                 u8 hmac[SHA256_DIGEST_SIZE];
 502
 503                 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ)) {
 504                         subflow->reset_reason = MPTCP_RST_EMPTCP;
 505                         goto do_reset;
 506                 }
 507
 508                 subflow->backup = mp_opt.backup;
 509                 subflow->thmac = mp_opt.thmac;
 510                 subflow->remote_nonce = mp_opt.nonce;
 511                 subflow->remote_id = mp_opt.join_id;
 512                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u backup=%d",
 513                          subflow, subflow->thmac, subflow->remote_nonce,
 514                          subflow->backup);
 515
 516                 if (!subflow_thmac_valid(subflow)) {
 517                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 518                         subflow->reset_reason = MPTCP_RST_EMPTCP;
 519                         goto do_reset;
 520                 }
 521
 522                 if (!mptcp_finish_join(sk))
 523                         goto do_reset;
 524
 525                 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 526                                       subflow->local_nonce,
 527                                       subflow->remote_nonce,
 528                                       hmac);
 529                 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
 530
 531                 subflow->mp_join = 1;
 532                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 533
 534                 if (subflow_use_different_dport(msk, sk)) {
 535                         pr_debug("synack inet_dport=%d %d",
 536                                  ntohs(inet_sk(sk)->inet_dport),
 537                                  ntohs(inet_sk(parent)->inet_dport));
 538                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
 539                 }
 540         } else if (mptcp_check_fallback(sk)) {
 541 fallback:
 542                 mptcp_rcv_space_init(msk, sk);
 543                 mptcp_set_connected(parent);
 544         }
 545         return;
 546
 547 do_reset:
 548         subflow->reset_transient = 0;
 549         mptcp_subflow_reset(sk);
 550 }
 551
 552 static void subflow_set_local_id(struct mptcp_subflow_context *subflow, int local_id)
 553 {
 554         subflow->local_id = local_id;
 555         subflow->local_id_valid = 1;
 556 }
 557
 558 static int subflow_chk_local_id(struct sock *sk)
 559 {
 560         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 561         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 562         int err;
 563
 564         if (likely(subflow->local_id_valid))
 565                 return 0;
 566
 567         err = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
 568         if (err < 0)
 569                 return err;
 570
 571         subflow_set_local_id(subflow, err);
 572         return 0;
 573 }
 574
 575 static int subflow_rebuild_header(struct sock *sk)
 576 {
 577         int err = subflow_chk_local_id(sk);
 578
 579         if (unlikely(err < 0))
 580                 return err;
 581
 582         return inet_sk_rebuild_header(sk);
 583 }
 584
 585 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 586 static int subflow_v6_rebuild_header(struct sock *sk)
 587 {
 588         int err = subflow_chk_local_id(sk);
 589
 590         if (unlikely(err < 0))
 591                 return err;
 592
 593         return inet6_sk_rebuild_header(sk);
 594 }
 595 #endif
 596
 597 static struct request_sock_ops mptcp_subflow_v4_request_sock_ops __ro_after_init;
 598 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops __ro_after_init;
 599
 600 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 601 {
 602         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 603
 604         pr_debug("subflow=%p", subflow);
 605
 606         /* Never answer to SYNs sent to broadcast or multicast */
 607         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 608                 goto drop;
 609
 610         return tcp_conn_request(&mptcp_subflow_v4_request_sock_ops,
 611                                 &subflow_request_sock_ipv4_ops,
 612                                 sk, skb);
 613 drop:
 614         tcp_listendrop(sk);
 615         return 0;
 616 }
 617
 618 static void subflow_v4_req_destructor(struct request_sock *req)
 619 {
 620         subflow_req_destructor(req);
 621         tcp_request_sock_ops.destructor(req);
 622 }
 623
 624 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 625 static struct request_sock_ops mptcp_subflow_v6_request_sock_ops __ro_after_init;
 626 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops __ro_after_init;
 627 static struct inet_connection_sock_af_ops subflow_v6_specific __ro_after_init;
 628 static struct inet_connection_sock_af_ops subflow_v6m_specific __ro_after_init;
 629 static struct proto tcpv6_prot_override __ro_after_init;
 630
 631 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 632 {
 633         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 634
 635         pr_debug("subflow=%p", subflow);
 636
 637         if (skb->protocol == htons(ETH_P_IP))
 638                 return subflow_v4_conn_request(sk, skb);
 639
 640         if (!ipv6_unicast_destination(skb))
 641                 goto drop;
 642
 643         if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
 644                 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
 645                 return 0;
 646         }
 647
 648         return tcp_conn_request(&mptcp_subflow_v6_request_sock_ops,
 649                                 &subflow_request_sock_ipv6_ops, sk, skb);
 650
 651 drop:
 652         tcp_listendrop(sk);
 653         return 0; /* don't send reset */
 654 }
 655
 656 static void subflow_v6_req_destructor(struct request_sock *req)
 657 {
 658         subflow_req_destructor(req);
 659         tcp6_request_sock_ops.destructor(req);
 660 }
 661 #endif
 662
 663 struct request_sock *mptcp_subflow_reqsk_alloc(const struct request_sock_ops *ops,
 664                                                struct sock *sk_listener,
 665                                                bool attach_listener)
 666 {
 667         if (ops->family == AF_INET)
 668                 ops = &mptcp_subflow_v4_request_sock_ops;
 669 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 670         else if (ops->family == AF_INET6)
 671                 ops = &mptcp_subflow_v6_request_sock_ops;
 672 #endif
 673
 674         return inet_reqsk_alloc(ops, sk_listener, attach_listener);
 675 }
 676 EXPORT_SYMBOL(mptcp_subflow_reqsk_alloc);
 677
 678 /* validate hmac received in third ACK */
 679 static bool subflow_hmac_valid(const struct request_sock *req,
 680                                const struct mptcp_options_received *mp_opt)
 681 {
 682         const struct mptcp_subflow_request_sock *subflow_req;
 683         u8 hmac[SHA256_DIGEST_SIZE];
 684         struct mptcp_sock *msk;
 685
 686         subflow_req = mptcp_subflow_rsk(req);
 687         msk = subflow_req->msk;
 688         if (!msk)
 689                 return false;
 690
 691         subflow_generate_hmac(msk->remote_key, msk->local_key,
 692                               subflow_req->remote_nonce,
 693                               subflow_req->local_nonce, hmac);
 694
 695         return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
 696 }
 697
 698 static void subflow_ulp_fallback(struct sock *sk,
 699                                  struct mptcp_subflow_context *old_ctx)
 700 {
 701         struct inet_connection_sock *icsk = inet_csk(sk);
 702
 703         mptcp_subflow_tcp_fallback(sk, old_ctx);
 704         icsk->icsk_ulp_ops = NULL;
 705         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 706         tcp_sk(sk)->is_mptcp = 0;
 707
 708         mptcp_subflow_ops_undo_override(sk);
 709 }
 710
 711 void mptcp_subflow_drop_ctx(struct sock *ssk)
 712 {
 713         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 714
 715         if (!ctx)
 716                 return;
 717
 718         list_del(&mptcp_subflow_ctx(ssk)->node);
 719         if (inet_csk(ssk)->icsk_ulp_ops) {
 720                 subflow_ulp_fallback(ssk, ctx);
 721                 if (ctx->conn)
 722                         sock_put(ctx->conn);
 723         }
 724
 725         kfree_rcu(ctx, rcu);
 726 }
 727
 728 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 729                                      const struct mptcp_options_received *mp_opt)
 730 {
 731         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 732
 733         subflow_set_remote_key(msk, subflow, mp_opt);
 734         subflow->fully_established = 1;
 735         WRITE_ONCE(msk->fully_established, true);
 736
 737         if (subflow->is_mptfo)
 738                 mptcp_fastopen_gen_msk_ackseq(msk, subflow, mp_opt);
 739 }
 740
 741 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 742                                           struct sk_buff *skb,
 743                                           struct request_sock *req,
 744                                           struct dst_entry *dst,
 745                                           struct request_sock *req_unhash,
 746                                           bool *own_req)
 747 {
 748         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 749         struct mptcp_subflow_request_sock *subflow_req;
 750         struct mptcp_options_received mp_opt;
 751         bool fallback, fallback_is_fatal;
 752         struct mptcp_sock *owner;
 753         struct sock *child;
 754
 755         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 756
 757         /* After child creation we must look for MPC even when options
 758          * are not parsed
 759          */
 760         mp_opt.suboptions = 0;
 761
 762         /* hopefully temporary handling for MP_JOIN+syncookie */
 763         subflow_req = mptcp_subflow_rsk(req);
 764         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 765         fallback = !tcp_rsk(req)->is_mptcp;
 766         if (fallback)
 767                 goto create_child;
 768
 769         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 770         if (subflow_req->mp_capable) {
 771                 /* we can receive and accept an in-window, out-of-order pkt,
 772                  * which may not carry the MP_CAPABLE opt even on mptcp enabled
 773                  * paths: always try to extract the peer key, and fallback
 774                  * for packets missing it.
 775                  * Even OoO DSS packets coming legitly after dropped or
 776                  * reordered MPC will cause fallback, but we don't have other
 777                  * options.
 778                  */
 779                 mptcp_get_options(skb, &mp_opt);
 780                 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPC))
 781                         fallback = true;
 782
 783         } else if (subflow_req->mp_join) {
 784                 mptcp_get_options(skb, &mp_opt);
 785                 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ) ||
 786                     !subflow_hmac_valid(req, &mp_opt) ||
 787                     !mptcp_can_accept_new_subflow(subflow_req->msk)) {
 788                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 789                         fallback = true;
 790                 }
 791         }
 792
 793 create_child:
 794         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 795                                                      req_unhash, own_req);
 796
 797         if (child && *own_req) {
 798                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 799
 800                 tcp_rsk(req)->drop_req = false;
 801
 802                 /* we need to fallback on ctx allocation failure and on pre-reqs
 803                  * checking above. In the latter scenario we additionally need
 804                  * to reset the context to non MPTCP status.
 805                  */
 806                 if (!ctx || fallback) {
 807                         if (fallback_is_fatal) {
 808                                 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 809                                 goto dispose_child;
 810                         }
 811                         goto fallback;
 812                 }
 813
 814                 /* ssk inherits options of listener sk */
 815                 ctx->setsockopt_seq = listener->setsockopt_seq;
 816
 817                 if (ctx->mp_capable) {
 818                         ctx->conn = mptcp_sk_clone_init(listener->conn, &mp_opt, child, req);
 819                         if (!ctx->conn)
 820                                 goto fallback;
 821
 822                         owner = mptcp_sk(ctx->conn);
 823                         mptcp_pm_new_connection(owner, child, 1);
 824
 825                         /* with OoO packets we can reach here without ingress
 826                          * mpc option
 827                          */
 828                         if (mp_opt.suboptions & OPTION_MPTCP_MPC_ACK) {
 829                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 830                                 mptcp_pm_fully_established(owner, child);
 831                                 ctx->pm_notified = 1;
 832                         }
 833                 } else if (ctx->mp_join) {
 834                         owner = subflow_req->msk;
 835                         if (!owner) {
 836                                 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 837                                 goto dispose_child;
 838                         }
 839
 840                         /* move the msk reference ownership to the subflow */
 841                         subflow_req->msk = NULL;
 842                         ctx->conn = (struct sock *)owner;
 843
 844                         if (subflow_use_different_sport(owner, sk)) {
 845                                 pr_debug("ack inet_sport=%d %d",
 846                                          ntohs(inet_sk(sk)->inet_sport),
 847                                          ntohs(inet_sk((struct sock *)owner)->inet_sport));
 848                                 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
 849                                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
 850                                         goto dispose_child;
 851                                 }
 852                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
 853                         }
 854
 855                         if (!mptcp_finish_join(child))
 856                                 goto dispose_child;
 857
 858                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 859                         tcp_rsk(req)->drop_req = true;
 860                 }
 861         }
 862
 863         /* check for expected invariant - should never trigger, just help
 864          * catching eariler subtle bugs
 865          */
 866         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 867                      (!mptcp_subflow_ctx(child) ||
 868                       !mptcp_subflow_ctx(child)->conn));
 869         return child;
 870
 871 dispose_child:
 872         mptcp_subflow_drop_ctx(child);
 873         tcp_rsk(req)->drop_req = true;
 874         inet_csk_prepare_for_destroy_sock(child);
 875         tcp_done(child);
 876         req->rsk_ops->send_reset(sk, skb);
 877
 878         /* The last child reference will be released by the caller */
 879         return child;
 880
 881 fallback:
 882         mptcp_subflow_drop_ctx(child);
 883         return child;
 884 }
 885
 886 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
 887 static struct proto tcp_prot_override __ro_after_init;
 888
 889 enum mapping_status {
 890         MAPPING_OK,
 891         MAPPING_INVALID,
 892         MAPPING_EMPTY,
 893         MAPPING_DATA_FIN,
 894         MAPPING_DUMMY,
 895         MAPPING_BAD_CSUM
 896 };
 897
 898 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 899 {
 900         pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 901                  ssn, subflow->map_subflow_seq, subflow->map_data_len);
 902 }
 903
 904 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 905 {
 906         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 907         unsigned int skb_consumed;
 908
 909         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 910         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 911                 return true;
 912
 913         return skb->len - skb_consumed <= subflow->map_data_len -
 914                                           mptcp_subflow_get_map_offset(subflow);
 915 }
 916
 917 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 918 {
 919         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 920         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 921
 922         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 923                 /* Mapping covers data later in the subflow stream,
 924                  * currently unsupported.
 925                  */
 926                 dbg_bad_map(subflow, ssn);
 927                 return false;
 928         }
 929         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 930                                   subflow->map_data_len))) {
 931                 /* Mapping does covers past subflow data, invalid */
 932                 dbg_bad_map(subflow, ssn);
 933                 return false;
 934         }
 935         return true;
 936 }
 937
 938 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
 939                                               bool csum_reqd)
 940 {
 941         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 942         u32 offset, seq, delta;
 943         __sum16 csum;
 944         int len;
 945
 946         if (!csum_reqd)
 947                 return MAPPING_OK;
 948
 949         /* mapping already validated on previous traversal */
 950         if (subflow->map_csum_len == subflow->map_data_len)
 951                 return MAPPING_OK;
 952
 953         /* traverse the receive queue, ensuring it contains a full
 954          * DSS mapping and accumulating the related csum.
 955          * Preserve the accoumlate csum across multiple calls, to compute
 956          * the csum only once
 957          */
 958         delta = subflow->map_data_len - subflow->map_csum_len;
 959         for (;;) {
 960                 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
 961                 offset = seq - TCP_SKB_CB(skb)->seq;
 962
 963                 /* if the current skb has not been accounted yet, csum its contents
 964                  * up to the amount covered by the current DSS
 965                  */
 966                 if (offset < skb->len) {
 967                         __wsum csum;
 968
 969                         len = min(skb->len - offset, delta);
 970                         csum = skb_checksum(skb, offset, len, 0);
 971                         subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
 972                                                                 subflow->map_csum_len);
 973
 974                         delta -= len;
 975                         subflow->map_csum_len += len;
 976                 }
 977                 if (delta == 0)
 978                         break;
 979
 980                 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
 981                         /* if this subflow is closed, the partial mapping
 982                          * will be never completed; flush the pending skbs, so
 983                          * that subflow_sched_work_if_closed() can kick in
 984                          */
 985                         if (unlikely(ssk->sk_state == TCP_CLOSE))
 986                                 while ((skb = skb_peek(&ssk->sk_receive_queue)))
 987                                         sk_eat_skb(ssk, skb);
 988
 989                         /* not enough data to validate the csum */
 990                         return MAPPING_EMPTY;
 991                 }
 992
 993                 /* the DSS mapping for next skbs will be validated later,
 994                  * when a get_mapping_status call will process such skb
 995                  */
 996                 skb = skb->next;
 997         }
 998
 999         /* note that 'map_data_len' accounts only for the carried data, does
1000          * not include the eventual seq increment due to the data fin,
1001          * while the pseudo header requires the original DSS data len,
1002          * including that
1003          */
1004         csum = __mptcp_make_csum(subflow->map_seq,
1005                                  subflow->map_subflow_seq,
1006                                  subflow->map_data_len + subflow->map_data_fin,
1007                                  subflow->map_data_csum);
1008         if (unlikely(csum)) {
1009                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
1010                 return MAPPING_BAD_CSUM;
1011         }
1012
1013         subflow->valid_csum_seen = 1;
1014         return MAPPING_OK;
1015 }
1016
1017 static enum mapping_status get_mapping_status(struct sock *ssk,
1018                                               struct mptcp_sock *msk)
1019 {
1020         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1021         bool csum_reqd = READ_ONCE(msk->csum_enabled);
1022         struct mptcp_ext *mpext;
1023         struct sk_buff *skb;
1024         u16 data_len;
1025         u64 map_seq;
1026
1027         skb = skb_peek(&ssk->sk_receive_queue);
1028         if (!skb)
1029                 return MAPPING_EMPTY;
1030
1031         if (mptcp_check_fallback(ssk))
1032                 return MAPPING_DUMMY;
1033
1034         mpext = mptcp_get_ext(skb);
1035         if (!mpext || !mpext->use_map) {
1036                 if (!subflow->map_valid && !skb->len) {
1037                         /* the TCP stack deliver 0 len FIN pkt to the receive
1038                          * queue, that is the only 0len pkts ever expected here,
1039                          * and we can admit no mapping only for 0 len pkts
1040                          */
1041                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
1042                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
1043                                           TCP_SKB_CB(skb)->seq,
1044                                           TCP_SKB_CB(skb)->end_seq,
1045                                           TCP_SKB_CB(skb)->tcp_flags);
1046                         sk_eat_skb(ssk, skb);
1047                         return MAPPING_EMPTY;
1048                 }
1049
1050                 if (!subflow->map_valid)
1051                         return MAPPING_INVALID;
1052
1053                 goto validate_seq;
1054         }
1055
1056         trace_get_mapping_status(mpext);
1057
1058         data_len = mpext->data_len;
1059         if (data_len == 0) {
1060                 pr_debug("infinite mapping received");
1061                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1062                 subflow->map_data_len = 0;
1063                 return MAPPING_INVALID;
1064         }
1065
1066         if (mpext->data_fin == 1) {
1067                 if (data_len == 1) {
1068                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1069                                                                  mpext->dsn64);
1070                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
1071                         if (subflow->map_valid) {
1072                                 /* A DATA_FIN might arrive in a DSS
1073                                  * option before the previous mapping
1074                                  * has been fully consumed. Continue
1075                                  * handling the existing mapping.
1076                                  */
1077                                 skb_ext_del(skb, SKB_EXT_MPTCP);
1078                                 return MAPPING_OK;
1079                         } else {
1080                                 if (updated)
1081                                         mptcp_schedule_work((struct sock *)msk);
1082
1083                                 return MAPPING_DATA_FIN;
1084                         }
1085                 } else {
1086                         u64 data_fin_seq = mpext->data_seq + data_len - 1;
1087
1088                         /* If mpext->data_seq is a 32-bit value, data_fin_seq
1089                          * must also be limited to 32 bits.
1090                          */
1091                         if (!mpext->dsn64)
1092                                 data_fin_seq &= GENMASK_ULL(31, 0);
1093
1094                         mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1095                         pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
1096                                  data_fin_seq, mpext->dsn64);
1097                 }
1098
1099                 /* Adjust for DATA_FIN using 1 byte of sequence space */
1100                 data_len--;
1101         }
1102
1103         map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1104         WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1105
1106         if (subflow->map_valid) {
1107                 /* Allow replacing only with an identical map */
1108                 if (subflow->map_seq == map_seq &&
1109                     subflow->map_subflow_seq == mpext->subflow_seq &&
1110                     subflow->map_data_len == data_len &&
1111                     subflow->map_csum_reqd == mpext->csum_reqd) {
1112                         skb_ext_del(skb, SKB_EXT_MPTCP);
1113                         goto validate_csum;
1114                 }
1115
1116                 /* If this skb data are fully covered by the current mapping,
1117                  * the new map would need caching, which is not supported
1118                  */
1119                 if (skb_is_fully_mapped(ssk, skb)) {
1120                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1121                         return MAPPING_INVALID;
1122                 }
1123
1124                 /* will validate the next map after consuming the current one */
1125                 goto validate_csum;
1126         }
1127
1128         subflow->map_seq = map_seq;
1129         subflow->map_subflow_seq = mpext->subflow_seq;
1130         subflow->map_data_len = data_len;
1131         subflow->map_valid = 1;
1132         subflow->map_data_fin = mpext->data_fin;
1133         subflow->mpc_map = mpext->mpc_map;
1134         subflow->map_csum_reqd = mpext->csum_reqd;
1135         subflow->map_csum_len = 0;
1136         subflow->map_data_csum = csum_unfold(mpext->csum);
1137
1138         /* Cfr RFC 8684 Section 3.3.0 */
1139         if (unlikely(subflow->map_csum_reqd != csum_reqd))
1140                 return MAPPING_INVALID;
1141
1142         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
1143                  subflow->map_seq, subflow->map_subflow_seq,
1144                  subflow->map_data_len, subflow->map_csum_reqd,
1145                  subflow->map_data_csum);
1146
1147 validate_seq:
1148         /* we revalidate valid mapping on new skb, because we must ensure
1149          * the current skb is completely covered by the available mapping
1150          */
1151         if (!validate_mapping(ssk, skb)) {
1152                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1153                 return MAPPING_INVALID;
1154         }
1155
1156         skb_ext_del(skb, SKB_EXT_MPTCP);
1157
1158 validate_csum:
1159         return validate_data_csum(ssk, skb, csum_reqd);
1160 }
1161
1162 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1163                                        u64 limit)
1164 {
1165         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1166         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1167         u32 incr;
1168
1169         incr = limit >= skb->len ? skb->len + fin : limit;
1170
1171         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
1172                  subflow->map_subflow_seq);
1173         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1174         tcp_sk(ssk)->copied_seq += incr;
1175         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1176                 sk_eat_skb(ssk, skb);
1177         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1178                 subflow->map_valid = 0;
1179 }
1180
1181 /* sched mptcp worker to remove the subflow if no more data is pending */
1182 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1183 {
1184         if (likely(ssk->sk_state != TCP_CLOSE))
1185                 return;
1186
1187         if (skb_queue_empty(&ssk->sk_receive_queue) &&
1188             !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
1189                 mptcp_schedule_work((struct sock *)msk);
1190 }
1191
1192 static bool subflow_can_fallback(struct mptcp_subflow_context *subflow)
1193 {
1194         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1195
1196         if (subflow->mp_join)
1197                 return false;
1198         else if (READ_ONCE(msk->csum_enabled))
1199                 return !subflow->valid_csum_seen;
1200         else
1201                 return !subflow->fully_established;
1202 }
1203
1204 static void mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk)
1205 {
1206         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1207         unsigned long fail_tout;
1208
1209         /* greceful failure can happen only on the MPC subflow */
1210         if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first)))
1211                 return;
1212
1213         /* since the close timeout take precedence on the fail one,
1214          * no need to start the latter when the first is already set
1215          */
1216         if (sock_flag((struct sock *)msk, SOCK_DEAD))
1217                 return;
1218
1219         /* we don't need extreme accuracy here, use a zero fail_tout as special
1220          * value meaning no fail timeout at all;
1221          */
1222         fail_tout = jiffies + TCP_RTO_MAX;
1223         if (!fail_tout)
1224                 fail_tout = 1;
1225         WRITE_ONCE(subflow->fail_tout, fail_tout);
1226         tcp_send_ack(ssk);
1227
1228         mptcp_reset_timeout(msk, subflow->fail_tout);
1229 }
1230
1231 static bool subflow_check_data_avail(struct sock *ssk)
1232 {
1233         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1234         enum mapping_status status;
1235         struct mptcp_sock *msk;
1236         struct sk_buff *skb;
1237
1238         if (!skb_peek(&ssk->sk_receive_queue))
1239                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1240         if (subflow->data_avail)
1241                 return true;
1242
1243         msk = mptcp_sk(subflow->conn);
1244         for (;;) {
1245                 u64 ack_seq;
1246                 u64 old_ack;
1247
1248                 status = get_mapping_status(ssk, msk);
1249                 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1250                 if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY ||
1251                              status == MAPPING_BAD_CSUM))
1252                         goto fallback;
1253
1254                 if (status != MAPPING_OK)
1255                         goto no_data;
1256
1257                 skb = skb_peek(&ssk->sk_receive_queue);
1258                 if (WARN_ON_ONCE(!skb))
1259                         goto no_data;
1260
1261                 if (unlikely(!READ_ONCE(msk->can_ack)))
1262                         goto fallback;
1263
1264                 old_ack = READ_ONCE(msk->ack_seq);
1265                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1266                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1267                          ack_seq);
1268                 if (unlikely(before64(ack_seq, old_ack))) {
1269                         mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1270                         continue;
1271                 }
1272
1273                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1274                 break;
1275         }
1276         return true;
1277
1278 no_data:
1279         subflow_sched_work_if_closed(msk, ssk);
1280         return false;
1281
1282 fallback:
1283         if (!__mptcp_check_fallback(msk)) {
1284                 /* RFC 8684 section 3.7. */
1285                 if (status == MAPPING_BAD_CSUM &&
1286                     (subflow->mp_join || subflow->valid_csum_seen)) {
1287                         subflow->send_mp_fail = 1;
1288
1289                         if (!READ_ONCE(msk->allow_infinite_fallback)) {
1290                                 subflow->reset_transient = 0;
1291                                 subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1292                                 goto reset;
1293                         }
1294                         mptcp_subflow_fail(msk, ssk);
1295                         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1296                         return true;
1297                 }
1298
1299                 if (!subflow_can_fallback(subflow) && subflow->map_data_len) {
1300                         /* fatal protocol error, close the socket.
1301                          * subflow_error_report() will introduce the appropriate barriers
1302                          */
1303                         subflow->reset_transient = 0;
1304                         subflow->reset_reason = MPTCP_RST_EMPTCP;
1305
1306 reset:
1307                         WRITE_ONCE(ssk->sk_err, EBADMSG);
1308                         tcp_set_state(ssk, TCP_CLOSE);
1309                         while ((skb = skb_peek(&ssk->sk_receive_queue)))
1310                                 sk_eat_skb(ssk, skb);
1311                         tcp_send_active_reset(ssk, GFP_ATOMIC);
1312                         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1313                         return false;
1314                 }
1315
1316                 mptcp_do_fallback(ssk);
1317         }
1318
1319         skb = skb_peek(&ssk->sk_receive_queue);
1320         subflow->map_valid = 1;
1321         subflow->map_seq = READ_ONCE(msk->ack_seq);
1322         subflow->map_data_len = skb->len;
1323         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1324         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1325         return true;
1326 }
1327
1328 bool mptcp_subflow_data_available(struct sock *sk)
1329 {
1330         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1331
1332         /* check if current mapping is still valid */
1333         if (subflow->map_valid &&
1334             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1335                 subflow->map_valid = 0;
1336                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1337
1338                 pr_debug("Done with mapping: seq=%u data_len=%u",
1339                          subflow->map_subflow_seq,
1340                          subflow->map_data_len);
1341         }
1342
1343         return subflow_check_data_avail(sk);
1344 }
1345
1346 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1347  * not the ssk one.
1348  *
1349  * In mptcp, rwin is about the mptcp-level connection data.
1350  *
1351  * Data that is still on the ssk rx queue can thus be ignored,
1352  * as far as mptcp peer is concerned that data is still inflight.
1353  * DSS ACK is updated when skb is moved to the mptcp rx queue.
1354  */
1355 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1356 {
1357         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1358         const struct sock *sk = subflow->conn;
1359
1360         *space = __mptcp_space(sk);
1361         *full_space = tcp_full_space(sk);
1362 }
1363
1364 void __mptcp_error_report(struct sock *sk)
1365 {
1366         struct mptcp_subflow_context *subflow;
1367         struct mptcp_sock *msk = mptcp_sk(sk);
1368
1369         mptcp_for_each_subflow(msk, subflow) {
1370                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1371                 int err = sock_error(ssk);
1372                 int ssk_state;
1373
1374                 if (!err)
1375                         continue;
1376
1377                 /* only propagate errors on fallen-back sockets or
1378                  * on MPC connect
1379                  */
1380                 if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(msk))
1381                         continue;
1382
1383                 /* We need to propagate only transition to CLOSE state.
1384                  * Orphaned socket will see such state change via
1385                  * subflow_sched_work_if_closed() and that path will properly
1386                  * destroy the msk as needed.
1387                  */
1388                 ssk_state = inet_sk_state_load(ssk);
1389                 if (ssk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DEAD))
1390                         inet_sk_state_store(sk, ssk_state);
1391                 WRITE_ONCE(sk->sk_err, -err);
1392
1393                 /* This barrier is coupled with smp_rmb() in mptcp_poll() */
1394                 smp_wmb();
1395                 sk_error_report(sk);
1396                 break;
1397         }
1398 }
1399
1400 static void subflow_error_report(struct sock *ssk)
1401 {
1402         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1403
1404         /* bail early if this is a no-op, so that we avoid introducing a
1405          * problematic lockdep dependency between TCP accept queue lock
1406          * and msk socket spinlock
1407          */
1408         if (!sk->sk_socket)
1409                 return;
1410
1411         mptcp_data_lock(sk);
1412         if (!sock_owned_by_user(sk))
1413                 __mptcp_error_report(sk);
1414         else
1415                 __set_bit(MPTCP_ERROR_REPORT,  &mptcp_sk(sk)->cb_flags);
1416         mptcp_data_unlock(sk);
1417 }
1418
1419 static void subflow_data_ready(struct sock *sk)
1420 {
1421         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1422         u16 state = 1 << inet_sk_state_load(sk);
1423         struct sock *parent = subflow->conn;
1424         struct mptcp_sock *msk;
1425
1426         trace_sk_data_ready(sk);
1427
1428         msk = mptcp_sk(parent);
1429         if (state & TCPF_LISTEN) {
1430                 /* MPJ subflow are removed from accept queue before reaching here,
1431                  * avoid stray wakeups
1432                  */
1433                 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1434                         return;
1435
1436                 parent->sk_data_ready(parent);
1437                 return;
1438         }
1439
1440         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1441                      !subflow->mp_join && !(state & TCPF_CLOSE));
1442
1443         if (mptcp_subflow_data_available(sk))
1444                 mptcp_data_ready(parent, sk);
1445         else if (unlikely(sk->sk_err))
1446                 subflow_error_report(sk);
1447 }
1448
1449 static void subflow_write_space(struct sock *ssk)
1450 {
1451         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1452
1453         mptcp_propagate_sndbuf(sk, ssk);
1454         mptcp_write_space(sk);
1455 }
1456
1457 static const struct inet_connection_sock_af_ops *
1458 subflow_default_af_ops(struct sock *sk)
1459 {
1460 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1461         if (sk->sk_family == AF_INET6)
1462                 return &subflow_v6_specific;
1463 #endif
1464         return &subflow_specific;
1465 }
1466
1467 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1468 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1469 {
1470         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1471         struct inet_connection_sock *icsk = inet_csk(sk);
1472         const struct inet_connection_sock_af_ops *target;
1473
1474         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1475
1476         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1477                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1478
1479         if (likely(icsk->icsk_af_ops == target))
1480                 return;
1481
1482         subflow->icsk_af_ops = icsk->icsk_af_ops;
1483         icsk->icsk_af_ops = target;
1484 }
1485 #endif
1486
1487 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1488                          struct sockaddr_storage *addr,
1489                          unsigned short family)
1490 {
1491         memset(addr, 0, sizeof(*addr));
1492         addr->ss_family = family;
1493         if (addr->ss_family == AF_INET) {
1494                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1495
1496                 if (info->family == AF_INET)
1497                         in_addr->sin_addr = info->addr;
1498 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1499                 else if (ipv6_addr_v4mapped(&info->addr6))
1500                         in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1501 #endif
1502                 in_addr->sin_port = info->port;
1503         }
1504 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1505         else if (addr->ss_family == AF_INET6) {
1506                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1507
1508                 if (info->family == AF_INET)
1509                         ipv6_addr_set_v4mapped(info->addr.s_addr,
1510                                                &in6_addr->sin6_addr);
1511                 else
1512                         in6_addr->sin6_addr = info->addr6;
1513                 in6_addr->sin6_port = info->port;
1514         }
1515 #endif
1516 }
1517
1518 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1519                             const struct mptcp_addr_info *remote)
1520 {
1521         struct mptcp_sock *msk = mptcp_sk(sk);
1522         struct mptcp_subflow_context *subflow;
1523         struct sockaddr_storage addr;
1524         int remote_id = remote->id;
1525         int local_id = loc->id;
1526         int err = -ENOTCONN;
1527         struct socket *sf;
1528         struct sock *ssk;
1529         u32 remote_token;
1530         int addrlen;
1531         int ifindex;
1532         u8 flags;
1533
1534         if (!mptcp_is_fully_established(sk))
1535                 goto err_out;
1536
1537         err = mptcp_subflow_create_socket(sk, loc->family, &sf);
1538         if (err)
1539                 goto err_out;
1540
1541         ssk = sf->sk;
1542         subflow = mptcp_subflow_ctx(ssk);
1543         do {
1544                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1545         } while (!subflow->local_nonce);
1546
1547         if (local_id)
1548                 subflow_set_local_id(subflow, local_id);
1549
1550         mptcp_pm_get_flags_and_ifindex_by_id(msk, local_id,
1551                                              &flags, &ifindex);
1552         subflow->remote_key_valid = 1;
1553         subflow->remote_key = msk->remote_key;
1554         subflow->local_key = msk->local_key;
1555         subflow->token = msk->token;
1556         mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1557
1558         addrlen = sizeof(struct sockaddr_in);
1559 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1560         if (addr.ss_family == AF_INET6)
1561                 addrlen = sizeof(struct sockaddr_in6);
1562 #endif
1563         mptcp_sockopt_sync(msk, ssk);
1564
1565         ssk->sk_bound_dev_if = ifindex;
1566         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1567         if (err)
1568                 goto failed;
1569
1570         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1571         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1572                  remote_token, local_id, remote_id);
1573         subflow->remote_token = remote_token;
1574         subflow->remote_id = remote_id;
1575         subflow->request_join = 1;
1576         subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1577         mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1578
1579         sock_hold(ssk);
1580         list_add_tail(&subflow->node, &msk->conn_list);
1581         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1582         if (err && err != -EINPROGRESS)
1583                 goto failed_unlink;
1584
1585         /* discard the subflow socket */
1586         mptcp_sock_graft(ssk, sk->sk_socket);
1587         iput(SOCK_INODE(sf));
1588         WRITE_ONCE(msk->allow_infinite_fallback, false);
1589         return 0;
1590
1591 failed_unlink:
1592         list_del(&subflow->node);
1593         sock_put(mptcp_subflow_tcp_sock(subflow));
1594
1595 failed:
1596         subflow->disposable = 1;
1597         sock_release(sf);
1598
1599 err_out:
1600         /* we account subflows before the creation, and this failures will not
1601          * be caught by sk_state_change()
1602          */
1603         mptcp_pm_close_subflow(msk);
1604         return err;
1605 }
1606
1607 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1608 {
1609 #ifdef CONFIG_SOCK_CGROUP_DATA
1610         struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1611                                 *child_skcd = &child->sk_cgrp_data;
1612
1613         /* only the additional subflows created by kworkers have to be modified */
1614         if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1615             cgroup_id(sock_cgroup_ptr(child_skcd))) {
1616 #ifdef CONFIG_MEMCG
1617                 struct mem_cgroup *memcg = parent->sk_memcg;
1618
1619                 mem_cgroup_sk_free(child);
1620                 if (memcg && css_tryget(&memcg->css))
1621                         child->sk_memcg = memcg;
1622 #endif /* CONFIG_MEMCG */
1623
1624                 cgroup_sk_free(child_skcd);
1625                 *child_skcd = *parent_skcd;
1626                 cgroup_sk_clone(child_skcd);
1627         }
1628 #endif /* CONFIG_SOCK_CGROUP_DATA */
1629 }
1630
1631 static void mptcp_subflow_ops_override(struct sock *ssk)
1632 {
1633 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1634         if (ssk->sk_prot == &tcpv6_prot)
1635                 ssk->sk_prot = &tcpv6_prot_override;
1636         else
1637 #endif
1638                 ssk->sk_prot = &tcp_prot_override;
1639 }
1640
1641 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1642 {
1643 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1644         if (ssk->sk_prot == &tcpv6_prot_override)
1645                 ssk->sk_prot = &tcpv6_prot;
1646         else
1647 #endif
1648                 ssk->sk_prot = &tcp_prot;
1649 }
1650
1651 int mptcp_subflow_create_socket(struct sock *sk, unsigned short family,
1652                                 struct socket **new_sock)
1653 {
1654         struct mptcp_subflow_context *subflow;
1655         struct net *net = sock_net(sk);
1656         struct socket *sf;
1657         int err;
1658
1659         /* un-accepted server sockets can reach here - on bad configuration
1660          * bail early to avoid greater trouble later
1661          */
1662         if (unlikely(!sk->sk_socket))
1663                 return -EINVAL;
1664
1665         err = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, &sf);
1666         if (err)
1667                 return err;
1668
1669         lock_sock_nested(sf->sk, SINGLE_DEPTH_NESTING);
1670
1671         err = security_mptcp_add_subflow(sk, sf->sk);
1672         if (err)
1673                 goto release_ssk;
1674
1675         /* the newly created socket has to be in the same cgroup as its parent */
1676         mptcp_attach_cgroup(sk, sf->sk);
1677
1678         /* kernel sockets do not by default acquire net ref, but TCP timer
1679          * needs it.
1680          * Update ns_tracker to current stack trace and refcounted tracker.
1681          */
1682         __netns_tracker_free(net, &sf->sk->ns_tracker, false);
1683         sf->sk->sk_net_refcnt = 1;
1684         get_net_track(net, &sf->sk->ns_tracker, GFP_KERNEL);
1685         sock_inuse_add(net, 1);
1686         err = tcp_set_ulp(sf->sk, "mptcp");
1687
1688 release_ssk:
1689         release_sock(sf->sk);
1690
1691         if (err) {
1692                 sock_release(sf);
1693                 return err;
1694         }
1695
1696         /* the newly created socket really belongs to the owning MPTCP master
1697          * socket, even if for additional subflows the allocation is performed
1698          * by a kernel workqueue. Adjust inode references, so that the
1699          * procfs/diag interfaces really show this one belonging to the correct
1700          * user.
1701          */
1702         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1703         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1704         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1705
1706         subflow = mptcp_subflow_ctx(sf->sk);
1707         pr_debug("subflow=%p", subflow);
1708
1709         *new_sock = sf;
1710         sock_hold(sk);
1711         subflow->conn = sk;
1712         mptcp_subflow_ops_override(sf->sk);
1713
1714         return 0;
1715 }
1716
1717 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1718                                                         gfp_t priority)
1719 {
1720         struct inet_connection_sock *icsk = inet_csk(sk);
1721         struct mptcp_subflow_context *ctx;
1722
1723         ctx = kzalloc(sizeof(*ctx), priority);
1724         if (!ctx)
1725                 return NULL;
1726
1727         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1728         INIT_LIST_HEAD(&ctx->node);
1729         INIT_LIST_HEAD(&ctx->delegated_node);
1730
1731         pr_debug("subflow=%p", ctx);
1732
1733         ctx->tcp_sock = sk;
1734
1735         return ctx;
1736 }
1737
1738 static void __subflow_state_change(struct sock *sk)
1739 {
1740         struct socket_wq *wq;
1741
1742         rcu_read_lock();
1743         wq = rcu_dereference(sk->sk_wq);
1744         if (skwq_has_sleeper(wq))
1745                 wake_up_interruptible_all(&wq->wait);
1746         rcu_read_unlock();
1747 }
1748
1749 static bool subflow_is_done(const struct sock *sk)
1750 {
1751         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1752 }
1753
1754 static void subflow_state_change(struct sock *sk)
1755 {
1756         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1757         struct sock *parent = subflow->conn;
1758         struct mptcp_sock *msk;
1759
1760         __subflow_state_change(sk);
1761
1762         msk = mptcp_sk(parent);
1763         if (subflow_simultaneous_connect(sk)) {
1764                 mptcp_propagate_sndbuf(parent, sk);
1765                 mptcp_do_fallback(sk);
1766                 mptcp_rcv_space_init(msk, sk);
1767                 pr_fallback(msk);
1768                 subflow->conn_finished = 1;
1769                 mptcp_set_connected(parent);
1770         }
1771
1772         /* as recvmsg() does not acquire the subflow socket for ssk selection
1773          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1774          * the data available machinery here.
1775          */
1776         if (mptcp_subflow_data_available(sk))
1777                 mptcp_data_ready(parent, sk);
1778         else if (unlikely(sk->sk_err))
1779                 subflow_error_report(sk);
1780
1781         subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1782
1783         /* when the fallback subflow closes the rx side, trigger a 'dummy'
1784          * ingress data fin, so that the msk state will follow along
1785          */
1786         if (__mptcp_check_fallback(msk) && subflow_is_done(sk) && msk->first == sk &&
1787             mptcp_update_rcv_data_fin(msk, READ_ONCE(msk->ack_seq), true))
1788                 mptcp_schedule_work(parent);
1789 }
1790
1791 void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_ssk)
1792 {
1793         struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
1794         struct mptcp_sock *msk, *next, *head = NULL;
1795         struct request_sock *req;
1796         struct sock *sk;
1797
1798         /* build a list of all unaccepted mptcp sockets */
1799         spin_lock_bh(&queue->rskq_lock);
1800         for (req = queue->rskq_accept_head; req; req = req->dl_next) {
1801                 struct mptcp_subflow_context *subflow;
1802                 struct sock *ssk = req->sk;
1803
1804                 if (!sk_is_mptcp(ssk))
1805                         continue;
1806
1807                 subflow = mptcp_subflow_ctx(ssk);
1808                 if (!subflow || !subflow->conn)
1809                         continue;
1810
1811                 /* skip if already in list */
1812                 sk = subflow->conn;
1813                 msk = mptcp_sk(sk);
1814                 if (msk->dl_next || msk == head)
1815                         continue;
1816
1817                 sock_hold(sk);
1818                 msk->dl_next = head;
1819                 head = msk;
1820         }
1821         spin_unlock_bh(&queue->rskq_lock);
1822         if (!head)
1823                 return;
1824
1825         /* can't acquire the msk socket lock under the subflow one,
1826          * or will cause ABBA deadlock
1827          */
1828         release_sock(listener_ssk);
1829
1830         for (msk = head; msk; msk = next) {
1831                 sk = (struct sock *)msk;
1832
1833                 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
1834                 next = msk->dl_next;
1835                 msk->dl_next = NULL;
1836
1837                 __mptcp_unaccepted_force_close(sk);
1838                 release_sock(sk);
1839
1840                 /* lockdep will report a false positive ABBA deadlock
1841                  * between cancel_work_sync and the listener socket.
1842                  * The involved locks belong to different sockets WRT
1843                  * the existing AB chain.
1844                  * Using a per socket key is problematic as key
1845                  * deregistration requires process context and must be
1846                  * performed at socket disposal time, in atomic
1847                  * context.
1848                  * Just tell lockdep to consider the listener socket
1849                  * released here.
1850                  */
1851                 mutex_release(&listener_sk->sk_lock.dep_map, _RET_IP_);
1852                 mptcp_cancel_work(sk);
1853                 mutex_acquire(&listener_sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1854
1855                 sock_put(sk);
1856         }
1857
1858         /* we are still under the listener msk socket lock */
1859         lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
1860 }
1861
1862 static int subflow_ulp_init(struct sock *sk)
1863 {
1864         struct inet_connection_sock *icsk = inet_csk(sk);
1865         struct mptcp_subflow_context *ctx;
1866         struct tcp_sock *tp = tcp_sk(sk);
1867         int err = 0;
1868
1869         /* disallow attaching ULP to a socket unless it has been
1870          * created with sock_create_kern()
1871          */
1872         if (!sk->sk_kern_sock) {
1873                 err = -EOPNOTSUPP;
1874                 goto out;
1875         }
1876
1877         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1878         if (!ctx) {
1879                 err = -ENOMEM;
1880                 goto out;
1881         }
1882
1883         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1884
1885         tp->is_mptcp = 1;
1886         ctx->icsk_af_ops = icsk->icsk_af_ops;
1887         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1888         ctx->tcp_state_change = sk->sk_state_change;
1889         ctx->tcp_error_report = sk->sk_error_report;
1890
1891         WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1892         WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
1893
1894         sk->sk_data_ready = subflow_data_ready;
1895         sk->sk_write_space = subflow_write_space;
1896         sk->sk_state_change = subflow_state_change;
1897         sk->sk_error_report = subflow_error_report;
1898 out:
1899         return err;
1900 }
1901
1902 static void subflow_ulp_release(struct sock *ssk)
1903 {
1904         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1905         bool release = true;
1906         struct sock *sk;
1907
1908         if (!ctx)
1909                 return;
1910
1911         sk = ctx->conn;
1912         if (sk) {
1913                 /* if the msk has been orphaned, keep the ctx
1914                  * alive, will be freed by __mptcp_close_ssk(),
1915                  * when the subflow is still unaccepted
1916                  */
1917                 release = ctx->disposable || list_empty(&ctx->node);
1918
1919                 /* inet_child_forget() does not call sk_state_change(),
1920                  * explicitly trigger the socket close machinery
1921                  */
1922                 if (!release && !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW,
1923                                                   &mptcp_sk(sk)->flags))
1924                         mptcp_schedule_work(sk);
1925                 sock_put(sk);
1926         }
1927
1928         mptcp_subflow_ops_undo_override(ssk);
1929         if (release)
1930                 kfree_rcu(ctx, rcu);
1931 }
1932
1933 static void subflow_ulp_clone(const struct request_sock *req,
1934                               struct sock *newsk,
1935                               const gfp_t priority)
1936 {
1937         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1938         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1939         struct mptcp_subflow_context *new_ctx;
1940
1941         if (!tcp_rsk(req)->is_mptcp ||
1942             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1943                 subflow_ulp_fallback(newsk, old_ctx);
1944                 return;
1945         }
1946
1947         new_ctx = subflow_create_ctx(newsk, priority);
1948         if (!new_ctx) {
1949                 subflow_ulp_fallback(newsk, old_ctx);
1950                 return;
1951         }
1952
1953         new_ctx->conn_finished = 1;
1954         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1955         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1956         new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1957         new_ctx->rel_write_seq = 1;
1958         new_ctx->tcp_sock = newsk;
1959
1960         if (subflow_req->mp_capable) {
1961                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1962                  * is fully established only after we receive the remote key
1963                  */
1964                 new_ctx->mp_capable = 1;
1965                 new_ctx->local_key = subflow_req->local_key;
1966                 new_ctx->token = subflow_req->token;
1967                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1968                 new_ctx->idsn = subflow_req->idsn;
1969
1970                 /* this is the first subflow, id is always 0 */
1971                 new_ctx->local_id_valid = 1;
1972         } else if (subflow_req->mp_join) {
1973                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1974                 new_ctx->mp_join = 1;
1975                 new_ctx->fully_established = 1;
1976                 new_ctx->remote_key_valid = 1;
1977                 new_ctx->backup = subflow_req->backup;
1978                 new_ctx->remote_id = subflow_req->remote_id;
1979                 new_ctx->token = subflow_req->token;
1980                 new_ctx->thmac = subflow_req->thmac;
1981
1982                 /* the subflow req id is valid, fetched via subflow_check_req()
1983                  * and subflow_token_join_request()
1984                  */
1985                 subflow_set_local_id(new_ctx, subflow_req->local_id);
1986         }
1987 }
1988
1989 static void tcp_release_cb_override(struct sock *ssk)
1990 {
1991         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1992
1993         if (mptcp_subflow_has_delegated_action(subflow))
1994                 mptcp_subflow_process_delegated(ssk);
1995
1996         tcp_release_cb(ssk);
1997 }
1998
1999 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
2000         .name           = "mptcp",
2001         .owner          = THIS_MODULE,
2002         .init           = subflow_ulp_init,
2003         .release        = subflow_ulp_release,
2004         .clone          = subflow_ulp_clone,
2005 };
2006
2007 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
2008 {
2009         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
2010
2011         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
2012                                               subflow_ops->obj_size, 0,
2013                                               SLAB_ACCOUNT |
2014                                               SLAB_TYPESAFE_BY_RCU,
2015                                               NULL);
2016         if (!subflow_ops->slab)
2017                 return -ENOMEM;
2018
2019         return 0;
2020 }
2021
2022 void __init mptcp_subflow_init(void)
2023 {
2024         mptcp_subflow_v4_request_sock_ops = tcp_request_sock_ops;
2025         mptcp_subflow_v4_request_sock_ops.slab_name = "request_sock_subflow_v4";
2026         mptcp_subflow_v4_request_sock_ops.destructor = subflow_v4_req_destructor;
2027
2028         if (subflow_ops_init(&mptcp_subflow_v4_request_sock_ops) != 0)
2029                 panic("MPTCP: failed to init subflow v4 request sock ops\n");
2030
2031         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
2032         subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
2033         subflow_request_sock_ipv4_ops.send_synack = subflow_v4_send_synack;
2034
2035         subflow_specific = ipv4_specific;
2036         subflow_specific.conn_request = subflow_v4_conn_request;
2037         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
2038         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
2039         subflow_specific.rebuild_header = subflow_rebuild_header;
2040
2041         tcp_prot_override = tcp_prot;
2042         tcp_prot_override.release_cb = tcp_release_cb_override;
2043
2044 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2045         /* In struct mptcp_subflow_request_sock, we assume the TCP request sock
2046          * structures for v4 and v6 have the same size. It should not changed in
2047          * the future but better to make sure to be warned if it is no longer
2048          * the case.
2049          */
2050         BUILD_BUG_ON(sizeof(struct tcp_request_sock) != sizeof(struct tcp6_request_sock));
2051
2052         mptcp_subflow_v6_request_sock_ops = tcp6_request_sock_ops;
2053         mptcp_subflow_v6_request_sock_ops.slab_name = "request_sock_subflow_v6";
2054         mptcp_subflow_v6_request_sock_ops.destructor = subflow_v6_req_destructor;
2055
2056         if (subflow_ops_init(&mptcp_subflow_v6_request_sock_ops) != 0)
2057                 panic("MPTCP: failed to init subflow v6 request sock ops\n");
2058
2059         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
2060         subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
2061         subflow_request_sock_ipv6_ops.send_synack = subflow_v6_send_synack;
2062
2063         subflow_v6_specific = ipv6_specific;
2064         subflow_v6_specific.conn_request = subflow_v6_conn_request;
2065         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
2066         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
2067         subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
2068
2069         subflow_v6m_specific = subflow_v6_specific;
2070         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
2071         subflow_v6m_specific.send_check = ipv4_specific.send_check;
2072         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
2073         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
2074         subflow_v6m_specific.net_frag_header_len = 0;
2075         subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
2076
2077         tcpv6_prot_override = tcpv6_prot;
2078         tcpv6_prot_override.release_cb = tcp_release_cb_override;
2079 #endif
2080
2081         mptcp_diag_subflow_init(&subflow_ulp_ops);
2082
2083         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
2084                 panic("MPTCP: failed to register subflows to ULP\n");
2085 }