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                         ctx->subflow_id = 1;
 823                         owner = mptcp_sk(ctx->conn);
 824                         mptcp_pm_new_connection(owner, child, 1);
 825
 826                         /* with OoO packets we can reach here without ingress
 827                          * mpc option
 828                          */
 829                         if (mp_opt.suboptions & OPTION_MPTCP_MPC_ACK) {
 830                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 831                                 mptcp_pm_fully_established(owner, child);
 832                                 ctx->pm_notified = 1;
 833                         }
 834                 } else if (ctx->mp_join) {
 835                         owner = subflow_req->msk;
 836                         if (!owner) {
 837                                 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 838                                 goto dispose_child;
 839                         }
 840
 841                         /* move the msk reference ownership to the subflow */
 842                         subflow_req->msk = NULL;
 843                         ctx->conn = (struct sock *)owner;
 844
 845                         if (subflow_use_different_sport(owner, sk)) {
 846                                 pr_debug("ack inet_sport=%d %d",
 847                                          ntohs(inet_sk(sk)->inet_sport),
 848                                          ntohs(inet_sk((struct sock *)owner)->inet_sport));
 849                                 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
 850                                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
 851                                         goto dispose_child;
 852                                 }
 853                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
 854                         }
 855
 856                         if (!mptcp_finish_join(child))
 857                                 goto dispose_child;
 858
 859                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 860                         tcp_rsk(req)->drop_req = true;
 861                 }
 862         }
 863
 864         /* check for expected invariant - should never trigger, just help
 865          * catching eariler subtle bugs
 866          */
 867         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 868                      (!mptcp_subflow_ctx(child) ||
 869                       !mptcp_subflow_ctx(child)->conn));
 870         return child;
 871
 872 dispose_child:
 873         mptcp_subflow_drop_ctx(child);
 874         tcp_rsk(req)->drop_req = true;
 875         inet_csk_prepare_for_destroy_sock(child);
 876         tcp_done(child);
 877         req->rsk_ops->send_reset(sk, skb);
 878
 879         /* The last child reference will be released by the caller */
 880         return child;
 881
 882 fallback:
 883         mptcp_subflow_drop_ctx(child);
 884         return child;
 885 }
 886
 887 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
 888 static struct proto tcp_prot_override __ro_after_init;
 889
 890 enum mapping_status {
 891         MAPPING_OK,
 892         MAPPING_INVALID,
 893         MAPPING_EMPTY,
 894         MAPPING_DATA_FIN,
 895         MAPPING_DUMMY,
 896         MAPPING_BAD_CSUM
 897 };
 898
 899 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 900 {
 901         pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 902                  ssn, subflow->map_subflow_seq, subflow->map_data_len);
 903 }
 904
 905 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 906 {
 907         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 908         unsigned int skb_consumed;
 909
 910         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 911         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 912                 return true;
 913
 914         return skb->len - skb_consumed <= subflow->map_data_len -
 915                                           mptcp_subflow_get_map_offset(subflow);
 916 }
 917
 918 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 919 {
 920         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 921         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 922
 923         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 924                 /* Mapping covers data later in the subflow stream,
 925                  * currently unsupported.
 926                  */
 927                 dbg_bad_map(subflow, ssn);
 928                 return false;
 929         }
 930         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 931                                   subflow->map_data_len))) {
 932                 /* Mapping does covers past subflow data, invalid */
 933                 dbg_bad_map(subflow, ssn);
 934                 return false;
 935         }
 936         return true;
 937 }
 938
 939 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
 940                                               bool csum_reqd)
 941 {
 942         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 943         u32 offset, seq, delta;
 944         __sum16 csum;
 945         int len;
 946
 947         if (!csum_reqd)
 948                 return MAPPING_OK;
 949
 950         /* mapping already validated on previous traversal */
 951         if (subflow->map_csum_len == subflow->map_data_len)
 952                 return MAPPING_OK;
 953
 954         /* traverse the receive queue, ensuring it contains a full
 955          * DSS mapping and accumulating the related csum.
 956          * Preserve the accoumlate csum across multiple calls, to compute
 957          * the csum only once
 958          */
 959         delta = subflow->map_data_len - subflow->map_csum_len;
 960         for (;;) {
 961                 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
 962                 offset = seq - TCP_SKB_CB(skb)->seq;
 963
 964                 /* if the current skb has not been accounted yet, csum its contents
 965                  * up to the amount covered by the current DSS
 966                  */
 967                 if (offset < skb->len) {
 968                         __wsum csum;
 969
 970                         len = min(skb->len - offset, delta);
 971                         csum = skb_checksum(skb, offset, len, 0);
 972                         subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
 973                                                                 subflow->map_csum_len);
 974
 975                         delta -= len;
 976                         subflow->map_csum_len += len;
 977                 }
 978                 if (delta == 0)
 979                         break;
 980
 981                 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
 982                         /* if this subflow is closed, the partial mapping
 983                          * will be never completed; flush the pending skbs, so
 984                          * that subflow_sched_work_if_closed() can kick in
 985                          */
 986                         if (unlikely(ssk->sk_state == TCP_CLOSE))
 987                                 while ((skb = skb_peek(&ssk->sk_receive_queue)))
 988                                         sk_eat_skb(ssk, skb);
 989
 990                         /* not enough data to validate the csum */
 991                         return MAPPING_EMPTY;
 992                 }
 993
 994                 /* the DSS mapping for next skbs will be validated later,
 995                  * when a get_mapping_status call will process such skb
 996                  */
 997                 skb = skb->next;
 998         }
 999
1000         /* note that 'map_data_len' accounts only for the carried data, does
1001          * not include the eventual seq increment due to the data fin,
1002          * while the pseudo header requires the original DSS data len,
1003          * including that
1004          */
1005         csum = __mptcp_make_csum(subflow->map_seq,
1006                                  subflow->map_subflow_seq,
1007                                  subflow->map_data_len + subflow->map_data_fin,
1008                                  subflow->map_data_csum);
1009         if (unlikely(csum)) {
1010                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
1011                 return MAPPING_BAD_CSUM;
1012         }
1013
1014         subflow->valid_csum_seen = 1;
1015         return MAPPING_OK;
1016 }
1017
1018 static enum mapping_status get_mapping_status(struct sock *ssk,
1019                                               struct mptcp_sock *msk)
1020 {
1021         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1022         bool csum_reqd = READ_ONCE(msk->csum_enabled);
1023         struct mptcp_ext *mpext;
1024         struct sk_buff *skb;
1025         u16 data_len;
1026         u64 map_seq;
1027
1028         skb = skb_peek(&ssk->sk_receive_queue);
1029         if (!skb)
1030                 return MAPPING_EMPTY;
1031
1032         if (mptcp_check_fallback(ssk))
1033                 return MAPPING_DUMMY;
1034
1035         mpext = mptcp_get_ext(skb);
1036         if (!mpext || !mpext->use_map) {
1037                 if (!subflow->map_valid && !skb->len) {
1038                         /* the TCP stack deliver 0 len FIN pkt to the receive
1039                          * queue, that is the only 0len pkts ever expected here,
1040                          * and we can admit no mapping only for 0 len pkts
1041                          */
1042                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
1043                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
1044                                           TCP_SKB_CB(skb)->seq,
1045                                           TCP_SKB_CB(skb)->end_seq,
1046                                           TCP_SKB_CB(skb)->tcp_flags);
1047                         sk_eat_skb(ssk, skb);
1048                         return MAPPING_EMPTY;
1049                 }
1050
1051                 if (!subflow->map_valid)
1052                         return MAPPING_INVALID;
1053
1054                 goto validate_seq;
1055         }
1056
1057         trace_get_mapping_status(mpext);
1058
1059         data_len = mpext->data_len;
1060         if (data_len == 0) {
1061                 pr_debug("infinite mapping received");
1062                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1063                 subflow->map_data_len = 0;
1064                 return MAPPING_INVALID;
1065         }
1066
1067         if (mpext->data_fin == 1) {
1068                 if (data_len == 1) {
1069                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1070                                                                  mpext->dsn64);
1071                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
1072                         if (subflow->map_valid) {
1073                                 /* A DATA_FIN might arrive in a DSS
1074                                  * option before the previous mapping
1075                                  * has been fully consumed. Continue
1076                                  * handling the existing mapping.
1077                                  */
1078                                 skb_ext_del(skb, SKB_EXT_MPTCP);
1079                                 return MAPPING_OK;
1080                         } else {
1081                                 if (updated)
1082                                         mptcp_schedule_work((struct sock *)msk);
1083
1084                                 return MAPPING_DATA_FIN;
1085                         }
1086                 } else {
1087                         u64 data_fin_seq = mpext->data_seq + data_len - 1;
1088
1089                         /* If mpext->data_seq is a 32-bit value, data_fin_seq
1090                          * must also be limited to 32 bits.
1091                          */
1092                         if (!mpext->dsn64)
1093                                 data_fin_seq &= GENMASK_ULL(31, 0);
1094
1095                         mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1096                         pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
1097                                  data_fin_seq, mpext->dsn64);
1098                 }
1099
1100                 /* Adjust for DATA_FIN using 1 byte of sequence space */
1101                 data_len--;
1102         }
1103
1104         map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1105         WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1106
1107         if (subflow->map_valid) {
1108                 /* Allow replacing only with an identical map */
1109                 if (subflow->map_seq == map_seq &&
1110                     subflow->map_subflow_seq == mpext->subflow_seq &&
1111                     subflow->map_data_len == data_len &&
1112                     subflow->map_csum_reqd == mpext->csum_reqd) {
1113                         skb_ext_del(skb, SKB_EXT_MPTCP);
1114                         goto validate_csum;
1115                 }
1116
1117                 /* If this skb data are fully covered by the current mapping,
1118                  * the new map would need caching, which is not supported
1119                  */
1120                 if (skb_is_fully_mapped(ssk, skb)) {
1121                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1122                         return MAPPING_INVALID;
1123                 }
1124
1125                 /* will validate the next map after consuming the current one */
1126                 goto validate_csum;
1127         }
1128
1129         subflow->map_seq = map_seq;
1130         subflow->map_subflow_seq = mpext->subflow_seq;
1131         subflow->map_data_len = data_len;
1132         subflow->map_valid = 1;
1133         subflow->map_data_fin = mpext->data_fin;
1134         subflow->mpc_map = mpext->mpc_map;
1135         subflow->map_csum_reqd = mpext->csum_reqd;
1136         subflow->map_csum_len = 0;
1137         subflow->map_data_csum = csum_unfold(mpext->csum);
1138
1139         /* Cfr RFC 8684 Section 3.3.0 */
1140         if (unlikely(subflow->map_csum_reqd != csum_reqd))
1141                 return MAPPING_INVALID;
1142
1143         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
1144                  subflow->map_seq, subflow->map_subflow_seq,
1145                  subflow->map_data_len, subflow->map_csum_reqd,
1146                  subflow->map_data_csum);
1147
1148 validate_seq:
1149         /* we revalidate valid mapping on new skb, because we must ensure
1150          * the current skb is completely covered by the available mapping
1151          */
1152         if (!validate_mapping(ssk, skb)) {
1153                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1154                 return MAPPING_INVALID;
1155         }
1156
1157         skb_ext_del(skb, SKB_EXT_MPTCP);
1158
1159 validate_csum:
1160         return validate_data_csum(ssk, skb, csum_reqd);
1161 }
1162
1163 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1164                                        u64 limit)
1165 {
1166         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1167         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1168         u32 incr;
1169
1170         incr = limit >= skb->len ? skb->len + fin : limit;
1171
1172         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
1173                  subflow->map_subflow_seq);
1174         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1175         tcp_sk(ssk)->copied_seq += incr;
1176         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1177                 sk_eat_skb(ssk, skb);
1178         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1179                 subflow->map_valid = 0;
1180 }
1181
1182 /* sched mptcp worker to remove the subflow if no more data is pending */
1183 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1184 {
1185         if (likely(ssk->sk_state != TCP_CLOSE))
1186                 return;
1187
1188         if (skb_queue_empty(&ssk->sk_receive_queue) &&
1189             !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
1190                 mptcp_schedule_work((struct sock *)msk);
1191 }
1192
1193 static bool subflow_can_fallback(struct mptcp_subflow_context *subflow)
1194 {
1195         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1196
1197         if (subflow->mp_join)
1198                 return false;
1199         else if (READ_ONCE(msk->csum_enabled))
1200                 return !subflow->valid_csum_seen;
1201         else
1202                 return !subflow->fully_established;
1203 }
1204
1205 static void mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk)
1206 {
1207         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1208         unsigned long fail_tout;
1209
1210         /* greceful failure can happen only on the MPC subflow */
1211         if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first)))
1212                 return;
1213
1214         /* since the close timeout take precedence on the fail one,
1215          * no need to start the latter when the first is already set
1216          */
1217         if (sock_flag((struct sock *)msk, SOCK_DEAD))
1218                 return;
1219
1220         /* we don't need extreme accuracy here, use a zero fail_tout as special
1221          * value meaning no fail timeout at all;
1222          */
1223         fail_tout = jiffies + TCP_RTO_MAX;
1224         if (!fail_tout)
1225                 fail_tout = 1;
1226         WRITE_ONCE(subflow->fail_tout, fail_tout);
1227         tcp_send_ack(ssk);
1228
1229         mptcp_reset_tout_timer(msk, subflow->fail_tout);
1230 }
1231
1232 static bool subflow_check_data_avail(struct sock *ssk)
1233 {
1234         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1235         enum mapping_status status;
1236         struct mptcp_sock *msk;
1237         struct sk_buff *skb;
1238
1239         if (!skb_peek(&ssk->sk_receive_queue))
1240                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1241         if (subflow->data_avail)
1242                 return true;
1243
1244         msk = mptcp_sk(subflow->conn);
1245         for (;;) {
1246                 u64 ack_seq;
1247                 u64 old_ack;
1248
1249                 status = get_mapping_status(ssk, msk);
1250                 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1251                 if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY ||
1252                              status == MAPPING_BAD_CSUM))
1253                         goto fallback;
1254
1255                 if (status != MAPPING_OK)
1256                         goto no_data;
1257
1258                 skb = skb_peek(&ssk->sk_receive_queue);
1259                 if (WARN_ON_ONCE(!skb))
1260                         goto no_data;
1261
1262                 if (unlikely(!READ_ONCE(msk->can_ack)))
1263                         goto fallback;
1264
1265                 old_ack = READ_ONCE(msk->ack_seq);
1266                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1267                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1268                          ack_seq);
1269                 if (unlikely(before64(ack_seq, old_ack))) {
1270                         mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1271                         continue;
1272                 }
1273
1274                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1275                 break;
1276         }
1277         return true;
1278
1279 no_data:
1280         subflow_sched_work_if_closed(msk, ssk);
1281         return false;
1282
1283 fallback:
1284         if (!__mptcp_check_fallback(msk)) {
1285                 /* RFC 8684 section 3.7. */
1286                 if (status == MAPPING_BAD_CSUM &&
1287                     (subflow->mp_join || subflow->valid_csum_seen)) {
1288                         subflow->send_mp_fail = 1;
1289
1290                         if (!READ_ONCE(msk->allow_infinite_fallback)) {
1291                                 subflow->reset_transient = 0;
1292                                 subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1293                                 goto reset;
1294                         }
1295                         mptcp_subflow_fail(msk, ssk);
1296                         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1297                         return true;
1298                 }
1299
1300                 if (!subflow_can_fallback(subflow) && subflow->map_data_len) {
1301                         /* fatal protocol error, close the socket.
1302                          * subflow_error_report() will introduce the appropriate barriers
1303                          */
1304                         subflow->reset_transient = 0;
1305                         subflow->reset_reason = MPTCP_RST_EMPTCP;
1306
1307 reset:
1308                         WRITE_ONCE(ssk->sk_err, EBADMSG);
1309                         tcp_set_state(ssk, TCP_CLOSE);
1310                         while ((skb = skb_peek(&ssk->sk_receive_queue)))
1311                                 sk_eat_skb(ssk, skb);
1312                         tcp_send_active_reset(ssk, GFP_ATOMIC);
1313                         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1314                         return false;
1315                 }
1316
1317                 mptcp_do_fallback(ssk);
1318         }
1319
1320         skb = skb_peek(&ssk->sk_receive_queue);
1321         subflow->map_valid = 1;
1322         subflow->map_seq = READ_ONCE(msk->ack_seq);
1323         subflow->map_data_len = skb->len;
1324         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1325         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1326         return true;
1327 }
1328
1329 bool mptcp_subflow_data_available(struct sock *sk)
1330 {
1331         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1332
1333         /* check if current mapping is still valid */
1334         if (subflow->map_valid &&
1335             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1336                 subflow->map_valid = 0;
1337                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1338
1339                 pr_debug("Done with mapping: seq=%u data_len=%u",
1340                          subflow->map_subflow_seq,
1341                          subflow->map_data_len);
1342         }
1343
1344         return subflow_check_data_avail(sk);
1345 }
1346
1347 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1348  * not the ssk one.
1349  *
1350  * In mptcp, rwin is about the mptcp-level connection data.
1351  *
1352  * Data that is still on the ssk rx queue can thus be ignored,
1353  * as far as mptcp peer is concerned that data is still inflight.
1354  * DSS ACK is updated when skb is moved to the mptcp rx queue.
1355  */
1356 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1357 {
1358         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1359         const struct sock *sk = subflow->conn;
1360
1361         *space = __mptcp_space(sk);
1362         *full_space = mptcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1363 }
1364
1365 static void subflow_error_report(struct sock *ssk)
1366 {
1367         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1368
1369         /* bail early if this is a no-op, so that we avoid introducing a
1370          * problematic lockdep dependency between TCP accept queue lock
1371          * and msk socket spinlock
1372          */
1373         if (!sk->sk_socket)
1374                 return;
1375
1376         mptcp_data_lock(sk);
1377         if (!sock_owned_by_user(sk))
1378                 __mptcp_error_report(sk);
1379         else
1380                 __set_bit(MPTCP_ERROR_REPORT,  &mptcp_sk(sk)->cb_flags);
1381         mptcp_data_unlock(sk);
1382 }
1383
1384 static void subflow_data_ready(struct sock *sk)
1385 {
1386         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1387         u16 state = 1 << inet_sk_state_load(sk);
1388         struct sock *parent = subflow->conn;
1389         struct mptcp_sock *msk;
1390
1391         trace_sk_data_ready(sk);
1392
1393         msk = mptcp_sk(parent);
1394         if (state & TCPF_LISTEN) {
1395                 /* MPJ subflow are removed from accept queue before reaching here,
1396                  * avoid stray wakeups
1397                  */
1398                 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1399                         return;
1400
1401                 parent->sk_data_ready(parent);
1402                 return;
1403         }
1404
1405         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1406                      !subflow->mp_join && !(state & TCPF_CLOSE));
1407
1408         if (mptcp_subflow_data_available(sk))
1409                 mptcp_data_ready(parent, sk);
1410         else if (unlikely(sk->sk_err))
1411                 subflow_error_report(sk);
1412 }
1413
1414 static void subflow_write_space(struct sock *ssk)
1415 {
1416         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1417
1418         mptcp_propagate_sndbuf(sk, ssk);
1419         mptcp_write_space(sk);
1420 }
1421
1422 static const struct inet_connection_sock_af_ops *
1423 subflow_default_af_ops(struct sock *sk)
1424 {
1425 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1426         if (sk->sk_family == AF_INET6)
1427                 return &subflow_v6_specific;
1428 #endif
1429         return &subflow_specific;
1430 }
1431
1432 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1433 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1434 {
1435         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1436         struct inet_connection_sock *icsk = inet_csk(sk);
1437         const struct inet_connection_sock_af_ops *target;
1438
1439         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1440
1441         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1442                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1443
1444         if (likely(icsk->icsk_af_ops == target))
1445                 return;
1446
1447         subflow->icsk_af_ops = icsk->icsk_af_ops;
1448         icsk->icsk_af_ops = target;
1449 }
1450 #endif
1451
1452 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1453                          struct sockaddr_storage *addr,
1454                          unsigned short family)
1455 {
1456         memset(addr, 0, sizeof(*addr));
1457         addr->ss_family = family;
1458         if (addr->ss_family == AF_INET) {
1459                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1460
1461                 if (info->family == AF_INET)
1462                         in_addr->sin_addr = info->addr;
1463 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1464                 else if (ipv6_addr_v4mapped(&info->addr6))
1465                         in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1466 #endif
1467                 in_addr->sin_port = info->port;
1468         }
1469 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1470         else if (addr->ss_family == AF_INET6) {
1471                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1472
1473                 if (info->family == AF_INET)
1474                         ipv6_addr_set_v4mapped(info->addr.s_addr,
1475                                                &in6_addr->sin6_addr);
1476                 else
1477                         in6_addr->sin6_addr = info->addr6;
1478                 in6_addr->sin6_port = info->port;
1479         }
1480 #endif
1481 }
1482
1483 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1484                             const struct mptcp_addr_info *remote)
1485 {
1486         struct mptcp_sock *msk = mptcp_sk(sk);
1487         struct mptcp_subflow_context *subflow;
1488         struct sockaddr_storage addr;
1489         int remote_id = remote->id;
1490         int local_id = loc->id;
1491         int err = -ENOTCONN;
1492         struct socket *sf;
1493         struct sock *ssk;
1494         u32 remote_token;
1495         int addrlen;
1496         int ifindex;
1497         u8 flags;
1498
1499         if (!mptcp_is_fully_established(sk))
1500                 goto err_out;
1501
1502         err = mptcp_subflow_create_socket(sk, loc->family, &sf);
1503         if (err)
1504                 goto err_out;
1505
1506         ssk = sf->sk;
1507         subflow = mptcp_subflow_ctx(ssk);
1508         do {
1509                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1510         } while (!subflow->local_nonce);
1511
1512         if (local_id)
1513                 subflow_set_local_id(subflow, local_id);
1514
1515         mptcp_pm_get_flags_and_ifindex_by_id(msk, local_id,
1516                                              &flags, &ifindex);
1517         subflow->remote_key_valid = 1;
1518         subflow->remote_key = msk->remote_key;
1519         subflow->local_key = msk->local_key;
1520         subflow->token = msk->token;
1521         mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1522
1523         addrlen = sizeof(struct sockaddr_in);
1524 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1525         if (addr.ss_family == AF_INET6)
1526                 addrlen = sizeof(struct sockaddr_in6);
1527 #endif
1528         mptcp_sockopt_sync(msk, ssk);
1529
1530         ssk->sk_bound_dev_if = ifindex;
1531         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1532         if (err)
1533                 goto failed;
1534
1535         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1536         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1537                  remote_token, local_id, remote_id);
1538         subflow->remote_token = remote_token;
1539         subflow->remote_id = remote_id;
1540         subflow->request_join = 1;
1541         subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1542         subflow->subflow_id = msk->subflow_id++;
1543         mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1544
1545         sock_hold(ssk);
1546         list_add_tail(&subflow->node, &msk->conn_list);
1547         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1548         if (err && err != -EINPROGRESS)
1549                 goto failed_unlink;
1550
1551         /* discard the subflow socket */
1552         mptcp_sock_graft(ssk, sk->sk_socket);
1553         iput(SOCK_INODE(sf));
1554         WRITE_ONCE(msk->allow_infinite_fallback, false);
1555         mptcp_stop_tout_timer(sk);
1556         return 0;
1557
1558 failed_unlink:
1559         list_del(&subflow->node);
1560         sock_put(mptcp_subflow_tcp_sock(subflow));
1561
1562 failed:
1563         subflow->disposable = 1;
1564         sock_release(sf);
1565
1566 err_out:
1567         /* we account subflows before the creation, and this failures will not
1568          * be caught by sk_state_change()
1569          */
1570         mptcp_pm_close_subflow(msk);
1571         return err;
1572 }
1573
1574 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1575 {
1576 #ifdef CONFIG_SOCK_CGROUP_DATA
1577         struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1578                                 *child_skcd = &child->sk_cgrp_data;
1579
1580         /* only the additional subflows created by kworkers have to be modified */
1581         if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1582             cgroup_id(sock_cgroup_ptr(child_skcd))) {
1583 #ifdef CONFIG_MEMCG
1584                 struct mem_cgroup *memcg = parent->sk_memcg;
1585
1586                 mem_cgroup_sk_free(child);
1587                 if (memcg && css_tryget(&memcg->css))
1588                         child->sk_memcg = memcg;
1589 #endif /* CONFIG_MEMCG */
1590
1591                 cgroup_sk_free(child_skcd);
1592                 *child_skcd = *parent_skcd;
1593                 cgroup_sk_clone(child_skcd);
1594         }
1595 #endif /* CONFIG_SOCK_CGROUP_DATA */
1596 }
1597
1598 static void mptcp_subflow_ops_override(struct sock *ssk)
1599 {
1600 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1601         if (ssk->sk_prot == &tcpv6_prot)
1602                 ssk->sk_prot = &tcpv6_prot_override;
1603         else
1604 #endif
1605                 ssk->sk_prot = &tcp_prot_override;
1606 }
1607
1608 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1609 {
1610 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1611         if (ssk->sk_prot == &tcpv6_prot_override)
1612                 ssk->sk_prot = &tcpv6_prot;
1613         else
1614 #endif
1615                 ssk->sk_prot = &tcp_prot;
1616 }
1617
1618 int mptcp_subflow_create_socket(struct sock *sk, unsigned short family,
1619                                 struct socket **new_sock)
1620 {
1621         struct mptcp_subflow_context *subflow;
1622         struct net *net = sock_net(sk);
1623         struct socket *sf;
1624         int err;
1625
1626         /* un-accepted server sockets can reach here - on bad configuration
1627          * bail early to avoid greater trouble later
1628          */
1629         if (unlikely(!sk->sk_socket))
1630                 return -EINVAL;
1631
1632         err = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, &sf);
1633         if (err)
1634                 return err;
1635
1636         lock_sock_nested(sf->sk, SINGLE_DEPTH_NESTING);
1637
1638         err = security_mptcp_add_subflow(sk, sf->sk);
1639         if (err)
1640                 goto release_ssk;
1641
1642         /* the newly created socket has to be in the same cgroup as its parent */
1643         mptcp_attach_cgroup(sk, sf->sk);
1644
1645         /* kernel sockets do not by default acquire net ref, but TCP timer
1646          * needs it.
1647          * Update ns_tracker to current stack trace and refcounted tracker.
1648          */
1649         __netns_tracker_free(net, &sf->sk->ns_tracker, false);
1650         sf->sk->sk_net_refcnt = 1;
1651         get_net_track(net, &sf->sk->ns_tracker, GFP_KERNEL);
1652         sock_inuse_add(net, 1);
1653         err = tcp_set_ulp(sf->sk, "mptcp");
1654
1655 release_ssk:
1656         release_sock(sf->sk);
1657
1658         if (err) {
1659                 sock_release(sf);
1660                 return err;
1661         }
1662
1663         /* the newly created socket really belongs to the owning MPTCP master
1664          * socket, even if for additional subflows the allocation is performed
1665          * by a kernel workqueue. Adjust inode references, so that the
1666          * procfs/diag interfaces really show this one belonging to the correct
1667          * user.
1668          */
1669         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1670         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1671         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1672
1673         subflow = mptcp_subflow_ctx(sf->sk);
1674         pr_debug("subflow=%p", subflow);
1675
1676         *new_sock = sf;
1677         sock_hold(sk);
1678         subflow->conn = sk;
1679         mptcp_subflow_ops_override(sf->sk);
1680
1681         return 0;
1682 }
1683
1684 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1685                                                         gfp_t priority)
1686 {
1687         struct inet_connection_sock *icsk = inet_csk(sk);
1688         struct mptcp_subflow_context *ctx;
1689
1690         ctx = kzalloc(sizeof(*ctx), priority);
1691         if (!ctx)
1692                 return NULL;
1693
1694         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1695         INIT_LIST_HEAD(&ctx->node);
1696         INIT_LIST_HEAD(&ctx->delegated_node);
1697
1698         pr_debug("subflow=%p", ctx);
1699
1700         ctx->tcp_sock = sk;
1701
1702         return ctx;
1703 }
1704
1705 static void __subflow_state_change(struct sock *sk)
1706 {
1707         struct socket_wq *wq;
1708
1709         rcu_read_lock();
1710         wq = rcu_dereference(sk->sk_wq);
1711         if (skwq_has_sleeper(wq))
1712                 wake_up_interruptible_all(&wq->wait);
1713         rcu_read_unlock();
1714 }
1715
1716 static bool subflow_is_done(const struct sock *sk)
1717 {
1718         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1719 }
1720
1721 static void subflow_state_change(struct sock *sk)
1722 {
1723         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1724         struct sock *parent = subflow->conn;
1725         struct mptcp_sock *msk;
1726
1727         __subflow_state_change(sk);
1728
1729         msk = mptcp_sk(parent);
1730         if (subflow_simultaneous_connect(sk)) {
1731                 mptcp_propagate_sndbuf(parent, sk);
1732                 mptcp_do_fallback(sk);
1733                 mptcp_rcv_space_init(msk, sk);
1734                 pr_fallback(msk);
1735                 subflow->conn_finished = 1;
1736                 mptcp_set_connected(parent);
1737         }
1738
1739         /* as recvmsg() does not acquire the subflow socket for ssk selection
1740          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1741          * the data available machinery here.
1742          */
1743         if (mptcp_subflow_data_available(sk))
1744                 mptcp_data_ready(parent, sk);
1745         else if (unlikely(sk->sk_err))
1746                 subflow_error_report(sk);
1747
1748         subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1749
1750         /* when the fallback subflow closes the rx side, trigger a 'dummy'
1751          * ingress data fin, so that the msk state will follow along
1752          */
1753         if (__mptcp_check_fallback(msk) && subflow_is_done(sk) && msk->first == sk &&
1754             mptcp_update_rcv_data_fin(msk, READ_ONCE(msk->ack_seq), true))
1755                 mptcp_schedule_work(parent);
1756 }
1757
1758 void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_ssk)
1759 {
1760         struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
1761         struct request_sock *req, *head, *tail;
1762         struct mptcp_subflow_context *subflow;
1763         struct sock *sk, *ssk;
1764
1765         /* Due to lock dependencies no relevant lock can be acquired under rskq_lock.
1766          * Splice the req list, so that accept() can not reach the pending ssk after
1767          * the listener socket is released below.
1768          */
1769         spin_lock_bh(&queue->rskq_lock);
1770         head = queue->rskq_accept_head;
1771         tail = queue->rskq_accept_tail;
1772         queue->rskq_accept_head = NULL;
1773         queue->rskq_accept_tail = NULL;
1774         spin_unlock_bh(&queue->rskq_lock);
1775
1776         if (!head)
1777                 return;
1778
1779         /* can't acquire the msk socket lock under the subflow one,
1780          * or will cause ABBA deadlock
1781          */
1782         release_sock(listener_ssk);
1783
1784         for (req = head; req; req = req->dl_next) {
1785                 ssk = req->sk;
1786                 if (!sk_is_mptcp(ssk))
1787                         continue;
1788
1789                 subflow = mptcp_subflow_ctx(ssk);
1790                 if (!subflow || !subflow->conn)
1791                         continue;
1792
1793                 sk = subflow->conn;
1794                 sock_hold(sk);
1795
1796                 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
1797                 __mptcp_unaccepted_force_close(sk);
1798                 release_sock(sk);
1799
1800                 /* lockdep will report a false positive ABBA deadlock
1801                  * between cancel_work_sync and the listener socket.
1802                  * The involved locks belong to different sockets WRT
1803                  * the existing AB chain.
1804                  * Using a per socket key is problematic as key
1805                  * deregistration requires process context and must be
1806                  * performed at socket disposal time, in atomic
1807                  * context.
1808                  * Just tell lockdep to consider the listener socket
1809                  * released here.
1810                  */
1811                 mutex_release(&listener_sk->sk_lock.dep_map, _RET_IP_);
1812                 mptcp_cancel_work(sk);
1813                 mutex_acquire(&listener_sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1814
1815                 sock_put(sk);
1816         }
1817
1818         /* we are still under the listener msk socket lock */
1819         lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
1820
1821         /* restore the listener queue, to let the TCP code clean it up */
1822         spin_lock_bh(&queue->rskq_lock);
1823         WARN_ON_ONCE(queue->rskq_accept_head);
1824         queue->rskq_accept_head = head;
1825         queue->rskq_accept_tail = tail;
1826         spin_unlock_bh(&queue->rskq_lock);
1827 }
1828
1829 static int subflow_ulp_init(struct sock *sk)
1830 {
1831         struct inet_connection_sock *icsk = inet_csk(sk);
1832         struct mptcp_subflow_context *ctx;
1833         struct tcp_sock *tp = tcp_sk(sk);
1834         int err = 0;
1835
1836         /* disallow attaching ULP to a socket unless it has been
1837          * created with sock_create_kern()
1838          */
1839         if (!sk->sk_kern_sock) {
1840                 err = -EOPNOTSUPP;
1841                 goto out;
1842         }
1843
1844         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1845         if (!ctx) {
1846                 err = -ENOMEM;
1847                 goto out;
1848         }
1849
1850         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1851
1852         tp->is_mptcp = 1;
1853         ctx->icsk_af_ops = icsk->icsk_af_ops;
1854         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1855         ctx->tcp_state_change = sk->sk_state_change;
1856         ctx->tcp_error_report = sk->sk_error_report;
1857
1858         WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1859         WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
1860
1861         sk->sk_data_ready = subflow_data_ready;
1862         sk->sk_write_space = subflow_write_space;
1863         sk->sk_state_change = subflow_state_change;
1864         sk->sk_error_report = subflow_error_report;
1865 out:
1866         return err;
1867 }
1868
1869 static void subflow_ulp_release(struct sock *ssk)
1870 {
1871         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1872         bool release = true;
1873         struct sock *sk;
1874
1875         if (!ctx)
1876                 return;
1877
1878         sk = ctx->conn;
1879         if (sk) {
1880                 /* if the msk has been orphaned, keep the ctx
1881                  * alive, will be freed by __mptcp_close_ssk(),
1882                  * when the subflow is still unaccepted
1883                  */
1884                 release = ctx->disposable || list_empty(&ctx->node);
1885
1886                 /* inet_child_forget() does not call sk_state_change(),
1887                  * explicitly trigger the socket close machinery
1888                  */
1889                 if (!release && !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW,
1890                                                   &mptcp_sk(sk)->flags))
1891                         mptcp_schedule_work(sk);
1892                 sock_put(sk);
1893         }
1894
1895         mptcp_subflow_ops_undo_override(ssk);
1896         if (release)
1897                 kfree_rcu(ctx, rcu);
1898 }
1899
1900 static void subflow_ulp_clone(const struct request_sock *req,
1901                               struct sock *newsk,
1902                               const gfp_t priority)
1903 {
1904         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1905         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1906         struct mptcp_subflow_context *new_ctx;
1907
1908         if (!tcp_rsk(req)->is_mptcp ||
1909             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1910                 subflow_ulp_fallback(newsk, old_ctx);
1911                 return;
1912         }
1913
1914         new_ctx = subflow_create_ctx(newsk, priority);
1915         if (!new_ctx) {
1916                 subflow_ulp_fallback(newsk, old_ctx);
1917                 return;
1918         }
1919
1920         new_ctx->conn_finished = 1;
1921         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1922         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1923         new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1924         new_ctx->rel_write_seq = 1;
1925         new_ctx->tcp_sock = newsk;
1926
1927         if (subflow_req->mp_capable) {
1928                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1929                  * is fully established only after we receive the remote key
1930                  */
1931                 new_ctx->mp_capable = 1;
1932                 new_ctx->local_key = subflow_req->local_key;
1933                 new_ctx->token = subflow_req->token;
1934                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1935                 new_ctx->idsn = subflow_req->idsn;
1936
1937                 /* this is the first subflow, id is always 0 */
1938                 new_ctx->local_id_valid = 1;
1939         } else if (subflow_req->mp_join) {
1940                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1941                 new_ctx->mp_join = 1;
1942                 new_ctx->fully_established = 1;
1943                 new_ctx->remote_key_valid = 1;
1944                 new_ctx->backup = subflow_req->backup;
1945                 new_ctx->remote_id = subflow_req->remote_id;
1946                 new_ctx->token = subflow_req->token;
1947                 new_ctx->thmac = subflow_req->thmac;
1948
1949                 /* the subflow req id is valid, fetched via subflow_check_req()
1950                  * and subflow_token_join_request()
1951                  */
1952                 subflow_set_local_id(new_ctx, subflow_req->local_id);
1953         }
1954 }
1955
1956 static void tcp_release_cb_override(struct sock *ssk)
1957 {
1958         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1959         long status;
1960
1961         /* process and clear all the pending actions, but leave the subflow into
1962          * the napi queue. To respect locking, only the same CPU that originated
1963          * the action can touch the list. mptcp_napi_poll will take care of it.
1964          */
1965         status = set_mask_bits(&subflow->delegated_status, MPTCP_DELEGATE_ACTIONS_MASK, 0);
1966         if (status)
1967                 mptcp_subflow_process_delegated(ssk, status);
1968
1969         tcp_release_cb(ssk);
1970 }
1971
1972 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1973         .name           = "mptcp",
1974         .owner          = THIS_MODULE,
1975         .init           = subflow_ulp_init,
1976         .release        = subflow_ulp_release,
1977         .clone          = subflow_ulp_clone,
1978 };
1979
1980 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1981 {
1982         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1983
1984         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1985                                               subflow_ops->obj_size, 0,
1986                                               SLAB_ACCOUNT |
1987                                               SLAB_TYPESAFE_BY_RCU,
1988                                               NULL);
1989         if (!subflow_ops->slab)
1990                 return -ENOMEM;
1991
1992         return 0;
1993 }
1994
1995 void __init mptcp_subflow_init(void)
1996 {
1997         mptcp_subflow_v4_request_sock_ops = tcp_request_sock_ops;
1998         mptcp_subflow_v4_request_sock_ops.slab_name = "request_sock_subflow_v4";
1999         mptcp_subflow_v4_request_sock_ops.destructor = subflow_v4_req_destructor;
2000
2001         if (subflow_ops_init(&mptcp_subflow_v4_request_sock_ops) != 0)
2002                 panic("MPTCP: failed to init subflow v4 request sock ops\n");
2003
2004         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
2005         subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
2006         subflow_request_sock_ipv4_ops.send_synack = subflow_v4_send_synack;
2007
2008         subflow_specific = ipv4_specific;
2009         subflow_specific.conn_request = subflow_v4_conn_request;
2010         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
2011         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
2012         subflow_specific.rebuild_header = subflow_rebuild_header;
2013
2014         tcp_prot_override = tcp_prot;
2015         tcp_prot_override.release_cb = tcp_release_cb_override;
2016
2017 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2018         /* In struct mptcp_subflow_request_sock, we assume the TCP request sock
2019          * structures for v4 and v6 have the same size. It should not changed in
2020          * the future but better to make sure to be warned if it is no longer
2021          * the case.
2022          */
2023         BUILD_BUG_ON(sizeof(struct tcp_request_sock) != sizeof(struct tcp6_request_sock));
2024
2025         mptcp_subflow_v6_request_sock_ops = tcp6_request_sock_ops;
2026         mptcp_subflow_v6_request_sock_ops.slab_name = "request_sock_subflow_v6";
2027         mptcp_subflow_v6_request_sock_ops.destructor = subflow_v6_req_destructor;
2028
2029         if (subflow_ops_init(&mptcp_subflow_v6_request_sock_ops) != 0)
2030                 panic("MPTCP: failed to init subflow v6 request sock ops\n");
2031
2032         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
2033         subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
2034         subflow_request_sock_ipv6_ops.send_synack = subflow_v6_send_synack;
2035
2036         subflow_v6_specific = ipv6_specific;
2037         subflow_v6_specific.conn_request = subflow_v6_conn_request;
2038         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
2039         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
2040         subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
2041
2042         subflow_v6m_specific = subflow_v6_specific;
2043         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
2044         subflow_v6m_specific.send_check = ipv4_specific.send_check;
2045         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
2046         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
2047         subflow_v6m_specific.net_frag_header_len = 0;
2048         subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
2049
2050         tcpv6_prot_override = tcpv6_prot;
2051         tcpv6_prot_override.release_cb = tcp_release_cb_override;
2052 #endif
2053
2054         mptcp_diag_subflow_init(&subflow_ulp_ops);
2055
2056         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
2057                 panic("MPTCP: failed to register subflows to ULP\n");
2058 }