net/mptcp/subflow.c

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