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