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_sock_destruct(struct sock *sk)
 606 {
 607         /* if new mptcp socket isn't accepted, it is free'd
 608          * from the tcp listener sockets request queue, linked
 609          * from req->sk.  The tcp socket is released.
 610          * This calls the ULP release function which will
 611          * also remove the mptcp socket, via
 612          * sock_put(ctx->conn).
 613          *
 614          * Problem is that the mptcp socket will be in
 615          * ESTABLISHED state and will not have the SOCK_DEAD flag.
 616          * Both result in warnings from inet_sock_destruct.
 617          */
 618         if ((1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)) {
 619                 sk->sk_state = TCP_CLOSE;
 620                 WARN_ON_ONCE(sk->sk_socket);
 621                 sock_orphan(sk);
 622         }
 623
 624         mptcp_destroy_common(mptcp_sk(sk));
 625         inet_sock_destruct(sk);
 626 }
 627
 628 static void mptcp_force_close(struct sock *sk)
 629 {
 630         /* the msk is not yet exposed to user-space */
 631         inet_sk_state_store(sk, TCP_CLOSE);
 632         sk_common_release(sk);
 633 }
 634
 635 static void subflow_ulp_fallback(struct sock *sk,
 636                                  struct mptcp_subflow_context *old_ctx)
 637 {
 638         struct inet_connection_sock *icsk = inet_csk(sk);
 639
 640         mptcp_subflow_tcp_fallback(sk, old_ctx);
 641         icsk->icsk_ulp_ops = NULL;
 642         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 643         tcp_sk(sk)->is_mptcp = 0;
 644
 645         mptcp_subflow_ops_undo_override(sk);
 646 }
 647
 648 static void subflow_drop_ctx(struct sock *ssk)
 649 {
 650         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 651
 652         if (!ctx)
 653                 return;
 654
 655         subflow_ulp_fallback(ssk, ctx);
 656         if (ctx->conn)
 657                 sock_put(ctx->conn);
 658
 659         kfree_rcu(ctx, rcu);
 660 }
 661
 662 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 663                                      struct mptcp_options_received *mp_opt)
 664 {
 665         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 666
 667         subflow->remote_key = mp_opt->sndr_key;
 668         subflow->fully_established = 1;
 669         subflow->can_ack = 1;
 670         WRITE_ONCE(msk->fully_established, true);
 671 }
 672
 673 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 674                                           struct sk_buff *skb,
 675                                           struct request_sock *req,
 676                                           struct dst_entry *dst,
 677                                           struct request_sock *req_unhash,
 678                                           bool *own_req)
 679 {
 680         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 681         struct mptcp_subflow_request_sock *subflow_req;
 682         struct mptcp_options_received mp_opt;
 683         bool fallback, fallback_is_fatal;
 684         struct sock *new_msk = NULL;
 685         struct sock *child;
 686
 687         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 688
 689         /* After child creation we must look for MPC even when options
 690          * are not parsed
 691          */
 692         mp_opt.suboptions = 0;
 693
 694         /* hopefully temporary handling for MP_JOIN+syncookie */
 695         subflow_req = mptcp_subflow_rsk(req);
 696         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 697         fallback = !tcp_rsk(req)->is_mptcp;
 698         if (fallback)
 699                 goto create_child;
 700
 701         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 702         if (subflow_req->mp_capable) {
 703                 /* we can receive and accept an in-window, out-of-order pkt,
 704                  * which may not carry the MP_CAPABLE opt even on mptcp enabled
 705                  * paths: always try to extract the peer key, and fallback
 706                  * for packets missing it.
 707                  * Even OoO DSS packets coming legitly after dropped or
 708                  * reordered MPC will cause fallback, but we don't have other
 709                  * options.
 710                  */
 711                 mptcp_get_options(skb, &mp_opt);
 712                 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPC)) {
 713                         fallback = true;
 714                         goto create_child;
 715                 }
 716
 717                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 718                 if (!new_msk)
 719                         fallback = true;
 720         } else if (subflow_req->mp_join) {
 721                 mptcp_get_options(skb, &mp_opt);
 722                 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ) ||
 723                     !subflow_hmac_valid(req, &mp_opt) ||
 724                     !mptcp_can_accept_new_subflow(subflow_req->msk)) {
 725                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 726                         fallback = true;
 727                 }
 728         }
 729
 730 create_child:
 731         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 732                                                      req_unhash, own_req);
 733
 734         if (child && *own_req) {
 735                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 736
 737                 tcp_rsk(req)->drop_req = false;
 738
 739                 /* we need to fallback on ctx allocation failure and on pre-reqs
 740                  * checking above. In the latter scenario we additionally need
 741                  * to reset the context to non MPTCP status.
 742                  */
 743                 if (!ctx || fallback) {
 744                         if (fallback_is_fatal) {
 745                                 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 746                                 goto dispose_child;
 747                         }
 748
 749                         subflow_drop_ctx(child);
 750                         goto out;
 751                 }
 752
 753                 /* ssk inherits options of listener sk */
 754                 ctx->setsockopt_seq = listener->setsockopt_seq;
 755
 756                 if (ctx->mp_capable) {
 757                         /* this can't race with mptcp_close(), as the msk is
 758                          * not yet exposted to user-space
 759                          */
 760                         inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
 761
 762                         /* record the newly created socket as the first msk
 763                          * subflow, but don't link it yet into conn_list
 764                          */
 765                         WRITE_ONCE(mptcp_sk(new_msk)->first, child);
 766
 767                         /* new mpc subflow takes ownership of the newly
 768                          * created mptcp socket
 769                          */
 770                         new_msk->sk_destruct = mptcp_sock_destruct;
 771                         mptcp_sk(new_msk)->setsockopt_seq = ctx->setsockopt_seq;
 772                         mptcp_pm_new_connection(mptcp_sk(new_msk), child, 1);
 773                         mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
 774                         ctx->conn = new_msk;
 775                         new_msk = NULL;
 776
 777                         /* with OoO packets we can reach here without ingress
 778                          * mpc option
 779                          */
 780                         if (mp_opt.suboptions & OPTIONS_MPTCP_MPC)
 781                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 782                 } else if (ctx->mp_join) {
 783                         struct mptcp_sock *owner;
 784
 785                         owner = subflow_req->msk;
 786                         if (!owner) {
 787                                 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 788                                 goto dispose_child;
 789                         }
 790
 791                         /* move the msk reference ownership to the subflow */
 792                         subflow_req->msk = NULL;
 793                         ctx->conn = (struct sock *)owner;
 794
 795                         if (subflow_use_different_sport(owner, sk)) {
 796                                 pr_debug("ack inet_sport=%d %d",
 797                                          ntohs(inet_sk(sk)->inet_sport),
 798                                          ntohs(inet_sk((struct sock *)owner)->inet_sport));
 799                                 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
 800                                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
 801                                         goto dispose_child;
 802                                 }
 803                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
 804                         }
 805
 806                         if (!mptcp_finish_join(child))
 807                                 goto dispose_child;
 808
 809                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 810                         tcp_rsk(req)->drop_req = true;
 811                 }
 812         }
 813
 814 out:
 815         /* dispose of the left over mptcp master, if any */
 816         if (unlikely(new_msk))
 817                 mptcp_force_close(new_msk);
 818
 819         /* check for expected invariant - should never trigger, just help
 820          * catching eariler subtle bugs
 821          */
 822         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 823                      (!mptcp_subflow_ctx(child) ||
 824                       !mptcp_subflow_ctx(child)->conn));
 825         return child;
 826
 827 dispose_child:
 828         subflow_drop_ctx(child);
 829         tcp_rsk(req)->drop_req = true;
 830         inet_csk_prepare_for_destroy_sock(child);
 831         tcp_done(child);
 832         req->rsk_ops->send_reset(sk, skb);
 833
 834         /* The last child reference will be released by the caller */
 835         return child;
 836 }
 837
 838 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
 839 static struct proto tcp_prot_override;
 840
 841 enum mapping_status {
 842         MAPPING_OK,
 843         MAPPING_INVALID,
 844         MAPPING_EMPTY,
 845         MAPPING_DATA_FIN,
 846         MAPPING_DUMMY,
 847         MAPPING_BAD_CSUM
 848 };
 849
 850 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 851 {
 852         pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 853                  ssn, subflow->map_subflow_seq, subflow->map_data_len);
 854 }
 855
 856 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 857 {
 858         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 859         unsigned int skb_consumed;
 860
 861         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 862         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 863                 return true;
 864
 865         return skb->len - skb_consumed <= subflow->map_data_len -
 866                                           mptcp_subflow_get_map_offset(subflow);
 867 }
 868
 869 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 870 {
 871         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 872         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 873
 874         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 875                 /* Mapping covers data later in the subflow stream,
 876                  * currently unsupported.
 877                  */
 878                 dbg_bad_map(subflow, ssn);
 879                 return false;
 880         }
 881         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 882                                   subflow->map_data_len))) {
 883                 /* Mapping does covers past subflow data, invalid */
 884                 dbg_bad_map(subflow, ssn);
 885                 return false;
 886         }
 887         return true;
 888 }
 889
 890 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
 891                                               bool csum_reqd)
 892 {
 893         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 894         u32 offset, seq, delta;
 895         __sum16 csum;
 896         int len;
 897
 898         if (!csum_reqd)
 899                 return MAPPING_OK;
 900
 901         /* mapping already validated on previous traversal */
 902         if (subflow->map_csum_len == subflow->map_data_len)
 903                 return MAPPING_OK;
 904
 905         /* traverse the receive queue, ensuring it contains a full
 906          * DSS mapping and accumulating the related csum.
 907          * Preserve the accoumlate csum across multiple calls, to compute
 908          * the csum only once
 909          */
 910         delta = subflow->map_data_len - subflow->map_csum_len;
 911         for (;;) {
 912                 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
 913                 offset = seq - TCP_SKB_CB(skb)->seq;
 914
 915                 /* if the current skb has not been accounted yet, csum its contents
 916                  * up to the amount covered by the current DSS
 917                  */
 918                 if (offset < skb->len) {
 919                         __wsum csum;
 920
 921                         len = min(skb->len - offset, delta);
 922                         csum = skb_checksum(skb, offset, len, 0);
 923                         subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
 924                                                                 subflow->map_csum_len);
 925
 926                         delta -= len;
 927                         subflow->map_csum_len += len;
 928                 }
 929                 if (delta == 0)
 930                         break;
 931
 932                 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
 933                         /* if this subflow is closed, the partial mapping
 934                          * will be never completed; flush the pending skbs, so
 935                          * that subflow_sched_work_if_closed() can kick in
 936                          */
 937                         if (unlikely(ssk->sk_state == TCP_CLOSE))
 938                                 while ((skb = skb_peek(&ssk->sk_receive_queue)))
 939                                         sk_eat_skb(ssk, skb);
 940
 941                         /* not enough data to validate the csum */
 942                         return MAPPING_EMPTY;
 943                 }
 944
 945                 /* the DSS mapping for next skbs will be validated later,
 946                  * when a get_mapping_status call will process such skb
 947                  */
 948                 skb = skb->next;
 949         }
 950
 951         /* note that 'map_data_len' accounts only for the carried data, does
 952          * not include the eventual seq increment due to the data fin,
 953          * while the pseudo header requires the original DSS data len,
 954          * including that
 955          */
 956         csum = __mptcp_make_csum(subflow->map_seq,
 957                                  subflow->map_subflow_seq,
 958                                  subflow->map_data_len + subflow->map_data_fin,
 959                                  subflow->map_data_csum);
 960         if (unlikely(csum)) {
 961                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
 962                 return MAPPING_BAD_CSUM;
 963         }
 964
 965         subflow->valid_csum_seen = 1;
 966         return MAPPING_OK;
 967 }
 968
 969 static enum mapping_status get_mapping_status(struct sock *ssk,
 970                                               struct mptcp_sock *msk)
 971 {
 972         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 973         bool csum_reqd = READ_ONCE(msk->csum_enabled);
 974         struct mptcp_ext *mpext;
 975         struct sk_buff *skb;
 976         u16 data_len;
 977         u64 map_seq;
 978
 979         skb = skb_peek(&ssk->sk_receive_queue);
 980         if (!skb)
 981                 return MAPPING_EMPTY;
 982
 983         if (mptcp_check_fallback(ssk))
 984                 return MAPPING_DUMMY;
 985
 986         mpext = mptcp_get_ext(skb);
 987         if (!mpext || !mpext->use_map) {
 988                 if (!subflow->map_valid && !skb->len) {
 989                         /* the TCP stack deliver 0 len FIN pkt to the receive
 990                          * queue, that is the only 0len pkts ever expected here,
 991                          * and we can admit no mapping only for 0 len pkts
 992                          */
 993                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 994                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 995                                           TCP_SKB_CB(skb)->seq,
 996                                           TCP_SKB_CB(skb)->end_seq,
 997                                           TCP_SKB_CB(skb)->tcp_flags);
 998                         sk_eat_skb(ssk, skb);
 999                         return MAPPING_EMPTY;
1000                 }
1001
1002                 if (!subflow->map_valid)
1003                         return MAPPING_INVALID;
1004
1005                 goto validate_seq;
1006         }
1007
1008         trace_get_mapping_status(mpext);
1009
1010         data_len = mpext->data_len;
1011         if (data_len == 0) {
1012                 pr_debug("infinite mapping received");
1013                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1014                 subflow->map_data_len = 0;
1015                 return MAPPING_INVALID;
1016         }
1017
1018         if (mpext->data_fin == 1) {
1019                 if (data_len == 1) {
1020                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1021                                                                  mpext->dsn64);
1022                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
1023                         if (subflow->map_valid) {
1024                                 /* A DATA_FIN might arrive in a DSS
1025                                  * option before the previous mapping
1026                                  * has been fully consumed. Continue
1027                                  * handling the existing mapping.
1028                                  */
1029                                 skb_ext_del(skb, SKB_EXT_MPTCP);
1030                                 return MAPPING_OK;
1031                         } else {
1032                                 if (updated && schedule_work(&msk->work))
1033                                         sock_hold((struct sock *)msk);
1034
1035                                 return MAPPING_DATA_FIN;
1036                         }
1037                 } else {
1038                         u64 data_fin_seq = mpext->data_seq + data_len - 1;
1039
1040                         /* If mpext->data_seq is a 32-bit value, data_fin_seq
1041                          * must also be limited to 32 bits.
1042                          */
1043                         if (!mpext->dsn64)
1044                                 data_fin_seq &= GENMASK_ULL(31, 0);
1045
1046                         mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1047                         pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
1048                                  data_fin_seq, mpext->dsn64);
1049                 }
1050
1051                 /* Adjust for DATA_FIN using 1 byte of sequence space */
1052                 data_len--;
1053         }
1054
1055         map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1056         WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1057
1058         if (subflow->map_valid) {
1059                 /* Allow replacing only with an identical map */
1060                 if (subflow->map_seq == map_seq &&
1061                     subflow->map_subflow_seq == mpext->subflow_seq &&
1062                     subflow->map_data_len == data_len &&
1063                     subflow->map_csum_reqd == mpext->csum_reqd) {
1064                         skb_ext_del(skb, SKB_EXT_MPTCP);
1065                         goto validate_csum;
1066                 }
1067
1068                 /* If this skb data are fully covered by the current mapping,
1069                  * the new map would need caching, which is not supported
1070                  */
1071                 if (skb_is_fully_mapped(ssk, skb)) {
1072                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1073                         return MAPPING_INVALID;
1074                 }
1075
1076                 /* will validate the next map after consuming the current one */
1077                 goto validate_csum;
1078         }
1079
1080         subflow->map_seq = map_seq;
1081         subflow->map_subflow_seq = mpext->subflow_seq;
1082         subflow->map_data_len = data_len;
1083         subflow->map_valid = 1;
1084         subflow->map_data_fin = mpext->data_fin;
1085         subflow->mpc_map = mpext->mpc_map;
1086         subflow->map_csum_reqd = mpext->csum_reqd;
1087         subflow->map_csum_len = 0;
1088         subflow->map_data_csum = csum_unfold(mpext->csum);
1089
1090         /* Cfr RFC 8684 Section 3.3.0 */
1091         if (unlikely(subflow->map_csum_reqd != csum_reqd))
1092                 return MAPPING_INVALID;
1093
1094         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
1095                  subflow->map_seq, subflow->map_subflow_seq,
1096                  subflow->map_data_len, subflow->map_csum_reqd,
1097                  subflow->map_data_csum);
1098
1099 validate_seq:
1100         /* we revalidate valid mapping on new skb, because we must ensure
1101          * the current skb is completely covered by the available mapping
1102          */
1103         if (!validate_mapping(ssk, skb)) {
1104                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1105                 return MAPPING_INVALID;
1106         }
1107
1108         skb_ext_del(skb, SKB_EXT_MPTCP);
1109
1110 validate_csum:
1111         return validate_data_csum(ssk, skb, csum_reqd);
1112 }
1113
1114 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1115                                        u64 limit)
1116 {
1117         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1118         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1119         u32 incr;
1120
1121         incr = limit >= skb->len ? skb->len + fin : limit;
1122
1123         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
1124                  subflow->map_subflow_seq);
1125         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1126         tcp_sk(ssk)->copied_seq += incr;
1127         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1128                 sk_eat_skb(ssk, skb);
1129         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1130                 subflow->map_valid = 0;
1131 }
1132
1133 /* sched mptcp worker to remove the subflow if no more data is pending */
1134 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1135 {
1136         struct sock *sk = (struct sock *)msk;
1137
1138         if (likely(ssk->sk_state != TCP_CLOSE))
1139                 return;
1140
1141         if (skb_queue_empty(&ssk->sk_receive_queue) &&
1142             !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags)) {
1143                 sock_hold(sk);
1144                 if (!schedule_work(&msk->work))
1145                         sock_put(sk);
1146         }
1147 }
1148
1149 static bool subflow_can_fallback(struct mptcp_subflow_context *subflow)
1150 {
1151         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1152
1153         if (subflow->mp_join)
1154                 return false;
1155         else if (READ_ONCE(msk->csum_enabled))
1156                 return !subflow->valid_csum_seen;
1157         else
1158                 return !subflow->fully_established;
1159 }
1160
1161 static void mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk)
1162 {
1163         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1164         unsigned long fail_tout;
1165
1166         /* greceful failure can happen only on the MPC subflow */
1167         if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first)))
1168                 return;
1169
1170         /* since the close timeout take precedence on the fail one,
1171          * no need to start the latter when the first is already set
1172          */
1173         if (sock_flag((struct sock *)msk, SOCK_DEAD))
1174                 return;
1175
1176         /* we don't need extreme accuracy here, use a zero fail_tout as special
1177          * value meaning no fail timeout at all;
1178          */
1179         fail_tout = jiffies + TCP_RTO_MAX;
1180         if (!fail_tout)
1181                 fail_tout = 1;
1182         WRITE_ONCE(subflow->fail_tout, fail_tout);
1183         tcp_send_ack(ssk);
1184
1185         mptcp_reset_timeout(msk, subflow->fail_tout);
1186 }
1187
1188 static bool subflow_check_data_avail(struct sock *ssk)
1189 {
1190         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1191         enum mapping_status status;
1192         struct mptcp_sock *msk;
1193         struct sk_buff *skb;
1194
1195         if (!skb_peek(&ssk->sk_receive_queue))
1196                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1197         if (subflow->data_avail)
1198                 return true;
1199
1200         msk = mptcp_sk(subflow->conn);
1201         for (;;) {
1202                 u64 ack_seq;
1203                 u64 old_ack;
1204
1205                 status = get_mapping_status(ssk, msk);
1206                 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1207                 if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY ||
1208                              status == MAPPING_BAD_CSUM))
1209                         goto fallback;
1210
1211                 if (status != MAPPING_OK)
1212                         goto no_data;
1213
1214                 skb = skb_peek(&ssk->sk_receive_queue);
1215                 if (WARN_ON_ONCE(!skb))
1216                         goto no_data;
1217
1218                 /* if msk lacks the remote key, this subflow must provide an
1219                  * MP_CAPABLE-based mapping
1220                  */
1221                 if (unlikely(!READ_ONCE(msk->can_ack))) {
1222                         if (!subflow->mpc_map)
1223                                 goto fallback;
1224                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
1225                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
1226                         WRITE_ONCE(msk->can_ack, true);
1227                 }
1228
1229                 old_ack = READ_ONCE(msk->ack_seq);
1230                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1231                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1232                          ack_seq);
1233                 if (unlikely(before64(ack_seq, old_ack))) {
1234                         mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1235                         continue;
1236                 }
1237
1238                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1239                 break;
1240         }
1241         return true;
1242
1243 no_data:
1244         subflow_sched_work_if_closed(msk, ssk);
1245         return false;
1246
1247 fallback:
1248         if (!__mptcp_check_fallback(msk)) {
1249                 /* RFC 8684 section 3.7. */
1250                 if (status == MAPPING_BAD_CSUM &&
1251                     (subflow->mp_join || subflow->valid_csum_seen)) {
1252                         subflow->send_mp_fail = 1;
1253
1254                         if (!READ_ONCE(msk->allow_infinite_fallback)) {
1255                                 subflow->reset_transient = 0;
1256                                 subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1257                                 goto reset;
1258                         }
1259                         mptcp_subflow_fail(msk, ssk);
1260                         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1261                         return true;
1262                 }
1263
1264                 if (!subflow_can_fallback(subflow) && subflow->map_data_len) {
1265                         /* fatal protocol error, close the socket.
1266                          * subflow_error_report() will introduce the appropriate barriers
1267                          */
1268                         subflow->reset_transient = 0;
1269                         subflow->reset_reason = MPTCP_RST_EMPTCP;
1270
1271 reset:
1272                         ssk->sk_err = EBADMSG;
1273                         tcp_set_state(ssk, TCP_CLOSE);
1274                         while ((skb = skb_peek(&ssk->sk_receive_queue)))
1275                                 sk_eat_skb(ssk, skb);
1276                         tcp_send_active_reset(ssk, GFP_ATOMIC);
1277                         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1278                         return false;
1279                 }
1280
1281                 mptcp_do_fallback(ssk);
1282         }
1283
1284         skb = skb_peek(&ssk->sk_receive_queue);
1285         subflow->map_valid = 1;
1286         subflow->map_seq = READ_ONCE(msk->ack_seq);
1287         subflow->map_data_len = skb->len;
1288         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1289         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1290         return true;
1291 }
1292
1293 bool mptcp_subflow_data_available(struct sock *sk)
1294 {
1295         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1296
1297         /* check if current mapping is still valid */
1298         if (subflow->map_valid &&
1299             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1300                 subflow->map_valid = 0;
1301                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1302
1303                 pr_debug("Done with mapping: seq=%u data_len=%u",
1304                          subflow->map_subflow_seq,
1305                          subflow->map_data_len);
1306         }
1307
1308         return subflow_check_data_avail(sk);
1309 }
1310
1311 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1312  * not the ssk one.
1313  *
1314  * In mptcp, rwin is about the mptcp-level connection data.
1315  *
1316  * Data that is still on the ssk rx queue can thus be ignored,
1317  * as far as mptcp peer is concerned that data is still inflight.
1318  * DSS ACK is updated when skb is moved to the mptcp rx queue.
1319  */
1320 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1321 {
1322         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1323         const struct sock *sk = subflow->conn;
1324
1325         *space = __mptcp_space(sk);
1326         *full_space = tcp_full_space(sk);
1327 }
1328
1329 void __mptcp_error_report(struct sock *sk)
1330 {
1331         struct mptcp_subflow_context *subflow;
1332         struct mptcp_sock *msk = mptcp_sk(sk);
1333
1334         mptcp_for_each_subflow(msk, subflow) {
1335                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1336                 int err = sock_error(ssk);
1337
1338                 if (!err)
1339                         continue;
1340
1341                 /* only propagate errors on fallen-back sockets or
1342                  * on MPC connect
1343                  */
1344                 if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(msk))
1345                         continue;
1346
1347                 inet_sk_state_store(sk, inet_sk_state_load(ssk));
1348                 sk->sk_err = -err;
1349
1350                 /* This barrier is coupled with smp_rmb() in mptcp_poll() */
1351                 smp_wmb();
1352                 sk_error_report(sk);
1353                 break;
1354         }
1355 }
1356
1357 static void subflow_error_report(struct sock *ssk)
1358 {
1359         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1360
1361         mptcp_data_lock(sk);
1362         if (!sock_owned_by_user(sk))
1363                 __mptcp_error_report(sk);
1364         else
1365                 __set_bit(MPTCP_ERROR_REPORT,  &mptcp_sk(sk)->cb_flags);
1366         mptcp_data_unlock(sk);
1367 }
1368
1369 static void subflow_data_ready(struct sock *sk)
1370 {
1371         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1372         u16 state = 1 << inet_sk_state_load(sk);
1373         struct sock *parent = subflow->conn;
1374         struct mptcp_sock *msk;
1375
1376         msk = mptcp_sk(parent);
1377         if (state & TCPF_LISTEN) {
1378                 /* MPJ subflow are removed from accept queue before reaching here,
1379                  * avoid stray wakeups
1380                  */
1381                 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1382                         return;
1383
1384                 parent->sk_data_ready(parent);
1385                 return;
1386         }
1387
1388         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1389                      !subflow->mp_join && !(state & TCPF_CLOSE));
1390
1391         if (mptcp_subflow_data_available(sk))
1392                 mptcp_data_ready(parent, sk);
1393         else if (unlikely(sk->sk_err))
1394                 subflow_error_report(sk);
1395 }
1396
1397 static void subflow_write_space(struct sock *ssk)
1398 {
1399         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1400
1401         mptcp_propagate_sndbuf(sk, ssk);
1402         mptcp_write_space(sk);
1403 }
1404
1405 static const struct inet_connection_sock_af_ops *
1406 subflow_default_af_ops(struct sock *sk)
1407 {
1408 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1409         if (sk->sk_family == AF_INET6)
1410                 return &subflow_v6_specific;
1411 #endif
1412         return &subflow_specific;
1413 }
1414
1415 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1416 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1417 {
1418         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1419         struct inet_connection_sock *icsk = inet_csk(sk);
1420         const struct inet_connection_sock_af_ops *target;
1421
1422         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1423
1424         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1425                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1426
1427         if (likely(icsk->icsk_af_ops == target))
1428                 return;
1429
1430         subflow->icsk_af_ops = icsk->icsk_af_ops;
1431         icsk->icsk_af_ops = target;
1432 }
1433 #endif
1434
1435 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1436                          struct sockaddr_storage *addr,
1437                          unsigned short family)
1438 {
1439         memset(addr, 0, sizeof(*addr));
1440         addr->ss_family = family;
1441         if (addr->ss_family == AF_INET) {
1442                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1443
1444                 if (info->family == AF_INET)
1445                         in_addr->sin_addr = info->addr;
1446 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1447                 else if (ipv6_addr_v4mapped(&info->addr6))
1448                         in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1449 #endif
1450                 in_addr->sin_port = info->port;
1451         }
1452 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1453         else if (addr->ss_family == AF_INET6) {
1454                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1455
1456                 if (info->family == AF_INET)
1457                         ipv6_addr_set_v4mapped(info->addr.s_addr,
1458                                                &in6_addr->sin6_addr);
1459                 else
1460                         in6_addr->sin6_addr = info->addr6;
1461                 in6_addr->sin6_port = info->port;
1462         }
1463 #endif
1464 }
1465
1466 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1467                             const struct mptcp_addr_info *remote)
1468 {
1469         struct mptcp_sock *msk = mptcp_sk(sk);
1470         struct mptcp_subflow_context *subflow;
1471         struct sockaddr_storage addr;
1472         int remote_id = remote->id;
1473         int local_id = loc->id;
1474         int err = -ENOTCONN;
1475         struct socket *sf;
1476         struct sock *ssk;
1477         u32 remote_token;
1478         int addrlen;
1479         int ifindex;
1480         u8 flags;
1481
1482         if (!mptcp_is_fully_established(sk))
1483                 goto err_out;
1484
1485         err = mptcp_subflow_create_socket(sk, &sf);
1486         if (err)
1487                 goto err_out;
1488
1489         ssk = sf->sk;
1490         subflow = mptcp_subflow_ctx(ssk);
1491         do {
1492                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1493         } while (!subflow->local_nonce);
1494
1495         if (local_id)
1496                 subflow_set_local_id(subflow, local_id);
1497
1498         mptcp_pm_get_flags_and_ifindex_by_id(msk, local_id,
1499                                              &flags, &ifindex);
1500         subflow->remote_key = msk->remote_key;
1501         subflow->local_key = msk->local_key;
1502         subflow->token = msk->token;
1503         mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1504
1505         addrlen = sizeof(struct sockaddr_in);
1506 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1507         if (addr.ss_family == AF_INET6)
1508                 addrlen = sizeof(struct sockaddr_in6);
1509 #endif
1510         mptcp_sockopt_sync(msk, ssk);
1511
1512         ssk->sk_bound_dev_if = ifindex;
1513         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1514         if (err)
1515                 goto failed;
1516
1517         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1518         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1519                  remote_token, local_id, remote_id);
1520         subflow->remote_token = remote_token;
1521         subflow->remote_id = remote_id;
1522         subflow->request_join = 1;
1523         subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1524         mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1525
1526         sock_hold(ssk);
1527         list_add_tail(&subflow->node, &msk->conn_list);
1528         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1529         if (err && err != -EINPROGRESS)
1530                 goto failed_unlink;
1531
1532         /* discard the subflow socket */
1533         mptcp_sock_graft(ssk, sk->sk_socket);
1534         iput(SOCK_INODE(sf));
1535         WRITE_ONCE(msk->allow_infinite_fallback, false);
1536         return 0;
1537
1538 failed_unlink:
1539         list_del(&subflow->node);
1540         sock_put(mptcp_subflow_tcp_sock(subflow));
1541
1542 failed:
1543         subflow->disposable = 1;
1544         sock_release(sf);
1545
1546 err_out:
1547         /* we account subflows before the creation, and this failures will not
1548          * be caught by sk_state_change()
1549          */
1550         mptcp_pm_close_subflow(msk);
1551         return err;
1552 }
1553
1554 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1555 {
1556 #ifdef CONFIG_SOCK_CGROUP_DATA
1557         struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1558                                 *child_skcd = &child->sk_cgrp_data;
1559
1560         /* only the additional subflows created by kworkers have to be modified */
1561         if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1562             cgroup_id(sock_cgroup_ptr(child_skcd))) {
1563 #ifdef CONFIG_MEMCG
1564                 struct mem_cgroup *memcg = parent->sk_memcg;
1565
1566                 mem_cgroup_sk_free(child);
1567                 if (memcg && css_tryget(&memcg->css))
1568                         child->sk_memcg = memcg;
1569 #endif /* CONFIG_MEMCG */
1570
1571                 cgroup_sk_free(child_skcd);
1572                 *child_skcd = *parent_skcd;
1573                 cgroup_sk_clone(child_skcd);
1574         }
1575 #endif /* CONFIG_SOCK_CGROUP_DATA */
1576 }
1577
1578 static void mptcp_subflow_ops_override(struct sock *ssk)
1579 {
1580 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1581         if (ssk->sk_prot == &tcpv6_prot)
1582                 ssk->sk_prot = &tcpv6_prot_override;
1583         else
1584 #endif
1585                 ssk->sk_prot = &tcp_prot_override;
1586 }
1587
1588 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1589 {
1590 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1591         if (ssk->sk_prot == &tcpv6_prot_override)
1592                 ssk->sk_prot = &tcpv6_prot;
1593         else
1594 #endif
1595                 ssk->sk_prot = &tcp_prot;
1596 }
1597 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1598 {
1599         struct mptcp_subflow_context *subflow;
1600         struct net *net = sock_net(sk);
1601         struct socket *sf;
1602         int err;
1603
1604         /* un-accepted server sockets can reach here - on bad configuration
1605          * bail early to avoid greater trouble later
1606          */
1607         if (unlikely(!sk->sk_socket))
1608                 return -EINVAL;
1609
1610         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1611                                &sf);
1612         if (err)
1613                 return err;
1614
1615         lock_sock(sf->sk);
1616
1617         /* the newly created socket has to be in the same cgroup as its parent */
1618         mptcp_attach_cgroup(sk, sf->sk);
1619
1620         /* kernel sockets do not by default acquire net ref, but TCP timer
1621          * needs it.
1622          */
1623         sf->sk->sk_net_refcnt = 1;
1624         get_net_track(net, &sf->sk->ns_tracker, GFP_KERNEL);
1625         sock_inuse_add(net, 1);
1626         err = tcp_set_ulp(sf->sk, "mptcp");
1627         release_sock(sf->sk);
1628
1629         if (err) {
1630                 sock_release(sf);
1631                 return err;
1632         }
1633
1634         /* the newly created socket really belongs to the owning MPTCP master
1635          * socket, even if for additional subflows the allocation is performed
1636          * by a kernel workqueue. Adjust inode references, so that the
1637          * procfs/diag interfaces really show this one belonging to the correct
1638          * user.
1639          */
1640         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1641         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1642         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1643
1644         subflow = mptcp_subflow_ctx(sf->sk);
1645         pr_debug("subflow=%p", subflow);
1646
1647         *new_sock = sf;
1648         sock_hold(sk);
1649         subflow->conn = sk;
1650         mptcp_subflow_ops_override(sf->sk);
1651
1652         return 0;
1653 }
1654
1655 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1656                                                         gfp_t priority)
1657 {
1658         struct inet_connection_sock *icsk = inet_csk(sk);
1659         struct mptcp_subflow_context *ctx;
1660
1661         ctx = kzalloc(sizeof(*ctx), priority);
1662         if (!ctx)
1663                 return NULL;
1664
1665         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1666         INIT_LIST_HEAD(&ctx->node);
1667         INIT_LIST_HEAD(&ctx->delegated_node);
1668
1669         pr_debug("subflow=%p", ctx);
1670
1671         ctx->tcp_sock = sk;
1672
1673         return ctx;
1674 }
1675
1676 static void __subflow_state_change(struct sock *sk)
1677 {
1678         struct socket_wq *wq;
1679
1680         rcu_read_lock();
1681         wq = rcu_dereference(sk->sk_wq);
1682         if (skwq_has_sleeper(wq))
1683                 wake_up_interruptible_all(&wq->wait);
1684         rcu_read_unlock();
1685 }
1686
1687 static bool subflow_is_done(const struct sock *sk)
1688 {
1689         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1690 }
1691
1692 static void subflow_state_change(struct sock *sk)
1693 {
1694         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1695         struct sock *parent = subflow->conn;
1696
1697         __subflow_state_change(sk);
1698
1699         if (subflow_simultaneous_connect(sk)) {
1700                 mptcp_propagate_sndbuf(parent, sk);
1701                 mptcp_do_fallback(sk);
1702                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1703                 pr_fallback(mptcp_sk(parent));
1704                 subflow->conn_finished = 1;
1705                 mptcp_set_connected(parent);
1706         }
1707
1708         /* as recvmsg() does not acquire the subflow socket for ssk selection
1709          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1710          * the data available machinery here.
1711          */
1712         if (mptcp_subflow_data_available(sk))
1713                 mptcp_data_ready(parent, sk);
1714         else if (unlikely(sk->sk_err))
1715                 subflow_error_report(sk);
1716
1717         subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1718
1719         if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1720             !subflow->rx_eof && subflow_is_done(sk)) {
1721                 subflow->rx_eof = 1;
1722                 mptcp_subflow_eof(parent);
1723         }
1724 }
1725
1726 void mptcp_subflow_queue_clean(struct sock *listener_ssk)
1727 {
1728         struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
1729         struct mptcp_sock *msk, *next, *head = NULL;
1730         struct request_sock *req;
1731
1732         /* build a list of all unaccepted mptcp sockets */
1733         spin_lock_bh(&queue->rskq_lock);
1734         for (req = queue->rskq_accept_head; req; req = req->dl_next) {
1735                 struct mptcp_subflow_context *subflow;
1736                 struct sock *ssk = req->sk;
1737                 struct mptcp_sock *msk;
1738
1739                 if (!sk_is_mptcp(ssk))
1740                         continue;
1741
1742                 subflow = mptcp_subflow_ctx(ssk);
1743                 if (!subflow || !subflow->conn)
1744                         continue;
1745
1746                 /* skip if already in list */
1747                 msk = mptcp_sk(subflow->conn);
1748                 if (msk->dl_next || msk == head)
1749                         continue;
1750
1751                 msk->dl_next = head;
1752                 head = msk;
1753         }
1754         spin_unlock_bh(&queue->rskq_lock);
1755         if (!head)
1756                 return;
1757
1758         /* can't acquire the msk socket lock under the subflow one,
1759          * or will cause ABBA deadlock
1760          */
1761         release_sock(listener_ssk);
1762
1763         for (msk = head; msk; msk = next) {
1764                 struct sock *sk = (struct sock *)msk;
1765                 bool slow;
1766
1767                 slow = lock_sock_fast_nested(sk);
1768                 next = msk->dl_next;
1769                 msk->first = NULL;
1770                 msk->dl_next = NULL;
1771                 unlock_sock_fast(sk, slow);
1772         }
1773
1774         /* we are still under the listener msk socket lock */
1775         lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
1776 }
1777
1778 static int subflow_ulp_init(struct sock *sk)
1779 {
1780         struct inet_connection_sock *icsk = inet_csk(sk);
1781         struct mptcp_subflow_context *ctx;
1782         struct tcp_sock *tp = tcp_sk(sk);
1783         int err = 0;
1784
1785         /* disallow attaching ULP to a socket unless it has been
1786          * created with sock_create_kern()
1787          */
1788         if (!sk->sk_kern_sock) {
1789                 err = -EOPNOTSUPP;
1790                 goto out;
1791         }
1792
1793         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1794         if (!ctx) {
1795                 err = -ENOMEM;
1796                 goto out;
1797         }
1798
1799         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1800
1801         tp->is_mptcp = 1;
1802         ctx->icsk_af_ops = icsk->icsk_af_ops;
1803         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1804         ctx->tcp_state_change = sk->sk_state_change;
1805         ctx->tcp_error_report = sk->sk_error_report;
1806
1807         WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1808         WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
1809
1810         sk->sk_data_ready = subflow_data_ready;
1811         sk->sk_write_space = subflow_write_space;
1812         sk->sk_state_change = subflow_state_change;
1813         sk->sk_error_report = subflow_error_report;
1814 out:
1815         return err;
1816 }
1817
1818 static void subflow_ulp_release(struct sock *ssk)
1819 {
1820         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1821         bool release = true;
1822         struct sock *sk;
1823
1824         if (!ctx)
1825                 return;
1826
1827         sk = ctx->conn;
1828         if (sk) {
1829                 /* if the msk has been orphaned, keep the ctx
1830                  * alive, will be freed by __mptcp_close_ssk(),
1831                  * when the subflow is still unaccepted
1832                  */
1833                 release = ctx->disposable || list_empty(&ctx->node);
1834                 sock_put(sk);
1835         }
1836
1837         mptcp_subflow_ops_undo_override(ssk);
1838         if (release)
1839                 kfree_rcu(ctx, rcu);
1840 }
1841
1842 static void subflow_ulp_clone(const struct request_sock *req,
1843                               struct sock *newsk,
1844                               const gfp_t priority)
1845 {
1846         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1847         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1848         struct mptcp_subflow_context *new_ctx;
1849
1850         if (!tcp_rsk(req)->is_mptcp ||
1851             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1852                 subflow_ulp_fallback(newsk, old_ctx);
1853                 return;
1854         }
1855
1856         new_ctx = subflow_create_ctx(newsk, priority);
1857         if (!new_ctx) {
1858                 subflow_ulp_fallback(newsk, old_ctx);
1859                 return;
1860         }
1861
1862         new_ctx->conn_finished = 1;
1863         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1864         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1865         new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1866         new_ctx->rel_write_seq = 1;
1867         new_ctx->tcp_sock = newsk;
1868
1869         if (subflow_req->mp_capable) {
1870                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1871                  * is fully established only after we receive the remote key
1872                  */
1873                 new_ctx->mp_capable = 1;
1874                 new_ctx->local_key = subflow_req->local_key;
1875                 new_ctx->token = subflow_req->token;
1876                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1877                 new_ctx->idsn = subflow_req->idsn;
1878
1879                 /* this is the first subflow, id is always 0 */
1880                 new_ctx->local_id_valid = 1;
1881         } else if (subflow_req->mp_join) {
1882                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1883                 new_ctx->mp_join = 1;
1884                 new_ctx->fully_established = 1;
1885                 new_ctx->backup = subflow_req->backup;
1886                 new_ctx->remote_id = subflow_req->remote_id;
1887                 new_ctx->token = subflow_req->token;
1888                 new_ctx->thmac = subflow_req->thmac;
1889
1890                 /* the subflow req id is valid, fetched via subflow_check_req()
1891                  * and subflow_token_join_request()
1892                  */
1893                 subflow_set_local_id(new_ctx, subflow_req->local_id);
1894         }
1895 }
1896
1897 static void tcp_release_cb_override(struct sock *ssk)
1898 {
1899         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1900
1901         if (mptcp_subflow_has_delegated_action(subflow))
1902                 mptcp_subflow_process_delegated(ssk);
1903
1904         tcp_release_cb(ssk);
1905 }
1906
1907 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1908         .name           = "mptcp",
1909         .owner          = THIS_MODULE,
1910         .init           = subflow_ulp_init,
1911         .release        = subflow_ulp_release,
1912         .clone          = subflow_ulp_clone,
1913 };
1914
1915 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1916 {
1917         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1918         subflow_ops->slab_name = "request_sock_subflow";
1919
1920         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1921                                               subflow_ops->obj_size, 0,
1922                                               SLAB_ACCOUNT |
1923                                               SLAB_TYPESAFE_BY_RCU,
1924                                               NULL);
1925         if (!subflow_ops->slab)
1926                 return -ENOMEM;
1927
1928         subflow_ops->destructor = subflow_req_destructor;
1929
1930         return 0;
1931 }
1932
1933 void __init mptcp_subflow_init(void)
1934 {
1935         mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1936         if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1937                 panic("MPTCP: failed to init subflow request sock ops\n");
1938
1939         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1940         subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
1941
1942         subflow_specific = ipv4_specific;
1943         subflow_specific.conn_request = subflow_v4_conn_request;
1944         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1945         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1946         subflow_specific.rebuild_header = subflow_rebuild_header;
1947
1948         tcp_prot_override = tcp_prot;
1949         tcp_prot_override.release_cb = tcp_release_cb_override;
1950
1951 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1952         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1953         subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
1954
1955         subflow_v6_specific = ipv6_specific;
1956         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1957         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1958         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1959         subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
1960
1961         subflow_v6m_specific = subflow_v6_specific;
1962         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1963         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1964         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1965         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1966         subflow_v6m_specific.net_frag_header_len = 0;
1967         subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
1968
1969         tcpv6_prot_override = tcpv6_prot;
1970         tcpv6_prot_override.release_cb = tcp_release_cb_override;
1971 #endif
1972
1973         mptcp_diag_subflow_init(&subflow_ulp_ops);
1974
1975         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1976                 panic("MPTCP: failed to register subflows to ULP\n");
1977 }