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         /* We don't need to clear msk->subflow, as it's still NULL at this point */
 625         mptcp_destroy_common(mptcp_sk(sk), 0);
 626         inet_sock_destruct(sk);
 627 }
 628
 629 static void mptcp_force_close(struct sock *sk)
 630 {
 631         /* the msk is not yet exposed to user-space */
 632         inet_sk_state_store(sk, TCP_CLOSE);
 633         sk_common_release(sk);
 634 }
 635
 636 static void subflow_ulp_fallback(struct sock *sk,
 637                                  struct mptcp_subflow_context *old_ctx)
 638 {
 639         struct inet_connection_sock *icsk = inet_csk(sk);
 640
 641         mptcp_subflow_tcp_fallback(sk, old_ctx);
 642         icsk->icsk_ulp_ops = NULL;
 643         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 644         tcp_sk(sk)->is_mptcp = 0;
 645
 646         mptcp_subflow_ops_undo_override(sk);
 647 }
 648
 649 static void subflow_drop_ctx(struct sock *ssk)
 650 {
 651         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 652
 653         if (!ctx)
 654                 return;
 655
 656         subflow_ulp_fallback(ssk, ctx);
 657         if (ctx->conn)
 658                 sock_put(ctx->conn);
 659
 660         kfree_rcu(ctx, rcu);
 661 }
 662
 663 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 664                                      struct mptcp_options_received *mp_opt)
 665 {
 666         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 667
 668         subflow->remote_key = mp_opt->sndr_key;
 669         subflow->fully_established = 1;
 670         subflow->can_ack = 1;
 671         WRITE_ONCE(msk->fully_established, true);
 672 }
 673
 674 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 675                                           struct sk_buff *skb,
 676                                           struct request_sock *req,
 677                                           struct dst_entry *dst,
 678                                           struct request_sock *req_unhash,
 679                                           bool *own_req)
 680 {
 681         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 682         struct mptcp_subflow_request_sock *subflow_req;
 683         struct mptcp_options_received mp_opt;
 684         bool fallback, fallback_is_fatal;
 685         struct sock *new_msk = NULL;
 686         struct sock *child;
 687
 688         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 689
 690         /* After child creation we must look for MPC even when options
 691          * are not parsed
 692          */
 693         mp_opt.suboptions = 0;
 694
 695         /* hopefully temporary handling for MP_JOIN+syncookie */
 696         subflow_req = mptcp_subflow_rsk(req);
 697         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 698         fallback = !tcp_rsk(req)->is_mptcp;
 699         if (fallback)
 700                 goto create_child;
 701
 702         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 703         if (subflow_req->mp_capable) {
 704                 /* we can receive and accept an in-window, out-of-order pkt,
 705                  * which may not carry the MP_CAPABLE opt even on mptcp enabled
 706                  * paths: always try to extract the peer key, and fallback
 707                  * for packets missing it.
 708                  * Even OoO DSS packets coming legitly after dropped or
 709                  * reordered MPC will cause fallback, but we don't have other
 710                  * options.
 711                  */
 712                 mptcp_get_options(skb, &mp_opt);
 713                 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPC)) {
 714                         fallback = true;
 715                         goto create_child;
 716                 }
 717
 718                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 719                 if (!new_msk)
 720                         fallback = true;
 721         } else if (subflow_req->mp_join) {
 722                 mptcp_get_options(skb, &mp_opt);
 723                 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ) ||
 724                     !subflow_hmac_valid(req, &mp_opt) ||
 725                     !mptcp_can_accept_new_subflow(subflow_req->msk)) {
 726                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 727                         fallback = true;
 728                 }
 729         }
 730
 731 create_child:
 732         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 733                                                      req_unhash, own_req);
 734
 735         if (child && *own_req) {
 736                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 737
 738                 tcp_rsk(req)->drop_req = false;
 739
 740                 /* we need to fallback on ctx allocation failure and on pre-reqs
 741                  * checking above. In the latter scenario we additionally need
 742                  * to reset the context to non MPTCP status.
 743                  */
 744                 if (!ctx || fallback) {
 745                         if (fallback_is_fatal) {
 746                                 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 747                                 goto dispose_child;
 748                         }
 749
 750                         subflow_drop_ctx(child);
 751                         goto out;
 752                 }
 753
 754                 /* ssk inherits options of listener sk */
 755                 ctx->setsockopt_seq = listener->setsockopt_seq;
 756
 757                 if (ctx->mp_capable) {
 758                         /* this can't race with mptcp_close(), as the msk is
 759                          * not yet exposted to user-space
 760                          */
 761                         inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
 762
 763                         /* record the newly created socket as the first msk
 764                          * subflow, but don't link it yet into conn_list
 765                          */
 766                         WRITE_ONCE(mptcp_sk(new_msk)->first, child);
 767
 768                         /* new mpc subflow takes ownership of the newly
 769                          * created mptcp socket
 770                          */
 771                         new_msk->sk_destruct = mptcp_sock_destruct;
 772                         mptcp_sk(new_msk)->setsockopt_seq = ctx->setsockopt_seq;
 773                         mptcp_pm_new_connection(mptcp_sk(new_msk), child, 1);
 774                         mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
 775                         ctx->conn = new_msk;
 776                         new_msk = NULL;
 777
 778                         /* with OoO packets we can reach here without ingress
 779                          * mpc option
 780                          */
 781                         if (mp_opt.suboptions & OPTIONS_MPTCP_MPC)
 782                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 783                 } else if (ctx->mp_join) {
 784                         struct mptcp_sock *owner;
 785
 786                         owner = subflow_req->msk;
 787                         if (!owner) {
 788                                 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 789                                 goto dispose_child;
 790                         }
 791
 792                         /* move the msk reference ownership to the subflow */
 793                         subflow_req->msk = NULL;
 794                         ctx->conn = (struct sock *)owner;
 795
 796                         if (subflow_use_different_sport(owner, sk)) {
 797                                 pr_debug("ack inet_sport=%d %d",
 798                                          ntohs(inet_sk(sk)->inet_sport),
 799                                          ntohs(inet_sk((struct sock *)owner)->inet_sport));
 800                                 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
 801                                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
 802                                         goto dispose_child;
 803                                 }
 804                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
 805                         }
 806
 807                         if (!mptcp_finish_join(child))
 808                                 goto dispose_child;
 809
 810                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 811                         tcp_rsk(req)->drop_req = true;
 812                 }
 813         }
 814
 815 out:
 816         /* dispose of the left over mptcp master, if any */
 817         if (unlikely(new_msk))
 818                 mptcp_force_close(new_msk);
 819
 820         /* check for expected invariant - should never trigger, just help
 821          * catching eariler subtle bugs
 822          */
 823         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 824                      (!mptcp_subflow_ctx(child) ||
 825                       !mptcp_subflow_ctx(child)->conn));
 826         return child;
 827
 828 dispose_child:
 829         subflow_drop_ctx(child);
 830         tcp_rsk(req)->drop_req = true;
 831         inet_csk_prepare_for_destroy_sock(child);
 832         tcp_done(child);
 833         req->rsk_ops->send_reset(sk, skb);
 834
 835         /* The last child reference will be released by the caller */
 836         return child;
 837 }
 838
 839 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
 840 static struct proto tcp_prot_override;
 841
 842 enum mapping_status {
 843         MAPPING_OK,
 844         MAPPING_INVALID,
 845         MAPPING_EMPTY,
 846         MAPPING_DATA_FIN,
 847         MAPPING_DUMMY,
 848         MAPPING_BAD_CSUM
 849 };
 850
 851 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 852 {
 853         pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 854                  ssn, subflow->map_subflow_seq, subflow->map_data_len);
 855 }
 856
 857 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 858 {
 859         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 860         unsigned int skb_consumed;
 861
 862         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 863         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 864                 return true;
 865
 866         return skb->len - skb_consumed <= subflow->map_data_len -
 867                                           mptcp_subflow_get_map_offset(subflow);
 868 }
 869
 870 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 871 {
 872         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 873         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 874
 875         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 876                 /* Mapping covers data later in the subflow stream,
 877                  * currently unsupported.
 878                  */
 879                 dbg_bad_map(subflow, ssn);
 880                 return false;
 881         }
 882         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 883                                   subflow->map_data_len))) {
 884                 /* Mapping does covers past subflow data, invalid */
 885                 dbg_bad_map(subflow, ssn);
 886                 return false;
 887         }
 888         return true;
 889 }
 890
 891 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
 892                                               bool csum_reqd)
 893 {
 894         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 895         u32 offset, seq, delta;
 896         __sum16 csum;
 897         int len;
 898
 899         if (!csum_reqd)
 900                 return MAPPING_OK;
 901
 902         /* mapping already validated on previous traversal */
 903         if (subflow->map_csum_len == subflow->map_data_len)
 904                 return MAPPING_OK;
 905
 906         /* traverse the receive queue, ensuring it contains a full
 907          * DSS mapping and accumulating the related csum.
 908          * Preserve the accoumlate csum across multiple calls, to compute
 909          * the csum only once
 910          */
 911         delta = subflow->map_data_len - subflow->map_csum_len;
 912         for (;;) {
 913                 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
 914                 offset = seq - TCP_SKB_CB(skb)->seq;
 915
 916                 /* if the current skb has not been accounted yet, csum its contents
 917                  * up to the amount covered by the current DSS
 918                  */
 919                 if (offset < skb->len) {
 920                         __wsum csum;
 921
 922                         len = min(skb->len - offset, delta);
 923                         csum = skb_checksum(skb, offset, len, 0);
 924                         subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
 925                                                                 subflow->map_csum_len);
 926
 927                         delta -= len;
 928                         subflow->map_csum_len += len;
 929                 }
 930                 if (delta == 0)
 931                         break;
 932
 933                 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
 934                         /* if this subflow is closed, the partial mapping
 935                          * will be never completed; flush the pending skbs, so
 936                          * that subflow_sched_work_if_closed() can kick in
 937                          */
 938                         if (unlikely(ssk->sk_state == TCP_CLOSE))
 939                                 while ((skb = skb_peek(&ssk->sk_receive_queue)))
 940                                         sk_eat_skb(ssk, skb);
 941
 942                         /* not enough data to validate the csum */
 943                         return MAPPING_EMPTY;
 944                 }
 945
 946                 /* the DSS mapping for next skbs will be validated later,
 947                  * when a get_mapping_status call will process such skb
 948                  */
 949                 skb = skb->next;
 950         }
 951
 952         /* note that 'map_data_len' accounts only for the carried data, does
 953          * not include the eventual seq increment due to the data fin,
 954          * while the pseudo header requires the original DSS data len,
 955          * including that
 956          */
 957         csum = __mptcp_make_csum(subflow->map_seq,
 958                                  subflow->map_subflow_seq,
 959                                  subflow->map_data_len + subflow->map_data_fin,
 960                                  subflow->map_data_csum);
 961         if (unlikely(csum)) {
 962                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
 963                 return MAPPING_BAD_CSUM;
 964         }
 965
 966         subflow->valid_csum_seen = 1;
 967         return MAPPING_OK;
 968 }
 969
 970 static enum mapping_status get_mapping_status(struct sock *ssk,
 971                                               struct mptcp_sock *msk)
 972 {
 973         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 974         bool csum_reqd = READ_ONCE(msk->csum_enabled);
 975         struct mptcp_ext *mpext;
 976         struct sk_buff *skb;
 977         u16 data_len;
 978         u64 map_seq;
 979
 980         skb = skb_peek(&ssk->sk_receive_queue);
 981         if (!skb)
 982                 return MAPPING_EMPTY;
 983
 984         if (mptcp_check_fallback(ssk))
 985                 return MAPPING_DUMMY;
 986
 987         mpext = mptcp_get_ext(skb);
 988         if (!mpext || !mpext->use_map) {
 989                 if (!subflow->map_valid && !skb->len) {
 990                         /* the TCP stack deliver 0 len FIN pkt to the receive
 991                          * queue, that is the only 0len pkts ever expected here,
 992                          * and we can admit no mapping only for 0 len pkts
 993                          */
 994                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 995                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 996                                           TCP_SKB_CB(skb)->seq,
 997                                           TCP_SKB_CB(skb)->end_seq,
 998                                           TCP_SKB_CB(skb)->tcp_flags);
 999                         sk_eat_skb(ssk, skb);
1000                         return MAPPING_EMPTY;
1001                 }
1002
1003                 if (!subflow->map_valid)
1004                         return MAPPING_INVALID;
1005
1006                 goto validate_seq;
1007         }
1008
1009         trace_get_mapping_status(mpext);
1010
1011         data_len = mpext->data_len;
1012         if (data_len == 0) {
1013                 pr_debug("infinite mapping received");
1014                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1015                 subflow->map_data_len = 0;
1016                 return MAPPING_INVALID;
1017         }
1018
1019         if (mpext->data_fin == 1) {
1020                 if (data_len == 1) {
1021                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1022                                                                  mpext->dsn64);
1023                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
1024                         if (subflow->map_valid) {
1025                                 /* A DATA_FIN might arrive in a DSS
1026                                  * option before the previous mapping
1027                                  * has been fully consumed. Continue
1028                                  * handling the existing mapping.
1029                                  */
1030                                 skb_ext_del(skb, SKB_EXT_MPTCP);
1031                                 return MAPPING_OK;
1032                         } else {
1033                                 if (updated && schedule_work(&msk->work))
1034                                         sock_hold((struct sock *)msk);
1035
1036                                 return MAPPING_DATA_FIN;
1037                         }
1038                 } else {
1039                         u64 data_fin_seq = mpext->data_seq + data_len - 1;
1040
1041                         /* If mpext->data_seq is a 32-bit value, data_fin_seq
1042                          * must also be limited to 32 bits.
1043                          */
1044                         if (!mpext->dsn64)
1045                                 data_fin_seq &= GENMASK_ULL(31, 0);
1046
1047                         mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1048                         pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
1049                                  data_fin_seq, mpext->dsn64);
1050                 }
1051
1052                 /* Adjust for DATA_FIN using 1 byte of sequence space */
1053                 data_len--;
1054         }
1055
1056         map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1057         WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1058
1059         if (subflow->map_valid) {
1060                 /* Allow replacing only with an identical map */
1061                 if (subflow->map_seq == map_seq &&
1062                     subflow->map_subflow_seq == mpext->subflow_seq &&
1063                     subflow->map_data_len == data_len &&
1064                     subflow->map_csum_reqd == mpext->csum_reqd) {
1065                         skb_ext_del(skb, SKB_EXT_MPTCP);
1066                         goto validate_csum;
1067                 }
1068
1069                 /* If this skb data are fully covered by the current mapping,
1070                  * the new map would need caching, which is not supported
1071                  */
1072                 if (skb_is_fully_mapped(ssk, skb)) {
1073                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1074                         return MAPPING_INVALID;
1075                 }
1076
1077                 /* will validate the next map after consuming the current one */
1078                 goto validate_csum;
1079         }
1080
1081         subflow->map_seq = map_seq;
1082         subflow->map_subflow_seq = mpext->subflow_seq;
1083         subflow->map_data_len = data_len;
1084         subflow->map_valid = 1;
1085         subflow->map_data_fin = mpext->data_fin;
1086         subflow->mpc_map = mpext->mpc_map;
1087         subflow->map_csum_reqd = mpext->csum_reqd;
1088         subflow->map_csum_len = 0;
1089         subflow->map_data_csum = csum_unfold(mpext->csum);
1090
1091         /* Cfr RFC 8684 Section 3.3.0 */
1092         if (unlikely(subflow->map_csum_reqd != csum_reqd))
1093                 return MAPPING_INVALID;
1094
1095         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
1096                  subflow->map_seq, subflow->map_subflow_seq,
1097                  subflow->map_data_len, subflow->map_csum_reqd,
1098                  subflow->map_data_csum);
1099
1100 validate_seq:
1101         /* we revalidate valid mapping on new skb, because we must ensure
1102          * the current skb is completely covered by the available mapping
1103          */
1104         if (!validate_mapping(ssk, skb)) {
1105                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1106                 return MAPPING_INVALID;
1107         }
1108
1109         skb_ext_del(skb, SKB_EXT_MPTCP);
1110
1111 validate_csum:
1112         return validate_data_csum(ssk, skb, csum_reqd);
1113 }
1114
1115 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1116                                        u64 limit)
1117 {
1118         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1119         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1120         u32 incr;
1121
1122         incr = limit >= skb->len ? skb->len + fin : limit;
1123
1124         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
1125                  subflow->map_subflow_seq);
1126         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1127         tcp_sk(ssk)->copied_seq += incr;
1128         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1129                 sk_eat_skb(ssk, skb);
1130         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1131                 subflow->map_valid = 0;
1132 }
1133
1134 /* sched mptcp worker to remove the subflow if no more data is pending */
1135 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1136 {
1137         struct sock *sk = (struct sock *)msk;
1138
1139         if (likely(ssk->sk_state != TCP_CLOSE))
1140                 return;
1141
1142         if (skb_queue_empty(&ssk->sk_receive_queue) &&
1143             !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags)) {
1144                 sock_hold(sk);
1145                 if (!schedule_work(&msk->work))
1146                         sock_put(sk);
1147         }
1148 }
1149
1150 static bool subflow_can_fallback(struct mptcp_subflow_context *subflow)
1151 {
1152         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1153
1154         if (subflow->mp_join)
1155                 return false;
1156         else if (READ_ONCE(msk->csum_enabled))
1157                 return !subflow->valid_csum_seen;
1158         else
1159                 return !subflow->fully_established;
1160 }
1161
1162 static void mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk)
1163 {
1164         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1165         unsigned long fail_tout;
1166
1167         /* greceful failure can happen only on the MPC subflow */
1168         if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first)))
1169                 return;
1170
1171         /* since the close timeout take precedence on the fail one,
1172          * no need to start the latter when the first is already set
1173          */
1174         if (sock_flag((struct sock *)msk, SOCK_DEAD))
1175                 return;
1176
1177         /* we don't need extreme accuracy here, use a zero fail_tout as special
1178          * value meaning no fail timeout at all;
1179          */
1180         fail_tout = jiffies + TCP_RTO_MAX;
1181         if (!fail_tout)
1182                 fail_tout = 1;
1183         WRITE_ONCE(subflow->fail_tout, fail_tout);
1184         tcp_send_ack(ssk);
1185
1186         mptcp_reset_timeout(msk, subflow->fail_tout);
1187 }
1188
1189 static bool subflow_check_data_avail(struct sock *ssk)
1190 {
1191         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1192         enum mapping_status status;
1193         struct mptcp_sock *msk;
1194         struct sk_buff *skb;
1195
1196         if (!skb_peek(&ssk->sk_receive_queue))
1197                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1198         if (subflow->data_avail)
1199                 return true;
1200
1201         msk = mptcp_sk(subflow->conn);
1202         for (;;) {
1203                 u64 ack_seq;
1204                 u64 old_ack;
1205
1206                 status = get_mapping_status(ssk, msk);
1207                 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1208                 if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY ||
1209                              status == MAPPING_BAD_CSUM))
1210                         goto fallback;
1211
1212                 if (status != MAPPING_OK)
1213                         goto no_data;
1214
1215                 skb = skb_peek(&ssk->sk_receive_queue);
1216                 if (WARN_ON_ONCE(!skb))
1217                         goto no_data;
1218
1219                 /* if msk lacks the remote key, this subflow must provide an
1220                  * MP_CAPABLE-based mapping
1221                  */
1222                 if (unlikely(!READ_ONCE(msk->can_ack))) {
1223                         if (!subflow->mpc_map)
1224                                 goto fallback;
1225                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
1226                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
1227                         WRITE_ONCE(msk->can_ack, true);
1228                 }
1229
1230                 old_ack = READ_ONCE(msk->ack_seq);
1231                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1232                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1233                          ack_seq);
1234                 if (unlikely(before64(ack_seq, old_ack))) {
1235                         mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1236                         continue;
1237                 }
1238
1239                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1240                 break;
1241         }
1242         return true;
1243
1244 no_data:
1245         subflow_sched_work_if_closed(msk, ssk);
1246         return false;
1247
1248 fallback:
1249         if (!__mptcp_check_fallback(msk)) {
1250                 /* RFC 8684 section 3.7. */
1251                 if (status == MAPPING_BAD_CSUM &&
1252                     (subflow->mp_join || subflow->valid_csum_seen)) {
1253                         subflow->send_mp_fail = 1;
1254
1255                         if (!READ_ONCE(msk->allow_infinite_fallback)) {
1256                                 subflow->reset_transient = 0;
1257                                 subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1258                                 goto reset;
1259                         }
1260                         mptcp_subflow_fail(msk, ssk);
1261                         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1262                         return true;
1263                 }
1264
1265                 if (!subflow_can_fallback(subflow) && subflow->map_data_len) {
1266                         /* fatal protocol error, close the socket.
1267                          * subflow_error_report() will introduce the appropriate barriers
1268                          */
1269                         subflow->reset_transient = 0;
1270                         subflow->reset_reason = MPTCP_RST_EMPTCP;
1271
1272 reset:
1273                         ssk->sk_err = EBADMSG;
1274                         tcp_set_state(ssk, TCP_CLOSE);
1275                         while ((skb = skb_peek(&ssk->sk_receive_queue)))
1276                                 sk_eat_skb(ssk, skb);
1277                         tcp_send_active_reset(ssk, GFP_ATOMIC);
1278                         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1279                         return false;
1280                 }
1281
1282                 mptcp_do_fallback(ssk);
1283         }
1284
1285         skb = skb_peek(&ssk->sk_receive_queue);
1286         subflow->map_valid = 1;
1287         subflow->map_seq = READ_ONCE(msk->ack_seq);
1288         subflow->map_data_len = skb->len;
1289         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1290         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1291         return true;
1292 }
1293
1294 bool mptcp_subflow_data_available(struct sock *sk)
1295 {
1296         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1297
1298         /* check if current mapping is still valid */
1299         if (subflow->map_valid &&
1300             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1301                 subflow->map_valid = 0;
1302                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1303
1304                 pr_debug("Done with mapping: seq=%u data_len=%u",
1305                          subflow->map_subflow_seq,
1306                          subflow->map_data_len);
1307         }
1308
1309         return subflow_check_data_avail(sk);
1310 }
1311
1312 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1313  * not the ssk one.
1314  *
1315  * In mptcp, rwin is about the mptcp-level connection data.
1316  *
1317  * Data that is still on the ssk rx queue can thus be ignored,
1318  * as far as mptcp peer is concerned that data is still inflight.
1319  * DSS ACK is updated when skb is moved to the mptcp rx queue.
1320  */
1321 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1322 {
1323         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1324         const struct sock *sk = subflow->conn;
1325
1326         *space = __mptcp_space(sk);
1327         *full_space = tcp_full_space(sk);
1328 }
1329
1330 void __mptcp_error_report(struct sock *sk)
1331 {
1332         struct mptcp_subflow_context *subflow;
1333         struct mptcp_sock *msk = mptcp_sk(sk);
1334
1335         mptcp_for_each_subflow(msk, subflow) {
1336                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1337                 int err = sock_error(ssk);
1338
1339                 if (!err)
1340                         continue;
1341
1342                 /* only propagate errors on fallen-back sockets or
1343                  * on MPC connect
1344                  */
1345                 if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(msk))
1346                         continue;
1347
1348                 inet_sk_state_store(sk, inet_sk_state_load(ssk));
1349                 sk->sk_err = -err;
1350
1351                 /* This barrier is coupled with smp_rmb() in mptcp_poll() */
1352                 smp_wmb();
1353                 sk_error_report(sk);
1354                 break;
1355         }
1356 }
1357
1358 static void subflow_error_report(struct sock *ssk)
1359 {
1360         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1361
1362         mptcp_data_lock(sk);
1363         if (!sock_owned_by_user(sk))
1364                 __mptcp_error_report(sk);
1365         else
1366                 __set_bit(MPTCP_ERROR_REPORT,  &mptcp_sk(sk)->cb_flags);
1367         mptcp_data_unlock(sk);
1368 }
1369
1370 static void subflow_data_ready(struct sock *sk)
1371 {
1372         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1373         u16 state = 1 << inet_sk_state_load(sk);
1374         struct sock *parent = subflow->conn;
1375         struct mptcp_sock *msk;
1376
1377         msk = mptcp_sk(parent);
1378         if (state & TCPF_LISTEN) {
1379                 /* MPJ subflow are removed from accept queue before reaching here,
1380                  * avoid stray wakeups
1381                  */
1382                 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1383                         return;
1384
1385                 parent->sk_data_ready(parent);
1386                 return;
1387         }
1388
1389         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1390                      !subflow->mp_join && !(state & TCPF_CLOSE));
1391
1392         if (mptcp_subflow_data_available(sk))
1393                 mptcp_data_ready(parent, sk);
1394         else if (unlikely(sk->sk_err))
1395                 subflow_error_report(sk);
1396 }
1397
1398 static void subflow_write_space(struct sock *ssk)
1399 {
1400         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1401
1402         mptcp_propagate_sndbuf(sk, ssk);
1403         mptcp_write_space(sk);
1404 }
1405
1406 static const struct inet_connection_sock_af_ops *
1407 subflow_default_af_ops(struct sock *sk)
1408 {
1409 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1410         if (sk->sk_family == AF_INET6)
1411                 return &subflow_v6_specific;
1412 #endif
1413         return &subflow_specific;
1414 }
1415
1416 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1417 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1418 {
1419         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1420         struct inet_connection_sock *icsk = inet_csk(sk);
1421         const struct inet_connection_sock_af_ops *target;
1422
1423         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1424
1425         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1426                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1427
1428         if (likely(icsk->icsk_af_ops == target))
1429                 return;
1430
1431         subflow->icsk_af_ops = icsk->icsk_af_ops;
1432         icsk->icsk_af_ops = target;
1433 }
1434 #endif
1435
1436 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1437                          struct sockaddr_storage *addr,
1438                          unsigned short family)
1439 {
1440         memset(addr, 0, sizeof(*addr));
1441         addr->ss_family = family;
1442         if (addr->ss_family == AF_INET) {
1443                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1444
1445                 if (info->family == AF_INET)
1446                         in_addr->sin_addr = info->addr;
1447 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1448                 else if (ipv6_addr_v4mapped(&info->addr6))
1449                         in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1450 #endif
1451                 in_addr->sin_port = info->port;
1452         }
1453 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1454         else if (addr->ss_family == AF_INET6) {
1455                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1456
1457                 if (info->family == AF_INET)
1458                         ipv6_addr_set_v4mapped(info->addr.s_addr,
1459                                                &in6_addr->sin6_addr);
1460                 else
1461                         in6_addr->sin6_addr = info->addr6;
1462                 in6_addr->sin6_port = info->port;
1463         }
1464 #endif
1465 }
1466
1467 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1468                             const struct mptcp_addr_info *remote)
1469 {
1470         struct mptcp_sock *msk = mptcp_sk(sk);
1471         struct mptcp_subflow_context *subflow;
1472         struct sockaddr_storage addr;
1473         int remote_id = remote->id;
1474         int local_id = loc->id;
1475         int err = -ENOTCONN;
1476         struct socket *sf;
1477         struct sock *ssk;
1478         u32 remote_token;
1479         int addrlen;
1480         int ifindex;
1481         u8 flags;
1482
1483         if (!mptcp_is_fully_established(sk))
1484                 goto err_out;
1485
1486         err = mptcp_subflow_create_socket(sk, &sf);
1487         if (err)
1488                 goto err_out;
1489
1490         ssk = sf->sk;
1491         subflow = mptcp_subflow_ctx(ssk);
1492         do {
1493                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1494         } while (!subflow->local_nonce);
1495
1496         if (local_id)
1497                 subflow_set_local_id(subflow, local_id);
1498
1499         mptcp_pm_get_flags_and_ifindex_by_id(msk, local_id,
1500                                              &flags, &ifindex);
1501         subflow->remote_key = msk->remote_key;
1502         subflow->local_key = msk->local_key;
1503         subflow->token = msk->token;
1504         mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1505
1506         addrlen = sizeof(struct sockaddr_in);
1507 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1508         if (addr.ss_family == AF_INET6)
1509                 addrlen = sizeof(struct sockaddr_in6);
1510 #endif
1511         mptcp_sockopt_sync(msk, ssk);
1512
1513         ssk->sk_bound_dev_if = ifindex;
1514         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1515         if (err)
1516                 goto failed;
1517
1518         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1519         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1520                  remote_token, local_id, remote_id);
1521         subflow->remote_token = remote_token;
1522         subflow->remote_id = remote_id;
1523         subflow->request_join = 1;
1524         subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1525         mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1526
1527         sock_hold(ssk);
1528         list_add_tail(&subflow->node, &msk->conn_list);
1529         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1530         if (err && err != -EINPROGRESS)
1531                 goto failed_unlink;
1532
1533         /* discard the subflow socket */
1534         mptcp_sock_graft(ssk, sk->sk_socket);
1535         iput(SOCK_INODE(sf));
1536         WRITE_ONCE(msk->allow_infinite_fallback, false);
1537         return 0;
1538
1539 failed_unlink:
1540         list_del(&subflow->node);
1541         sock_put(mptcp_subflow_tcp_sock(subflow));
1542
1543 failed:
1544         subflow->disposable = 1;
1545         sock_release(sf);
1546
1547 err_out:
1548         /* we account subflows before the creation, and this failures will not
1549          * be caught by sk_state_change()
1550          */
1551         mptcp_pm_close_subflow(msk);
1552         return err;
1553 }
1554
1555 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1556 {
1557 #ifdef CONFIG_SOCK_CGROUP_DATA
1558         struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1559                                 *child_skcd = &child->sk_cgrp_data;
1560
1561         /* only the additional subflows created by kworkers have to be modified */
1562         if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1563             cgroup_id(sock_cgroup_ptr(child_skcd))) {
1564 #ifdef CONFIG_MEMCG
1565                 struct mem_cgroup *memcg = parent->sk_memcg;
1566
1567                 mem_cgroup_sk_free(child);
1568                 if (memcg && css_tryget(&memcg->css))
1569                         child->sk_memcg = memcg;
1570 #endif /* CONFIG_MEMCG */
1571
1572                 cgroup_sk_free(child_skcd);
1573                 *child_skcd = *parent_skcd;
1574                 cgroup_sk_clone(child_skcd);
1575         }
1576 #endif /* CONFIG_SOCK_CGROUP_DATA */
1577 }
1578
1579 static void mptcp_subflow_ops_override(struct sock *ssk)
1580 {
1581 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1582         if (ssk->sk_prot == &tcpv6_prot)
1583                 ssk->sk_prot = &tcpv6_prot_override;
1584         else
1585 #endif
1586                 ssk->sk_prot = &tcp_prot_override;
1587 }
1588
1589 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1590 {
1591 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1592         if (ssk->sk_prot == &tcpv6_prot_override)
1593                 ssk->sk_prot = &tcpv6_prot;
1594         else
1595 #endif
1596                 ssk->sk_prot = &tcp_prot;
1597 }
1598 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1599 {
1600         struct mptcp_subflow_context *subflow;
1601         struct net *net = sock_net(sk);
1602         struct socket *sf;
1603         int err;
1604
1605         /* un-accepted server sockets can reach here - on bad configuration
1606          * bail early to avoid greater trouble later
1607          */
1608         if (unlikely(!sk->sk_socket))
1609                 return -EINVAL;
1610
1611         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1612                                &sf);
1613         if (err)
1614                 return err;
1615
1616         lock_sock(sf->sk);
1617
1618         /* the newly created socket has to be in the same cgroup as its parent */
1619         mptcp_attach_cgroup(sk, sf->sk);
1620
1621         /* kernel sockets do not by default acquire net ref, but TCP timer
1622          * needs it.
1623          */
1624         sf->sk->sk_net_refcnt = 1;
1625         get_net_track(net, &sf->sk->ns_tracker, GFP_KERNEL);
1626         sock_inuse_add(net, 1);
1627         err = tcp_set_ulp(sf->sk, "mptcp");
1628         release_sock(sf->sk);
1629
1630         if (err) {
1631                 sock_release(sf);
1632                 return err;
1633         }
1634
1635         /* the newly created socket really belongs to the owning MPTCP master
1636          * socket, even if for additional subflows the allocation is performed
1637          * by a kernel workqueue. Adjust inode references, so that the
1638          * procfs/diag interaces really show this one belonging to the correct
1639          * user.
1640          */
1641         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1642         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1643         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1644
1645         subflow = mptcp_subflow_ctx(sf->sk);
1646         pr_debug("subflow=%p", subflow);
1647
1648         *new_sock = sf;
1649         sock_hold(sk);
1650         subflow->conn = sk;
1651         mptcp_subflow_ops_override(sf->sk);
1652
1653         return 0;
1654 }
1655
1656 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1657                                                         gfp_t priority)
1658 {
1659         struct inet_connection_sock *icsk = inet_csk(sk);
1660         struct mptcp_subflow_context *ctx;
1661
1662         ctx = kzalloc(sizeof(*ctx), priority);
1663         if (!ctx)
1664                 return NULL;
1665
1666         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1667         INIT_LIST_HEAD(&ctx->node);
1668         INIT_LIST_HEAD(&ctx->delegated_node);
1669
1670         pr_debug("subflow=%p", ctx);
1671
1672         ctx->tcp_sock = sk;
1673
1674         return ctx;
1675 }
1676
1677 static void __subflow_state_change(struct sock *sk)
1678 {
1679         struct socket_wq *wq;
1680
1681         rcu_read_lock();
1682         wq = rcu_dereference(sk->sk_wq);
1683         if (skwq_has_sleeper(wq))
1684                 wake_up_interruptible_all(&wq->wait);
1685         rcu_read_unlock();
1686 }
1687
1688 static bool subflow_is_done(const struct sock *sk)
1689 {
1690         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1691 }
1692
1693 static void subflow_state_change(struct sock *sk)
1694 {
1695         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1696         struct sock *parent = subflow->conn;
1697
1698         __subflow_state_change(sk);
1699
1700         if (subflow_simultaneous_connect(sk)) {
1701                 mptcp_propagate_sndbuf(parent, sk);
1702                 mptcp_do_fallback(sk);
1703                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1704                 pr_fallback(mptcp_sk(parent));
1705                 subflow->conn_finished = 1;
1706                 mptcp_set_connected(parent);
1707         }
1708
1709         /* as recvmsg() does not acquire the subflow socket for ssk selection
1710          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1711          * the data available machinery here.
1712          */
1713         if (mptcp_subflow_data_available(sk))
1714                 mptcp_data_ready(parent, sk);
1715         else if (unlikely(sk->sk_err))
1716                 subflow_error_report(sk);
1717
1718         subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1719
1720         if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1721             !subflow->rx_eof && subflow_is_done(sk)) {
1722                 subflow->rx_eof = 1;
1723                 mptcp_subflow_eof(parent);
1724         }
1725 }
1726
1727 void mptcp_subflow_queue_clean(struct sock *listener_ssk)
1728 {
1729         struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
1730         struct mptcp_sock *msk, *next, *head = NULL;
1731         struct request_sock *req;
1732
1733         /* build a list of all unaccepted mptcp sockets */
1734         spin_lock_bh(&queue->rskq_lock);
1735         for (req = queue->rskq_accept_head; req; req = req->dl_next) {
1736                 struct mptcp_subflow_context *subflow;
1737                 struct sock *ssk = req->sk;
1738                 struct mptcp_sock *msk;
1739
1740                 if (!sk_is_mptcp(ssk))
1741                         continue;
1742
1743                 subflow = mptcp_subflow_ctx(ssk);
1744                 if (!subflow || !subflow->conn)
1745                         continue;
1746
1747                 /* skip if already in list */
1748                 msk = mptcp_sk(subflow->conn);
1749                 if (msk->dl_next || msk == head)
1750                         continue;
1751
1752                 msk->dl_next = head;
1753                 head = msk;
1754         }
1755         spin_unlock_bh(&queue->rskq_lock);
1756         if (!head)
1757                 return;
1758
1759         /* can't acquire the msk socket lock under the subflow one,
1760          * or will cause ABBA deadlock
1761          */
1762         release_sock(listener_ssk);
1763
1764         for (msk = head; msk; msk = next) {
1765                 struct sock *sk = (struct sock *)msk;
1766                 bool slow;
1767
1768                 slow = lock_sock_fast_nested(sk);
1769                 next = msk->dl_next;
1770                 msk->first = NULL;
1771                 msk->dl_next = NULL;
1772                 unlock_sock_fast(sk, slow);
1773         }
1774
1775         /* we are still under the listener msk socket lock */
1776         lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
1777 }
1778
1779 static int subflow_ulp_init(struct sock *sk)
1780 {
1781         struct inet_connection_sock *icsk = inet_csk(sk);
1782         struct mptcp_subflow_context *ctx;
1783         struct tcp_sock *tp = tcp_sk(sk);
1784         int err = 0;
1785
1786         /* disallow attaching ULP to a socket unless it has been
1787          * created with sock_create_kern()
1788          */
1789         if (!sk->sk_kern_sock) {
1790                 err = -EOPNOTSUPP;
1791                 goto out;
1792         }
1793
1794         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1795         if (!ctx) {
1796                 err = -ENOMEM;
1797                 goto out;
1798         }
1799
1800         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1801
1802         tp->is_mptcp = 1;
1803         ctx->icsk_af_ops = icsk->icsk_af_ops;
1804         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1805         ctx->tcp_state_change = sk->sk_state_change;
1806         ctx->tcp_error_report = sk->sk_error_report;
1807
1808         WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1809         WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
1810
1811         sk->sk_data_ready = subflow_data_ready;
1812         sk->sk_write_space = subflow_write_space;
1813         sk->sk_state_change = subflow_state_change;
1814         sk->sk_error_report = subflow_error_report;
1815 out:
1816         return err;
1817 }
1818
1819 static void subflow_ulp_release(struct sock *ssk)
1820 {
1821         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1822         bool release = true;
1823         struct sock *sk;
1824
1825         if (!ctx)
1826                 return;
1827
1828         sk = ctx->conn;
1829         if (sk) {
1830                 /* if the msk has been orphaned, keep the ctx
1831                  * alive, will be freed by __mptcp_close_ssk(),
1832                  * when the subflow is still unaccepted
1833                  */
1834                 release = ctx->disposable || list_empty(&ctx->node);
1835                 sock_put(sk);
1836         }
1837
1838         mptcp_subflow_ops_undo_override(ssk);
1839         if (release)
1840                 kfree_rcu(ctx, rcu);
1841 }
1842
1843 static void subflow_ulp_clone(const struct request_sock *req,
1844                               struct sock *newsk,
1845                               const gfp_t priority)
1846 {
1847         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1848         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1849         struct mptcp_subflow_context *new_ctx;
1850
1851         if (!tcp_rsk(req)->is_mptcp ||
1852             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1853                 subflow_ulp_fallback(newsk, old_ctx);
1854                 return;
1855         }
1856
1857         new_ctx = subflow_create_ctx(newsk, priority);
1858         if (!new_ctx) {
1859                 subflow_ulp_fallback(newsk, old_ctx);
1860                 return;
1861         }
1862
1863         new_ctx->conn_finished = 1;
1864         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1865         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1866         new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1867         new_ctx->rel_write_seq = 1;
1868         new_ctx->tcp_sock = newsk;
1869
1870         if (subflow_req->mp_capable) {
1871                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1872                  * is fully established only after we receive the remote key
1873                  */
1874                 new_ctx->mp_capable = 1;
1875                 new_ctx->local_key = subflow_req->local_key;
1876                 new_ctx->token = subflow_req->token;
1877                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1878                 new_ctx->idsn = subflow_req->idsn;
1879
1880                 /* this is the first subflow, id is always 0 */
1881                 new_ctx->local_id_valid = 1;
1882         } else if (subflow_req->mp_join) {
1883                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1884                 new_ctx->mp_join = 1;
1885                 new_ctx->fully_established = 1;
1886                 new_ctx->backup = subflow_req->backup;
1887                 new_ctx->remote_id = subflow_req->remote_id;
1888                 new_ctx->token = subflow_req->token;
1889                 new_ctx->thmac = subflow_req->thmac;
1890
1891                 /* the subflow req id is valid, fetched via subflow_check_req()
1892                  * and subflow_token_join_request()
1893                  */
1894                 subflow_set_local_id(new_ctx, subflow_req->local_id);
1895         }
1896 }
1897
1898 static void tcp_release_cb_override(struct sock *ssk)
1899 {
1900         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1901
1902         if (mptcp_subflow_has_delegated_action(subflow))
1903                 mptcp_subflow_process_delegated(ssk);
1904
1905         tcp_release_cb(ssk);
1906 }
1907
1908 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1909         .name           = "mptcp",
1910         .owner          = THIS_MODULE,
1911         .init           = subflow_ulp_init,
1912         .release        = subflow_ulp_release,
1913         .clone          = subflow_ulp_clone,
1914 };
1915
1916 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1917 {
1918         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1919         subflow_ops->slab_name = "request_sock_subflow";
1920
1921         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1922                                               subflow_ops->obj_size, 0,
1923                                               SLAB_ACCOUNT |
1924                                               SLAB_TYPESAFE_BY_RCU,
1925                                               NULL);
1926         if (!subflow_ops->slab)
1927                 return -ENOMEM;
1928
1929         subflow_ops->destructor = subflow_req_destructor;
1930
1931         return 0;
1932 }
1933
1934 void __init mptcp_subflow_init(void)
1935 {
1936         mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1937         if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1938                 panic("MPTCP: failed to init subflow request sock ops\n");
1939
1940         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1941         subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
1942
1943         subflow_specific = ipv4_specific;
1944         subflow_specific.conn_request = subflow_v4_conn_request;
1945         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1946         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1947         subflow_specific.rebuild_header = subflow_rebuild_header;
1948
1949         tcp_prot_override = tcp_prot;
1950         tcp_prot_override.release_cb = tcp_release_cb_override;
1951
1952 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1953         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1954         subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
1955
1956         subflow_v6_specific = ipv6_specific;
1957         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1958         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1959         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1960         subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
1961
1962         subflow_v6m_specific = subflow_v6_specific;
1963         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1964         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1965         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1966         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1967         subflow_v6m_specific.net_frag_header_len = 0;
1968         subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
1969
1970         tcpv6_prot_override = tcpv6_prot;
1971         tcpv6_prot_override.release_cb = tcp_release_cb_override;
1972 #endif
1973
1974         mptcp_diag_subflow_init(&subflow_ulp_ops);
1975
1976         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1977                 panic("MPTCP: failed to register subflows to ULP\n");
1978 }