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 static struct request_sock_ops mptcp_subflow_request_sock_ops __ro_after_init;
 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 struct request_sock *mptcp_subflow_reqsk_alloc(const struct request_sock_ops *ops,
 586                                                struct sock *sk_listener,
 587                                                bool attach_listener)
 588 {
 589         ops = &mptcp_subflow_request_sock_ops;
 590
 591         return inet_reqsk_alloc(ops, sk_listener, attach_listener);
 592 }
 593 EXPORT_SYMBOL(mptcp_subflow_reqsk_alloc);
 594
 595 /* validate hmac received in third ACK */
 596 static bool subflow_hmac_valid(const struct request_sock *req,
 597                                const struct mptcp_options_received *mp_opt)
 598 {
 599         const struct mptcp_subflow_request_sock *subflow_req;
 600         u8 hmac[SHA256_DIGEST_SIZE];
 601         struct mptcp_sock *msk;
 602
 603         subflow_req = mptcp_subflow_rsk(req);
 604         msk = subflow_req->msk;
 605         if (!msk)
 606                 return false;
 607
 608         subflow_generate_hmac(msk->remote_key, msk->local_key,
 609                               subflow_req->remote_nonce,
 610                               subflow_req->local_nonce, hmac);
 611
 612         return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
 613 }
 614
 615 static void mptcp_force_close(struct sock *sk)
 616 {
 617         /* the msk is not yet exposed to user-space */
 618         inet_sk_state_store(sk, TCP_CLOSE);
 619         sk_common_release(sk);
 620 }
 621
 622 static void subflow_ulp_fallback(struct sock *sk,
 623                                  struct mptcp_subflow_context *old_ctx)
 624 {
 625         struct inet_connection_sock *icsk = inet_csk(sk);
 626
 627         mptcp_subflow_tcp_fallback(sk, old_ctx);
 628         icsk->icsk_ulp_ops = NULL;
 629         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 630         tcp_sk(sk)->is_mptcp = 0;
 631
 632         mptcp_subflow_ops_undo_override(sk);
 633 }
 634
 635 static void subflow_drop_ctx(struct sock *ssk)
 636 {
 637         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 638
 639         if (!ctx)
 640                 return;
 641
 642         subflow_ulp_fallback(ssk, ctx);
 643         if (ctx->conn)
 644                 sock_put(ctx->conn);
 645
 646         kfree_rcu(ctx, rcu);
 647 }
 648
 649 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 650                                      struct mptcp_options_received *mp_opt)
 651 {
 652         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 653
 654         subflow->remote_key = mp_opt->sndr_key;
 655         subflow->fully_established = 1;
 656         subflow->can_ack = 1;
 657         WRITE_ONCE(msk->fully_established, true);
 658 }
 659
 660 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 661                                           struct sk_buff *skb,
 662                                           struct request_sock *req,
 663                                           struct dst_entry *dst,
 664                                           struct request_sock *req_unhash,
 665                                           bool *own_req)
 666 {
 667         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 668         struct mptcp_subflow_request_sock *subflow_req;
 669         struct mptcp_options_received mp_opt;
 670         bool fallback, fallback_is_fatal;
 671         struct sock *new_msk = NULL;
 672         struct sock *child;
 673
 674         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 675
 676         /* After child creation we must look for MPC even when options
 677          * are not parsed
 678          */
 679         mp_opt.suboptions = 0;
 680
 681         /* hopefully temporary handling for MP_JOIN+syncookie */
 682         subflow_req = mptcp_subflow_rsk(req);
 683         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 684         fallback = !tcp_rsk(req)->is_mptcp;
 685         if (fallback)
 686                 goto create_child;
 687
 688         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 689         if (subflow_req->mp_capable) {
 690                 /* we can receive and accept an in-window, out-of-order pkt,
 691                  * which may not carry the MP_CAPABLE opt even on mptcp enabled
 692                  * paths: always try to extract the peer key, and fallback
 693                  * for packets missing it.
 694                  * Even OoO DSS packets coming legitly after dropped or
 695                  * reordered MPC will cause fallback, but we don't have other
 696                  * options.
 697                  */
 698                 mptcp_get_options(skb, &mp_opt);
 699                 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPC)) {
 700                         fallback = true;
 701                         goto create_child;
 702                 }
 703
 704                 new_msk = mptcp_sk_clone(listener->conn, &mp_opt, req);
 705                 if (!new_msk)
 706                         fallback = true;
 707         } else if (subflow_req->mp_join) {
 708                 mptcp_get_options(skb, &mp_opt);
 709                 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ) ||
 710                     !subflow_hmac_valid(req, &mp_opt) ||
 711                     !mptcp_can_accept_new_subflow(subflow_req->msk)) {
 712                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 713                         fallback = true;
 714                 }
 715         }
 716
 717 create_child:
 718         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 719                                                      req_unhash, own_req);
 720
 721         if (child && *own_req) {
 722                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 723
 724                 tcp_rsk(req)->drop_req = false;
 725
 726                 /* we need to fallback on ctx allocation failure and on pre-reqs
 727                  * checking above. In the latter scenario we additionally need
 728                  * to reset the context to non MPTCP status.
 729                  */
 730                 if (!ctx || fallback) {
 731                         if (fallback_is_fatal) {
 732                                 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 733                                 goto dispose_child;
 734                         }
 735
 736                         if (new_msk)
 737                                 mptcp_copy_inaddrs(new_msk, child);
 738                         subflow_drop_ctx(child);
 739                         goto out;
 740                 }
 741
 742                 /* ssk inherits options of listener sk */
 743                 ctx->setsockopt_seq = listener->setsockopt_seq;
 744
 745                 if (ctx->mp_capable) {
 746                         /* this can't race with mptcp_close(), as the msk is
 747                          * not yet exposted to user-space
 748                          */
 749                         inet_sk_state_store((void *)new_msk, TCP_ESTABLISHED);
 750
 751                         /* record the newly created socket as the first msk
 752                          * subflow, but don't link it yet into conn_list
 753                          */
 754                         WRITE_ONCE(mptcp_sk(new_msk)->first, child);
 755
 756                         /* new mpc subflow takes ownership of the newly
 757                          * created mptcp socket
 758                          */
 759                         mptcp_sk(new_msk)->setsockopt_seq = ctx->setsockopt_seq;
 760                         mptcp_pm_new_connection(mptcp_sk(new_msk), child, 1);
 761                         mptcp_token_accept(subflow_req, mptcp_sk(new_msk));
 762                         ctx->conn = new_msk;
 763                         new_msk = NULL;
 764
 765                         /* set msk addresses early to ensure mptcp_pm_get_local_id()
 766                          * uses the correct data
 767                          */
 768                         mptcp_copy_inaddrs(ctx->conn, child);
 769
 770                         /* with OoO packets we can reach here without ingress
 771                          * mpc option
 772                          */
 773                         if (mp_opt.suboptions & OPTIONS_MPTCP_MPC)
 774                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 775                 } else if (ctx->mp_join) {
 776                         struct mptcp_sock *owner;
 777
 778                         owner = subflow_req->msk;
 779                         if (!owner) {
 780                                 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 781                                 goto dispose_child;
 782                         }
 783
 784                         /* move the msk reference ownership to the subflow */
 785                         subflow_req->msk = NULL;
 786                         ctx->conn = (struct sock *)owner;
 787
 788                         if (subflow_use_different_sport(owner, sk)) {
 789                                 pr_debug("ack inet_sport=%d %d",
 790                                          ntohs(inet_sk(sk)->inet_sport),
 791                                          ntohs(inet_sk((struct sock *)owner)->inet_sport));
 792                                 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
 793                                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
 794                                         goto dispose_child;
 795                                 }
 796                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
 797                         }
 798
 799                         if (!mptcp_finish_join(child))
 800                                 goto dispose_child;
 801
 802                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 803                         tcp_rsk(req)->drop_req = true;
 804                 }
 805         }
 806
 807 out:
 808         /* dispose of the left over mptcp master, if any */
 809         if (unlikely(new_msk))
 810                 mptcp_force_close(new_msk);
 811
 812         /* check for expected invariant - should never trigger, just help
 813          * catching eariler subtle bugs
 814          */
 815         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 816                      (!mptcp_subflow_ctx(child) ||
 817                       !mptcp_subflow_ctx(child)->conn));
 818         return child;
 819
 820 dispose_child:
 821         subflow_drop_ctx(child);
 822         tcp_rsk(req)->drop_req = true;
 823         inet_csk_prepare_for_destroy_sock(child);
 824         tcp_done(child);
 825         req->rsk_ops->send_reset(sk, skb);
 826
 827         /* The last child reference will be released by the caller */
 828         return child;
 829 }
 830
 831 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
 832 static struct proto tcp_prot_override;
 833
 834 enum mapping_status {
 835         MAPPING_OK,
 836         MAPPING_INVALID,
 837         MAPPING_EMPTY,
 838         MAPPING_DATA_FIN,
 839         MAPPING_DUMMY,
 840         MAPPING_BAD_CSUM
 841 };
 842
 843 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 844 {
 845         pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 846                  ssn, subflow->map_subflow_seq, subflow->map_data_len);
 847 }
 848
 849 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 850 {
 851         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 852         unsigned int skb_consumed;
 853
 854         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 855         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 856                 return true;
 857
 858         return skb->len - skb_consumed <= subflow->map_data_len -
 859                                           mptcp_subflow_get_map_offset(subflow);
 860 }
 861
 862 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 863 {
 864         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 865         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 866
 867         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 868                 /* Mapping covers data later in the subflow stream,
 869                  * currently unsupported.
 870                  */
 871                 dbg_bad_map(subflow, ssn);
 872                 return false;
 873         }
 874         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 875                                   subflow->map_data_len))) {
 876                 /* Mapping does covers past subflow data, invalid */
 877                 dbg_bad_map(subflow, ssn);
 878                 return false;
 879         }
 880         return true;
 881 }
 882
 883 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
 884                                               bool csum_reqd)
 885 {
 886         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 887         u32 offset, seq, delta;
 888         __sum16 csum;
 889         int len;
 890
 891         if (!csum_reqd)
 892                 return MAPPING_OK;
 893
 894         /* mapping already validated on previous traversal */
 895         if (subflow->map_csum_len == subflow->map_data_len)
 896                 return MAPPING_OK;
 897
 898         /* traverse the receive queue, ensuring it contains a full
 899          * DSS mapping and accumulating the related csum.
 900          * Preserve the accoumlate csum across multiple calls, to compute
 901          * the csum only once
 902          */
 903         delta = subflow->map_data_len - subflow->map_csum_len;
 904         for (;;) {
 905                 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
 906                 offset = seq - TCP_SKB_CB(skb)->seq;
 907
 908                 /* if the current skb has not been accounted yet, csum its contents
 909                  * up to the amount covered by the current DSS
 910                  */
 911                 if (offset < skb->len) {
 912                         __wsum csum;
 913
 914                         len = min(skb->len - offset, delta);
 915                         csum = skb_checksum(skb, offset, len, 0);
 916                         subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
 917                                                                 subflow->map_csum_len);
 918
 919                         delta -= len;
 920                         subflow->map_csum_len += len;
 921                 }
 922                 if (delta == 0)
 923                         break;
 924
 925                 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
 926                         /* if this subflow is closed, the partial mapping
 927                          * will be never completed; flush the pending skbs, so
 928                          * that subflow_sched_work_if_closed() can kick in
 929                          */
 930                         if (unlikely(ssk->sk_state == TCP_CLOSE))
 931                                 while ((skb = skb_peek(&ssk->sk_receive_queue)))
 932                                         sk_eat_skb(ssk, skb);
 933
 934                         /* not enough data to validate the csum */
 935                         return MAPPING_EMPTY;
 936                 }
 937
 938                 /* the DSS mapping for next skbs will be validated later,
 939                  * when a get_mapping_status call will process such skb
 940                  */
 941                 skb = skb->next;
 942         }
 943
 944         /* note that 'map_data_len' accounts only for the carried data, does
 945          * not include the eventual seq increment due to the data fin,
 946          * while the pseudo header requires the original DSS data len,
 947          * including that
 948          */
 949         csum = __mptcp_make_csum(subflow->map_seq,
 950                                  subflow->map_subflow_seq,
 951                                  subflow->map_data_len + subflow->map_data_fin,
 952                                  subflow->map_data_csum);
 953         if (unlikely(csum)) {
 954                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
 955                 return MAPPING_BAD_CSUM;
 956         }
 957
 958         subflow->valid_csum_seen = 1;
 959         return MAPPING_OK;
 960 }
 961
 962 static enum mapping_status get_mapping_status(struct sock *ssk,
 963                                               struct mptcp_sock *msk)
 964 {
 965         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 966         bool csum_reqd = READ_ONCE(msk->csum_enabled);
 967         struct mptcp_ext *mpext;
 968         struct sk_buff *skb;
 969         u16 data_len;
 970         u64 map_seq;
 971
 972         skb = skb_peek(&ssk->sk_receive_queue);
 973         if (!skb)
 974                 return MAPPING_EMPTY;
 975
 976         if (mptcp_check_fallback(ssk))
 977                 return MAPPING_DUMMY;
 978
 979         mpext = mptcp_get_ext(skb);
 980         if (!mpext || !mpext->use_map) {
 981                 if (!subflow->map_valid && !skb->len) {
 982                         /* the TCP stack deliver 0 len FIN pkt to the receive
 983                          * queue, that is the only 0len pkts ever expected here,
 984                          * and we can admit no mapping only for 0 len pkts
 985                          */
 986                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
 987                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
 988                                           TCP_SKB_CB(skb)->seq,
 989                                           TCP_SKB_CB(skb)->end_seq,
 990                                           TCP_SKB_CB(skb)->tcp_flags);
 991                         sk_eat_skb(ssk, skb);
 992                         return MAPPING_EMPTY;
 993                 }
 994
 995                 if (!subflow->map_valid)
 996                         return MAPPING_INVALID;
 997
 998                 goto validate_seq;
 999         }
1000
1001         trace_get_mapping_status(mpext);
1002
1003         data_len = mpext->data_len;
1004         if (data_len == 0) {
1005                 pr_debug("infinite mapping received");
1006                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1007                 subflow->map_data_len = 0;
1008                 return MAPPING_INVALID;
1009         }
1010
1011         if (mpext->data_fin == 1) {
1012                 if (data_len == 1) {
1013                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1014                                                                  mpext->dsn64);
1015                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
1016                         if (subflow->map_valid) {
1017                                 /* A DATA_FIN might arrive in a DSS
1018                                  * option before the previous mapping
1019                                  * has been fully consumed. Continue
1020                                  * handling the existing mapping.
1021                                  */
1022                                 skb_ext_del(skb, SKB_EXT_MPTCP);
1023                                 return MAPPING_OK;
1024                         } else {
1025                                 if (updated && schedule_work(&msk->work))
1026                                         sock_hold((struct sock *)msk);
1027
1028                                 return MAPPING_DATA_FIN;
1029                         }
1030                 } else {
1031                         u64 data_fin_seq = mpext->data_seq + data_len - 1;
1032
1033                         /* If mpext->data_seq is a 32-bit value, data_fin_seq
1034                          * must also be limited to 32 bits.
1035                          */
1036                         if (!mpext->dsn64)
1037                                 data_fin_seq &= GENMASK_ULL(31, 0);
1038
1039                         mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1040                         pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
1041                                  data_fin_seq, mpext->dsn64);
1042                 }
1043
1044                 /* Adjust for DATA_FIN using 1 byte of sequence space */
1045                 data_len--;
1046         }
1047
1048         map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1049         WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1050
1051         if (subflow->map_valid) {
1052                 /* Allow replacing only with an identical map */
1053                 if (subflow->map_seq == map_seq &&
1054                     subflow->map_subflow_seq == mpext->subflow_seq &&
1055                     subflow->map_data_len == data_len &&
1056                     subflow->map_csum_reqd == mpext->csum_reqd) {
1057                         skb_ext_del(skb, SKB_EXT_MPTCP);
1058                         goto validate_csum;
1059                 }
1060
1061                 /* If this skb data are fully covered by the current mapping,
1062                  * the new map would need caching, which is not supported
1063                  */
1064                 if (skb_is_fully_mapped(ssk, skb)) {
1065                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1066                         return MAPPING_INVALID;
1067                 }
1068
1069                 /* will validate the next map after consuming the current one */
1070                 goto validate_csum;
1071         }
1072
1073         subflow->map_seq = map_seq;
1074         subflow->map_subflow_seq = mpext->subflow_seq;
1075         subflow->map_data_len = data_len;
1076         subflow->map_valid = 1;
1077         subflow->map_data_fin = mpext->data_fin;
1078         subflow->mpc_map = mpext->mpc_map;
1079         subflow->map_csum_reqd = mpext->csum_reqd;
1080         subflow->map_csum_len = 0;
1081         subflow->map_data_csum = csum_unfold(mpext->csum);
1082
1083         /* Cfr RFC 8684 Section 3.3.0 */
1084         if (unlikely(subflow->map_csum_reqd != csum_reqd))
1085                 return MAPPING_INVALID;
1086
1087         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
1088                  subflow->map_seq, subflow->map_subflow_seq,
1089                  subflow->map_data_len, subflow->map_csum_reqd,
1090                  subflow->map_data_csum);
1091
1092 validate_seq:
1093         /* we revalidate valid mapping on new skb, because we must ensure
1094          * the current skb is completely covered by the available mapping
1095          */
1096         if (!validate_mapping(ssk, skb)) {
1097                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1098                 return MAPPING_INVALID;
1099         }
1100
1101         skb_ext_del(skb, SKB_EXT_MPTCP);
1102
1103 validate_csum:
1104         return validate_data_csum(ssk, skb, csum_reqd);
1105 }
1106
1107 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1108                                        u64 limit)
1109 {
1110         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1111         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1112         u32 incr;
1113
1114         incr = limit >= skb->len ? skb->len + fin : limit;
1115
1116         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
1117                  subflow->map_subflow_seq);
1118         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1119         tcp_sk(ssk)->copied_seq += incr;
1120         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1121                 sk_eat_skb(ssk, skb);
1122         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1123                 subflow->map_valid = 0;
1124 }
1125
1126 /* sched mptcp worker to remove the subflow if no more data is pending */
1127 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1128 {
1129         struct sock *sk = (struct sock *)msk;
1130
1131         if (likely(ssk->sk_state != TCP_CLOSE))
1132                 return;
1133
1134         if (skb_queue_empty(&ssk->sk_receive_queue) &&
1135             !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags)) {
1136                 sock_hold(sk);
1137                 if (!schedule_work(&msk->work))
1138                         sock_put(sk);
1139         }
1140 }
1141
1142 static bool subflow_can_fallback(struct mptcp_subflow_context *subflow)
1143 {
1144         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1145
1146         if (subflow->mp_join)
1147                 return false;
1148         else if (READ_ONCE(msk->csum_enabled))
1149                 return !subflow->valid_csum_seen;
1150         else
1151                 return !subflow->fully_established;
1152 }
1153
1154 static void mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk)
1155 {
1156         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1157         unsigned long fail_tout;
1158
1159         /* greceful failure can happen only on the MPC subflow */
1160         if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first)))
1161                 return;
1162
1163         /* since the close timeout take precedence on the fail one,
1164          * no need to start the latter when the first is already set
1165          */
1166         if (sock_flag((struct sock *)msk, SOCK_DEAD))
1167                 return;
1168
1169         /* we don't need extreme accuracy here, use a zero fail_tout as special
1170          * value meaning no fail timeout at all;
1171          */
1172         fail_tout = jiffies + TCP_RTO_MAX;
1173         if (!fail_tout)
1174                 fail_tout = 1;
1175         WRITE_ONCE(subflow->fail_tout, fail_tout);
1176         tcp_send_ack(ssk);
1177
1178         mptcp_reset_timeout(msk, subflow->fail_tout);
1179 }
1180
1181 static bool subflow_check_data_avail(struct sock *ssk)
1182 {
1183         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1184         enum mapping_status status;
1185         struct mptcp_sock *msk;
1186         struct sk_buff *skb;
1187
1188         if (!skb_peek(&ssk->sk_receive_queue))
1189                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1190         if (subflow->data_avail)
1191                 return true;
1192
1193         msk = mptcp_sk(subflow->conn);
1194         for (;;) {
1195                 u64 ack_seq;
1196                 u64 old_ack;
1197
1198                 status = get_mapping_status(ssk, msk);
1199                 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1200                 if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY ||
1201                              status == MAPPING_BAD_CSUM))
1202                         goto fallback;
1203
1204                 if (status != MAPPING_OK)
1205                         goto no_data;
1206
1207                 skb = skb_peek(&ssk->sk_receive_queue);
1208                 if (WARN_ON_ONCE(!skb))
1209                         goto no_data;
1210
1211                 /* if msk lacks the remote key, this subflow must provide an
1212                  * MP_CAPABLE-based mapping
1213                  */
1214                 if (unlikely(!READ_ONCE(msk->can_ack))) {
1215                         if (!subflow->mpc_map)
1216                                 goto fallback;
1217                         WRITE_ONCE(msk->remote_key, subflow->remote_key);
1218                         WRITE_ONCE(msk->ack_seq, subflow->map_seq);
1219                         WRITE_ONCE(msk->can_ack, true);
1220                 }
1221
1222                 old_ack = READ_ONCE(msk->ack_seq);
1223                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1224                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1225                          ack_seq);
1226                 if (unlikely(before64(ack_seq, old_ack))) {
1227                         mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1228                         continue;
1229                 }
1230
1231                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1232                 break;
1233         }
1234         return true;
1235
1236 no_data:
1237         subflow_sched_work_if_closed(msk, ssk);
1238         return false;
1239
1240 fallback:
1241         if (!__mptcp_check_fallback(msk)) {
1242                 /* RFC 8684 section 3.7. */
1243                 if (status == MAPPING_BAD_CSUM &&
1244                     (subflow->mp_join || subflow->valid_csum_seen)) {
1245                         subflow->send_mp_fail = 1;
1246
1247                         if (!READ_ONCE(msk->allow_infinite_fallback)) {
1248                                 subflow->reset_transient = 0;
1249                                 subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1250                                 goto reset;
1251                         }
1252                         mptcp_subflow_fail(msk, ssk);
1253                         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1254                         return true;
1255                 }
1256
1257                 if (!subflow_can_fallback(subflow) && subflow->map_data_len) {
1258                         /* fatal protocol error, close the socket.
1259                          * subflow_error_report() will introduce the appropriate barriers
1260                          */
1261                         subflow->reset_transient = 0;
1262                         subflow->reset_reason = MPTCP_RST_EMPTCP;
1263
1264 reset:
1265                         ssk->sk_err = EBADMSG;
1266                         tcp_set_state(ssk, TCP_CLOSE);
1267                         while ((skb = skb_peek(&ssk->sk_receive_queue)))
1268                                 sk_eat_skb(ssk, skb);
1269                         tcp_send_active_reset(ssk, GFP_ATOMIC);
1270                         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1271                         return false;
1272                 }
1273
1274                 mptcp_do_fallback(ssk);
1275         }
1276
1277         skb = skb_peek(&ssk->sk_receive_queue);
1278         subflow->map_valid = 1;
1279         subflow->map_seq = READ_ONCE(msk->ack_seq);
1280         subflow->map_data_len = skb->len;
1281         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1282         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1283         return true;
1284 }
1285
1286 bool mptcp_subflow_data_available(struct sock *sk)
1287 {
1288         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1289
1290         /* check if current mapping is still valid */
1291         if (subflow->map_valid &&
1292             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1293                 subflow->map_valid = 0;
1294                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1295
1296                 pr_debug("Done with mapping: seq=%u data_len=%u",
1297                          subflow->map_subflow_seq,
1298                          subflow->map_data_len);
1299         }
1300
1301         return subflow_check_data_avail(sk);
1302 }
1303
1304 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1305  * not the ssk one.
1306  *
1307  * In mptcp, rwin is about the mptcp-level connection data.
1308  *
1309  * Data that is still on the ssk rx queue can thus be ignored,
1310  * as far as mptcp peer is concerned that data is still inflight.
1311  * DSS ACK is updated when skb is moved to the mptcp rx queue.
1312  */
1313 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1314 {
1315         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1316         const struct sock *sk = subflow->conn;
1317
1318         *space = __mptcp_space(sk);
1319         *full_space = tcp_full_space(sk);
1320 }
1321
1322 void __mptcp_error_report(struct sock *sk)
1323 {
1324         struct mptcp_subflow_context *subflow;
1325         struct mptcp_sock *msk = mptcp_sk(sk);
1326
1327         mptcp_for_each_subflow(msk, subflow) {
1328                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1329                 int err = sock_error(ssk);
1330
1331                 if (!err)
1332                         continue;
1333
1334                 /* only propagate errors on fallen-back sockets or
1335                  * on MPC connect
1336                  */
1337                 if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(msk))
1338                         continue;
1339
1340                 inet_sk_state_store(sk, inet_sk_state_load(ssk));
1341                 sk->sk_err = -err;
1342
1343                 /* This barrier is coupled with smp_rmb() in mptcp_poll() */
1344                 smp_wmb();
1345                 sk_error_report(sk);
1346                 break;
1347         }
1348 }
1349
1350 static void subflow_error_report(struct sock *ssk)
1351 {
1352         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1353
1354         mptcp_data_lock(sk);
1355         if (!sock_owned_by_user(sk))
1356                 __mptcp_error_report(sk);
1357         else
1358                 __set_bit(MPTCP_ERROR_REPORT,  &mptcp_sk(sk)->cb_flags);
1359         mptcp_data_unlock(sk);
1360 }
1361
1362 static void subflow_data_ready(struct sock *sk)
1363 {
1364         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1365         u16 state = 1 << inet_sk_state_load(sk);
1366         struct sock *parent = subflow->conn;
1367         struct mptcp_sock *msk;
1368
1369         msk = mptcp_sk(parent);
1370         if (state & TCPF_LISTEN) {
1371                 /* MPJ subflow are removed from accept queue before reaching here,
1372                  * avoid stray wakeups
1373                  */
1374                 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1375                         return;
1376
1377                 parent->sk_data_ready(parent);
1378                 return;
1379         }
1380
1381         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1382                      !subflow->mp_join && !(state & TCPF_CLOSE));
1383
1384         if (mptcp_subflow_data_available(sk))
1385                 mptcp_data_ready(parent, sk);
1386         else if (unlikely(sk->sk_err))
1387                 subflow_error_report(sk);
1388 }
1389
1390 static void subflow_write_space(struct sock *ssk)
1391 {
1392         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1393
1394         mptcp_propagate_sndbuf(sk, ssk);
1395         mptcp_write_space(sk);
1396 }
1397
1398 static const struct inet_connection_sock_af_ops *
1399 subflow_default_af_ops(struct sock *sk)
1400 {
1401 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1402         if (sk->sk_family == AF_INET6)
1403                 return &subflow_v6_specific;
1404 #endif
1405         return &subflow_specific;
1406 }
1407
1408 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1409 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1410 {
1411         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1412         struct inet_connection_sock *icsk = inet_csk(sk);
1413         const struct inet_connection_sock_af_ops *target;
1414
1415         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1416
1417         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1418                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1419
1420         if (likely(icsk->icsk_af_ops == target))
1421                 return;
1422
1423         subflow->icsk_af_ops = icsk->icsk_af_ops;
1424         icsk->icsk_af_ops = target;
1425 }
1426 #endif
1427
1428 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1429                          struct sockaddr_storage *addr,
1430                          unsigned short family)
1431 {
1432         memset(addr, 0, sizeof(*addr));
1433         addr->ss_family = family;
1434         if (addr->ss_family == AF_INET) {
1435                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1436
1437                 if (info->family == AF_INET)
1438                         in_addr->sin_addr = info->addr;
1439 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1440                 else if (ipv6_addr_v4mapped(&info->addr6))
1441                         in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1442 #endif
1443                 in_addr->sin_port = info->port;
1444         }
1445 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1446         else if (addr->ss_family == AF_INET6) {
1447                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1448
1449                 if (info->family == AF_INET)
1450                         ipv6_addr_set_v4mapped(info->addr.s_addr,
1451                                                &in6_addr->sin6_addr);
1452                 else
1453                         in6_addr->sin6_addr = info->addr6;
1454                 in6_addr->sin6_port = info->port;
1455         }
1456 #endif
1457 }
1458
1459 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1460                             const struct mptcp_addr_info *remote)
1461 {
1462         struct mptcp_sock *msk = mptcp_sk(sk);
1463         struct mptcp_subflow_context *subflow;
1464         struct sockaddr_storage addr;
1465         int remote_id = remote->id;
1466         int local_id = loc->id;
1467         int err = -ENOTCONN;
1468         struct socket *sf;
1469         struct sock *ssk;
1470         u32 remote_token;
1471         int addrlen;
1472         int ifindex;
1473         u8 flags;
1474
1475         if (!mptcp_is_fully_established(sk))
1476                 goto err_out;
1477
1478         err = mptcp_subflow_create_socket(sk, &sf);
1479         if (err)
1480                 goto err_out;
1481
1482         ssk = sf->sk;
1483         subflow = mptcp_subflow_ctx(ssk);
1484         do {
1485                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1486         } while (!subflow->local_nonce);
1487
1488         if (local_id)
1489                 subflow_set_local_id(subflow, local_id);
1490
1491         mptcp_pm_get_flags_and_ifindex_by_id(msk, local_id,
1492                                              &flags, &ifindex);
1493         subflow->remote_key = msk->remote_key;
1494         subflow->local_key = msk->local_key;
1495         subflow->token = msk->token;
1496         mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1497
1498         addrlen = sizeof(struct sockaddr_in);
1499 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1500         if (addr.ss_family == AF_INET6)
1501                 addrlen = sizeof(struct sockaddr_in6);
1502 #endif
1503         mptcp_sockopt_sync(msk, ssk);
1504
1505         ssk->sk_bound_dev_if = ifindex;
1506         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1507         if (err)
1508                 goto failed;
1509
1510         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1511         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1512                  remote_token, local_id, remote_id);
1513         subflow->remote_token = remote_token;
1514         subflow->remote_id = remote_id;
1515         subflow->request_join = 1;
1516         subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1517         mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1518
1519         sock_hold(ssk);
1520         list_add_tail(&subflow->node, &msk->conn_list);
1521         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1522         if (err && err != -EINPROGRESS)
1523                 goto failed_unlink;
1524
1525         /* discard the subflow socket */
1526         mptcp_sock_graft(ssk, sk->sk_socket);
1527         iput(SOCK_INODE(sf));
1528         WRITE_ONCE(msk->allow_infinite_fallback, false);
1529         return 0;
1530
1531 failed_unlink:
1532         list_del(&subflow->node);
1533         sock_put(mptcp_subflow_tcp_sock(subflow));
1534
1535 failed:
1536         subflow->disposable = 1;
1537         sock_release(sf);
1538
1539 err_out:
1540         /* we account subflows before the creation, and this failures will not
1541          * be caught by sk_state_change()
1542          */
1543         mptcp_pm_close_subflow(msk);
1544         return err;
1545 }
1546
1547 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1548 {
1549 #ifdef CONFIG_SOCK_CGROUP_DATA
1550         struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1551                                 *child_skcd = &child->sk_cgrp_data;
1552
1553         /* only the additional subflows created by kworkers have to be modified */
1554         if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1555             cgroup_id(sock_cgroup_ptr(child_skcd))) {
1556 #ifdef CONFIG_MEMCG
1557                 struct mem_cgroup *memcg = parent->sk_memcg;
1558
1559                 mem_cgroup_sk_free(child);
1560                 if (memcg && css_tryget(&memcg->css))
1561                         child->sk_memcg = memcg;
1562 #endif /* CONFIG_MEMCG */
1563
1564                 cgroup_sk_free(child_skcd);
1565                 *child_skcd = *parent_skcd;
1566                 cgroup_sk_clone(child_skcd);
1567         }
1568 #endif /* CONFIG_SOCK_CGROUP_DATA */
1569 }
1570
1571 static void mptcp_subflow_ops_override(struct sock *ssk)
1572 {
1573 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1574         if (ssk->sk_prot == &tcpv6_prot)
1575                 ssk->sk_prot = &tcpv6_prot_override;
1576         else
1577 #endif
1578                 ssk->sk_prot = &tcp_prot_override;
1579 }
1580
1581 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1582 {
1583 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1584         if (ssk->sk_prot == &tcpv6_prot_override)
1585                 ssk->sk_prot = &tcpv6_prot;
1586         else
1587 #endif
1588                 ssk->sk_prot = &tcp_prot;
1589 }
1590 int mptcp_subflow_create_socket(struct sock *sk, struct socket **new_sock)
1591 {
1592         struct mptcp_subflow_context *subflow;
1593         struct net *net = sock_net(sk);
1594         struct socket *sf;
1595         int err;
1596
1597         /* un-accepted server sockets can reach here - on bad configuration
1598          * bail early to avoid greater trouble later
1599          */
1600         if (unlikely(!sk->sk_socket))
1601                 return -EINVAL;
1602
1603         err = sock_create_kern(net, sk->sk_family, SOCK_STREAM, IPPROTO_TCP,
1604                                &sf);
1605         if (err)
1606                 return err;
1607
1608         lock_sock(sf->sk);
1609
1610         /* the newly created socket has to be in the same cgroup as its parent */
1611         mptcp_attach_cgroup(sk, sf->sk);
1612
1613         /* kernel sockets do not by default acquire net ref, but TCP timer
1614          * needs it.
1615          */
1616         sf->sk->sk_net_refcnt = 1;
1617         get_net_track(net, &sf->sk->ns_tracker, GFP_KERNEL);
1618         sock_inuse_add(net, 1);
1619         err = tcp_set_ulp(sf->sk, "mptcp");
1620         release_sock(sf->sk);
1621
1622         if (err) {
1623                 sock_release(sf);
1624                 return err;
1625         }
1626
1627         /* the newly created socket really belongs to the owning MPTCP master
1628          * socket, even if for additional subflows the allocation is performed
1629          * by a kernel workqueue. Adjust inode references, so that the
1630          * procfs/diag interfaces really show this one belonging to the correct
1631          * user.
1632          */
1633         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1634         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1635         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1636
1637         subflow = mptcp_subflow_ctx(sf->sk);
1638         pr_debug("subflow=%p", subflow);
1639
1640         *new_sock = sf;
1641         sock_hold(sk);
1642         subflow->conn = sk;
1643         mptcp_subflow_ops_override(sf->sk);
1644
1645         return 0;
1646 }
1647
1648 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1649                                                         gfp_t priority)
1650 {
1651         struct inet_connection_sock *icsk = inet_csk(sk);
1652         struct mptcp_subflow_context *ctx;
1653
1654         ctx = kzalloc(sizeof(*ctx), priority);
1655         if (!ctx)
1656                 return NULL;
1657
1658         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1659         INIT_LIST_HEAD(&ctx->node);
1660         INIT_LIST_HEAD(&ctx->delegated_node);
1661
1662         pr_debug("subflow=%p", ctx);
1663
1664         ctx->tcp_sock = sk;
1665
1666         return ctx;
1667 }
1668
1669 static void __subflow_state_change(struct sock *sk)
1670 {
1671         struct socket_wq *wq;
1672
1673         rcu_read_lock();
1674         wq = rcu_dereference(sk->sk_wq);
1675         if (skwq_has_sleeper(wq))
1676                 wake_up_interruptible_all(&wq->wait);
1677         rcu_read_unlock();
1678 }
1679
1680 static bool subflow_is_done(const struct sock *sk)
1681 {
1682         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1683 }
1684
1685 static void subflow_state_change(struct sock *sk)
1686 {
1687         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1688         struct sock *parent = subflow->conn;
1689
1690         __subflow_state_change(sk);
1691
1692         if (subflow_simultaneous_connect(sk)) {
1693                 mptcp_propagate_sndbuf(parent, sk);
1694                 mptcp_do_fallback(sk);
1695                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1696                 pr_fallback(mptcp_sk(parent));
1697                 subflow->conn_finished = 1;
1698                 mptcp_set_connected(parent);
1699         }
1700
1701         /* as recvmsg() does not acquire the subflow socket for ssk selection
1702          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1703          * the data available machinery here.
1704          */
1705         if (mptcp_subflow_data_available(sk))
1706                 mptcp_data_ready(parent, sk);
1707         else if (unlikely(sk->sk_err))
1708                 subflow_error_report(sk);
1709
1710         subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1711
1712         if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1713             !subflow->rx_eof && subflow_is_done(sk)) {
1714                 subflow->rx_eof = 1;
1715                 mptcp_subflow_eof(parent);
1716         }
1717 }
1718
1719 void mptcp_subflow_queue_clean(struct sock *listener_ssk)
1720 {
1721         struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
1722         struct mptcp_sock *msk, *next, *head = NULL;
1723         struct request_sock *req;
1724
1725         /* build a list of all unaccepted mptcp sockets */
1726         spin_lock_bh(&queue->rskq_lock);
1727         for (req = queue->rskq_accept_head; req; req = req->dl_next) {
1728                 struct mptcp_subflow_context *subflow;
1729                 struct sock *ssk = req->sk;
1730                 struct mptcp_sock *msk;
1731
1732                 if (!sk_is_mptcp(ssk))
1733                         continue;
1734
1735                 subflow = mptcp_subflow_ctx(ssk);
1736                 if (!subflow || !subflow->conn)
1737                         continue;
1738
1739                 /* skip if already in list */
1740                 msk = mptcp_sk(subflow->conn);
1741                 if (msk->dl_next || msk == head)
1742                         continue;
1743
1744                 msk->dl_next = head;
1745                 head = msk;
1746         }
1747         spin_unlock_bh(&queue->rskq_lock);
1748         if (!head)
1749                 return;
1750
1751         /* can't acquire the msk socket lock under the subflow one,
1752          * or will cause ABBA deadlock
1753          */
1754         release_sock(listener_ssk);
1755
1756         for (msk = head; msk; msk = next) {
1757                 struct sock *sk = (struct sock *)msk;
1758                 bool do_cancel_work;
1759
1760                 sock_hold(sk);
1761                 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
1762                 next = msk->dl_next;
1763                 msk->first = NULL;
1764                 msk->dl_next = NULL;
1765
1766                 do_cancel_work = __mptcp_close(sk, 0);
1767                 release_sock(sk);
1768                 if (do_cancel_work)
1769                         mptcp_cancel_work(sk);
1770                 sock_put(sk);
1771         }
1772
1773         /* we are still under the listener msk socket lock */
1774         lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
1775 }
1776
1777 static int subflow_ulp_init(struct sock *sk)
1778 {
1779         struct inet_connection_sock *icsk = inet_csk(sk);
1780         struct mptcp_subflow_context *ctx;
1781         struct tcp_sock *tp = tcp_sk(sk);
1782         int err = 0;
1783
1784         /* disallow attaching ULP to a socket unless it has been
1785          * created with sock_create_kern()
1786          */
1787         if (!sk->sk_kern_sock) {
1788                 err = -EOPNOTSUPP;
1789                 goto out;
1790         }
1791
1792         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1793         if (!ctx) {
1794                 err = -ENOMEM;
1795                 goto out;
1796         }
1797
1798         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1799
1800         tp->is_mptcp = 1;
1801         ctx->icsk_af_ops = icsk->icsk_af_ops;
1802         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1803         ctx->tcp_state_change = sk->sk_state_change;
1804         ctx->tcp_error_report = sk->sk_error_report;
1805
1806         WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1807         WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
1808
1809         sk->sk_data_ready = subflow_data_ready;
1810         sk->sk_write_space = subflow_write_space;
1811         sk->sk_state_change = subflow_state_change;
1812         sk->sk_error_report = subflow_error_report;
1813 out:
1814         return err;
1815 }
1816
1817 static void subflow_ulp_release(struct sock *ssk)
1818 {
1819         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1820         bool release = true;
1821         struct sock *sk;
1822
1823         if (!ctx)
1824                 return;
1825
1826         sk = ctx->conn;
1827         if (sk) {
1828                 /* if the msk has been orphaned, keep the ctx
1829                  * alive, will be freed by __mptcp_close_ssk(),
1830                  * when the subflow is still unaccepted
1831                  */
1832                 release = ctx->disposable || list_empty(&ctx->node);
1833                 sock_put(sk);
1834         }
1835
1836         mptcp_subflow_ops_undo_override(ssk);
1837         if (release)
1838                 kfree_rcu(ctx, rcu);
1839 }
1840
1841 static void subflow_ulp_clone(const struct request_sock *req,
1842                               struct sock *newsk,
1843                               const gfp_t priority)
1844 {
1845         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1846         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1847         struct mptcp_subflow_context *new_ctx;
1848
1849         if (!tcp_rsk(req)->is_mptcp ||
1850             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1851                 subflow_ulp_fallback(newsk, old_ctx);
1852                 return;
1853         }
1854
1855         new_ctx = subflow_create_ctx(newsk, priority);
1856         if (!new_ctx) {
1857                 subflow_ulp_fallback(newsk, old_ctx);
1858                 return;
1859         }
1860
1861         new_ctx->conn_finished = 1;
1862         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1863         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1864         new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1865         new_ctx->rel_write_seq = 1;
1866         new_ctx->tcp_sock = newsk;
1867
1868         if (subflow_req->mp_capable) {
1869                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1870                  * is fully established only after we receive the remote key
1871                  */
1872                 new_ctx->mp_capable = 1;
1873                 new_ctx->local_key = subflow_req->local_key;
1874                 new_ctx->token = subflow_req->token;
1875                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1876                 new_ctx->idsn = subflow_req->idsn;
1877
1878                 /* this is the first subflow, id is always 0 */
1879                 new_ctx->local_id_valid = 1;
1880         } else if (subflow_req->mp_join) {
1881                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1882                 new_ctx->mp_join = 1;
1883                 new_ctx->fully_established = 1;
1884                 new_ctx->backup = subflow_req->backup;
1885                 new_ctx->remote_id = subflow_req->remote_id;
1886                 new_ctx->token = subflow_req->token;
1887                 new_ctx->thmac = subflow_req->thmac;
1888
1889                 /* the subflow req id is valid, fetched via subflow_check_req()
1890                  * and subflow_token_join_request()
1891                  */
1892                 subflow_set_local_id(new_ctx, subflow_req->local_id);
1893         }
1894 }
1895
1896 static void tcp_release_cb_override(struct sock *ssk)
1897 {
1898         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1899
1900         if (mptcp_subflow_has_delegated_action(subflow))
1901                 mptcp_subflow_process_delegated(ssk);
1902
1903         tcp_release_cb(ssk);
1904 }
1905
1906 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
1907         .name           = "mptcp",
1908         .owner          = THIS_MODULE,
1909         .init           = subflow_ulp_init,
1910         .release        = subflow_ulp_release,
1911         .clone          = subflow_ulp_clone,
1912 };
1913
1914 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
1915 {
1916         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
1917         subflow_ops->slab_name = "request_sock_subflow";
1918
1919         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
1920                                               subflow_ops->obj_size, 0,
1921                                               SLAB_ACCOUNT |
1922                                               SLAB_TYPESAFE_BY_RCU,
1923                                               NULL);
1924         if (!subflow_ops->slab)
1925                 return -ENOMEM;
1926
1927         subflow_ops->destructor = subflow_req_destructor;
1928
1929         return 0;
1930 }
1931
1932 void __init mptcp_subflow_init(void)
1933 {
1934         mptcp_subflow_request_sock_ops = tcp_request_sock_ops;
1935         if (subflow_ops_init(&mptcp_subflow_request_sock_ops) != 0)
1936                 panic("MPTCP: failed to init subflow request sock ops\n");
1937
1938         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
1939         subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
1940
1941         subflow_specific = ipv4_specific;
1942         subflow_specific.conn_request = subflow_v4_conn_request;
1943         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
1944         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
1945         subflow_specific.rebuild_header = subflow_rebuild_header;
1946
1947         tcp_prot_override = tcp_prot;
1948         tcp_prot_override.release_cb = tcp_release_cb_override;
1949
1950 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1951         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
1952         subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
1953
1954         subflow_v6_specific = ipv6_specific;
1955         subflow_v6_specific.conn_request = subflow_v6_conn_request;
1956         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
1957         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
1958         subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
1959
1960         subflow_v6m_specific = subflow_v6_specific;
1961         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
1962         subflow_v6m_specific.send_check = ipv4_specific.send_check;
1963         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
1964         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
1965         subflow_v6m_specific.net_frag_header_len = 0;
1966         subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
1967
1968         tcpv6_prot_override = tcpv6_prot;
1969         tcpv6_prot_override.release_cb = tcp_release_cb_override;
1970 #endif
1971
1972         mptcp_diag_subflow_init(&subflow_ulp_ops);
1973
1974         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
1975                 panic("MPTCP: failed to register subflows to ULP\n");
1976 }