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 #include <trace/events/sock.h>
  30
  31 static void mptcp_subflow_ops_undo_override(struct sock *ssk);
  32
  33 static void SUBFLOW_REQ_INC_STATS(struct request_sock *req,
  34                                   enum linux_mptcp_mib_field field)
  35 {
  36         MPTCP_INC_STATS(sock_net(req_to_sk(req)), field);
  37 }
  38
  39 static void subflow_req_destructor(struct request_sock *req)
  40 {
  41         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
  42
  43         pr_debug("subflow_req=%p", subflow_req);
  44
  45         if (subflow_req->msk)
  46                 sock_put((struct sock *)subflow_req->msk);
  47
  48         mptcp_token_destroy_request(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 static void subflow_prep_synack(const struct sock *sk, struct request_sock *req,
 311                                 struct tcp_fastopen_cookie *foc,
 312                                 enum tcp_synack_type synack_type)
 313 {
 314         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 315         struct inet_request_sock *ireq = inet_rsk(req);
 316
 317         /* clear tstamp_ok, as needed depending on cookie */
 318         if (foc && foc->len > -1)
 319                 ireq->tstamp_ok = 0;
 320
 321         if (synack_type == TCP_SYNACK_FASTOPEN)
 322                 mptcp_fastopen_subflow_synack_set_params(subflow, req);
 323 }
 324
 325 static int subflow_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
 326                                   struct flowi *fl,
 327                                   struct request_sock *req,
 328                                   struct tcp_fastopen_cookie *foc,
 329                                   enum tcp_synack_type synack_type,
 330                                   struct sk_buff *syn_skb)
 331 {
 332         subflow_prep_synack(sk, req, foc, synack_type);
 333
 334         return tcp_request_sock_ipv4_ops.send_synack(sk, dst, fl, req, foc,
 335                                                      synack_type, syn_skb);
 336 }
 337
 338 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 339 static int subflow_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
 340                                   struct flowi *fl,
 341                                   struct request_sock *req,
 342                                   struct tcp_fastopen_cookie *foc,
 343                                   enum tcp_synack_type synack_type,
 344                                   struct sk_buff *syn_skb)
 345 {
 346         subflow_prep_synack(sk, req, foc, synack_type);
 347
 348         return tcp_request_sock_ipv6_ops.send_synack(sk, dst, fl, req, foc,
 349                                                      synack_type, syn_skb);
 350 }
 351
 352 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
 353                                               struct sk_buff *skb,
 354                                               struct flowi *fl,
 355                                               struct request_sock *req)
 356 {
 357         struct dst_entry *dst;
 358         int err;
 359
 360         tcp_rsk(req)->is_mptcp = 1;
 361         subflow_init_req(req, sk);
 362
 363         dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req);
 364         if (!dst)
 365                 return NULL;
 366
 367         err = subflow_check_req(req, sk, skb);
 368         if (err == 0)
 369                 return dst;
 370
 371         dst_release(dst);
 372         if (!req->syncookie)
 373                 tcp6_request_sock_ops.send_reset(sk, skb);
 374         return NULL;
 375 }
 376 #endif
 377
 378 /* validate received truncated hmac and create hmac for third ACK */
 379 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
 380 {
 381         u8 hmac[SHA256_DIGEST_SIZE];
 382         u64 thmac;
 383
 384         subflow_generate_hmac(subflow->remote_key, subflow->local_key,
 385                               subflow->remote_nonce, subflow->local_nonce,
 386                               hmac);
 387
 388         thmac = get_unaligned_be64(hmac);
 389         pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
 390                  subflow, subflow->token, thmac, subflow->thmac);
 391
 392         return thmac == subflow->thmac;
 393 }
 394
 395 void mptcp_subflow_reset(struct sock *ssk)
 396 {
 397         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 398         struct sock *sk = subflow->conn;
 399
 400         /* mptcp_mp_fail_no_response() can reach here on an already closed
 401          * socket
 402          */
 403         if (ssk->sk_state == TCP_CLOSE)
 404                 return;
 405
 406         /* must hold: tcp_done() could drop last reference on parent */
 407         sock_hold(sk);
 408
 409         tcp_send_active_reset(ssk, GFP_ATOMIC);
 410         tcp_done(ssk);
 411         if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags))
 412                 mptcp_schedule_work(sk);
 413
 414         sock_put(sk);
 415 }
 416
 417 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
 418 {
 419         return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
 420 }
 421
 422 void __mptcp_set_connected(struct sock *sk)
 423 {
 424         if (sk->sk_state == TCP_SYN_SENT) {
 425                 inet_sk_state_store(sk, TCP_ESTABLISHED);
 426                 sk->sk_state_change(sk);
 427         }
 428 }
 429
 430 static void mptcp_set_connected(struct sock *sk)
 431 {
 432         mptcp_data_lock(sk);
 433         if (!sock_owned_by_user(sk))
 434                 __mptcp_set_connected(sk);
 435         else
 436                 __set_bit(MPTCP_CONNECTED, &mptcp_sk(sk)->cb_flags);
 437         mptcp_data_unlock(sk);
 438 }
 439
 440 static void subflow_set_remote_key(struct mptcp_sock *msk,
 441                                    struct mptcp_subflow_context *subflow,
 442                                    const struct mptcp_options_received *mp_opt)
 443 {
 444         /* active MPC subflow will reach here multiple times:
 445          * at subflow_finish_connect() time and at 4th ack time
 446          */
 447         if (subflow->remote_key_valid)
 448                 return;
 449
 450         subflow->remote_key_valid = 1;
 451         subflow->remote_key = mp_opt->sndr_key;
 452         mptcp_crypto_key_sha(subflow->remote_key, NULL, &subflow->iasn);
 453         subflow->iasn++;
 454
 455         WRITE_ONCE(msk->remote_key, subflow->remote_key);
 456         WRITE_ONCE(msk->ack_seq, subflow->iasn);
 457         WRITE_ONCE(msk->can_ack, true);
 458         atomic64_set(&msk->rcv_wnd_sent, subflow->iasn);
 459 }
 460
 461 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 462 {
 463         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 464         struct mptcp_options_received mp_opt;
 465         struct sock *parent = subflow->conn;
 466         struct mptcp_sock *msk;
 467
 468         subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 469
 470         /* be sure no special action on any packet other than syn-ack */
 471         if (subflow->conn_finished)
 472                 return;
 473
 474         msk = mptcp_sk(parent);
 475         mptcp_propagate_sndbuf(parent, sk);
 476         subflow->rel_write_seq = 1;
 477         subflow->conn_finished = 1;
 478         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 479         pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
 480
 481         mptcp_get_options(skb, &mp_opt);
 482         if (subflow->request_mptcp) {
 483                 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPC)) {
 484                         MPTCP_INC_STATS(sock_net(sk),
 485                                         MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
 486                         mptcp_do_fallback(sk);
 487                         pr_fallback(msk);
 488                         goto fallback;
 489                 }
 490
 491                 if (mp_opt.suboptions & OPTION_MPTCP_CSUMREQD)
 492                         WRITE_ONCE(msk->csum_enabled, true);
 493                 if (mp_opt.deny_join_id0)
 494                         WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
 495                 subflow->mp_capable = 1;
 496                 subflow_set_remote_key(msk, subflow, &mp_opt);
 497                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
 498                 mptcp_finish_connect(sk);
 499                 mptcp_set_connected(parent);
 500         } else if (subflow->request_join) {
 501                 u8 hmac[SHA256_DIGEST_SIZE];
 502
 503                 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ)) {
 504                         subflow->reset_reason = MPTCP_RST_EMPTCP;
 505                         goto do_reset;
 506                 }
 507
 508                 subflow->backup = mp_opt.backup;
 509                 subflow->thmac = mp_opt.thmac;
 510                 subflow->remote_nonce = mp_opt.nonce;
 511                 subflow->remote_id = mp_opt.join_id;
 512                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u backup=%d",
 513                          subflow, subflow->thmac, subflow->remote_nonce,
 514                          subflow->backup);
 515
 516                 if (!subflow_thmac_valid(subflow)) {
 517                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 518                         subflow->reset_reason = MPTCP_RST_EMPTCP;
 519                         goto do_reset;
 520                 }
 521
 522                 if (!mptcp_finish_join(sk))
 523                         goto do_reset;
 524
 525                 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 526                                       subflow->local_nonce,
 527                                       subflow->remote_nonce,
 528                                       hmac);
 529                 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
 530
 531                 subflow->mp_join = 1;
 532                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 533
 534                 if (subflow_use_different_dport(msk, sk)) {
 535                         pr_debug("synack inet_dport=%d %d",
 536                                  ntohs(inet_sk(sk)->inet_dport),
 537                                  ntohs(inet_sk(parent)->inet_dport));
 538                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
 539                 }
 540         } else if (mptcp_check_fallback(sk)) {
 541 fallback:
 542                 mptcp_rcv_space_init(msk, sk);
 543                 mptcp_set_connected(parent);
 544         }
 545         return;
 546
 547 do_reset:
 548         subflow->reset_transient = 0;
 549         mptcp_subflow_reset(sk);
 550 }
 551
 552 static void subflow_set_local_id(struct mptcp_subflow_context *subflow, int local_id)
 553 {
 554         subflow->local_id = local_id;
 555         subflow->local_id_valid = 1;
 556 }
 557
 558 static int subflow_chk_local_id(struct sock *sk)
 559 {
 560         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 561         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 562         int err;
 563
 564         if (likely(subflow->local_id_valid))
 565                 return 0;
 566
 567         err = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
 568         if (err < 0)
 569                 return err;
 570
 571         subflow_set_local_id(subflow, err);
 572         return 0;
 573 }
 574
 575 static int subflow_rebuild_header(struct sock *sk)
 576 {
 577         int err = subflow_chk_local_id(sk);
 578
 579         if (unlikely(err < 0))
 580                 return err;
 581
 582         return inet_sk_rebuild_header(sk);
 583 }
 584
 585 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 586 static int subflow_v6_rebuild_header(struct sock *sk)
 587 {
 588         int err = subflow_chk_local_id(sk);
 589
 590         if (unlikely(err < 0))
 591                 return err;
 592
 593         return inet6_sk_rebuild_header(sk);
 594 }
 595 #endif
 596
 597 static struct request_sock_ops mptcp_subflow_v4_request_sock_ops __ro_after_init;
 598 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops __ro_after_init;
 599
 600 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 601 {
 602         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 603
 604         pr_debug("subflow=%p", subflow);
 605
 606         /* Never answer to SYNs sent to broadcast or multicast */
 607         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 608                 goto drop;
 609
 610         return tcp_conn_request(&mptcp_subflow_v4_request_sock_ops,
 611                                 &subflow_request_sock_ipv4_ops,
 612                                 sk, skb);
 613 drop:
 614         tcp_listendrop(sk);
 615         return 0;
 616 }
 617
 618 static void subflow_v4_req_destructor(struct request_sock *req)
 619 {
 620         subflow_req_destructor(req);
 621         tcp_request_sock_ops.destructor(req);
 622 }
 623
 624 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 625 static struct request_sock_ops mptcp_subflow_v6_request_sock_ops __ro_after_init;
 626 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops __ro_after_init;
 627 static struct inet_connection_sock_af_ops subflow_v6_specific __ro_after_init;
 628 static struct inet_connection_sock_af_ops subflow_v6m_specific __ro_after_init;
 629 static struct proto tcpv6_prot_override __ro_after_init;
 630
 631 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 632 {
 633         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 634
 635         pr_debug("subflow=%p", subflow);
 636
 637         if (skb->protocol == htons(ETH_P_IP))
 638                 return subflow_v4_conn_request(sk, skb);
 639
 640         if (!ipv6_unicast_destination(skb))
 641                 goto drop;
 642
 643         if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
 644                 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
 645                 return 0;
 646         }
 647
 648         return tcp_conn_request(&mptcp_subflow_v6_request_sock_ops,
 649                                 &subflow_request_sock_ipv6_ops, sk, skb);
 650
 651 drop:
 652         tcp_listendrop(sk);
 653         return 0; /* don't send reset */
 654 }
 655
 656 static void subflow_v6_req_destructor(struct request_sock *req)
 657 {
 658         subflow_req_destructor(req);
 659         tcp6_request_sock_ops.destructor(req);
 660 }
 661 #endif
 662
 663 struct request_sock *mptcp_subflow_reqsk_alloc(const struct request_sock_ops *ops,
 664                                                struct sock *sk_listener,
 665                                                bool attach_listener)
 666 {
 667         if (ops->family == AF_INET)
 668                 ops = &mptcp_subflow_v4_request_sock_ops;
 669 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 670         else if (ops->family == AF_INET6)
 671                 ops = &mptcp_subflow_v6_request_sock_ops;
 672 #endif
 673
 674         return inet_reqsk_alloc(ops, sk_listener, attach_listener);
 675 }
 676 EXPORT_SYMBOL(mptcp_subflow_reqsk_alloc);
 677
 678 /* validate hmac received in third ACK */
 679 static bool subflow_hmac_valid(const struct request_sock *req,
 680                                const struct mptcp_options_received *mp_opt)
 681 {
 682         const struct mptcp_subflow_request_sock *subflow_req;
 683         u8 hmac[SHA256_DIGEST_SIZE];
 684         struct mptcp_sock *msk;
 685
 686         subflow_req = mptcp_subflow_rsk(req);
 687         msk = subflow_req->msk;
 688         if (!msk)
 689                 return false;
 690
 691         subflow_generate_hmac(msk->remote_key, msk->local_key,
 692                               subflow_req->remote_nonce,
 693                               subflow_req->local_nonce, hmac);
 694
 695         return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
 696 }
 697
 698 static void subflow_ulp_fallback(struct sock *sk,
 699                                  struct mptcp_subflow_context *old_ctx)
 700 {
 701         struct inet_connection_sock *icsk = inet_csk(sk);
 702
 703         mptcp_subflow_tcp_fallback(sk, old_ctx);
 704         icsk->icsk_ulp_ops = NULL;
 705         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 706         tcp_sk(sk)->is_mptcp = 0;
 707
 708         mptcp_subflow_ops_undo_override(sk);
 709 }
 710
 711 void mptcp_subflow_drop_ctx(struct sock *ssk)
 712 {
 713         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 714
 715         if (!ctx)
 716                 return;
 717
 718         list_del(&mptcp_subflow_ctx(ssk)->node);
 719         if (inet_csk(ssk)->icsk_ulp_ops) {
 720                 subflow_ulp_fallback(ssk, ctx);
 721                 if (ctx->conn)
 722                         sock_put(ctx->conn);
 723         }
 724
 725         kfree_rcu(ctx, rcu);
 726 }
 727
 728 void mptcp_subflow_fully_established(struct mptcp_subflow_context *subflow,
 729                                      const struct mptcp_options_received *mp_opt)
 730 {
 731         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 732
 733         subflow_set_remote_key(msk, subflow, mp_opt);
 734         subflow->fully_established = 1;
 735         WRITE_ONCE(msk->fully_established, true);
 736
 737         if (subflow->is_mptfo)
 738                 mptcp_fastopen_gen_msk_ackseq(msk, subflow, mp_opt);
 739 }
 740
 741 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 742                                           struct sk_buff *skb,
 743                                           struct request_sock *req,
 744                                           struct dst_entry *dst,
 745                                           struct request_sock *req_unhash,
 746                                           bool *own_req)
 747 {
 748         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 749         struct mptcp_subflow_request_sock *subflow_req;
 750         struct mptcp_options_received mp_opt;
 751         bool fallback, fallback_is_fatal;
 752         struct mptcp_sock *owner;
 753         struct sock *child;
 754
 755         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 756
 757         /* After child creation we must look for MPC even when options
 758          * are not parsed
 759          */
 760         mp_opt.suboptions = 0;
 761
 762         /* hopefully temporary handling for MP_JOIN+syncookie */
 763         subflow_req = mptcp_subflow_rsk(req);
 764         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 765         fallback = !tcp_rsk(req)->is_mptcp;
 766         if (fallback)
 767                 goto create_child;
 768
 769         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 770         if (subflow_req->mp_capable) {
 771                 /* we can receive and accept an in-window, out-of-order pkt,
 772                  * which may not carry the MP_CAPABLE opt even on mptcp enabled
 773                  * paths: always try to extract the peer key, and fallback
 774                  * for packets missing it.
 775                  * Even OoO DSS packets coming legitly after dropped or
 776                  * reordered MPC will cause fallback, but we don't have other
 777                  * options.
 778                  */
 779                 mptcp_get_options(skb, &mp_opt);
 780                 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPC))
 781                         fallback = true;
 782
 783         } else if (subflow_req->mp_join) {
 784                 mptcp_get_options(skb, &mp_opt);
 785                 if (!(mp_opt.suboptions & OPTIONS_MPTCP_MPJ) ||
 786                     !subflow_hmac_valid(req, &mp_opt) ||
 787                     !mptcp_can_accept_new_subflow(subflow_req->msk)) {
 788                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 789                         fallback = true;
 790                 }
 791         }
 792
 793 create_child:
 794         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 795                                                      req_unhash, own_req);
 796
 797         if (child && *own_req) {
 798                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 799
 800                 tcp_rsk(req)->drop_req = false;
 801
 802                 /* we need to fallback on ctx allocation failure and on pre-reqs
 803                  * checking above. In the latter scenario we additionally need
 804                  * to reset the context to non MPTCP status.
 805                  */
 806                 if (!ctx || fallback) {
 807                         if (fallback_is_fatal) {
 808                                 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 809                                 goto dispose_child;
 810                         }
 811                         goto fallback;
 812                 }
 813
 814                 /* ssk inherits options of listener sk */
 815                 ctx->setsockopt_seq = listener->setsockopt_seq;
 816
 817                 if (ctx->mp_capable) {
 818                         ctx->conn = mptcp_sk_clone(listener->conn, &mp_opt, req);
 819                         if (!ctx->conn)
 820                                 goto fallback;
 821
 822                         owner = mptcp_sk(ctx->conn);
 823
 824                         /* this can't race with mptcp_close(), as the msk is
 825                          * not yet exposted to user-space
 826                          */
 827                         inet_sk_state_store(ctx->conn, TCP_ESTABLISHED);
 828
 829                         /* record the newly created socket as the first msk
 830                          * subflow, but don't link it yet into conn_list
 831                          */
 832                         WRITE_ONCE(owner->first, child);
 833
 834                         /* new mpc subflow takes ownership of the newly
 835                          * created mptcp socket
 836                          */
 837                         owner->setsockopt_seq = ctx->setsockopt_seq;
 838                         mptcp_pm_new_connection(owner, child, 1);
 839                         mptcp_token_accept(subflow_req, owner);
 840
 841                         /* set msk addresses early to ensure mptcp_pm_get_local_id()
 842                          * uses the correct data
 843                          */
 844                         mptcp_copy_inaddrs(ctx->conn, child);
 845                         mptcp_propagate_sndbuf(ctx->conn, child);
 846
 847                         mptcp_rcv_space_init(owner, child);
 848                         list_add(&ctx->node, &owner->conn_list);
 849                         sock_hold(child);
 850
 851                         /* with OoO packets we can reach here without ingress
 852                          * mpc option
 853                          */
 854                         if (mp_opt.suboptions & OPTION_MPTCP_MPC_ACK) {
 855                                 mptcp_subflow_fully_established(ctx, &mp_opt);
 856                                 mptcp_pm_fully_established(owner, child);
 857                                 ctx->pm_notified = 1;
 858                         }
 859                 } else if (ctx->mp_join) {
 860                         owner = subflow_req->msk;
 861                         if (!owner) {
 862                                 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 863                                 goto dispose_child;
 864                         }
 865
 866                         /* move the msk reference ownership to the subflow */
 867                         subflow_req->msk = NULL;
 868                         ctx->conn = (struct sock *)owner;
 869
 870                         if (subflow_use_different_sport(owner, sk)) {
 871                                 pr_debug("ack inet_sport=%d %d",
 872                                          ntohs(inet_sk(sk)->inet_sport),
 873                                          ntohs(inet_sk((struct sock *)owner)->inet_sport));
 874                                 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
 875                                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
 876                                         goto dispose_child;
 877                                 }
 878                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
 879                         }
 880
 881                         if (!mptcp_finish_join(child))
 882                                 goto dispose_child;
 883
 884                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 885                         tcp_rsk(req)->drop_req = true;
 886                 }
 887         }
 888
 889         /* check for expected invariant - should never trigger, just help
 890          * catching eariler subtle bugs
 891          */
 892         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 893                      (!mptcp_subflow_ctx(child) ||
 894                       !mptcp_subflow_ctx(child)->conn));
 895         return child;
 896
 897 dispose_child:
 898         mptcp_subflow_drop_ctx(child);
 899         tcp_rsk(req)->drop_req = true;
 900         inet_csk_prepare_for_destroy_sock(child);
 901         tcp_done(child);
 902         req->rsk_ops->send_reset(sk, skb);
 903
 904         /* The last child reference will be released by the caller */
 905         return child;
 906
 907 fallback:
 908         mptcp_subflow_drop_ctx(child);
 909         return child;
 910 }
 911
 912 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
 913 static struct proto tcp_prot_override __ro_after_init;
 914
 915 enum mapping_status {
 916         MAPPING_OK,
 917         MAPPING_INVALID,
 918         MAPPING_EMPTY,
 919         MAPPING_DATA_FIN,
 920         MAPPING_DUMMY,
 921         MAPPING_BAD_CSUM
 922 };
 923
 924 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 925 {
 926         pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 927                  ssn, subflow->map_subflow_seq, subflow->map_data_len);
 928 }
 929
 930 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 931 {
 932         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 933         unsigned int skb_consumed;
 934
 935         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 936         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 937                 return true;
 938
 939         return skb->len - skb_consumed <= subflow->map_data_len -
 940                                           mptcp_subflow_get_map_offset(subflow);
 941 }
 942
 943 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 944 {
 945         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 946         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 947
 948         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 949                 /* Mapping covers data later in the subflow stream,
 950                  * currently unsupported.
 951                  */
 952                 dbg_bad_map(subflow, ssn);
 953                 return false;
 954         }
 955         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 956                                   subflow->map_data_len))) {
 957                 /* Mapping does covers past subflow data, invalid */
 958                 dbg_bad_map(subflow, ssn);
 959                 return false;
 960         }
 961         return true;
 962 }
 963
 964 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
 965                                               bool csum_reqd)
 966 {
 967         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 968         u32 offset, seq, delta;
 969         __sum16 csum;
 970         int len;
 971
 972         if (!csum_reqd)
 973                 return MAPPING_OK;
 974
 975         /* mapping already validated on previous traversal */
 976         if (subflow->map_csum_len == subflow->map_data_len)
 977                 return MAPPING_OK;
 978
 979         /* traverse the receive queue, ensuring it contains a full
 980          * DSS mapping and accumulating the related csum.
 981          * Preserve the accoumlate csum across multiple calls, to compute
 982          * the csum only once
 983          */
 984         delta = subflow->map_data_len - subflow->map_csum_len;
 985         for (;;) {
 986                 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
 987                 offset = seq - TCP_SKB_CB(skb)->seq;
 988
 989                 /* if the current skb has not been accounted yet, csum its contents
 990                  * up to the amount covered by the current DSS
 991                  */
 992                 if (offset < skb->len) {
 993                         __wsum csum;
 994
 995                         len = min(skb->len - offset, delta);
 996                         csum = skb_checksum(skb, offset, len, 0);
 997                         subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
 998                                                                 subflow->map_csum_len);
 999
1000                         delta -= len;
1001                         subflow->map_csum_len += len;
1002                 }
1003                 if (delta == 0)
1004                         break;
1005
1006                 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
1007                         /* if this subflow is closed, the partial mapping
1008                          * will be never completed; flush the pending skbs, so
1009                          * that subflow_sched_work_if_closed() can kick in
1010                          */
1011                         if (unlikely(ssk->sk_state == TCP_CLOSE))
1012                                 while ((skb = skb_peek(&ssk->sk_receive_queue)))
1013                                         sk_eat_skb(ssk, skb);
1014
1015                         /* not enough data to validate the csum */
1016                         return MAPPING_EMPTY;
1017                 }
1018
1019                 /* the DSS mapping for next skbs will be validated later,
1020                  * when a get_mapping_status call will process such skb
1021                  */
1022                 skb = skb->next;
1023         }
1024
1025         /* note that 'map_data_len' accounts only for the carried data, does
1026          * not include the eventual seq increment due to the data fin,
1027          * while the pseudo header requires the original DSS data len,
1028          * including that
1029          */
1030         csum = __mptcp_make_csum(subflow->map_seq,
1031                                  subflow->map_subflow_seq,
1032                                  subflow->map_data_len + subflow->map_data_fin,
1033                                  subflow->map_data_csum);
1034         if (unlikely(csum)) {
1035                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
1036                 return MAPPING_BAD_CSUM;
1037         }
1038
1039         subflow->valid_csum_seen = 1;
1040         return MAPPING_OK;
1041 }
1042
1043 static enum mapping_status get_mapping_status(struct sock *ssk,
1044                                               struct mptcp_sock *msk)
1045 {
1046         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1047         bool csum_reqd = READ_ONCE(msk->csum_enabled);
1048         struct mptcp_ext *mpext;
1049         struct sk_buff *skb;
1050         u16 data_len;
1051         u64 map_seq;
1052
1053         skb = skb_peek(&ssk->sk_receive_queue);
1054         if (!skb)
1055                 return MAPPING_EMPTY;
1056
1057         if (mptcp_check_fallback(ssk))
1058                 return MAPPING_DUMMY;
1059
1060         mpext = mptcp_get_ext(skb);
1061         if (!mpext || !mpext->use_map) {
1062                 if (!subflow->map_valid && !skb->len) {
1063                         /* the TCP stack deliver 0 len FIN pkt to the receive
1064                          * queue, that is the only 0len pkts ever expected here,
1065                          * and we can admit no mapping only for 0 len pkts
1066                          */
1067                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
1068                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
1069                                           TCP_SKB_CB(skb)->seq,
1070                                           TCP_SKB_CB(skb)->end_seq,
1071                                           TCP_SKB_CB(skb)->tcp_flags);
1072                         sk_eat_skb(ssk, skb);
1073                         return MAPPING_EMPTY;
1074                 }
1075
1076                 if (!subflow->map_valid)
1077                         return MAPPING_INVALID;
1078
1079                 goto validate_seq;
1080         }
1081
1082         trace_get_mapping_status(mpext);
1083
1084         data_len = mpext->data_len;
1085         if (data_len == 0) {
1086                 pr_debug("infinite mapping received");
1087                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1088                 subflow->map_data_len = 0;
1089                 return MAPPING_INVALID;
1090         }
1091
1092         if (mpext->data_fin == 1) {
1093                 if (data_len == 1) {
1094                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1095                                                                  mpext->dsn64);
1096                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
1097                         if (subflow->map_valid) {
1098                                 /* A DATA_FIN might arrive in a DSS
1099                                  * option before the previous mapping
1100                                  * has been fully consumed. Continue
1101                                  * handling the existing mapping.
1102                                  */
1103                                 skb_ext_del(skb, SKB_EXT_MPTCP);
1104                                 return MAPPING_OK;
1105                         } else {
1106                                 if (updated)
1107                                         mptcp_schedule_work((struct sock *)msk);
1108
1109                                 return MAPPING_DATA_FIN;
1110                         }
1111                 } else {
1112                         u64 data_fin_seq = mpext->data_seq + data_len - 1;
1113
1114                         /* If mpext->data_seq is a 32-bit value, data_fin_seq
1115                          * must also be limited to 32 bits.
1116                          */
1117                         if (!mpext->dsn64)
1118                                 data_fin_seq &= GENMASK_ULL(31, 0);
1119
1120                         mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1121                         pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
1122                                  data_fin_seq, mpext->dsn64);
1123                 }
1124
1125                 /* Adjust for DATA_FIN using 1 byte of sequence space */
1126                 data_len--;
1127         }
1128
1129         map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1130         WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1131
1132         if (subflow->map_valid) {
1133                 /* Allow replacing only with an identical map */
1134                 if (subflow->map_seq == map_seq &&
1135                     subflow->map_subflow_seq == mpext->subflow_seq &&
1136                     subflow->map_data_len == data_len &&
1137                     subflow->map_csum_reqd == mpext->csum_reqd) {
1138                         skb_ext_del(skb, SKB_EXT_MPTCP);
1139                         goto validate_csum;
1140                 }
1141
1142                 /* If this skb data are fully covered by the current mapping,
1143                  * the new map would need caching, which is not supported
1144                  */
1145                 if (skb_is_fully_mapped(ssk, skb)) {
1146                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1147                         return MAPPING_INVALID;
1148                 }
1149
1150                 /* will validate the next map after consuming the current one */
1151                 goto validate_csum;
1152         }
1153
1154         subflow->map_seq = map_seq;
1155         subflow->map_subflow_seq = mpext->subflow_seq;
1156         subflow->map_data_len = data_len;
1157         subflow->map_valid = 1;
1158         subflow->map_data_fin = mpext->data_fin;
1159         subflow->mpc_map = mpext->mpc_map;
1160         subflow->map_csum_reqd = mpext->csum_reqd;
1161         subflow->map_csum_len = 0;
1162         subflow->map_data_csum = csum_unfold(mpext->csum);
1163
1164         /* Cfr RFC 8684 Section 3.3.0 */
1165         if (unlikely(subflow->map_csum_reqd != csum_reqd))
1166                 return MAPPING_INVALID;
1167
1168         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
1169                  subflow->map_seq, subflow->map_subflow_seq,
1170                  subflow->map_data_len, subflow->map_csum_reqd,
1171                  subflow->map_data_csum);
1172
1173 validate_seq:
1174         /* we revalidate valid mapping on new skb, because we must ensure
1175          * the current skb is completely covered by the available mapping
1176          */
1177         if (!validate_mapping(ssk, skb)) {
1178                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1179                 return MAPPING_INVALID;
1180         }
1181
1182         skb_ext_del(skb, SKB_EXT_MPTCP);
1183
1184 validate_csum:
1185         return validate_data_csum(ssk, skb, csum_reqd);
1186 }
1187
1188 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1189                                        u64 limit)
1190 {
1191         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1192         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1193         u32 incr;
1194
1195         incr = limit >= skb->len ? skb->len + fin : limit;
1196
1197         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
1198                  subflow->map_subflow_seq);
1199         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1200         tcp_sk(ssk)->copied_seq += incr;
1201         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1202                 sk_eat_skb(ssk, skb);
1203         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1204                 subflow->map_valid = 0;
1205 }
1206
1207 /* sched mptcp worker to remove the subflow if no more data is pending */
1208 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1209 {
1210         if (likely(ssk->sk_state != TCP_CLOSE))
1211                 return;
1212
1213         if (skb_queue_empty(&ssk->sk_receive_queue) &&
1214             !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
1215                 mptcp_schedule_work((struct sock *)msk);
1216 }
1217
1218 static bool subflow_can_fallback(struct mptcp_subflow_context *subflow)
1219 {
1220         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1221
1222         if (subflow->mp_join)
1223                 return false;
1224         else if (READ_ONCE(msk->csum_enabled))
1225                 return !subflow->valid_csum_seen;
1226         else
1227                 return !subflow->fully_established;
1228 }
1229
1230 static void mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk)
1231 {
1232         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1233         unsigned long fail_tout;
1234
1235         /* greceful failure can happen only on the MPC subflow */
1236         if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first)))
1237                 return;
1238
1239         /* since the close timeout take precedence on the fail one,
1240          * no need to start the latter when the first is already set
1241          */
1242         if (sock_flag((struct sock *)msk, SOCK_DEAD))
1243                 return;
1244
1245         /* we don't need extreme accuracy here, use a zero fail_tout as special
1246          * value meaning no fail timeout at all;
1247          */
1248         fail_tout = jiffies + TCP_RTO_MAX;
1249         if (!fail_tout)
1250                 fail_tout = 1;
1251         WRITE_ONCE(subflow->fail_tout, fail_tout);
1252         tcp_send_ack(ssk);
1253
1254         mptcp_reset_timeout(msk, subflow->fail_tout);
1255 }
1256
1257 static bool subflow_check_data_avail(struct sock *ssk)
1258 {
1259         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1260         enum mapping_status status;
1261         struct mptcp_sock *msk;
1262         struct sk_buff *skb;
1263
1264         if (!skb_peek(&ssk->sk_receive_queue))
1265                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1266         if (subflow->data_avail)
1267                 return true;
1268
1269         msk = mptcp_sk(subflow->conn);
1270         for (;;) {
1271                 u64 ack_seq;
1272                 u64 old_ack;
1273
1274                 status = get_mapping_status(ssk, msk);
1275                 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1276                 if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY ||
1277                              status == MAPPING_BAD_CSUM))
1278                         goto fallback;
1279
1280                 if (status != MAPPING_OK)
1281                         goto no_data;
1282
1283                 skb = skb_peek(&ssk->sk_receive_queue);
1284                 if (WARN_ON_ONCE(!skb))
1285                         goto no_data;
1286
1287                 if (unlikely(!READ_ONCE(msk->can_ack)))
1288                         goto fallback;
1289
1290                 old_ack = READ_ONCE(msk->ack_seq);
1291                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1292                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1293                          ack_seq);
1294                 if (unlikely(before64(ack_seq, old_ack))) {
1295                         mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1296                         continue;
1297                 }
1298
1299                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1300                 break;
1301         }
1302         return true;
1303
1304 no_data:
1305         subflow_sched_work_if_closed(msk, ssk);
1306         return false;
1307
1308 fallback:
1309         if (!__mptcp_check_fallback(msk)) {
1310                 /* RFC 8684 section 3.7. */
1311                 if (status == MAPPING_BAD_CSUM &&
1312                     (subflow->mp_join || subflow->valid_csum_seen)) {
1313                         subflow->send_mp_fail = 1;
1314
1315                         if (!READ_ONCE(msk->allow_infinite_fallback)) {
1316                                 subflow->reset_transient = 0;
1317                                 subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1318                                 goto reset;
1319                         }
1320                         mptcp_subflow_fail(msk, ssk);
1321                         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1322                         return true;
1323                 }
1324
1325                 if (!subflow_can_fallback(subflow) && subflow->map_data_len) {
1326                         /* fatal protocol error, close the socket.
1327                          * subflow_error_report() will introduce the appropriate barriers
1328                          */
1329                         subflow->reset_transient = 0;
1330                         subflow->reset_reason = MPTCP_RST_EMPTCP;
1331
1332 reset:
1333                         WRITE_ONCE(ssk->sk_err, EBADMSG);
1334                         tcp_set_state(ssk, TCP_CLOSE);
1335                         while ((skb = skb_peek(&ssk->sk_receive_queue)))
1336                                 sk_eat_skb(ssk, skb);
1337                         tcp_send_active_reset(ssk, GFP_ATOMIC);
1338                         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1339                         return false;
1340                 }
1341
1342                 mptcp_do_fallback(ssk);
1343         }
1344
1345         skb = skb_peek(&ssk->sk_receive_queue);
1346         subflow->map_valid = 1;
1347         subflow->map_seq = READ_ONCE(msk->ack_seq);
1348         subflow->map_data_len = skb->len;
1349         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1350         WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_DATA_AVAIL);
1351         return true;
1352 }
1353
1354 bool mptcp_subflow_data_available(struct sock *sk)
1355 {
1356         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1357
1358         /* check if current mapping is still valid */
1359         if (subflow->map_valid &&
1360             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1361                 subflow->map_valid = 0;
1362                 WRITE_ONCE(subflow->data_avail, MPTCP_SUBFLOW_NODATA);
1363
1364                 pr_debug("Done with mapping: seq=%u data_len=%u",
1365                          subflow->map_subflow_seq,
1366                          subflow->map_data_len);
1367         }
1368
1369         return subflow_check_data_avail(sk);
1370 }
1371
1372 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1373  * not the ssk one.
1374  *
1375  * In mptcp, rwin is about the mptcp-level connection data.
1376  *
1377  * Data that is still on the ssk rx queue can thus be ignored,
1378  * as far as mptcp peer is concerned that data is still inflight.
1379  * DSS ACK is updated when skb is moved to the mptcp rx queue.
1380  */
1381 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1382 {
1383         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1384         const struct sock *sk = subflow->conn;
1385
1386         *space = __mptcp_space(sk);
1387         *full_space = tcp_full_space(sk);
1388 }
1389
1390 void __mptcp_error_report(struct sock *sk)
1391 {
1392         struct mptcp_subflow_context *subflow;
1393         struct mptcp_sock *msk = mptcp_sk(sk);
1394
1395         mptcp_for_each_subflow(msk, subflow) {
1396                 struct sock *ssk = mptcp_subflow_tcp_sock(subflow);
1397                 int err = sock_error(ssk);
1398                 int ssk_state;
1399
1400                 if (!err)
1401                         continue;
1402
1403                 /* only propagate errors on fallen-back sockets or
1404                  * on MPC connect
1405                  */
1406                 if (sk->sk_state != TCP_SYN_SENT && !__mptcp_check_fallback(msk))
1407                         continue;
1408
1409                 /* We need to propagate only transition to CLOSE state.
1410                  * Orphaned socket will see such state change via
1411                  * subflow_sched_work_if_closed() and that path will properly
1412                  * destroy the msk as needed.
1413                  */
1414                 ssk_state = inet_sk_state_load(ssk);
1415                 if (ssk_state == TCP_CLOSE && !sock_flag(sk, SOCK_DEAD))
1416                         inet_sk_state_store(sk, ssk_state);
1417                 WRITE_ONCE(sk->sk_err, -err);
1418
1419                 /* This barrier is coupled with smp_rmb() in mptcp_poll() */
1420                 smp_wmb();
1421                 sk_error_report(sk);
1422                 break;
1423         }
1424 }
1425
1426 static void subflow_error_report(struct sock *ssk)
1427 {
1428         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1429
1430         /* bail early if this is a no-op, so that we avoid introducing a
1431          * problematic lockdep dependency between TCP accept queue lock
1432          * and msk socket spinlock
1433          */
1434         if (!sk->sk_socket)
1435                 return;
1436
1437         mptcp_data_lock(sk);
1438         if (!sock_owned_by_user(sk))
1439                 __mptcp_error_report(sk);
1440         else
1441                 __set_bit(MPTCP_ERROR_REPORT,  &mptcp_sk(sk)->cb_flags);
1442         mptcp_data_unlock(sk);
1443 }
1444
1445 static void subflow_data_ready(struct sock *sk)
1446 {
1447         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1448         u16 state = 1 << inet_sk_state_load(sk);
1449         struct sock *parent = subflow->conn;
1450         struct mptcp_sock *msk;
1451
1452         trace_sk_data_ready(sk);
1453
1454         msk = mptcp_sk(parent);
1455         if (state & TCPF_LISTEN) {
1456                 /* MPJ subflow are removed from accept queue before reaching here,
1457                  * avoid stray wakeups
1458                  */
1459                 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1460                         return;
1461
1462                 parent->sk_data_ready(parent);
1463                 return;
1464         }
1465
1466         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1467                      !subflow->mp_join && !(state & TCPF_CLOSE));
1468
1469         if (mptcp_subflow_data_available(sk))
1470                 mptcp_data_ready(parent, sk);
1471         else if (unlikely(sk->sk_err))
1472                 subflow_error_report(sk);
1473 }
1474
1475 static void subflow_write_space(struct sock *ssk)
1476 {
1477         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1478
1479         mptcp_propagate_sndbuf(sk, ssk);
1480         mptcp_write_space(sk);
1481 }
1482
1483 static const struct inet_connection_sock_af_ops *
1484 subflow_default_af_ops(struct sock *sk)
1485 {
1486 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1487         if (sk->sk_family == AF_INET6)
1488                 return &subflow_v6_specific;
1489 #endif
1490         return &subflow_specific;
1491 }
1492
1493 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1494 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1495 {
1496         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1497         struct inet_connection_sock *icsk = inet_csk(sk);
1498         const struct inet_connection_sock_af_ops *target;
1499
1500         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1501
1502         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1503                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1504
1505         if (likely(icsk->icsk_af_ops == target))
1506                 return;
1507
1508         subflow->icsk_af_ops = icsk->icsk_af_ops;
1509         icsk->icsk_af_ops = target;
1510 }
1511 #endif
1512
1513 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1514                          struct sockaddr_storage *addr,
1515                          unsigned short family)
1516 {
1517         memset(addr, 0, sizeof(*addr));
1518         addr->ss_family = family;
1519         if (addr->ss_family == AF_INET) {
1520                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1521
1522                 if (info->family == AF_INET)
1523                         in_addr->sin_addr = info->addr;
1524 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1525                 else if (ipv6_addr_v4mapped(&info->addr6))
1526                         in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1527 #endif
1528                 in_addr->sin_port = info->port;
1529         }
1530 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1531         else if (addr->ss_family == AF_INET6) {
1532                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1533
1534                 if (info->family == AF_INET)
1535                         ipv6_addr_set_v4mapped(info->addr.s_addr,
1536                                                &in6_addr->sin6_addr);
1537                 else
1538                         in6_addr->sin6_addr = info->addr6;
1539                 in6_addr->sin6_port = info->port;
1540         }
1541 #endif
1542 }
1543
1544 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1545                             const struct mptcp_addr_info *remote)
1546 {
1547         struct mptcp_sock *msk = mptcp_sk(sk);
1548         struct mptcp_subflow_context *subflow;
1549         struct sockaddr_storage addr;
1550         int remote_id = remote->id;
1551         int local_id = loc->id;
1552         int err = -ENOTCONN;
1553         struct socket *sf;
1554         struct sock *ssk;
1555         u32 remote_token;
1556         int addrlen;
1557         int ifindex;
1558         u8 flags;
1559
1560         if (!mptcp_is_fully_established(sk))
1561                 goto err_out;
1562
1563         err = mptcp_subflow_create_socket(sk, loc->family, &sf);
1564         if (err)
1565                 goto err_out;
1566
1567         ssk = sf->sk;
1568         subflow = mptcp_subflow_ctx(ssk);
1569         do {
1570                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1571         } while (!subflow->local_nonce);
1572
1573         if (local_id)
1574                 subflow_set_local_id(subflow, local_id);
1575
1576         mptcp_pm_get_flags_and_ifindex_by_id(msk, local_id,
1577                                              &flags, &ifindex);
1578         subflow->remote_key_valid = 1;
1579         subflow->remote_key = msk->remote_key;
1580         subflow->local_key = msk->local_key;
1581         subflow->token = msk->token;
1582         mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1583
1584         addrlen = sizeof(struct sockaddr_in);
1585 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1586         if (addr.ss_family == AF_INET6)
1587                 addrlen = sizeof(struct sockaddr_in6);
1588 #endif
1589         mptcp_sockopt_sync(msk, ssk);
1590
1591         ssk->sk_bound_dev_if = ifindex;
1592         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1593         if (err)
1594                 goto failed;
1595
1596         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1597         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1598                  remote_token, local_id, remote_id);
1599         subflow->remote_token = remote_token;
1600         subflow->remote_id = remote_id;
1601         subflow->request_join = 1;
1602         subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1603         mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1604
1605         sock_hold(ssk);
1606         list_add_tail(&subflow->node, &msk->conn_list);
1607         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1608         if (err && err != -EINPROGRESS)
1609                 goto failed_unlink;
1610
1611         /* discard the subflow socket */
1612         mptcp_sock_graft(ssk, sk->sk_socket);
1613         iput(SOCK_INODE(sf));
1614         WRITE_ONCE(msk->allow_infinite_fallback, false);
1615         return 0;
1616
1617 failed_unlink:
1618         list_del(&subflow->node);
1619         sock_put(mptcp_subflow_tcp_sock(subflow));
1620
1621 failed:
1622         subflow->disposable = 1;
1623         sock_release(sf);
1624
1625 err_out:
1626         /* we account subflows before the creation, and this failures will not
1627          * be caught by sk_state_change()
1628          */
1629         mptcp_pm_close_subflow(msk);
1630         return err;
1631 }
1632
1633 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1634 {
1635 #ifdef CONFIG_SOCK_CGROUP_DATA
1636         struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1637                                 *child_skcd = &child->sk_cgrp_data;
1638
1639         /* only the additional subflows created by kworkers have to be modified */
1640         if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1641             cgroup_id(sock_cgroup_ptr(child_skcd))) {
1642 #ifdef CONFIG_MEMCG
1643                 struct mem_cgroup *memcg = parent->sk_memcg;
1644
1645                 mem_cgroup_sk_free(child);
1646                 if (memcg && css_tryget(&memcg->css))
1647                         child->sk_memcg = memcg;
1648 #endif /* CONFIG_MEMCG */
1649
1650                 cgroup_sk_free(child_skcd);
1651                 *child_skcd = *parent_skcd;
1652                 cgroup_sk_clone(child_skcd);
1653         }
1654 #endif /* CONFIG_SOCK_CGROUP_DATA */
1655 }
1656
1657 static void mptcp_subflow_ops_override(struct sock *ssk)
1658 {
1659 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1660         if (ssk->sk_prot == &tcpv6_prot)
1661                 ssk->sk_prot = &tcpv6_prot_override;
1662         else
1663 #endif
1664                 ssk->sk_prot = &tcp_prot_override;
1665 }
1666
1667 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1668 {
1669 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1670         if (ssk->sk_prot == &tcpv6_prot_override)
1671                 ssk->sk_prot = &tcpv6_prot;
1672         else
1673 #endif
1674                 ssk->sk_prot = &tcp_prot;
1675 }
1676
1677 int mptcp_subflow_create_socket(struct sock *sk, unsigned short family,
1678                                 struct socket **new_sock)
1679 {
1680         struct mptcp_subflow_context *subflow;
1681         struct net *net = sock_net(sk);
1682         struct socket *sf;
1683         int err;
1684
1685         /* un-accepted server sockets can reach here - on bad configuration
1686          * bail early to avoid greater trouble later
1687          */
1688         if (unlikely(!sk->sk_socket))
1689                 return -EINVAL;
1690
1691         err = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, &sf);
1692         if (err)
1693                 return err;
1694
1695         lock_sock_nested(sf->sk, SINGLE_DEPTH_NESTING);
1696
1697         /* the newly created socket has to be in the same cgroup as its parent */
1698         mptcp_attach_cgroup(sk, sf->sk);
1699
1700         /* kernel sockets do not by default acquire net ref, but TCP timer
1701          * needs it.
1702          * Update ns_tracker to current stack trace and refcounted tracker.
1703          */
1704         __netns_tracker_free(net, &sf->sk->ns_tracker, false);
1705         sf->sk->sk_net_refcnt = 1;
1706         get_net_track(net, &sf->sk->ns_tracker, GFP_KERNEL);
1707         sock_inuse_add(net, 1);
1708         err = tcp_set_ulp(sf->sk, "mptcp");
1709         release_sock(sf->sk);
1710
1711         if (err) {
1712                 sock_release(sf);
1713                 return err;
1714         }
1715
1716         /* the newly created socket really belongs to the owning MPTCP master
1717          * socket, even if for additional subflows the allocation is performed
1718          * by a kernel workqueue. Adjust inode references, so that the
1719          * procfs/diag interfaces really show this one belonging to the correct
1720          * user.
1721          */
1722         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1723         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1724         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1725
1726         subflow = mptcp_subflow_ctx(sf->sk);
1727         pr_debug("subflow=%p", subflow);
1728
1729         *new_sock = sf;
1730         sock_hold(sk);
1731         subflow->conn = sk;
1732         mptcp_subflow_ops_override(sf->sk);
1733
1734         return 0;
1735 }
1736
1737 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1738                                                         gfp_t priority)
1739 {
1740         struct inet_connection_sock *icsk = inet_csk(sk);
1741         struct mptcp_subflow_context *ctx;
1742
1743         ctx = kzalloc(sizeof(*ctx), priority);
1744         if (!ctx)
1745                 return NULL;
1746
1747         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1748         INIT_LIST_HEAD(&ctx->node);
1749         INIT_LIST_HEAD(&ctx->delegated_node);
1750
1751         pr_debug("subflow=%p", ctx);
1752
1753         ctx->tcp_sock = sk;
1754
1755         return ctx;
1756 }
1757
1758 static void __subflow_state_change(struct sock *sk)
1759 {
1760         struct socket_wq *wq;
1761
1762         rcu_read_lock();
1763         wq = rcu_dereference(sk->sk_wq);
1764         if (skwq_has_sleeper(wq))
1765                 wake_up_interruptible_all(&wq->wait);
1766         rcu_read_unlock();
1767 }
1768
1769 static bool subflow_is_done(const struct sock *sk)
1770 {
1771         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1772 }
1773
1774 static void subflow_state_change(struct sock *sk)
1775 {
1776         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1777         struct sock *parent = subflow->conn;
1778
1779         __subflow_state_change(sk);
1780
1781         if (subflow_simultaneous_connect(sk)) {
1782                 mptcp_propagate_sndbuf(parent, sk);
1783                 mptcp_do_fallback(sk);
1784                 mptcp_rcv_space_init(mptcp_sk(parent), sk);
1785                 pr_fallback(mptcp_sk(parent));
1786                 subflow->conn_finished = 1;
1787                 mptcp_set_connected(parent);
1788         }
1789
1790         /* as recvmsg() does not acquire the subflow socket for ssk selection
1791          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1792          * the data available machinery here.
1793          */
1794         if (mptcp_subflow_data_available(sk))
1795                 mptcp_data_ready(parent, sk);
1796         else if (unlikely(sk->sk_err))
1797                 subflow_error_report(sk);
1798
1799         subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1800
1801         if (__mptcp_check_fallback(mptcp_sk(parent)) &&
1802             !subflow->rx_eof && subflow_is_done(sk)) {
1803                 subflow->rx_eof = 1;
1804                 mptcp_subflow_eof(parent);
1805         }
1806 }
1807
1808 void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_ssk)
1809 {
1810         struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
1811         struct mptcp_sock *msk, *next, *head = NULL;
1812         struct request_sock *req;
1813         struct sock *sk;
1814
1815         /* build a list of all unaccepted mptcp sockets */
1816         spin_lock_bh(&queue->rskq_lock);
1817         for (req = queue->rskq_accept_head; req; req = req->dl_next) {
1818                 struct mptcp_subflow_context *subflow;
1819                 struct sock *ssk = req->sk;
1820
1821                 if (!sk_is_mptcp(ssk))
1822                         continue;
1823
1824                 subflow = mptcp_subflow_ctx(ssk);
1825                 if (!subflow || !subflow->conn)
1826                         continue;
1827
1828                 /* skip if already in list */
1829                 sk = subflow->conn;
1830                 msk = mptcp_sk(sk);
1831                 if (msk->dl_next || msk == head)
1832                         continue;
1833
1834                 sock_hold(sk);
1835                 msk->dl_next = head;
1836                 head = msk;
1837         }
1838         spin_unlock_bh(&queue->rskq_lock);
1839         if (!head)
1840                 return;
1841
1842         /* can't acquire the msk socket lock under the subflow one,
1843          * or will cause ABBA deadlock
1844          */
1845         release_sock(listener_ssk);
1846
1847         for (msk = head; msk; msk = next) {
1848                 sk = (struct sock *)msk;
1849
1850                 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
1851                 next = msk->dl_next;
1852                 msk->dl_next = NULL;
1853
1854                 __mptcp_unaccepted_force_close(sk);
1855                 release_sock(sk);
1856
1857                 /* lockdep will report a false positive ABBA deadlock
1858                  * between cancel_work_sync and the listener socket.
1859                  * The involved locks belong to different sockets WRT
1860                  * the existing AB chain.
1861                  * Using a per socket key is problematic as key
1862                  * deregistration requires process context and must be
1863                  * performed at socket disposal time, in atomic
1864                  * context.
1865                  * Just tell lockdep to consider the listener socket
1866                  * released here.
1867                  */
1868                 mutex_release(&listener_sk->sk_lock.dep_map, _RET_IP_);
1869                 mptcp_cancel_work(sk);
1870                 mutex_acquire(&listener_sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1871
1872                 sock_put(sk);
1873         }
1874
1875         /* we are still under the listener msk socket lock */
1876         lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
1877 }
1878
1879 static int subflow_ulp_init(struct sock *sk)
1880 {
1881         struct inet_connection_sock *icsk = inet_csk(sk);
1882         struct mptcp_subflow_context *ctx;
1883         struct tcp_sock *tp = tcp_sk(sk);
1884         int err = 0;
1885
1886         /* disallow attaching ULP to a socket unless it has been
1887          * created with sock_create_kern()
1888          */
1889         if (!sk->sk_kern_sock) {
1890                 err = -EOPNOTSUPP;
1891                 goto out;
1892         }
1893
1894         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1895         if (!ctx) {
1896                 err = -ENOMEM;
1897                 goto out;
1898         }
1899
1900         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1901
1902         tp->is_mptcp = 1;
1903         ctx->icsk_af_ops = icsk->icsk_af_ops;
1904         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1905         ctx->tcp_state_change = sk->sk_state_change;
1906         ctx->tcp_error_report = sk->sk_error_report;
1907
1908         WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1909         WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
1910
1911         sk->sk_data_ready = subflow_data_ready;
1912         sk->sk_write_space = subflow_write_space;
1913         sk->sk_state_change = subflow_state_change;
1914         sk->sk_error_report = subflow_error_report;
1915 out:
1916         return err;
1917 }
1918
1919 static void subflow_ulp_release(struct sock *ssk)
1920 {
1921         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1922         bool release = true;
1923         struct sock *sk;
1924
1925         if (!ctx)
1926                 return;
1927
1928         sk = ctx->conn;
1929         if (sk) {
1930                 /* if the msk has been orphaned, keep the ctx
1931                  * alive, will be freed by __mptcp_close_ssk(),
1932                  * when the subflow is still unaccepted
1933                  */
1934                 release = ctx->disposable || list_empty(&ctx->node);
1935
1936                 /* inet_child_forget() does not call sk_state_change(),
1937                  * explicitly trigger the socket close machinery
1938                  */
1939                 if (!release && !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW,
1940                                                   &mptcp_sk(sk)->flags))
1941                         mptcp_schedule_work(sk);
1942                 sock_put(sk);
1943         }
1944
1945         mptcp_subflow_ops_undo_override(ssk);
1946         if (release)
1947                 kfree_rcu(ctx, rcu);
1948 }
1949
1950 static void subflow_ulp_clone(const struct request_sock *req,
1951                               struct sock *newsk,
1952                               const gfp_t priority)
1953 {
1954         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1955         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1956         struct mptcp_subflow_context *new_ctx;
1957
1958         if (!tcp_rsk(req)->is_mptcp ||
1959             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1960                 subflow_ulp_fallback(newsk, old_ctx);
1961                 return;
1962         }
1963
1964         new_ctx = subflow_create_ctx(newsk, priority);
1965         if (!new_ctx) {
1966                 subflow_ulp_fallback(newsk, old_ctx);
1967                 return;
1968         }
1969
1970         new_ctx->conn_finished = 1;
1971         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1972         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1973         new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1974         new_ctx->rel_write_seq = 1;
1975         new_ctx->tcp_sock = newsk;
1976
1977         if (subflow_req->mp_capable) {
1978                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1979                  * is fully established only after we receive the remote key
1980                  */
1981                 new_ctx->mp_capable = 1;
1982                 new_ctx->local_key = subflow_req->local_key;
1983                 new_ctx->token = subflow_req->token;
1984                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1985                 new_ctx->idsn = subflow_req->idsn;
1986
1987                 /* this is the first subflow, id is always 0 */
1988                 new_ctx->local_id_valid = 1;
1989         } else if (subflow_req->mp_join) {
1990                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1991                 new_ctx->mp_join = 1;
1992                 new_ctx->fully_established = 1;
1993                 new_ctx->remote_key_valid = 1;
1994                 new_ctx->backup = subflow_req->backup;
1995                 new_ctx->remote_id = subflow_req->remote_id;
1996                 new_ctx->token = subflow_req->token;
1997                 new_ctx->thmac = subflow_req->thmac;
1998
1999                 /* the subflow req id is valid, fetched via subflow_check_req()
2000                  * and subflow_token_join_request()
2001                  */
2002                 subflow_set_local_id(new_ctx, subflow_req->local_id);
2003         }
2004 }
2005
2006 static void tcp_release_cb_override(struct sock *ssk)
2007 {
2008         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2009
2010         if (mptcp_subflow_has_delegated_action(subflow))
2011                 mptcp_subflow_process_delegated(ssk);
2012
2013         tcp_release_cb(ssk);
2014 }
2015
2016 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
2017         .name           = "mptcp",
2018         .owner          = THIS_MODULE,
2019         .init           = subflow_ulp_init,
2020         .release        = subflow_ulp_release,
2021         .clone          = subflow_ulp_clone,
2022 };
2023
2024 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
2025 {
2026         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
2027
2028         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
2029                                               subflow_ops->obj_size, 0,
2030                                               SLAB_ACCOUNT |
2031                                               SLAB_TYPESAFE_BY_RCU,
2032                                               NULL);
2033         if (!subflow_ops->slab)
2034                 return -ENOMEM;
2035
2036         return 0;
2037 }
2038
2039 void __init mptcp_subflow_init(void)
2040 {
2041         mptcp_subflow_v4_request_sock_ops = tcp_request_sock_ops;
2042         mptcp_subflow_v4_request_sock_ops.slab_name = "request_sock_subflow_v4";
2043         mptcp_subflow_v4_request_sock_ops.destructor = subflow_v4_req_destructor;
2044
2045         if (subflow_ops_init(&mptcp_subflow_v4_request_sock_ops) != 0)
2046                 panic("MPTCP: failed to init subflow v4 request sock ops\n");
2047
2048         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
2049         subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
2050         subflow_request_sock_ipv4_ops.send_synack = subflow_v4_send_synack;
2051
2052         subflow_specific = ipv4_specific;
2053         subflow_specific.conn_request = subflow_v4_conn_request;
2054         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
2055         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
2056         subflow_specific.rebuild_header = subflow_rebuild_header;
2057
2058         tcp_prot_override = tcp_prot;
2059         tcp_prot_override.release_cb = tcp_release_cb_override;
2060
2061 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2062         /* In struct mptcp_subflow_request_sock, we assume the TCP request sock
2063          * structures for v4 and v6 have the same size. It should not changed in
2064          * the future but better to make sure to be warned if it is no longer
2065          * the case.
2066          */
2067         BUILD_BUG_ON(sizeof(struct tcp_request_sock) != sizeof(struct tcp6_request_sock));
2068
2069         mptcp_subflow_v6_request_sock_ops = tcp6_request_sock_ops;
2070         mptcp_subflow_v6_request_sock_ops.slab_name = "request_sock_subflow_v6";
2071         mptcp_subflow_v6_request_sock_ops.destructor = subflow_v6_req_destructor;
2072
2073         if (subflow_ops_init(&mptcp_subflow_v6_request_sock_ops) != 0)
2074                 panic("MPTCP: failed to init subflow v6 request sock ops\n");
2075
2076         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
2077         subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
2078         subflow_request_sock_ipv6_ops.send_synack = subflow_v6_send_synack;
2079
2080         subflow_v6_specific = ipv6_specific;
2081         subflow_v6_specific.conn_request = subflow_v6_conn_request;
2082         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
2083         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
2084         subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
2085
2086         subflow_v6m_specific = subflow_v6_specific;
2087         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
2088         subflow_v6m_specific.send_check = ipv4_specific.send_check;
2089         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
2090         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
2091         subflow_v6m_specific.net_frag_header_len = 0;
2092         subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
2093
2094         tcpv6_prot_override = tcpv6_prot;
2095         tcpv6_prot_override.release_cb = tcp_release_cb_override;
2096 #endif
2097
2098         mptcp_diag_subflow_init(&subflow_ulp_ops);
2099
2100         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
2101                 panic("MPTCP: failed to register subflows to ULP\n");
2102 }