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/sha2.h>
  13 #include <crypto/utils.h>
  14 #include <net/sock.h>
  15 #include <net/inet_common.h>
  16 #include <net/inet_hashtables.h>
  17 #include <net/protocol.h>
  18 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
  19 #include <net/ip6_route.h>
  20 #include <net/transp_v6.h>
  21 #endif
  22 #include <net/mptcp.h>
  23
  24 #include "protocol.h"
  25 #include "mib.h"
  26
  27 #include <trace/events/mptcp.h>
  28 #include <trace/events/sock.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 }
  49
  50 static void subflow_generate_hmac(u64 key1, u64 key2, u32 nonce1, u32 nonce2,
  51                                   void *hmac)
  52 {
  53         u8 msg[8];
  54
  55         put_unaligned_be32(nonce1, &msg[0]);
  56         put_unaligned_be32(nonce2, &msg[4]);
  57
  58         mptcp_crypto_hmac_sha(key1, key2, msg, 8, hmac);
  59 }
  60
  61 static bool mptcp_can_accept_new_subflow(const struct mptcp_sock *msk)
  62 {
  63         return mptcp_is_fully_established((void *)msk) &&
  64                 ((mptcp_pm_is_userspace(msk) &&
  65                   mptcp_userspace_pm_active(msk)) ||
  66                  READ_ONCE(msk->pm.accept_subflow));
  67 }
  68
  69 /* validate received token and create truncated hmac and nonce for SYN-ACK */
  70 static void subflow_req_create_thmac(struct mptcp_subflow_request_sock *subflow_req)
  71 {
  72         struct mptcp_sock *msk = subflow_req->msk;
  73         u8 hmac[SHA256_DIGEST_SIZE];
  74
  75         get_random_bytes(&subflow_req->local_nonce, sizeof(u32));
  76
  77         subflow_generate_hmac(READ_ONCE(msk->local_key),
  78                               READ_ONCE(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                 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 156                 return -EINVAL;
 157         }
 158 #endif
 159
 160         mptcp_get_options(skb, &mp_opt);
 161
 162         opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYN);
 163         opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYN);
 164         if (opt_mp_capable) {
 165                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVE);
 166
 167                 if (opt_mp_join)
 168                         return 0;
 169         } else if (opt_mp_join) {
 170                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINSYNRX);
 171         }
 172
 173         if (opt_mp_capable && listener->request_mptcp) {
 174                 int err, retries = MPTCP_TOKEN_MAX_RETRIES;
 175
 176                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 177 again:
 178                 do {
 179                         get_random_bytes(&subflow_req->local_key, sizeof(subflow_req->local_key));
 180                 } while (subflow_req->local_key == 0);
 181
 182                 if (unlikely(req->syncookie)) {
 183                         mptcp_crypto_key_sha(subflow_req->local_key,
 184                                              &subflow_req->token,
 185                                              &subflow_req->idsn);
 186                         if (mptcp_token_exists(subflow_req->token)) {
 187                                 if (retries-- > 0)
 188                                         goto again;
 189                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
 190                         } else {
 191                                 subflow_req->mp_capable = 1;
 192                         }
 193                         return 0;
 194                 }
 195
 196                 err = mptcp_token_new_request(req);
 197                 if (err == 0)
 198                         subflow_req->mp_capable = 1;
 199                 else if (retries-- > 0)
 200                         goto again;
 201                 else
 202                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_TOKENFALLBACKINIT);
 203
 204         } else if (opt_mp_join && listener->request_mptcp) {
 205                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq;
 206                 subflow_req->mp_join = 1;
 207                 subflow_req->backup = mp_opt.backup;
 208                 subflow_req->remote_id = mp_opt.join_id;
 209                 subflow_req->token = mp_opt.token;
 210                 subflow_req->remote_nonce = mp_opt.nonce;
 211                 subflow_req->msk = subflow_token_join_request(req);
 212
 213                 /* Can't fall back to TCP in this case. */
 214                 if (!subflow_req->msk) {
 215                         subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 216                         return -EPERM;
 217                 }
 218
 219                 if (subflow_use_different_sport(subflow_req->msk, sk_listener)) {
 220                         pr_debug("syn inet_sport=%d %d",
 221                                  ntohs(inet_sk(sk_listener)->inet_sport),
 222                                  ntohs(inet_sk((struct sock *)subflow_req->msk)->inet_sport));
 223                         if (!mptcp_pm_sport_in_anno_list(subflow_req->msk, sk_listener)) {
 224                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTSYNRX);
 225                                 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 226                                 return -EPERM;
 227                         }
 228                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTSYNRX);
 229                 }
 230
 231                 subflow_req_create_thmac(subflow_req);
 232
 233                 if (unlikely(req->syncookie)) {
 234                         if (!mptcp_can_accept_new_subflow(subflow_req->msk)) {
 235                                 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 236                                 return -EPERM;
 237                         }
 238
 239                         subflow_init_req_cookie_join_save(subflow_req, skb);
 240                 }
 241
 242                 pr_debug("token=%u, remote_nonce=%u msk=%p", subflow_req->token,
 243                          subflow_req->remote_nonce, subflow_req->msk);
 244         }
 245
 246         return 0;
 247 }
 248
 249 int mptcp_subflow_init_cookie_req(struct request_sock *req,
 250                                   const struct sock *sk_listener,
 251                                   struct sk_buff *skb)
 252 {
 253         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk_listener);
 254         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 255         struct mptcp_options_received mp_opt;
 256         bool opt_mp_capable, opt_mp_join;
 257         int err;
 258
 259         subflow_init_req(req, sk_listener);
 260         mptcp_get_options(skb, &mp_opt);
 261
 262         opt_mp_capable = !!(mp_opt.suboptions & OPTION_MPTCP_MPC_ACK);
 263         opt_mp_join = !!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK);
 264         if (opt_mp_capable && opt_mp_join)
 265                 return -EINVAL;
 266
 267         if (opt_mp_capable && listener->request_mptcp) {
 268                 if (mp_opt.sndr_key == 0)
 269                         return -EINVAL;
 270
 271                 subflow_req->local_key = mp_opt.rcvr_key;
 272                 err = mptcp_token_new_request(req);
 273                 if (err)
 274                         return err;
 275
 276                 subflow_req->mp_capable = 1;
 277                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 278         } else if (opt_mp_join && listener->request_mptcp) {
 279                 if (!mptcp_token_join_cookie_init_state(subflow_req, skb))
 280                         return -EINVAL;
 281
 282                 subflow_req->mp_join = 1;
 283                 subflow_req->ssn_offset = TCP_SKB_CB(skb)->seq - 1;
 284         }
 285
 286         return 0;
 287 }
 288 EXPORT_SYMBOL_GPL(mptcp_subflow_init_cookie_req);
 289
 290 static enum sk_rst_reason mptcp_get_rst_reason(const struct sk_buff *skb)
 291 {
 292         const struct mptcp_ext *mpext = mptcp_get_ext(skb);
 293
 294         if (!mpext)
 295                 return SK_RST_REASON_NOT_SPECIFIED;
 296
 297         return sk_rst_convert_mptcp_reason(mpext->reset_reason);
 298 }
 299
 300 static struct dst_entry *subflow_v4_route_req(const struct sock *sk,
 301                                               struct sk_buff *skb,
 302                                               struct flowi *fl,
 303                                               struct request_sock *req,
 304                                               u32 tw_isn)
 305 {
 306         struct dst_entry *dst;
 307         int err;
 308
 309         tcp_rsk(req)->is_mptcp = 1;
 310         subflow_init_req(req, sk);
 311
 312         dst = tcp_request_sock_ipv4_ops.route_req(sk, skb, fl, req, tw_isn);
 313         if (!dst)
 314                 return NULL;
 315
 316         err = subflow_check_req(req, sk, skb);
 317         if (err == 0)
 318                 return dst;
 319
 320         dst_release(dst);
 321         if (!req->syncookie)
 322                 tcp_request_sock_ops.send_reset(sk, skb,
 323                                                 mptcp_get_rst_reason(skb));
 324         return NULL;
 325 }
 326
 327 static void subflow_prep_synack(const struct sock *sk, struct request_sock *req,
 328                                 struct tcp_fastopen_cookie *foc,
 329                                 enum tcp_synack_type synack_type)
 330 {
 331         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 332         struct inet_request_sock *ireq = inet_rsk(req);
 333
 334         /* clear tstamp_ok, as needed depending on cookie */
 335         if (foc && foc->len > -1)
 336                 ireq->tstamp_ok = 0;
 337
 338         if (synack_type == TCP_SYNACK_FASTOPEN)
 339                 mptcp_fastopen_subflow_synack_set_params(subflow, req);
 340 }
 341
 342 static int subflow_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
 343                                   struct flowi *fl,
 344                                   struct request_sock *req,
 345                                   struct tcp_fastopen_cookie *foc,
 346                                   enum tcp_synack_type synack_type,
 347                                   struct sk_buff *syn_skb)
 348 {
 349         subflow_prep_synack(sk, req, foc, synack_type);
 350
 351         return tcp_request_sock_ipv4_ops.send_synack(sk, dst, fl, req, foc,
 352                                                      synack_type, syn_skb);
 353 }
 354
 355 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 356 static int subflow_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
 357                                   struct flowi *fl,
 358                                   struct request_sock *req,
 359                                   struct tcp_fastopen_cookie *foc,
 360                                   enum tcp_synack_type synack_type,
 361                                   struct sk_buff *syn_skb)
 362 {
 363         subflow_prep_synack(sk, req, foc, synack_type);
 364
 365         return tcp_request_sock_ipv6_ops.send_synack(sk, dst, fl, req, foc,
 366                                                      synack_type, syn_skb);
 367 }
 368
 369 static struct dst_entry *subflow_v6_route_req(const struct sock *sk,
 370                                               struct sk_buff *skb,
 371                                               struct flowi *fl,
 372                                               struct request_sock *req,
 373                                               u32 tw_isn)
 374 {
 375         struct dst_entry *dst;
 376         int err;
 377
 378         tcp_rsk(req)->is_mptcp = 1;
 379         subflow_init_req(req, sk);
 380
 381         dst = tcp_request_sock_ipv6_ops.route_req(sk, skb, fl, req, tw_isn);
 382         if (!dst)
 383                 return NULL;
 384
 385         err = subflow_check_req(req, sk, skb);
 386         if (err == 0)
 387                 return dst;
 388
 389         dst_release(dst);
 390         if (!req->syncookie)
 391                 tcp6_request_sock_ops.send_reset(sk, skb,
 392                                                  mptcp_get_rst_reason(skb));
 393         return NULL;
 394 }
 395 #endif
 396
 397 /* validate received truncated hmac and create hmac for third ACK */
 398 static bool subflow_thmac_valid(struct mptcp_subflow_context *subflow)
 399 {
 400         u8 hmac[SHA256_DIGEST_SIZE];
 401         u64 thmac;
 402
 403         subflow_generate_hmac(subflow->remote_key, subflow->local_key,
 404                               subflow->remote_nonce, subflow->local_nonce,
 405                               hmac);
 406
 407         thmac = get_unaligned_be64(hmac);
 408         pr_debug("subflow=%p, token=%u, thmac=%llu, subflow->thmac=%llu\n",
 409                  subflow, subflow->token, thmac, subflow->thmac);
 410
 411         return thmac == subflow->thmac;
 412 }
 413
 414 void mptcp_subflow_reset(struct sock *ssk)
 415 {
 416         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 417         struct sock *sk = subflow->conn;
 418
 419         /* mptcp_mp_fail_no_response() can reach here on an already closed
 420          * socket
 421          */
 422         if (ssk->sk_state == TCP_CLOSE)
 423                 return;
 424
 425         /* must hold: tcp_done() could drop last reference on parent */
 426         sock_hold(sk);
 427
 428         mptcp_send_active_reset_reason(ssk);
 429         tcp_done(ssk);
 430         if (!test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &mptcp_sk(sk)->flags))
 431                 mptcp_schedule_work(sk);
 432
 433         sock_put(sk);
 434 }
 435
 436 static bool subflow_use_different_dport(struct mptcp_sock *msk, const struct sock *sk)
 437 {
 438         return inet_sk(sk)->inet_dport != inet_sk((struct sock *)msk)->inet_dport;
 439 }
 440
 441 void __mptcp_sync_state(struct sock *sk, int state)
 442 {
 443         struct mptcp_subflow_context *subflow;
 444         struct mptcp_sock *msk = mptcp_sk(sk);
 445         struct sock *ssk = msk->first;
 446
 447         subflow = mptcp_subflow_ctx(ssk);
 448         __mptcp_propagate_sndbuf(sk, ssk);
 449         if (!msk->rcvspace_init)
 450                 mptcp_rcv_space_init(msk, ssk);
 451
 452         if (sk->sk_state == TCP_SYN_SENT) {
 453                 /* subflow->idsn is always available is TCP_SYN_SENT state,
 454                  * even for the FASTOPEN scenarios
 455                  */
 456                 WRITE_ONCE(msk->write_seq, subflow->idsn + 1);
 457                 WRITE_ONCE(msk->snd_nxt, msk->write_seq);
 458                 mptcp_set_state(sk, state);
 459                 sk->sk_state_change(sk);
 460         }
 461 }
 462
 463 static void subflow_set_remote_key(struct mptcp_sock *msk,
 464                                    struct mptcp_subflow_context *subflow,
 465                                    const struct mptcp_options_received *mp_opt)
 466 {
 467         /* active MPC subflow will reach here multiple times:
 468          * at subflow_finish_connect() time and at 4th ack time
 469          */
 470         if (subflow->remote_key_valid)
 471                 return;
 472
 473         subflow->remote_key_valid = 1;
 474         subflow->remote_key = mp_opt->sndr_key;
 475         mptcp_crypto_key_sha(subflow->remote_key, NULL, &subflow->iasn);
 476         subflow->iasn++;
 477
 478         WRITE_ONCE(msk->remote_key, subflow->remote_key);
 479         WRITE_ONCE(msk->ack_seq, subflow->iasn);
 480         WRITE_ONCE(msk->can_ack, true);
 481         atomic64_set(&msk->rcv_wnd_sent, subflow->iasn);
 482 }
 483
 484 static void mptcp_propagate_state(struct sock *sk, struct sock *ssk,
 485                                   struct mptcp_subflow_context *subflow,
 486                                   const struct mptcp_options_received *mp_opt)
 487 {
 488         struct mptcp_sock *msk = mptcp_sk(sk);
 489
 490         mptcp_data_lock(sk);
 491         if (mp_opt) {
 492                 /* Options are available only in the non fallback cases
 493                  * avoid updating rx path fields otherwise
 494                  */
 495                 WRITE_ONCE(msk->snd_una, subflow->idsn + 1);
 496                 WRITE_ONCE(msk->wnd_end, subflow->idsn + 1 + tcp_sk(ssk)->snd_wnd);
 497                 subflow_set_remote_key(msk, subflow, mp_opt);
 498         }
 499
 500         if (!sock_owned_by_user(sk)) {
 501                 __mptcp_sync_state(sk, ssk->sk_state);
 502         } else {
 503                 msk->pending_state = ssk->sk_state;
 504                 __set_bit(MPTCP_SYNC_STATE, &msk->cb_flags);
 505         }
 506         mptcp_data_unlock(sk);
 507 }
 508
 509 static void subflow_finish_connect(struct sock *sk, const struct sk_buff *skb)
 510 {
 511         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 512         struct mptcp_options_received mp_opt;
 513         struct sock *parent = subflow->conn;
 514         struct mptcp_sock *msk;
 515
 516         subflow->icsk_af_ops->sk_rx_dst_set(sk, skb);
 517
 518         /* be sure no special action on any packet other than syn-ack */
 519         if (subflow->conn_finished)
 520                 return;
 521
 522         msk = mptcp_sk(parent);
 523         subflow->rel_write_seq = 1;
 524         subflow->conn_finished = 1;
 525         subflow->ssn_offset = TCP_SKB_CB(skb)->seq;
 526         pr_debug("subflow=%p synack seq=%x", subflow, subflow->ssn_offset);
 527
 528         mptcp_get_options(skb, &mp_opt);
 529         if (subflow->request_mptcp) {
 530                 if (!(mp_opt.suboptions & OPTION_MPTCP_MPC_SYNACK)) {
 531                         MPTCP_INC_STATS(sock_net(sk),
 532                                         MPTCP_MIB_MPCAPABLEACTIVEFALLBACK);
 533                         mptcp_do_fallback(sk);
 534                         pr_fallback(msk);
 535                         goto fallback;
 536                 }
 537
 538                 if (mp_opt.suboptions & OPTION_MPTCP_CSUMREQD)
 539                         WRITE_ONCE(msk->csum_enabled, true);
 540                 if (mp_opt.deny_join_id0)
 541                         WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
 542                 subflow->mp_capable = 1;
 543                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPCAPABLEACTIVEACK);
 544                 mptcp_finish_connect(sk);
 545                 mptcp_propagate_state(parent, sk, subflow, &mp_opt);
 546         } else if (subflow->request_join) {
 547                 u8 hmac[SHA256_DIGEST_SIZE];
 548
 549                 if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_SYNACK)) {
 550                         subflow->reset_reason = MPTCP_RST_EMPTCP;
 551                         goto do_reset;
 552                 }
 553
 554                 subflow->backup = mp_opt.backup;
 555                 subflow->thmac = mp_opt.thmac;
 556                 subflow->remote_nonce = mp_opt.nonce;
 557                 WRITE_ONCE(subflow->remote_id, mp_opt.join_id);
 558                 pr_debug("subflow=%p, thmac=%llu, remote_nonce=%u backup=%d",
 559                          subflow, subflow->thmac, subflow->remote_nonce,
 560                          subflow->backup);
 561
 562                 if (!subflow_thmac_valid(subflow)) {
 563                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINACKMAC);
 564                         subflow->reset_reason = MPTCP_RST_EMPTCP;
 565                         goto do_reset;
 566                 }
 567
 568                 if (!mptcp_finish_join(sk))
 569                         goto do_reset;
 570
 571                 subflow_generate_hmac(subflow->local_key, subflow->remote_key,
 572                                       subflow->local_nonce,
 573                                       subflow->remote_nonce,
 574                                       hmac);
 575                 memcpy(subflow->hmac, hmac, MPTCPOPT_HMAC_LEN);
 576
 577                 subflow->mp_join = 1;
 578                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINSYNACKRX);
 579
 580                 if (subflow_use_different_dport(msk, sk)) {
 581                         pr_debug("synack inet_dport=%d %d",
 582                                  ntohs(inet_sk(sk)->inet_dport),
 583                                  ntohs(inet_sk(parent)->inet_dport));
 584                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_JOINPORTSYNACKRX);
 585                 }
 586         } else if (mptcp_check_fallback(sk)) {
 587 fallback:
 588                 mptcp_propagate_state(parent, sk, subflow, NULL);
 589         }
 590         return;
 591
 592 do_reset:
 593         subflow->reset_transient = 0;
 594         mptcp_subflow_reset(sk);
 595 }
 596
 597 static void subflow_set_local_id(struct mptcp_subflow_context *subflow, int local_id)
 598 {
 599         WARN_ON_ONCE(local_id < 0 || local_id > 255);
 600         WRITE_ONCE(subflow->local_id, local_id);
 601 }
 602
 603 static int subflow_chk_local_id(struct sock *sk)
 604 {
 605         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 606         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 607         int err;
 608
 609         if (likely(subflow->local_id >= 0))
 610                 return 0;
 611
 612         err = mptcp_pm_get_local_id(msk, (struct sock_common *)sk);
 613         if (err < 0)
 614                 return err;
 615
 616         subflow_set_local_id(subflow, err);
 617         return 0;
 618 }
 619
 620 static int subflow_rebuild_header(struct sock *sk)
 621 {
 622         int err = subflow_chk_local_id(sk);
 623
 624         if (unlikely(err < 0))
 625                 return err;
 626
 627         return inet_sk_rebuild_header(sk);
 628 }
 629
 630 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 631 static int subflow_v6_rebuild_header(struct sock *sk)
 632 {
 633         int err = subflow_chk_local_id(sk);
 634
 635         if (unlikely(err < 0))
 636                 return err;
 637
 638         return inet6_sk_rebuild_header(sk);
 639 }
 640 #endif
 641
 642 static struct request_sock_ops mptcp_subflow_v4_request_sock_ops __ro_after_init;
 643 static struct tcp_request_sock_ops subflow_request_sock_ipv4_ops __ro_after_init;
 644
 645 static int subflow_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 646 {
 647         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 648
 649         pr_debug("subflow=%p", subflow);
 650
 651         /* Never answer to SYNs sent to broadcast or multicast */
 652         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 653                 goto drop;
 654
 655         return tcp_conn_request(&mptcp_subflow_v4_request_sock_ops,
 656                                 &subflow_request_sock_ipv4_ops,
 657                                 sk, skb);
 658 drop:
 659         tcp_listendrop(sk);
 660         return 0;
 661 }
 662
 663 static void subflow_v4_req_destructor(struct request_sock *req)
 664 {
 665         subflow_req_destructor(req);
 666         tcp_request_sock_ops.destructor(req);
 667 }
 668
 669 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 670 static struct request_sock_ops mptcp_subflow_v6_request_sock_ops __ro_after_init;
 671 static struct tcp_request_sock_ops subflow_request_sock_ipv6_ops __ro_after_init;
 672 static struct inet_connection_sock_af_ops subflow_v6_specific __ro_after_init;
 673 static struct inet_connection_sock_af_ops subflow_v6m_specific __ro_after_init;
 674 static struct proto tcpv6_prot_override __ro_after_init;
 675
 676 static int subflow_v6_conn_request(struct sock *sk, struct sk_buff *skb)
 677 {
 678         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 679
 680         pr_debug("subflow=%p", subflow);
 681
 682         if (skb->protocol == htons(ETH_P_IP))
 683                 return subflow_v4_conn_request(sk, skb);
 684
 685         if (!ipv6_unicast_destination(skb))
 686                 goto drop;
 687
 688         if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
 689                 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
 690                 return 0;
 691         }
 692
 693         return tcp_conn_request(&mptcp_subflow_v6_request_sock_ops,
 694                                 &subflow_request_sock_ipv6_ops, sk, skb);
 695
 696 drop:
 697         tcp_listendrop(sk);
 698         return 0; /* don't send reset */
 699 }
 700
 701 static void subflow_v6_req_destructor(struct request_sock *req)
 702 {
 703         subflow_req_destructor(req);
 704         tcp6_request_sock_ops.destructor(req);
 705 }
 706 #endif
 707
 708 struct request_sock *mptcp_subflow_reqsk_alloc(const struct request_sock_ops *ops,
 709                                                struct sock *sk_listener,
 710                                                bool attach_listener)
 711 {
 712         if (ops->family == AF_INET)
 713                 ops = &mptcp_subflow_v4_request_sock_ops;
 714 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 715         else if (ops->family == AF_INET6)
 716                 ops = &mptcp_subflow_v6_request_sock_ops;
 717 #endif
 718
 719         return inet_reqsk_alloc(ops, sk_listener, attach_listener);
 720 }
 721 EXPORT_SYMBOL(mptcp_subflow_reqsk_alloc);
 722
 723 /* validate hmac received in third ACK */
 724 static bool subflow_hmac_valid(const struct request_sock *req,
 725                                const struct mptcp_options_received *mp_opt)
 726 {
 727         const struct mptcp_subflow_request_sock *subflow_req;
 728         u8 hmac[SHA256_DIGEST_SIZE];
 729         struct mptcp_sock *msk;
 730
 731         subflow_req = mptcp_subflow_rsk(req);
 732         msk = subflow_req->msk;
 733         if (!msk)
 734                 return false;
 735
 736         subflow_generate_hmac(READ_ONCE(msk->remote_key),
 737                               READ_ONCE(msk->local_key),
 738                               subflow_req->remote_nonce,
 739                               subflow_req->local_nonce, hmac);
 740
 741         return !crypto_memneq(hmac, mp_opt->hmac, MPTCPOPT_HMAC_LEN);
 742 }
 743
 744 static void subflow_ulp_fallback(struct sock *sk,
 745                                  struct mptcp_subflow_context *old_ctx)
 746 {
 747         struct inet_connection_sock *icsk = inet_csk(sk);
 748
 749         mptcp_subflow_tcp_fallback(sk, old_ctx);
 750         icsk->icsk_ulp_ops = NULL;
 751         rcu_assign_pointer(icsk->icsk_ulp_data, NULL);
 752         tcp_sk(sk)->is_mptcp = 0;
 753
 754         mptcp_subflow_ops_undo_override(sk);
 755 }
 756
 757 void mptcp_subflow_drop_ctx(struct sock *ssk)
 758 {
 759         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
 760
 761         if (!ctx)
 762                 return;
 763
 764         list_del(&mptcp_subflow_ctx(ssk)->node);
 765         if (inet_csk(ssk)->icsk_ulp_ops) {
 766                 subflow_ulp_fallback(ssk, ctx);
 767                 if (ctx->conn)
 768                         sock_put(ctx->conn);
 769         }
 770
 771         kfree_rcu(ctx, rcu);
 772 }
 773
 774 void __mptcp_subflow_fully_established(struct mptcp_sock *msk,
 775                                        struct mptcp_subflow_context *subflow,
 776                                        const struct mptcp_options_received *mp_opt)
 777 {
 778         subflow_set_remote_key(msk, subflow, mp_opt);
 779         subflow->fully_established = 1;
 780         WRITE_ONCE(msk->fully_established, true);
 781
 782         if (subflow->is_mptfo)
 783                 __mptcp_fastopen_gen_msk_ackseq(msk, subflow, mp_opt);
 784 }
 785
 786 static struct sock *subflow_syn_recv_sock(const struct sock *sk,
 787                                           struct sk_buff *skb,
 788                                           struct request_sock *req,
 789                                           struct dst_entry *dst,
 790                                           struct request_sock *req_unhash,
 791                                           bool *own_req)
 792 {
 793         struct mptcp_subflow_context *listener = mptcp_subflow_ctx(sk);
 794         struct mptcp_subflow_request_sock *subflow_req;
 795         struct mptcp_options_received mp_opt;
 796         bool fallback, fallback_is_fatal;
 797         enum sk_rst_reason reason;
 798         struct mptcp_sock *owner;
 799         struct sock *child;
 800
 801         pr_debug("listener=%p, req=%p, conn=%p", listener, req, listener->conn);
 802
 803         /* After child creation we must look for MPC even when options
 804          * are not parsed
 805          */
 806         mp_opt.suboptions = 0;
 807
 808         /* hopefully temporary handling for MP_JOIN+syncookie */
 809         subflow_req = mptcp_subflow_rsk(req);
 810         fallback_is_fatal = tcp_rsk(req)->is_mptcp && subflow_req->mp_join;
 811         fallback = !tcp_rsk(req)->is_mptcp;
 812         if (fallback)
 813                 goto create_child;
 814
 815         /* if the sk is MP_CAPABLE, we try to fetch the client key */
 816         if (subflow_req->mp_capable) {
 817                 /* we can receive and accept an in-window, out-of-order pkt,
 818                  * which may not carry the MP_CAPABLE opt even on mptcp enabled
 819                  * paths: always try to extract the peer key, and fallback
 820                  * for packets missing it.
 821                  * Even OoO DSS packets coming legitly after dropped or
 822                  * reordered MPC will cause fallback, but we don't have other
 823                  * options.
 824                  */
 825                 mptcp_get_options(skb, &mp_opt);
 826                 if (!(mp_opt.suboptions &
 827                       (OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_ACK)))
 828                         fallback = true;
 829
 830         } else if (subflow_req->mp_join) {
 831                 mptcp_get_options(skb, &mp_opt);
 832                 if (!(mp_opt.suboptions & OPTION_MPTCP_MPJ_ACK) ||
 833                     !subflow_hmac_valid(req, &mp_opt) ||
 834                     !mptcp_can_accept_new_subflow(subflow_req->msk)) {
 835                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKMAC);
 836                         fallback = true;
 837                 }
 838         }
 839
 840 create_child:
 841         child = listener->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 842                                                      req_unhash, own_req);
 843
 844         if (child && *own_req) {
 845                 struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(child);
 846
 847                 tcp_rsk(req)->drop_req = false;
 848
 849                 /* we need to fallback on ctx allocation failure and on pre-reqs
 850                  * checking above. In the latter scenario we additionally need
 851                  * to reset the context to non MPTCP status.
 852                  */
 853                 if (!ctx || fallback) {
 854                         if (fallback_is_fatal) {
 855                                 subflow_add_reset_reason(skb, MPTCP_RST_EMPTCP);
 856                                 goto dispose_child;
 857                         }
 858                         goto fallback;
 859                 }
 860
 861                 /* ssk inherits options of listener sk */
 862                 ctx->setsockopt_seq = listener->setsockopt_seq;
 863
 864                 if (ctx->mp_capable) {
 865                         ctx->conn = mptcp_sk_clone_init(listener->conn, &mp_opt, child, req);
 866                         if (!ctx->conn)
 867                                 goto fallback;
 868
 869                         ctx->subflow_id = 1;
 870                         owner = mptcp_sk(ctx->conn);
 871                         mptcp_pm_new_connection(owner, child, 1);
 872
 873                         /* with OoO packets we can reach here without ingress
 874                          * mpc option
 875                          */
 876                         if (mp_opt.suboptions & OPTION_MPTCP_MPC_ACK) {
 877                                 mptcp_pm_fully_established(owner, child);
 878                                 ctx->pm_notified = 1;
 879                         }
 880                 } else if (ctx->mp_join) {
 881                         owner = subflow_req->msk;
 882                         if (!owner) {
 883                                 subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 884                                 goto dispose_child;
 885                         }
 886
 887                         /* move the msk reference ownership to the subflow */
 888                         subflow_req->msk = NULL;
 889                         ctx->conn = (struct sock *)owner;
 890
 891                         if (subflow_use_different_sport(owner, sk)) {
 892                                 pr_debug("ack inet_sport=%d %d",
 893                                          ntohs(inet_sk(sk)->inet_sport),
 894                                          ntohs(inet_sk((struct sock *)owner)->inet_sport));
 895                                 if (!mptcp_pm_sport_in_anno_list(owner, sk)) {
 896                                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MISMATCHPORTACKRX);
 897                                         subflow_add_reset_reason(skb, MPTCP_RST_EPROHIBIT);
 898                                         goto dispose_child;
 899                                 }
 900                                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINPORTACKRX);
 901                         }
 902
 903                         if (!mptcp_finish_join(child)) {
 904                                 struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(child);
 905
 906                                 subflow_add_reset_reason(skb, subflow->reset_reason);
 907                                 goto dispose_child;
 908                         }
 909
 910                         SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_JOINACKRX);
 911                         tcp_rsk(req)->drop_req = true;
 912                 }
 913         }
 914
 915         /* check for expected invariant - should never trigger, just help
 916          * catching earlier subtle bugs
 917          */
 918         WARN_ON_ONCE(child && *own_req && tcp_sk(child)->is_mptcp &&
 919                      (!mptcp_subflow_ctx(child) ||
 920                       !mptcp_subflow_ctx(child)->conn));
 921         return child;
 922
 923 dispose_child:
 924         mptcp_subflow_drop_ctx(child);
 925         tcp_rsk(req)->drop_req = true;
 926         inet_csk_prepare_for_destroy_sock(child);
 927         tcp_done(child);
 928         reason = mptcp_get_rst_reason(skb);
 929         req->rsk_ops->send_reset(sk, skb, reason);
 930
 931         /* The last child reference will be released by the caller */
 932         return child;
 933
 934 fallback:
 935         if (fallback)
 936                 SUBFLOW_REQ_INC_STATS(req, MPTCP_MIB_MPCAPABLEPASSIVEFALLBACK);
 937         mptcp_subflow_drop_ctx(child);
 938         return child;
 939 }
 940
 941 static struct inet_connection_sock_af_ops subflow_specific __ro_after_init;
 942 static struct proto tcp_prot_override __ro_after_init;
 943
 944 enum mapping_status {
 945         MAPPING_OK,
 946         MAPPING_INVALID,
 947         MAPPING_EMPTY,
 948         MAPPING_DATA_FIN,
 949         MAPPING_DUMMY,
 950         MAPPING_BAD_CSUM
 951 };
 952
 953 static void dbg_bad_map(struct mptcp_subflow_context *subflow, u32 ssn)
 954 {
 955         pr_debug("Bad mapping: ssn=%d map_seq=%d map_data_len=%d",
 956                  ssn, subflow->map_subflow_seq, subflow->map_data_len);
 957 }
 958
 959 static bool skb_is_fully_mapped(struct sock *ssk, struct sk_buff *skb)
 960 {
 961         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 962         unsigned int skb_consumed;
 963
 964         skb_consumed = tcp_sk(ssk)->copied_seq - TCP_SKB_CB(skb)->seq;
 965         if (WARN_ON_ONCE(skb_consumed >= skb->len))
 966                 return true;
 967
 968         return skb->len - skb_consumed <= subflow->map_data_len -
 969                                           mptcp_subflow_get_map_offset(subflow);
 970 }
 971
 972 static bool validate_mapping(struct sock *ssk, struct sk_buff *skb)
 973 {
 974         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 975         u32 ssn = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
 976
 977         if (unlikely(before(ssn, subflow->map_subflow_seq))) {
 978                 /* Mapping covers data later in the subflow stream,
 979                  * currently unsupported.
 980                  */
 981                 dbg_bad_map(subflow, ssn);
 982                 return false;
 983         }
 984         if (unlikely(!before(ssn, subflow->map_subflow_seq +
 985                                   subflow->map_data_len))) {
 986                 /* Mapping does covers past subflow data, invalid */
 987                 dbg_bad_map(subflow, ssn);
 988                 return false;
 989         }
 990         return true;
 991 }
 992
 993 static enum mapping_status validate_data_csum(struct sock *ssk, struct sk_buff *skb,
 994                                               bool csum_reqd)
 995 {
 996         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
 997         u32 offset, seq, delta;
 998         __sum16 csum;
 999         int len;
1000
1001         if (!csum_reqd)
1002                 return MAPPING_OK;
1003
1004         /* mapping already validated on previous traversal */
1005         if (subflow->map_csum_len == subflow->map_data_len)
1006                 return MAPPING_OK;
1007
1008         /* traverse the receive queue, ensuring it contains a full
1009          * DSS mapping and accumulating the related csum.
1010          * Preserve the accoumlate csum across multiple calls, to compute
1011          * the csum only once
1012          */
1013         delta = subflow->map_data_len - subflow->map_csum_len;
1014         for (;;) {
1015                 seq = tcp_sk(ssk)->copied_seq + subflow->map_csum_len;
1016                 offset = seq - TCP_SKB_CB(skb)->seq;
1017
1018                 /* if the current skb has not been accounted yet, csum its contents
1019                  * up to the amount covered by the current DSS
1020                  */
1021                 if (offset < skb->len) {
1022                         __wsum csum;
1023
1024                         len = min(skb->len - offset, delta);
1025                         csum = skb_checksum(skb, offset, len, 0);
1026                         subflow->map_data_csum = csum_block_add(subflow->map_data_csum, csum,
1027                                                                 subflow->map_csum_len);
1028
1029                         delta -= len;
1030                         subflow->map_csum_len += len;
1031                 }
1032                 if (delta == 0)
1033                         break;
1034
1035                 if (skb_queue_is_last(&ssk->sk_receive_queue, skb)) {
1036                         /* if this subflow is closed, the partial mapping
1037                          * will be never completed; flush the pending skbs, so
1038                          * that subflow_sched_work_if_closed() can kick in
1039                          */
1040                         if (unlikely(ssk->sk_state == TCP_CLOSE))
1041                                 while ((skb = skb_peek(&ssk->sk_receive_queue)))
1042                                         sk_eat_skb(ssk, skb);
1043
1044                         /* not enough data to validate the csum */
1045                         return MAPPING_EMPTY;
1046                 }
1047
1048                 /* the DSS mapping for next skbs will be validated later,
1049                  * when a get_mapping_status call will process such skb
1050                  */
1051                 skb = skb->next;
1052         }
1053
1054         /* note that 'map_data_len' accounts only for the carried data, does
1055          * not include the eventual seq increment due to the data fin,
1056          * while the pseudo header requires the original DSS data len,
1057          * including that
1058          */
1059         csum = __mptcp_make_csum(subflow->map_seq,
1060                                  subflow->map_subflow_seq,
1061                                  subflow->map_data_len + subflow->map_data_fin,
1062                                  subflow->map_data_csum);
1063         if (unlikely(csum)) {
1064                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DATACSUMERR);
1065                 return MAPPING_BAD_CSUM;
1066         }
1067
1068         subflow->valid_csum_seen = 1;
1069         return MAPPING_OK;
1070 }
1071
1072 static enum mapping_status get_mapping_status(struct sock *ssk,
1073                                               struct mptcp_sock *msk)
1074 {
1075         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1076         bool csum_reqd = READ_ONCE(msk->csum_enabled);
1077         struct mptcp_ext *mpext;
1078         struct sk_buff *skb;
1079         u16 data_len;
1080         u64 map_seq;
1081
1082         skb = skb_peek(&ssk->sk_receive_queue);
1083         if (!skb)
1084                 return MAPPING_EMPTY;
1085
1086         if (mptcp_check_fallback(ssk))
1087                 return MAPPING_DUMMY;
1088
1089         mpext = mptcp_get_ext(skb);
1090         if (!mpext || !mpext->use_map) {
1091                 if (!subflow->map_valid && !skb->len) {
1092                         /* the TCP stack deliver 0 len FIN pkt to the receive
1093                          * queue, that is the only 0len pkts ever expected here,
1094                          * and we can admit no mapping only for 0 len pkts
1095                          */
1096                         if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
1097                                 WARN_ONCE(1, "0len seq %d:%d flags %x",
1098                                           TCP_SKB_CB(skb)->seq,
1099                                           TCP_SKB_CB(skb)->end_seq,
1100                                           TCP_SKB_CB(skb)->tcp_flags);
1101                         sk_eat_skb(ssk, skb);
1102                         return MAPPING_EMPTY;
1103                 }
1104
1105                 if (!subflow->map_valid)
1106                         return MAPPING_INVALID;
1107
1108                 goto validate_seq;
1109         }
1110
1111         trace_get_mapping_status(mpext);
1112
1113         data_len = mpext->data_len;
1114         if (data_len == 0) {
1115                 pr_debug("infinite mapping received");
1116                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_INFINITEMAPRX);
1117                 subflow->map_data_len = 0;
1118                 return MAPPING_INVALID;
1119         }
1120
1121         if (mpext->data_fin == 1) {
1122                 u64 data_fin_seq;
1123
1124                 if (data_len == 1) {
1125                         bool updated = mptcp_update_rcv_data_fin(msk, mpext->data_seq,
1126                                                                  mpext->dsn64);
1127                         pr_debug("DATA_FIN with no payload seq=%llu", mpext->data_seq);
1128                         if (subflow->map_valid) {
1129                                 /* A DATA_FIN might arrive in a DSS
1130                                  * option before the previous mapping
1131                                  * has been fully consumed. Continue
1132                                  * handling the existing mapping.
1133                                  */
1134                                 skb_ext_del(skb, SKB_EXT_MPTCP);
1135                                 return MAPPING_OK;
1136                         }
1137
1138                         if (updated)
1139                                 mptcp_schedule_work((struct sock *)msk);
1140
1141                         return MAPPING_DATA_FIN;
1142                 }
1143
1144                 data_fin_seq = mpext->data_seq + data_len - 1;
1145
1146                 /* If mpext->data_seq is a 32-bit value, data_fin_seq must also
1147                  * be limited to 32 bits.
1148                  */
1149                 if (!mpext->dsn64)
1150                         data_fin_seq &= GENMASK_ULL(31, 0);
1151
1152                 mptcp_update_rcv_data_fin(msk, data_fin_seq, mpext->dsn64);
1153                 pr_debug("DATA_FIN with mapping seq=%llu dsn64=%d",
1154                          data_fin_seq, mpext->dsn64);
1155
1156                 /* Adjust for DATA_FIN using 1 byte of sequence space */
1157                 data_len--;
1158         }
1159
1160         map_seq = mptcp_expand_seq(READ_ONCE(msk->ack_seq), mpext->data_seq, mpext->dsn64);
1161         WRITE_ONCE(mptcp_sk(subflow->conn)->use_64bit_ack, !!mpext->dsn64);
1162
1163         if (subflow->map_valid) {
1164                 /* Allow replacing only with an identical map */
1165                 if (subflow->map_seq == map_seq &&
1166                     subflow->map_subflow_seq == mpext->subflow_seq &&
1167                     subflow->map_data_len == data_len &&
1168                     subflow->map_csum_reqd == mpext->csum_reqd) {
1169                         skb_ext_del(skb, SKB_EXT_MPTCP);
1170                         goto validate_csum;
1171                 }
1172
1173                 /* If this skb data are fully covered by the current mapping,
1174                  * the new map would need caching, which is not supported
1175                  */
1176                 if (skb_is_fully_mapped(ssk, skb)) {
1177                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSNOMATCH);
1178                         return MAPPING_INVALID;
1179                 }
1180
1181                 /* will validate the next map after consuming the current one */
1182                 goto validate_csum;
1183         }
1184
1185         subflow->map_seq = map_seq;
1186         subflow->map_subflow_seq = mpext->subflow_seq;
1187         subflow->map_data_len = data_len;
1188         subflow->map_valid = 1;
1189         subflow->map_data_fin = mpext->data_fin;
1190         subflow->mpc_map = mpext->mpc_map;
1191         subflow->map_csum_reqd = mpext->csum_reqd;
1192         subflow->map_csum_len = 0;
1193         subflow->map_data_csum = csum_unfold(mpext->csum);
1194
1195         /* Cfr RFC 8684 Section 3.3.0 */
1196         if (unlikely(subflow->map_csum_reqd != csum_reqd))
1197                 return MAPPING_INVALID;
1198
1199         pr_debug("new map seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
1200                  subflow->map_seq, subflow->map_subflow_seq,
1201                  subflow->map_data_len, subflow->map_csum_reqd,
1202                  subflow->map_data_csum);
1203
1204 validate_seq:
1205         /* we revalidate valid mapping on new skb, because we must ensure
1206          * the current skb is completely covered by the available mapping
1207          */
1208         if (!validate_mapping(ssk, skb)) {
1209                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DSSTCPMISMATCH);
1210                 return MAPPING_INVALID;
1211         }
1212
1213         skb_ext_del(skb, SKB_EXT_MPTCP);
1214
1215 validate_csum:
1216         return validate_data_csum(ssk, skb, csum_reqd);
1217 }
1218
1219 static void mptcp_subflow_discard_data(struct sock *ssk, struct sk_buff *skb,
1220                                        u64 limit)
1221 {
1222         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1223         bool fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;
1224         u32 incr;
1225
1226         incr = limit >= skb->len ? skb->len + fin : limit;
1227
1228         pr_debug("discarding=%d len=%d seq=%d", incr, skb->len,
1229                  subflow->map_subflow_seq);
1230         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_DUPDATA);
1231         tcp_sk(ssk)->copied_seq += incr;
1232         if (!before(tcp_sk(ssk)->copied_seq, TCP_SKB_CB(skb)->end_seq))
1233                 sk_eat_skb(ssk, skb);
1234         if (mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len)
1235                 subflow->map_valid = 0;
1236 }
1237
1238 /* sched mptcp worker to remove the subflow if no more data is pending */
1239 static void subflow_sched_work_if_closed(struct mptcp_sock *msk, struct sock *ssk)
1240 {
1241         if (likely(ssk->sk_state != TCP_CLOSE))
1242                 return;
1243
1244         if (skb_queue_empty(&ssk->sk_receive_queue) &&
1245             !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW, &msk->flags))
1246                 mptcp_schedule_work((struct sock *)msk);
1247 }
1248
1249 static bool subflow_can_fallback(struct mptcp_subflow_context *subflow)
1250 {
1251         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1252
1253         if (subflow->mp_join)
1254                 return false;
1255         else if (READ_ONCE(msk->csum_enabled))
1256                 return !subflow->valid_csum_seen;
1257         else
1258                 return !subflow->fully_established;
1259 }
1260
1261 static void mptcp_subflow_fail(struct mptcp_sock *msk, struct sock *ssk)
1262 {
1263         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1264         unsigned long fail_tout;
1265
1266         /* graceful failure can happen only on the MPC subflow */
1267         if (WARN_ON_ONCE(ssk != READ_ONCE(msk->first)))
1268                 return;
1269
1270         /* since the close timeout take precedence on the fail one,
1271          * no need to start the latter when the first is already set
1272          */
1273         if (sock_flag((struct sock *)msk, SOCK_DEAD))
1274                 return;
1275
1276         /* we don't need extreme accuracy here, use a zero fail_tout as special
1277          * value meaning no fail timeout at all;
1278          */
1279         fail_tout = jiffies + TCP_RTO_MAX;
1280         if (!fail_tout)
1281                 fail_tout = 1;
1282         WRITE_ONCE(subflow->fail_tout, fail_tout);
1283         tcp_send_ack(ssk);
1284
1285         mptcp_reset_tout_timer(msk, subflow->fail_tout);
1286 }
1287
1288 static bool subflow_check_data_avail(struct sock *ssk)
1289 {
1290         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1291         enum mapping_status status;
1292         struct mptcp_sock *msk;
1293         struct sk_buff *skb;
1294
1295         if (!skb_peek(&ssk->sk_receive_queue))
1296                 WRITE_ONCE(subflow->data_avail, false);
1297         if (subflow->data_avail)
1298                 return true;
1299
1300         msk = mptcp_sk(subflow->conn);
1301         for (;;) {
1302                 u64 ack_seq;
1303                 u64 old_ack;
1304
1305                 status = get_mapping_status(ssk, msk);
1306                 trace_subflow_check_data_avail(status, skb_peek(&ssk->sk_receive_queue));
1307                 if (unlikely(status == MAPPING_INVALID || status == MAPPING_DUMMY ||
1308                              status == MAPPING_BAD_CSUM))
1309                         goto fallback;
1310
1311                 if (status != MAPPING_OK)
1312                         goto no_data;
1313
1314                 skb = skb_peek(&ssk->sk_receive_queue);
1315                 if (WARN_ON_ONCE(!skb))
1316                         goto no_data;
1317
1318                 if (unlikely(!READ_ONCE(msk->can_ack)))
1319                         goto fallback;
1320
1321                 old_ack = READ_ONCE(msk->ack_seq);
1322                 ack_seq = mptcp_subflow_get_mapped_dsn(subflow);
1323                 pr_debug("msk ack_seq=%llx subflow ack_seq=%llx", old_ack,
1324                          ack_seq);
1325                 if (unlikely(before64(ack_seq, old_ack))) {
1326                         mptcp_subflow_discard_data(ssk, skb, old_ack - ack_seq);
1327                         continue;
1328                 }
1329
1330                 WRITE_ONCE(subflow->data_avail, true);
1331                 break;
1332         }
1333         return true;
1334
1335 no_data:
1336         subflow_sched_work_if_closed(msk, ssk);
1337         return false;
1338
1339 fallback:
1340         if (!__mptcp_check_fallback(msk)) {
1341                 /* RFC 8684 section 3.7. */
1342                 if (status == MAPPING_BAD_CSUM &&
1343                     (subflow->mp_join || subflow->valid_csum_seen)) {
1344                         subflow->send_mp_fail = 1;
1345
1346                         if (!READ_ONCE(msk->allow_infinite_fallback)) {
1347                                 subflow->reset_transient = 0;
1348                                 subflow->reset_reason = MPTCP_RST_EMIDDLEBOX;
1349                                 goto reset;
1350                         }
1351                         mptcp_subflow_fail(msk, ssk);
1352                         WRITE_ONCE(subflow->data_avail, true);
1353                         return true;
1354                 }
1355
1356                 if (!subflow_can_fallback(subflow) && subflow->map_data_len) {
1357                         /* fatal protocol error, close the socket.
1358                          * subflow_error_report() will introduce the appropriate barriers
1359                          */
1360                         subflow->reset_transient = 0;
1361                         subflow->reset_reason = MPTCP_RST_EMPTCP;
1362
1363 reset:
1364                         WRITE_ONCE(ssk->sk_err, EBADMSG);
1365                         tcp_set_state(ssk, TCP_CLOSE);
1366                         while ((skb = skb_peek(&ssk->sk_receive_queue)))
1367                                 sk_eat_skb(ssk, skb);
1368                         mptcp_send_active_reset_reason(ssk);
1369                         WRITE_ONCE(subflow->data_avail, false);
1370                         return false;
1371                 }
1372
1373                 mptcp_do_fallback(ssk);
1374         }
1375
1376         skb = skb_peek(&ssk->sk_receive_queue);
1377         subflow->map_valid = 1;
1378         subflow->map_seq = READ_ONCE(msk->ack_seq);
1379         subflow->map_data_len = skb->len;
1380         subflow->map_subflow_seq = tcp_sk(ssk)->copied_seq - subflow->ssn_offset;
1381         WRITE_ONCE(subflow->data_avail, true);
1382         return true;
1383 }
1384
1385 bool mptcp_subflow_data_available(struct sock *sk)
1386 {
1387         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1388
1389         /* check if current mapping is still valid */
1390         if (subflow->map_valid &&
1391             mptcp_subflow_get_map_offset(subflow) >= subflow->map_data_len) {
1392                 subflow->map_valid = 0;
1393                 WRITE_ONCE(subflow->data_avail, false);
1394
1395                 pr_debug("Done with mapping: seq=%u data_len=%u",
1396                          subflow->map_subflow_seq,
1397                          subflow->map_data_len);
1398         }
1399
1400         return subflow_check_data_avail(sk);
1401 }
1402
1403 /* If ssk has an mptcp parent socket, use the mptcp rcvbuf occupancy,
1404  * not the ssk one.
1405  *
1406  * In mptcp, rwin is about the mptcp-level connection data.
1407  *
1408  * Data that is still on the ssk rx queue can thus be ignored,
1409  * as far as mptcp peer is concerned that data is still inflight.
1410  * DSS ACK is updated when skb is moved to the mptcp rx queue.
1411  */
1412 void mptcp_space(const struct sock *ssk, int *space, int *full_space)
1413 {
1414         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
1415         const struct sock *sk = subflow->conn;
1416
1417         *space = __mptcp_space(sk);
1418         *full_space = mptcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1419 }
1420
1421 static void subflow_error_report(struct sock *ssk)
1422 {
1423         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1424
1425         /* bail early if this is a no-op, so that we avoid introducing a
1426          * problematic lockdep dependency between TCP accept queue lock
1427          * and msk socket spinlock
1428          */
1429         if (!sk->sk_socket)
1430                 return;
1431
1432         mptcp_data_lock(sk);
1433         if (!sock_owned_by_user(sk))
1434                 __mptcp_error_report(sk);
1435         else
1436                 __set_bit(MPTCP_ERROR_REPORT,  &mptcp_sk(sk)->cb_flags);
1437         mptcp_data_unlock(sk);
1438 }
1439
1440 static void subflow_data_ready(struct sock *sk)
1441 {
1442         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1443         u16 state = 1 << inet_sk_state_load(sk);
1444         struct sock *parent = subflow->conn;
1445         struct mptcp_sock *msk;
1446
1447         trace_sk_data_ready(sk);
1448
1449         msk = mptcp_sk(parent);
1450         if (state & TCPF_LISTEN) {
1451                 /* MPJ subflow are removed from accept queue before reaching here,
1452                  * avoid stray wakeups
1453                  */
1454                 if (reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue))
1455                         return;
1456
1457                 parent->sk_data_ready(parent);
1458                 return;
1459         }
1460
1461         WARN_ON_ONCE(!__mptcp_check_fallback(msk) && !subflow->mp_capable &&
1462                      !subflow->mp_join && !(state & TCPF_CLOSE));
1463
1464         if (mptcp_subflow_data_available(sk)) {
1465                 mptcp_data_ready(parent, sk);
1466
1467                 /* subflow-level lowat test are not relevant.
1468                  * respect the msk-level threshold eventually mandating an immediate ack
1469                  */
1470                 if (mptcp_data_avail(msk) < parent->sk_rcvlowat &&
1471                     (tcp_sk(sk)->rcv_nxt - tcp_sk(sk)->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss)
1472                         inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW;
1473         } else if (unlikely(sk->sk_err)) {
1474                 subflow_error_report(sk);
1475         }
1476 }
1477
1478 static void subflow_write_space(struct sock *ssk)
1479 {
1480         struct sock *sk = mptcp_subflow_ctx(ssk)->conn;
1481
1482         mptcp_propagate_sndbuf(sk, ssk);
1483         mptcp_write_space(sk);
1484 }
1485
1486 static const struct inet_connection_sock_af_ops *
1487 subflow_default_af_ops(struct sock *sk)
1488 {
1489 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1490         if (sk->sk_family == AF_INET6)
1491                 return &subflow_v6_specific;
1492 #endif
1493         return &subflow_specific;
1494 }
1495
1496 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1497 void mptcpv6_handle_mapped(struct sock *sk, bool mapped)
1498 {
1499         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1500         struct inet_connection_sock *icsk = inet_csk(sk);
1501         const struct inet_connection_sock_af_ops *target;
1502
1503         target = mapped ? &subflow_v6m_specific : subflow_default_af_ops(sk);
1504
1505         pr_debug("subflow=%p family=%d ops=%p target=%p mapped=%d",
1506                  subflow, sk->sk_family, icsk->icsk_af_ops, target, mapped);
1507
1508         if (likely(icsk->icsk_af_ops == target))
1509                 return;
1510
1511         subflow->icsk_af_ops = icsk->icsk_af_ops;
1512         icsk->icsk_af_ops = target;
1513 }
1514 #endif
1515
1516 void mptcp_info2sockaddr(const struct mptcp_addr_info *info,
1517                          struct sockaddr_storage *addr,
1518                          unsigned short family)
1519 {
1520         memset(addr, 0, sizeof(*addr));
1521         addr->ss_family = family;
1522         if (addr->ss_family == AF_INET) {
1523                 struct sockaddr_in *in_addr = (struct sockaddr_in *)addr;
1524
1525                 if (info->family == AF_INET)
1526                         in_addr->sin_addr = info->addr;
1527 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1528                 else if (ipv6_addr_v4mapped(&info->addr6))
1529                         in_addr->sin_addr.s_addr = info->addr6.s6_addr32[3];
1530 #endif
1531                 in_addr->sin_port = info->port;
1532         }
1533 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1534         else if (addr->ss_family == AF_INET6) {
1535                 struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)addr;
1536
1537                 if (info->family == AF_INET)
1538                         ipv6_addr_set_v4mapped(info->addr.s_addr,
1539                                                &in6_addr->sin6_addr);
1540                 else
1541                         in6_addr->sin6_addr = info->addr6;
1542                 in6_addr->sin6_port = info->port;
1543         }
1544 #endif
1545 }
1546
1547 int __mptcp_subflow_connect(struct sock *sk, const struct mptcp_addr_info *loc,
1548                             const struct mptcp_addr_info *remote)
1549 {
1550         struct mptcp_sock *msk = mptcp_sk(sk);
1551         struct mptcp_subflow_context *subflow;
1552         struct sockaddr_storage addr;
1553         int remote_id = remote->id;
1554         int local_id = loc->id;
1555         int err = -ENOTCONN;
1556         struct socket *sf;
1557         struct sock *ssk;
1558         u32 remote_token;
1559         int addrlen;
1560         int ifindex;
1561         u8 flags;
1562
1563         if (!mptcp_is_fully_established(sk))
1564                 goto err_out;
1565
1566         err = mptcp_subflow_create_socket(sk, loc->family, &sf);
1567         if (err)
1568                 goto err_out;
1569
1570         ssk = sf->sk;
1571         subflow = mptcp_subflow_ctx(ssk);
1572         do {
1573                 get_random_bytes(&subflow->local_nonce, sizeof(u32));
1574         } while (!subflow->local_nonce);
1575
1576         if (local_id)
1577                 subflow_set_local_id(subflow, local_id);
1578
1579         mptcp_pm_get_flags_and_ifindex_by_id(msk, local_id,
1580                                              &flags, &ifindex);
1581         subflow->remote_key_valid = 1;
1582         subflow->remote_key = READ_ONCE(msk->remote_key);
1583         subflow->local_key = READ_ONCE(msk->local_key);
1584         subflow->token = msk->token;
1585         mptcp_info2sockaddr(loc, &addr, ssk->sk_family);
1586
1587         addrlen = sizeof(struct sockaddr_in);
1588 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1589         if (addr.ss_family == AF_INET6)
1590                 addrlen = sizeof(struct sockaddr_in6);
1591 #endif
1592         ssk->sk_bound_dev_if = ifindex;
1593         err = kernel_bind(sf, (struct sockaddr *)&addr, addrlen);
1594         if (err)
1595                 goto failed;
1596
1597         mptcp_crypto_key_sha(subflow->remote_key, &remote_token, NULL);
1598         pr_debug("msk=%p remote_token=%u local_id=%d remote_id=%d", msk,
1599                  remote_token, local_id, remote_id);
1600         subflow->remote_token = remote_token;
1601         WRITE_ONCE(subflow->remote_id, remote_id);
1602         subflow->request_join = 1;
1603         subflow->request_bkup = !!(flags & MPTCP_PM_ADDR_FLAG_BACKUP);
1604         subflow->subflow_id = msk->subflow_id++;
1605         mptcp_info2sockaddr(remote, &addr, ssk->sk_family);
1606
1607         sock_hold(ssk);
1608         list_add_tail(&subflow->node, &msk->conn_list);
1609         err = kernel_connect(sf, (struct sockaddr *)&addr, addrlen, O_NONBLOCK);
1610         if (err && err != -EINPROGRESS)
1611                 goto failed_unlink;
1612
1613         /* discard the subflow socket */
1614         mptcp_sock_graft(ssk, sk->sk_socket);
1615         iput(SOCK_INODE(sf));
1616         WRITE_ONCE(msk->allow_infinite_fallback, false);
1617         mptcp_stop_tout_timer(sk);
1618         return 0;
1619
1620 failed_unlink:
1621         list_del(&subflow->node);
1622         sock_put(mptcp_subflow_tcp_sock(subflow));
1623
1624 failed:
1625         subflow->disposable = 1;
1626         sock_release(sf);
1627
1628 err_out:
1629         /* we account subflows before the creation, and this failures will not
1630          * be caught by sk_state_change()
1631          */
1632         mptcp_pm_close_subflow(msk);
1633         return err;
1634 }
1635
1636 static void mptcp_attach_cgroup(struct sock *parent, struct sock *child)
1637 {
1638 #ifdef CONFIG_SOCK_CGROUP_DATA
1639         struct sock_cgroup_data *parent_skcd = &parent->sk_cgrp_data,
1640                                 *child_skcd = &child->sk_cgrp_data;
1641
1642         /* only the additional subflows created by kworkers have to be modified */
1643         if (cgroup_id(sock_cgroup_ptr(parent_skcd)) !=
1644             cgroup_id(sock_cgroup_ptr(child_skcd))) {
1645 #ifdef CONFIG_MEMCG
1646                 struct mem_cgroup *memcg = parent->sk_memcg;
1647
1648                 mem_cgroup_sk_free(child);
1649                 if (memcg && css_tryget(&memcg->css))
1650                         child->sk_memcg = memcg;
1651 #endif /* CONFIG_MEMCG */
1652
1653                 cgroup_sk_free(child_skcd);
1654                 *child_skcd = *parent_skcd;
1655                 cgroup_sk_clone(child_skcd);
1656         }
1657 #endif /* CONFIG_SOCK_CGROUP_DATA */
1658 }
1659
1660 static void mptcp_subflow_ops_override(struct sock *ssk)
1661 {
1662 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1663         if (ssk->sk_prot == &tcpv6_prot)
1664                 ssk->sk_prot = &tcpv6_prot_override;
1665         else
1666 #endif
1667                 ssk->sk_prot = &tcp_prot_override;
1668 }
1669
1670 static void mptcp_subflow_ops_undo_override(struct sock *ssk)
1671 {
1672 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1673         if (ssk->sk_prot == &tcpv6_prot_override)
1674                 ssk->sk_prot = &tcpv6_prot;
1675         else
1676 #endif
1677                 ssk->sk_prot = &tcp_prot;
1678 }
1679
1680 int mptcp_subflow_create_socket(struct sock *sk, unsigned short family,
1681                                 struct socket **new_sock)
1682 {
1683         struct mptcp_subflow_context *subflow;
1684         struct net *net = sock_net(sk);
1685         struct socket *sf;
1686         int err;
1687
1688         /* un-accepted server sockets can reach here - on bad configuration
1689          * bail early to avoid greater trouble later
1690          */
1691         if (unlikely(!sk->sk_socket))
1692                 return -EINVAL;
1693
1694         err = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, &sf);
1695         if (err)
1696                 return err;
1697
1698         lock_sock_nested(sf->sk, SINGLE_DEPTH_NESTING);
1699
1700         err = security_mptcp_add_subflow(sk, sf->sk);
1701         if (err)
1702                 goto err_free;
1703
1704         /* the newly created socket has to be in the same cgroup as its parent */
1705         mptcp_attach_cgroup(sk, sf->sk);
1706
1707         /* kernel sockets do not by default acquire net ref, but TCP timer
1708          * needs it.
1709          * Update ns_tracker to current stack trace and refcounted tracker.
1710          */
1711         __netns_tracker_free(net, &sf->sk->ns_tracker, false);
1712         sf->sk->sk_net_refcnt = 1;
1713         get_net_track(net, &sf->sk->ns_tracker, GFP_KERNEL);
1714         sock_inuse_add(net, 1);
1715         err = tcp_set_ulp(sf->sk, "mptcp");
1716         if (err)
1717                 goto err_free;
1718
1719         mptcp_sockopt_sync_locked(mptcp_sk(sk), sf->sk);
1720         release_sock(sf->sk);
1721
1722         /* the newly created socket really belongs to the owning MPTCP
1723          * socket, even if for additional subflows the allocation is performed
1724          * by a kernel workqueue. Adjust inode references, so that the
1725          * procfs/diag interfaces really show this one belonging to the correct
1726          * user.
1727          */
1728         SOCK_INODE(sf)->i_ino = SOCK_INODE(sk->sk_socket)->i_ino;
1729         SOCK_INODE(sf)->i_uid = SOCK_INODE(sk->sk_socket)->i_uid;
1730         SOCK_INODE(sf)->i_gid = SOCK_INODE(sk->sk_socket)->i_gid;
1731
1732         subflow = mptcp_subflow_ctx(sf->sk);
1733         pr_debug("subflow=%p", subflow);
1734
1735         *new_sock = sf;
1736         sock_hold(sk);
1737         subflow->conn = sk;
1738         mptcp_subflow_ops_override(sf->sk);
1739
1740         return 0;
1741
1742 err_free:
1743         release_sock(sf->sk);
1744         sock_release(sf);
1745         return err;
1746 }
1747
1748 static struct mptcp_subflow_context *subflow_create_ctx(struct sock *sk,
1749                                                         gfp_t priority)
1750 {
1751         struct inet_connection_sock *icsk = inet_csk(sk);
1752         struct mptcp_subflow_context *ctx;
1753
1754         ctx = kzalloc(sizeof(*ctx), priority);
1755         if (!ctx)
1756                 return NULL;
1757
1758         rcu_assign_pointer(icsk->icsk_ulp_data, ctx);
1759         INIT_LIST_HEAD(&ctx->node);
1760         INIT_LIST_HEAD(&ctx->delegated_node);
1761
1762         pr_debug("subflow=%p", ctx);
1763
1764         ctx->tcp_sock = sk;
1765         WRITE_ONCE(ctx->local_id, -1);
1766
1767         return ctx;
1768 }
1769
1770 static void __subflow_state_change(struct sock *sk)
1771 {
1772         struct socket_wq *wq;
1773
1774         rcu_read_lock();
1775         wq = rcu_dereference(sk->sk_wq);
1776         if (skwq_has_sleeper(wq))
1777                 wake_up_interruptible_all(&wq->wait);
1778         rcu_read_unlock();
1779 }
1780
1781 static bool subflow_is_done(const struct sock *sk)
1782 {
1783         return sk->sk_shutdown & RCV_SHUTDOWN || sk->sk_state == TCP_CLOSE;
1784 }
1785
1786 static void subflow_state_change(struct sock *sk)
1787 {
1788         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1789         struct sock *parent = subflow->conn;
1790         struct mptcp_sock *msk;
1791
1792         __subflow_state_change(sk);
1793
1794         msk = mptcp_sk(parent);
1795         if (subflow_simultaneous_connect(sk)) {
1796                 mptcp_do_fallback(sk);
1797                 pr_fallback(msk);
1798                 subflow->conn_finished = 1;
1799                 mptcp_propagate_state(parent, sk, subflow, NULL);
1800         }
1801
1802         /* as recvmsg() does not acquire the subflow socket for ssk selection
1803          * a fin packet carrying a DSS can be unnoticed if we don't trigger
1804          * the data available machinery here.
1805          */
1806         if (mptcp_subflow_data_available(sk))
1807                 mptcp_data_ready(parent, sk);
1808         else if (unlikely(sk->sk_err))
1809                 subflow_error_report(sk);
1810
1811         subflow_sched_work_if_closed(mptcp_sk(parent), sk);
1812
1813         /* when the fallback subflow closes the rx side, trigger a 'dummy'
1814          * ingress data fin, so that the msk state will follow along
1815          */
1816         if (__mptcp_check_fallback(msk) && subflow_is_done(sk) && msk->first == sk &&
1817             mptcp_update_rcv_data_fin(msk, READ_ONCE(msk->ack_seq), true))
1818                 mptcp_schedule_work(parent);
1819 }
1820
1821 void mptcp_subflow_queue_clean(struct sock *listener_sk, struct sock *listener_ssk)
1822 {
1823         struct request_sock_queue *queue = &inet_csk(listener_ssk)->icsk_accept_queue;
1824         struct request_sock *req, *head, *tail;
1825         struct mptcp_subflow_context *subflow;
1826         struct sock *sk, *ssk;
1827
1828         /* Due to lock dependencies no relevant lock can be acquired under rskq_lock.
1829          * Splice the req list, so that accept() can not reach the pending ssk after
1830          * the listener socket is released below.
1831          */
1832         spin_lock_bh(&queue->rskq_lock);
1833         head = queue->rskq_accept_head;
1834         tail = queue->rskq_accept_tail;
1835         queue->rskq_accept_head = NULL;
1836         queue->rskq_accept_tail = NULL;
1837         spin_unlock_bh(&queue->rskq_lock);
1838
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 (req = head; req; req = req->dl_next) {
1848                 ssk = req->sk;
1849                 if (!sk_is_mptcp(ssk))
1850                         continue;
1851
1852                 subflow = mptcp_subflow_ctx(ssk);
1853                 if (!subflow || !subflow->conn)
1854                         continue;
1855
1856                 sk = subflow->conn;
1857                 sock_hold(sk);
1858
1859                 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
1860                 __mptcp_unaccepted_force_close(sk);
1861                 release_sock(sk);
1862
1863                 /* lockdep will report a false positive ABBA deadlock
1864                  * between cancel_work_sync and the listener socket.
1865                  * The involved locks belong to different sockets WRT
1866                  * the existing AB chain.
1867                  * Using a per socket key is problematic as key
1868                  * deregistration requires process context and must be
1869                  * performed at socket disposal time, in atomic
1870                  * context.
1871                  * Just tell lockdep to consider the listener socket
1872                  * released here.
1873                  */
1874                 mutex_release(&listener_sk->sk_lock.dep_map, _RET_IP_);
1875                 mptcp_cancel_work(sk);
1876                 mutex_acquire(&listener_sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1877
1878                 sock_put(sk);
1879         }
1880
1881         /* we are still under the listener msk socket lock */
1882         lock_sock_nested(listener_ssk, SINGLE_DEPTH_NESTING);
1883
1884         /* restore the listener queue, to let the TCP code clean it up */
1885         spin_lock_bh(&queue->rskq_lock);
1886         WARN_ON_ONCE(queue->rskq_accept_head);
1887         queue->rskq_accept_head = head;
1888         queue->rskq_accept_tail = tail;
1889         spin_unlock_bh(&queue->rskq_lock);
1890 }
1891
1892 static int subflow_ulp_init(struct sock *sk)
1893 {
1894         struct inet_connection_sock *icsk = inet_csk(sk);
1895         struct mptcp_subflow_context *ctx;
1896         struct tcp_sock *tp = tcp_sk(sk);
1897         int err = 0;
1898
1899         /* disallow attaching ULP to a socket unless it has been
1900          * created with sock_create_kern()
1901          */
1902         if (!sk->sk_kern_sock) {
1903                 err = -EOPNOTSUPP;
1904                 goto out;
1905         }
1906
1907         ctx = subflow_create_ctx(sk, GFP_KERNEL);
1908         if (!ctx) {
1909                 err = -ENOMEM;
1910                 goto out;
1911         }
1912
1913         pr_debug("subflow=%p, family=%d", ctx, sk->sk_family);
1914
1915         tp->is_mptcp = 1;
1916         ctx->icsk_af_ops = icsk->icsk_af_ops;
1917         icsk->icsk_af_ops = subflow_default_af_ops(sk);
1918         ctx->tcp_state_change = sk->sk_state_change;
1919         ctx->tcp_error_report = sk->sk_error_report;
1920
1921         WARN_ON_ONCE(sk->sk_data_ready != sock_def_readable);
1922         WARN_ON_ONCE(sk->sk_write_space != sk_stream_write_space);
1923
1924         sk->sk_data_ready = subflow_data_ready;
1925         sk->sk_write_space = subflow_write_space;
1926         sk->sk_state_change = subflow_state_change;
1927         sk->sk_error_report = subflow_error_report;
1928 out:
1929         return err;
1930 }
1931
1932 static void subflow_ulp_release(struct sock *ssk)
1933 {
1934         struct mptcp_subflow_context *ctx = mptcp_subflow_ctx(ssk);
1935         bool release = true;
1936         struct sock *sk;
1937
1938         if (!ctx)
1939                 return;
1940
1941         sk = ctx->conn;
1942         if (sk) {
1943                 /* if the msk has been orphaned, keep the ctx
1944                  * alive, will be freed by __mptcp_close_ssk(),
1945                  * when the subflow is still unaccepted
1946                  */
1947                 release = ctx->disposable || list_empty(&ctx->node);
1948
1949                 /* inet_child_forget() does not call sk_state_change(),
1950                  * explicitly trigger the socket close machinery
1951                  */
1952                 if (!release && !test_and_set_bit(MPTCP_WORK_CLOSE_SUBFLOW,
1953                                                   &mptcp_sk(sk)->flags))
1954                         mptcp_schedule_work(sk);
1955                 sock_put(sk);
1956         }
1957
1958         mptcp_subflow_ops_undo_override(ssk);
1959         if (release)
1960                 kfree_rcu(ctx, rcu);
1961 }
1962
1963 static void subflow_ulp_clone(const struct request_sock *req,
1964                               struct sock *newsk,
1965                               const gfp_t priority)
1966 {
1967         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
1968         struct mptcp_subflow_context *old_ctx = mptcp_subflow_ctx(newsk);
1969         struct mptcp_subflow_context *new_ctx;
1970
1971         if (!tcp_rsk(req)->is_mptcp ||
1972             (!subflow_req->mp_capable && !subflow_req->mp_join)) {
1973                 subflow_ulp_fallback(newsk, old_ctx);
1974                 return;
1975         }
1976
1977         new_ctx = subflow_create_ctx(newsk, priority);
1978         if (!new_ctx) {
1979                 subflow_ulp_fallback(newsk, old_ctx);
1980                 return;
1981         }
1982
1983         new_ctx->conn_finished = 1;
1984         new_ctx->icsk_af_ops = old_ctx->icsk_af_ops;
1985         new_ctx->tcp_state_change = old_ctx->tcp_state_change;
1986         new_ctx->tcp_error_report = old_ctx->tcp_error_report;
1987         new_ctx->rel_write_seq = 1;
1988         new_ctx->tcp_sock = newsk;
1989
1990         if (subflow_req->mp_capable) {
1991                 /* see comments in subflow_syn_recv_sock(), MPTCP connection
1992                  * is fully established only after we receive the remote key
1993                  */
1994                 new_ctx->mp_capable = 1;
1995                 new_ctx->local_key = subflow_req->local_key;
1996                 new_ctx->token = subflow_req->token;
1997                 new_ctx->ssn_offset = subflow_req->ssn_offset;
1998                 new_ctx->idsn = subflow_req->idsn;
1999
2000                 /* this is the first subflow, id is always 0 */
2001                 subflow_set_local_id(new_ctx, 0);
2002         } else if (subflow_req->mp_join) {
2003                 new_ctx->ssn_offset = subflow_req->ssn_offset;
2004                 new_ctx->mp_join = 1;
2005                 new_ctx->fully_established = 1;
2006                 new_ctx->remote_key_valid = 1;
2007                 new_ctx->backup = subflow_req->backup;
2008                 WRITE_ONCE(new_ctx->remote_id, subflow_req->remote_id);
2009                 new_ctx->token = subflow_req->token;
2010                 new_ctx->thmac = subflow_req->thmac;
2011
2012                 /* the subflow req id is valid, fetched via subflow_check_req()
2013                  * and subflow_token_join_request()
2014                  */
2015                 subflow_set_local_id(new_ctx, subflow_req->local_id);
2016         }
2017 }
2018
2019 static void tcp_release_cb_override(struct sock *ssk)
2020 {
2021         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(ssk);
2022         long status;
2023
2024         /* process and clear all the pending actions, but leave the subflow into
2025          * the napi queue. To respect locking, only the same CPU that originated
2026          * the action can touch the list. mptcp_napi_poll will take care of it.
2027          */
2028         status = set_mask_bits(&subflow->delegated_status, MPTCP_DELEGATE_ACTIONS_MASK, 0);
2029         if (status)
2030                 mptcp_subflow_process_delegated(ssk, status);
2031
2032         tcp_release_cb(ssk);
2033 }
2034
2035 static int tcp_abort_override(struct sock *ssk, int err)
2036 {
2037         /* closing a listener subflow requires a great deal of care.
2038          * keep it simple and just prevent such operation
2039          */
2040         if (inet_sk_state_load(ssk) == TCP_LISTEN)
2041                 return -EINVAL;
2042
2043         return tcp_abort(ssk, err);
2044 }
2045
2046 static struct tcp_ulp_ops subflow_ulp_ops __read_mostly = {
2047         .name           = "mptcp",
2048         .owner          = THIS_MODULE,
2049         .init           = subflow_ulp_init,
2050         .release        = subflow_ulp_release,
2051         .clone          = subflow_ulp_clone,
2052 };
2053
2054 static int subflow_ops_init(struct request_sock_ops *subflow_ops)
2055 {
2056         subflow_ops->obj_size = sizeof(struct mptcp_subflow_request_sock);
2057
2058         subflow_ops->slab = kmem_cache_create(subflow_ops->slab_name,
2059                                               subflow_ops->obj_size, 0,
2060                                               SLAB_ACCOUNT |
2061                                               SLAB_TYPESAFE_BY_RCU,
2062                                               NULL);
2063         if (!subflow_ops->slab)
2064                 return -ENOMEM;
2065
2066         return 0;
2067 }
2068
2069 void __init mptcp_subflow_init(void)
2070 {
2071         mptcp_subflow_v4_request_sock_ops = tcp_request_sock_ops;
2072         mptcp_subflow_v4_request_sock_ops.slab_name = "request_sock_subflow_v4";
2073         mptcp_subflow_v4_request_sock_ops.destructor = subflow_v4_req_destructor;
2074
2075         if (subflow_ops_init(&mptcp_subflow_v4_request_sock_ops) != 0)
2076                 panic("MPTCP: failed to init subflow v4 request sock ops\n");
2077
2078         subflow_request_sock_ipv4_ops = tcp_request_sock_ipv4_ops;
2079         subflow_request_sock_ipv4_ops.route_req = subflow_v4_route_req;
2080         subflow_request_sock_ipv4_ops.send_synack = subflow_v4_send_synack;
2081
2082         subflow_specific = ipv4_specific;
2083         subflow_specific.conn_request = subflow_v4_conn_request;
2084         subflow_specific.syn_recv_sock = subflow_syn_recv_sock;
2085         subflow_specific.sk_rx_dst_set = subflow_finish_connect;
2086         subflow_specific.rebuild_header = subflow_rebuild_header;
2087
2088         tcp_prot_override = tcp_prot;
2089         tcp_prot_override.release_cb = tcp_release_cb_override;
2090         tcp_prot_override.diag_destroy = tcp_abort_override;
2091
2092 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
2093         /* In struct mptcp_subflow_request_sock, we assume the TCP request sock
2094          * structures for v4 and v6 have the same size. It should not changed in
2095          * the future but better to make sure to be warned if it is no longer
2096          * the case.
2097          */
2098         BUILD_BUG_ON(sizeof(struct tcp_request_sock) != sizeof(struct tcp6_request_sock));
2099
2100         mptcp_subflow_v6_request_sock_ops = tcp6_request_sock_ops;
2101         mptcp_subflow_v6_request_sock_ops.slab_name = "request_sock_subflow_v6";
2102         mptcp_subflow_v6_request_sock_ops.destructor = subflow_v6_req_destructor;
2103
2104         if (subflow_ops_init(&mptcp_subflow_v6_request_sock_ops) != 0)
2105                 panic("MPTCP: failed to init subflow v6 request sock ops\n");
2106
2107         subflow_request_sock_ipv6_ops = tcp_request_sock_ipv6_ops;
2108         subflow_request_sock_ipv6_ops.route_req = subflow_v6_route_req;
2109         subflow_request_sock_ipv6_ops.send_synack = subflow_v6_send_synack;
2110
2111         subflow_v6_specific = ipv6_specific;
2112         subflow_v6_specific.conn_request = subflow_v6_conn_request;
2113         subflow_v6_specific.syn_recv_sock = subflow_syn_recv_sock;
2114         subflow_v6_specific.sk_rx_dst_set = subflow_finish_connect;
2115         subflow_v6_specific.rebuild_header = subflow_v6_rebuild_header;
2116
2117         subflow_v6m_specific = subflow_v6_specific;
2118         subflow_v6m_specific.queue_xmit = ipv4_specific.queue_xmit;
2119         subflow_v6m_specific.send_check = ipv4_specific.send_check;
2120         subflow_v6m_specific.net_header_len = ipv4_specific.net_header_len;
2121         subflow_v6m_specific.mtu_reduced = ipv4_specific.mtu_reduced;
2122         subflow_v6m_specific.rebuild_header = subflow_rebuild_header;
2123
2124         tcpv6_prot_override = tcpv6_prot;
2125         tcpv6_prot_override.release_cb = tcp_release_cb_override;
2126         tcpv6_prot_override.diag_destroy = tcp_abort_override;
2127 #endif
2128
2129         mptcp_diag_subflow_init(&subflow_ulp_ops);
2130
2131         if (tcp_register_ulp(&subflow_ulp_ops) != 0)
2132                 panic("MPTCP: failed to register subflows to ULP\n");
2133 }