net/mptcp/options.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 <crypto/sha2.h>
  11 #include <net/tcp.h>
  12 #include <net/mptcp.h>
  13 #include "protocol.h"
  14 #include "mib.h"
  15
  16 #include <trace/events/mptcp.h>
  17
  18 static bool mptcp_cap_flag_sha256(u8 flags)
  19 {
  20         return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256;
  21 }
  22
  23 static void mptcp_parse_option(const struct sk_buff *skb,
  24                                const unsigned char *ptr, int opsize,
  25                                struct mptcp_options_received *mp_opt)
  26 {
  27         u8 subtype = *ptr >> 4;
  28         int expected_opsize;
  29         u8 version;
  30         u8 flags;
  31         u8 i;
  32
  33         switch (subtype) {
  34         case MPTCPOPT_MP_CAPABLE:
  35                 /* strict size checking */
  36                 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
  37                         if (skb->len > tcp_hdr(skb)->doff << 2)
  38                                 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA;
  39                         else
  40                                 expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
  41                 } else {
  42                         if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)
  43                                 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
  44                         else
  45                                 expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
  46                 }
  47
  48                 /* Cfr RFC 8684 Section 3.3.0:
  49                  * If a checksum is present but its use had
  50                  * not been negotiated in the MP_CAPABLE handshake, the receiver MUST
  51                  * close the subflow with a RST, as it is not behaving as negotiated.
  52                  * If a checksum is not present when its use has been negotiated, the
  53                  * receiver MUST close the subflow with a RST, as it is considered
  54                  * broken
  55                  * We parse even option with mismatching csum presence, so that
  56                  * later in subflow_data_ready we can trigger the reset.
  57                  */
  58                 if (opsize != expected_opsize &&
  59                     (expected_opsize != TCPOLEN_MPTCP_MPC_ACK_DATA ||
  60                      opsize != TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM))
  61                         break;
  62
  63                 /* try to be gentle vs future versions on the initial syn */
  64                 version = *ptr++ & MPTCP_VERSION_MASK;
  65                 if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
  66                         if (version != MPTCP_SUPPORTED_VERSION)
  67                                 break;
  68                 } else if (version < MPTCP_SUPPORTED_VERSION) {
  69                         break;
  70                 }
  71
  72                 flags = *ptr++;
  73                 if (!mptcp_cap_flag_sha256(flags) ||
  74                     (flags & MPTCP_CAP_EXTENSIBILITY))
  75                         break;
  76
  77                 /* RFC 6824, Section 3.1:
  78                  * "For the Checksum Required bit (labeled "A"), if either
  79                  * host requires the use of checksums, checksums MUST be used.
  80                  * In other words, the only way for checksums not to be used
  81                  * is if both hosts in their SYNs set A=0."
  82                  */
  83                 if (flags & MPTCP_CAP_CHECKSUM_REQD)
  84                         mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
  85
  86                 mp_opt->deny_join_id0 = !!(flags & MPTCP_CAP_DENY_JOIN_ID0);
  87
  88                 mp_opt->suboptions |= OPTIONS_MPTCP_MPC;
  89                 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
  90                         mp_opt->sndr_key = get_unaligned_be64(ptr);
  91                         ptr += 8;
  92                 }
  93                 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
  94                         mp_opt->rcvr_key = get_unaligned_be64(ptr);
  95                         ptr += 8;
  96                 }
  97                 if (opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA) {
  98                         /* Section 3.1.:
  99                          * "the data parameters in a MP_CAPABLE are semantically
 100                          * equivalent to those in a DSS option and can be used
 101                          * interchangeably."
 102                          */
 103                         mp_opt->suboptions |= OPTION_MPTCP_DSS;
 104                         mp_opt->use_map = 1;
 105                         mp_opt->mpc_map = 1;
 106                         mp_opt->use_ack = 0;
 107                         mp_opt->data_len = get_unaligned_be16(ptr);
 108                         ptr += 2;
 109                 }
 110                 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM) {
 111                         mp_opt->csum = get_unaligned((__force __sum16 *)ptr);
 112                         mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
 113                         ptr += 2;
 114                 }
 115                 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d csum=%u",
 116                          version, flags, opsize, mp_opt->sndr_key,
 117                          mp_opt->rcvr_key, mp_opt->data_len, mp_opt->csum);
 118                 break;
 119
 120         case MPTCPOPT_MP_JOIN:
 121                 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
 122                         mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYN;
 123                         mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
 124                         mp_opt->join_id = *ptr++;
 125                         mp_opt->token = get_unaligned_be32(ptr);
 126                         ptr += 4;
 127                         mp_opt->nonce = get_unaligned_be32(ptr);
 128                         ptr += 4;
 129                         pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
 130                                  mp_opt->backup, mp_opt->join_id,
 131                                  mp_opt->token, mp_opt->nonce);
 132                 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
 133                         mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYNACK;
 134                         mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
 135                         mp_opt->join_id = *ptr++;
 136                         mp_opt->thmac = get_unaligned_be64(ptr);
 137                         ptr += 8;
 138                         mp_opt->nonce = get_unaligned_be32(ptr);
 139                         ptr += 4;
 140                         pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
 141                                  mp_opt->backup, mp_opt->join_id,
 142                                  mp_opt->thmac, mp_opt->nonce);
 143                 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
 144                         mp_opt->suboptions |= OPTION_MPTCP_MPJ_ACK;
 145                         ptr += 2;
 146                         memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
 147                         pr_debug("MP_JOIN hmac");
 148                 }
 149                 break;
 150
 151         case MPTCPOPT_DSS:
 152                 pr_debug("DSS");
 153                 ptr++;
 154
 155                 /* we must clear 'mpc_map' be able to detect MP_CAPABLE
 156                  * map vs DSS map in mptcp_incoming_options(), and reconstruct
 157                  * map info accordingly
 158                  */
 159                 mp_opt->mpc_map = 0;
 160                 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
 161                 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
 162                 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
 163                 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
 164                 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
 165                 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
 166
 167                 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
 168                          mp_opt->data_fin, mp_opt->dsn64,
 169                          mp_opt->use_map, mp_opt->ack64,
 170                          mp_opt->use_ack);
 171
 172                 expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
 173
 174                 if (mp_opt->use_ack) {
 175                         if (mp_opt->ack64)
 176                                 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
 177                         else
 178                                 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
 179                 }
 180
 181                 if (mp_opt->use_map) {
 182                         if (mp_opt->dsn64)
 183                                 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
 184                         else
 185                                 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
 186                 }
 187
 188                 /* Always parse any csum presence combination, we will enforce
 189                  * RFC 8684 Section 3.3.0 checks later in subflow_data_ready
 190                  */
 191                 if (opsize != expected_opsize &&
 192                     opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
 193                         break;
 194
 195                 mp_opt->suboptions |= OPTION_MPTCP_DSS;
 196                 if (mp_opt->use_ack) {
 197                         if (mp_opt->ack64) {
 198                                 mp_opt->data_ack = get_unaligned_be64(ptr);
 199                                 ptr += 8;
 200                         } else {
 201                                 mp_opt->data_ack = get_unaligned_be32(ptr);
 202                                 ptr += 4;
 203                         }
 204
 205                         pr_debug("data_ack=%llu", mp_opt->data_ack);
 206                 }
 207
 208                 if (mp_opt->use_map) {
 209                         if (mp_opt->dsn64) {
 210                                 mp_opt->data_seq = get_unaligned_be64(ptr);
 211                                 ptr += 8;
 212                         } else {
 213                                 mp_opt->data_seq = get_unaligned_be32(ptr);
 214                                 ptr += 4;
 215                         }
 216
 217                         mp_opt->subflow_seq = get_unaligned_be32(ptr);
 218                         ptr += 4;
 219
 220                         mp_opt->data_len = get_unaligned_be16(ptr);
 221                         ptr += 2;
 222
 223                         if (opsize == expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) {
 224                                 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
 225                                 mp_opt->csum = get_unaligned((__force __sum16 *)ptr);
 226                                 ptr += 2;
 227                         }
 228
 229                         pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
 230                                  mp_opt->data_seq, mp_opt->subflow_seq,
 231                                  mp_opt->data_len, !!(mp_opt->suboptions & OPTION_MPTCP_CSUMREQD),
 232                                  mp_opt->csum);
 233                 }
 234
 235                 break;
 236
 237         case MPTCPOPT_ADD_ADDR:
 238                 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
 239                 if (!mp_opt->echo) {
 240                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
 241                             opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
 242                                 mp_opt->addr.family = AF_INET;
 243 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 244                         else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
 245                                  opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
 246                                 mp_opt->addr.family = AF_INET6;
 247 #endif
 248                         else
 249                                 break;
 250                 } else {
 251                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
 252                             opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
 253                                 mp_opt->addr.family = AF_INET;
 254 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 255                         else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
 256                                  opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
 257                                 mp_opt->addr.family = AF_INET6;
 258 #endif
 259                         else
 260                                 break;
 261                 }
 262
 263                 mp_opt->suboptions |= OPTION_MPTCP_ADD_ADDR;
 264                 mp_opt->addr.id = *ptr++;
 265                 mp_opt->addr.port = 0;
 266                 mp_opt->ahmac = 0;
 267                 if (mp_opt->addr.family == AF_INET) {
 268                         memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4);
 269                         ptr += 4;
 270                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
 271                             opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
 272                                 mp_opt->addr.port = htons(get_unaligned_be16(ptr));
 273                                 ptr += 2;
 274                         }
 275                 }
 276 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 277                 else {
 278                         memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16);
 279                         ptr += 16;
 280                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
 281                             opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
 282                                 mp_opt->addr.port = htons(get_unaligned_be16(ptr));
 283                                 ptr += 2;
 284                         }
 285                 }
 286 #endif
 287                 if (!mp_opt->echo) {
 288                         mp_opt->ahmac = get_unaligned_be64(ptr);
 289                         ptr += 8;
 290                 }
 291                 pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d",
 292                          (mp_opt->addr.family == AF_INET6) ? "6" : "",
 293                          mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port));
 294                 break;
 295
 296         case MPTCPOPT_RM_ADDR:
 297                 if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 ||
 298                     opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX)
 299                         break;
 300
 301                 ptr++;
 302
 303                 mp_opt->suboptions |= OPTION_MPTCP_RM_ADDR;
 304                 mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE;
 305                 for (i = 0; i < mp_opt->rm_list.nr; i++)
 306                         mp_opt->rm_list.ids[i] = *ptr++;
 307                 pr_debug("RM_ADDR: rm_list_nr=%d", mp_opt->rm_list.nr);
 308                 break;
 309
 310         case MPTCPOPT_MP_PRIO:
 311                 if (opsize != TCPOLEN_MPTCP_PRIO)
 312                         break;
 313
 314                 mp_opt->suboptions |= OPTION_MPTCP_PRIO;
 315                 mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP;
 316                 pr_debug("MP_PRIO: prio=%d", mp_opt->backup);
 317                 break;
 318
 319         case MPTCPOPT_MP_FASTCLOSE:
 320                 if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
 321                         break;
 322
 323                 ptr += 2;
 324                 mp_opt->rcvr_key = get_unaligned_be64(ptr);
 325                 ptr += 8;
 326                 mp_opt->suboptions |= OPTION_MPTCP_FASTCLOSE;
 327                 pr_debug("MP_FASTCLOSE: recv_key=%llu", mp_opt->rcvr_key);
 328                 break;
 329
 330         case MPTCPOPT_RST:
 331                 if (opsize != TCPOLEN_MPTCP_RST)
 332                         break;
 333
 334                 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
 335                         break;
 336
 337                 mp_opt->suboptions |= OPTION_MPTCP_RST;
 338                 flags = *ptr++;
 339                 mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT;
 340                 mp_opt->reset_reason = *ptr;
 341                 pr_debug("MP_RST: transient=%u reason=%u",
 342                          mp_opt->reset_transient, mp_opt->reset_reason);
 343                 break;
 344
 345         case MPTCPOPT_MP_FAIL:
 346                 if (opsize != TCPOLEN_MPTCP_FAIL)
 347                         break;
 348
 349                 ptr += 2;
 350                 mp_opt->suboptions |= OPTION_MPTCP_FAIL;
 351                 mp_opt->fail_seq = get_unaligned_be64(ptr);
 352                 pr_debug("MP_FAIL: data_seq=%llu", mp_opt->fail_seq);
 353                 break;
 354
 355         default:
 356                 break;
 357         }
 358 }
 359
 360 void mptcp_get_options(const struct sk_buff *skb,
 361                        struct mptcp_options_received *mp_opt)
 362 {
 363         const struct tcphdr *th = tcp_hdr(skb);
 364         const unsigned char *ptr;
 365         int length;
 366
 367         /* initialize option status */
 368         mp_opt->suboptions = 0;
 369
 370         length = (th->doff * 4) - sizeof(struct tcphdr);
 371         ptr = (const unsigned char *)(th + 1);
 372
 373         while (length > 0) {
 374                 int opcode = *ptr++;
 375                 int opsize;
 376
 377                 switch (opcode) {
 378                 case TCPOPT_EOL:
 379                         return;
 380                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 381                         length--;
 382                         continue;
 383                 default:
 384                         if (length < 2)
 385                                 return;
 386                         opsize = *ptr++;
 387                         if (opsize < 2) /* "silly options" */
 388                                 return;
 389                         if (opsize > length)
 390                                 return; /* don't parse partial options */
 391                         if (opcode == TCPOPT_MPTCP)
 392                                 mptcp_parse_option(skb, ptr, opsize, mp_opt);
 393                         ptr += opsize - 2;
 394                         length -= opsize;
 395                 }
 396         }
 397 }
 398
 399 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
 400                        unsigned int *size, struct mptcp_out_options *opts)
 401 {
 402         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 403
 404         /* we will use snd_isn to detect first pkt [re]transmission
 405          * in mptcp_established_options_mp()
 406          */
 407         subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
 408         if (subflow->request_mptcp) {
 409                 opts->suboptions = OPTION_MPTCP_MPC_SYN;
 410                 opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk));
 411                 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
 412                 *size = TCPOLEN_MPTCP_MPC_SYN;
 413                 return true;
 414         } else if (subflow->request_join) {
 415                 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
 416                          subflow->local_nonce);
 417                 opts->suboptions = OPTION_MPTCP_MPJ_SYN;
 418                 opts->join_id = subflow->local_id;
 419                 opts->token = subflow->remote_token;
 420                 opts->nonce = subflow->local_nonce;
 421                 opts->backup = subflow->request_bkup;
 422                 *size = TCPOLEN_MPTCP_MPJ_SYN;
 423                 return true;
 424         }
 425         return false;
 426 }
 427
 428 static void clear_3rdack_retransmission(struct sock *sk)
 429 {
 430         struct inet_connection_sock *icsk = inet_csk(sk);
 431
 432         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 433         icsk->icsk_ack.timeout = 0;
 434         icsk->icsk_ack.ato = 0;
 435         icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
 436 }
 437
 438 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
 439                                          bool snd_data_fin_enable,
 440                                          unsigned int *size,
 441                                          unsigned int remaining,
 442                                          struct mptcp_out_options *opts)
 443 {
 444         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 445         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 446         struct mptcp_ext *mpext;
 447         unsigned int data_len;
 448         u8 len;
 449
 450         /* When skb is not available, we better over-estimate the emitted
 451          * options len. A full DSS option (28 bytes) is longer than
 452          * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
 453          * tell the caller to defer the estimate to
 454          * mptcp_established_options_dss(), which will reserve enough space.
 455          */
 456         if (!skb)
 457                 return false;
 458
 459         /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
 460         if (subflow->fully_established || snd_data_fin_enable ||
 461             subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
 462             sk->sk_state != TCP_ESTABLISHED)
 463                 return false;
 464
 465         if (subflow->mp_capable) {
 466                 mpext = mptcp_get_ext(skb);
 467                 data_len = mpext ? mpext->data_len : 0;
 468
 469                 /* we will check ops->data_len in mptcp_write_options() to
 470                  * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
 471                  * TCPOLEN_MPTCP_MPC_ACK
 472                  */
 473                 opts->data_len = data_len;
 474                 opts->suboptions = OPTION_MPTCP_MPC_ACK;
 475                 opts->sndr_key = subflow->local_key;
 476                 opts->rcvr_key = subflow->remote_key;
 477                 opts->csum_reqd = READ_ONCE(msk->csum_enabled);
 478                 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
 479
 480                 /* Section 3.1.
 481                  * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
 482                  * packets that start the first subflow of an MPTCP connection,
 483                  * as well as the first packet that carries data
 484                  */
 485                 if (data_len > 0) {
 486                         len = TCPOLEN_MPTCP_MPC_ACK_DATA;
 487                         if (opts->csum_reqd) {
 488                                 /* we need to propagate more info to csum the pseudo hdr */
 489                                 opts->data_seq = mpext->data_seq;
 490                                 opts->subflow_seq = mpext->subflow_seq;
 491                                 opts->csum = mpext->csum;
 492                                 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
 493                         }
 494                         *size = ALIGN(len, 4);
 495                 } else {
 496                         *size = TCPOLEN_MPTCP_MPC_ACK;
 497                 }
 498
 499                 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
 500                          subflow, subflow->local_key, subflow->remote_key,
 501                          data_len);
 502
 503                 return true;
 504         } else if (subflow->mp_join) {
 505                 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
 506                 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
 507                 *size = TCPOLEN_MPTCP_MPJ_ACK;
 508                 pr_debug("subflow=%p", subflow);
 509
 510                 /* we can use the full delegate action helper only from BH context
 511                  * If we are in process context - sk is flushing the backlog at
 512                  * socket lock release time - just set the appropriate flag, will
 513                  * be handled by the release callback
 514                  */
 515                 if (sock_owned_by_user(sk))
 516                         set_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status);
 517                 else
 518                         mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_ACK);
 519                 return true;
 520         }
 521         return false;
 522 }
 523
 524 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
 525                                  struct sk_buff *skb, struct mptcp_ext *ext)
 526 {
 527         /* The write_seq value has already been incremented, so the actual
 528          * sequence number for the DATA_FIN is one less.
 529          */
 530         u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
 531
 532         if (!ext->use_map || !skb->len) {
 533                 /* RFC6824 requires a DSS mapping with specific values
 534                  * if DATA_FIN is set but no data payload is mapped
 535                  */
 536                 ext->data_fin = 1;
 537                 ext->use_map = 1;
 538                 ext->dsn64 = 1;
 539                 ext->data_seq = data_fin_tx_seq;
 540                 ext->subflow_seq = 0;
 541                 ext->data_len = 1;
 542         } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
 543                 /* If there's an existing DSS mapping and it is the
 544                  * final mapping, DATA_FIN consumes 1 additional byte of
 545                  * mapping space.
 546                  */
 547                 ext->data_fin = 1;
 548                 ext->data_len++;
 549         }
 550 }
 551
 552 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
 553                                           bool snd_data_fin_enable,
 554                                           unsigned int *size,
 555                                           unsigned int remaining,
 556                                           struct mptcp_out_options *opts)
 557 {
 558         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 559         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 560         unsigned int dss_size = 0;
 561         struct mptcp_ext *mpext;
 562         unsigned int ack_size;
 563         bool ret = false;
 564         u64 ack_seq;
 565
 566         opts->csum_reqd = READ_ONCE(msk->csum_enabled);
 567         mpext = skb ? mptcp_get_ext(skb) : NULL;
 568
 569         if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
 570                 unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
 571
 572                 if (mpext) {
 573                         if (opts->csum_reqd)
 574                                 map_size += TCPOLEN_MPTCP_DSS_CHECKSUM;
 575
 576                         opts->ext_copy = *mpext;
 577                 }
 578
 579                 remaining -= map_size;
 580                 dss_size = map_size;
 581                 if (skb && snd_data_fin_enable)
 582                         mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
 583                 opts->suboptions = OPTION_MPTCP_DSS;
 584                 ret = true;
 585         }
 586
 587         /* passive sockets msk will set the 'can_ack' after accept(), even
 588          * if the first subflow may have the already the remote key handy
 589          */
 590         opts->ext_copy.use_ack = 0;
 591         if (!READ_ONCE(msk->can_ack)) {
 592                 *size = ALIGN(dss_size, 4);
 593                 return ret;
 594         }
 595
 596         ack_seq = READ_ONCE(msk->ack_seq);
 597         if (READ_ONCE(msk->use_64bit_ack)) {
 598                 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
 599                 opts->ext_copy.data_ack = ack_seq;
 600                 opts->ext_copy.ack64 = 1;
 601         } else {
 602                 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
 603                 opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
 604                 opts->ext_copy.ack64 = 0;
 605         }
 606         opts->ext_copy.use_ack = 1;
 607         opts->suboptions = OPTION_MPTCP_DSS;
 608         WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
 609
 610         /* Add kind/length/subtype/flag overhead if mapping is not populated */
 611         if (dss_size == 0)
 612                 ack_size += TCPOLEN_MPTCP_DSS_BASE;
 613
 614         dss_size += ack_size;
 615
 616         *size = ALIGN(dss_size, 4);
 617         return true;
 618 }
 619
 620 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
 621                                   struct mptcp_addr_info *addr)
 622 {
 623         u16 port = ntohs(addr->port);
 624         u8 hmac[SHA256_DIGEST_SIZE];
 625         u8 msg[19];
 626         int i = 0;
 627
 628         msg[i++] = addr->id;
 629         if (addr->family == AF_INET) {
 630                 memcpy(&msg[i], &addr->addr.s_addr, 4);
 631                 i += 4;
 632         }
 633 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 634         else if (addr->family == AF_INET6) {
 635                 memcpy(&msg[i], &addr->addr6.s6_addr, 16);
 636                 i += 16;
 637         }
 638 #endif
 639         msg[i++] = port >> 8;
 640         msg[i++] = port & 0xFF;
 641
 642         mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
 643
 644         return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
 645 }
 646
 647 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
 648                                                unsigned int *size,
 649                                                unsigned int remaining,
 650                                                struct mptcp_out_options *opts)
 651 {
 652         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 653         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 654         bool drop_other_suboptions = false;
 655         unsigned int opt_size = *size;
 656         bool echo;
 657         int len;
 658
 659         /* add addr will strip the existing options, be sure to avoid breaking
 660          * MPC/MPJ handshakes
 661          */
 662         if (!mptcp_pm_should_add_signal(msk) ||
 663             (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) ||
 664             !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, &opts->addr,
 665                     &echo, &drop_other_suboptions))
 666                 return false;
 667
 668         if (drop_other_suboptions)
 669                 remaining += opt_size;
 670         len = mptcp_add_addr_len(opts->addr.family, echo, !!opts->addr.port);
 671         if (remaining < len)
 672                 return false;
 673
 674         *size = len;
 675         if (drop_other_suboptions) {
 676                 pr_debug("drop other suboptions");
 677                 opts->suboptions = 0;
 678
 679                 /* note that e.g. DSS could have written into the memory
 680                  * aliased by ahmac, we must reset the field here
 681                  * to avoid appending the hmac even for ADD_ADDR echo
 682                  * options
 683                  */
 684                 opts->ahmac = 0;
 685                 *size -= opt_size;
 686         }
 687         opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
 688         if (!echo) {
 689                 opts->ahmac = add_addr_generate_hmac(msk->local_key,
 690                                                      msk->remote_key,
 691                                                      &opts->addr);
 692         }
 693         pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
 694                  opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
 695
 696         return true;
 697 }
 698
 699 static bool mptcp_established_options_rm_addr(struct sock *sk,
 700                                               unsigned int *size,
 701                                               unsigned int remaining,
 702                                               struct mptcp_out_options *opts)
 703 {
 704         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 705         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 706         struct mptcp_rm_list rm_list;
 707         int i, len;
 708
 709         if (!mptcp_pm_should_rm_signal(msk) ||
 710             !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
 711                 return false;
 712
 713         len = mptcp_rm_addr_len(&rm_list);
 714         if (len < 0)
 715                 return false;
 716         if (remaining < len)
 717                 return false;
 718
 719         *size = len;
 720         opts->suboptions |= OPTION_MPTCP_RM_ADDR;
 721         opts->rm_list = rm_list;
 722
 723         for (i = 0; i < opts->rm_list.nr; i++)
 724                 pr_debug("rm_list_ids[%d]=%d", i, opts->rm_list.ids[i]);
 725
 726         return true;
 727 }
 728
 729 static bool mptcp_established_options_mp_prio(struct sock *sk,
 730                                               unsigned int *size,
 731                                               unsigned int remaining,
 732                                               struct mptcp_out_options *opts)
 733 {
 734         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 735
 736         /* can't send MP_PRIO with MPC, as they share the same option space:
 737          * 'backup'. Also it makes no sense at all
 738          */
 739         if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC))
 740                 return false;
 741
 742         /* account for the trailing 'nop' option */
 743         if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
 744                 return false;
 745
 746         *size = TCPOLEN_MPTCP_PRIO_ALIGN;
 747         opts->suboptions |= OPTION_MPTCP_PRIO;
 748         opts->backup = subflow->request_bkup;
 749
 750         pr_debug("prio=%d", opts->backup);
 751
 752         return true;
 753 }
 754
 755 static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
 756                                                    unsigned int *size,
 757                                                    unsigned int remaining,
 758                                                    struct mptcp_out_options *opts)
 759 {
 760         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 761
 762         if (remaining < TCPOLEN_MPTCP_RST)
 763                 return false;
 764
 765         *size = TCPOLEN_MPTCP_RST;
 766         opts->suboptions |= OPTION_MPTCP_RST;
 767         opts->reset_transient = subflow->reset_transient;
 768         opts->reset_reason = subflow->reset_reason;
 769         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTTX);
 770
 771         return true;
 772 }
 773
 774 static bool mptcp_established_options_fastclose(struct sock *sk,
 775                                                 unsigned int *size,
 776                                                 unsigned int remaining,
 777                                                 struct mptcp_out_options *opts)
 778 {
 779         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 780         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 781
 782         if (likely(!subflow->send_fastclose))
 783                 return false;
 784
 785         if (remaining < TCPOLEN_MPTCP_FASTCLOSE)
 786                 return false;
 787
 788         *size = TCPOLEN_MPTCP_FASTCLOSE;
 789         opts->suboptions |= OPTION_MPTCP_FASTCLOSE;
 790         opts->rcvr_key = msk->remote_key;
 791
 792         pr_debug("FASTCLOSE key=%llu", opts->rcvr_key);
 793         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSETX);
 794         return true;
 795 }
 796
 797 static bool mptcp_established_options_mp_fail(struct sock *sk,
 798                                               unsigned int *size,
 799                                               unsigned int remaining,
 800                                               struct mptcp_out_options *opts)
 801 {
 802         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 803
 804         if (likely(!subflow->send_mp_fail))
 805                 return false;
 806
 807         if (remaining < TCPOLEN_MPTCP_FAIL)
 808                 return false;
 809
 810         *size = TCPOLEN_MPTCP_FAIL;
 811         opts->suboptions |= OPTION_MPTCP_FAIL;
 812         opts->fail_seq = subflow->map_seq;
 813
 814         pr_debug("MP_FAIL fail_seq=%llu", opts->fail_seq);
 815         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILTX);
 816
 817         return true;
 818 }
 819
 820 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
 821                                unsigned int *size, unsigned int remaining,
 822                                struct mptcp_out_options *opts)
 823 {
 824         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 825         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 826         unsigned int opt_size = 0;
 827         bool snd_data_fin;
 828         bool ret = false;
 829
 830         opts->suboptions = 0;
 831
 832         if (unlikely(__mptcp_check_fallback(msk) && !mptcp_check_infinite_map(skb)))
 833                 return false;
 834
 835         if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
 836                 if (mptcp_established_options_fastclose(sk, &opt_size, remaining, opts) ||
 837                     mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) {
 838                         *size += opt_size;
 839                         remaining -= opt_size;
 840                 }
 841                 /* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */
 842                 if (mptcp_established_options_rst(sk, skb, &opt_size, remaining, opts)) {
 843                         *size += opt_size;
 844                         remaining -= opt_size;
 845                 }
 846                 return true;
 847         }
 848
 849         snd_data_fin = mptcp_data_fin_enabled(msk);
 850         if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, remaining, opts))
 851                 ret = true;
 852         else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, remaining, opts)) {
 853                 unsigned int mp_fail_size;
 854
 855                 ret = true;
 856                 if (mptcp_established_options_mp_fail(sk, &mp_fail_size,
 857                                                       remaining - opt_size, opts)) {
 858                         *size += opt_size + mp_fail_size;
 859                         remaining -= opt_size - mp_fail_size;
 860                         return true;
 861                 }
 862         }
 863
 864         /* we reserved enough space for the above options, and exceeding the
 865          * TCP option space would be fatal
 866          */
 867         if (WARN_ON_ONCE(opt_size > remaining))
 868                 return false;
 869
 870         *size += opt_size;
 871         remaining -= opt_size;
 872         if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
 873                 *size += opt_size;
 874                 remaining -= opt_size;
 875                 ret = true;
 876         } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
 877                 *size += opt_size;
 878                 remaining -= opt_size;
 879                 ret = true;
 880         }
 881
 882         if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
 883                 *size += opt_size;
 884                 remaining -= opt_size;
 885                 ret = true;
 886         }
 887
 888         return ret;
 889 }
 890
 891 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
 892                           struct mptcp_out_options *opts)
 893 {
 894         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 895
 896         if (subflow_req->mp_capable) {
 897                 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
 898                 opts->sndr_key = subflow_req->local_key;
 899                 opts->csum_reqd = subflow_req->csum_reqd;
 900                 opts->allow_join_id0 = subflow_req->allow_join_id0;
 901                 *size = TCPOLEN_MPTCP_MPC_SYNACK;
 902                 pr_debug("subflow_req=%p, local_key=%llu",
 903                          subflow_req, subflow_req->local_key);
 904                 return true;
 905         } else if (subflow_req->mp_join) {
 906                 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
 907                 opts->backup = subflow_req->backup;
 908                 opts->join_id = subflow_req->local_id;
 909                 opts->thmac = subflow_req->thmac;
 910                 opts->nonce = subflow_req->local_nonce;
 911                 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
 912                          subflow_req, opts->backup, opts->join_id,
 913                          opts->thmac, opts->nonce);
 914                 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
 915                 return true;
 916         }
 917         return false;
 918 }
 919
 920 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
 921                                     struct mptcp_subflow_context *subflow,
 922                                     struct sk_buff *skb,
 923                                     struct mptcp_options_received *mp_opt)
 924 {
 925         /* here we can process OoO, in-window pkts, only in-sequence 4th ack
 926          * will make the subflow fully established
 927          */
 928         if (likely(subflow->fully_established)) {
 929                 /* on passive sockets, check for 3rd ack retransmission
 930                  * note that msk is always set by subflow_syn_recv_sock()
 931                  * for mp_join subflows
 932                  */
 933                 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
 934                     TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
 935                     subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) &&
 936                     !subflow->request_join)
 937                         tcp_send_ack(ssk);
 938                 goto fully_established;
 939         }
 940
 941         /* we must process OoO packets before the first subflow is fully
 942          * established. OoO packets are instead a protocol violation
 943          * for MP_JOIN subflows as the peer must not send any data
 944          * before receiving the forth ack - cfr. RFC 8684 section 3.2.
 945          */
 946         if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
 947                 if (subflow->mp_join)
 948                         goto reset;
 949                 return subflow->mp_capable;
 950         }
 951
 952         if (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) ||
 953             ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) && !mp_opt->echo)) {
 954                 /* subflows are fully established as soon as we get any
 955                  * additional ack, including ADD_ADDR.
 956                  */
 957                 subflow->fully_established = 1;
 958                 WRITE_ONCE(msk->fully_established, true);
 959                 goto fully_established;
 960         }
 961
 962         /* If the first established packet does not contain MP_CAPABLE + data
 963          * then fallback to TCP. Fallback scenarios requires a reset for
 964          * MP_JOIN subflows.
 965          */
 966         if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) {
 967                 if (subflow->mp_join)
 968                         goto reset;
 969                 subflow->mp_capable = 0;
 970                 pr_fallback(msk);
 971                 mptcp_do_fallback(ssk);
 972                 return false;
 973         }
 974
 975         if (mp_opt->deny_join_id0)
 976                 WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
 977
 978         if (unlikely(!READ_ONCE(msk->pm.server_side)))
 979                 pr_warn_once("bogus mpc option on established client sk");
 980         mptcp_subflow_fully_established(subflow, mp_opt);
 981
 982 fully_established:
 983         /* if the subflow is not already linked into the conn_list, we can't
 984          * notify the PM: this subflow is still on the listener queue
 985          * and the PM possibly acquiring the subflow lock could race with
 986          * the listener close
 987          */
 988         if (likely(subflow->pm_notified) || list_empty(&subflow->node))
 989                 return true;
 990
 991         subflow->pm_notified = 1;
 992         if (subflow->mp_join) {
 993                 clear_3rdack_retransmission(ssk);
 994                 mptcp_pm_subflow_established(msk);
 995         } else {
 996                 mptcp_pm_fully_established(msk, ssk, GFP_ATOMIC);
 997         }
 998         return true;
 999
1000 reset:
1001         mptcp_subflow_reset(ssk);
1002         return false;
1003 }
1004
1005 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq)
1006 {
1007         u32 old_seq32, cur_seq32;
1008
1009         old_seq32 = (u32)old_seq;
1010         cur_seq32 = (u32)cur_seq;
1011         cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32;
1012         if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32)))
1013                 return cur_seq + (1LL << 32);
1014
1015         /* reverse wrap could happen, too */
1016         if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32)))
1017                 return cur_seq - (1LL << 32);
1018         return cur_seq;
1019 }
1020
1021 static void ack_update_msk(struct mptcp_sock *msk,
1022                            struct sock *ssk,
1023                            struct mptcp_options_received *mp_opt)
1024 {
1025         u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
1026         struct sock *sk = (struct sock *)msk;
1027         u64 old_snd_una;
1028
1029         mptcp_data_lock(sk);
1030
1031         /* avoid ack expansion on update conflict, to reduce the risk of
1032          * wrongly expanding to a future ack sequence number, which is way
1033          * more dangerous than missing an ack
1034          */
1035         old_snd_una = msk->snd_una;
1036         new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
1037
1038         /* ACK for data not even sent yet? Ignore.*/
1039         if (unlikely(after64(new_snd_una, snd_nxt)))
1040                 new_snd_una = old_snd_una;
1041
1042         new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
1043
1044         if (after64(new_wnd_end, msk->wnd_end))
1045                 msk->wnd_end = new_wnd_end;
1046
1047         /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
1048         if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
1049                 __mptcp_check_push(sk, ssk);
1050
1051         if (after64(new_snd_una, old_snd_una)) {
1052                 msk->snd_una = new_snd_una;
1053                 __mptcp_data_acked(sk);
1054         }
1055         mptcp_data_unlock(sk);
1056
1057         trace_ack_update_msk(mp_opt->data_ack,
1058                              old_snd_una, new_snd_una,
1059                              new_wnd_end, msk->wnd_end);
1060 }
1061
1062 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
1063 {
1064         /* Skip if DATA_FIN was already received.
1065          * If updating simultaneously with the recvmsg loop, values
1066          * should match. If they mismatch, the peer is misbehaving and
1067          * we will prefer the most recent information.
1068          */
1069         if (READ_ONCE(msk->rcv_data_fin))
1070                 return false;
1071
1072         WRITE_ONCE(msk->rcv_data_fin_seq,
1073                    mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
1074         WRITE_ONCE(msk->rcv_data_fin, 1);
1075
1076         return true;
1077 }
1078
1079 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
1080                                 struct mptcp_options_received *mp_opt)
1081 {
1082         u64 hmac = 0;
1083
1084         if (mp_opt->echo)
1085                 return true;
1086
1087         hmac = add_addr_generate_hmac(msk->remote_key,
1088                                       msk->local_key,
1089                                       &mp_opt->addr);
1090
1091         pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
1092                  msk, hmac, mp_opt->ahmac);
1093
1094         return hmac == mp_opt->ahmac;
1095 }
1096
1097 /* Return false if a subflow has been reset, else return true */
1098 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
1099 {
1100         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1101         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1102         struct mptcp_options_received mp_opt;
1103         struct mptcp_ext *mpext;
1104
1105         if (__mptcp_check_fallback(msk)) {
1106                 /* Keep it simple and unconditionally trigger send data cleanup and
1107                  * pending queue spooling. We will need to acquire the data lock
1108                  * for more accurate checks, and once the lock is acquired, such
1109                  * helpers are cheap.
1110                  */
1111                 mptcp_data_lock(subflow->conn);
1112                 if (sk_stream_memory_free(sk))
1113                         __mptcp_check_push(subflow->conn, sk);
1114                 __mptcp_data_acked(subflow->conn);
1115                 mptcp_data_unlock(subflow->conn);
1116                 return true;
1117         }
1118
1119         mptcp_get_options(skb, &mp_opt);
1120
1121         /* The subflow can be in close state only if check_fully_established()
1122          * just sent a reset. If so, tell the caller to ignore the current packet.
1123          */
1124         if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1125                 return sk->sk_state != TCP_CLOSE;
1126
1127         if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) {
1128                 if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) &&
1129                     msk->local_key == mp_opt.rcvr_key) {
1130                         WRITE_ONCE(msk->rcv_fastclose, true);
1131                         mptcp_schedule_work((struct sock *)msk);
1132                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSERX);
1133                 }
1134
1135                 if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) &&
1136                     add_addr_hmac_valid(msk, &mp_opt)) {
1137                         if (!mp_opt.echo) {
1138                                 mptcp_pm_add_addr_received(sk, &mp_opt.addr);
1139                                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1140                         } else {
1141                                 mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1142                                 mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1143                                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1144                         }
1145
1146                         if (mp_opt.addr.port)
1147                                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1148                 }
1149
1150                 if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR)
1151                         mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1152
1153                 if (mp_opt.suboptions & OPTION_MPTCP_PRIO) {
1154                         mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1155                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1156                 }
1157
1158                 if (mp_opt.suboptions & OPTION_MPTCP_FAIL) {
1159                         mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq);
1160                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX);
1161                 }
1162
1163                 if (mp_opt.suboptions & OPTION_MPTCP_RST) {
1164                         subflow->reset_seen = 1;
1165                         subflow->reset_reason = mp_opt.reset_reason;
1166                         subflow->reset_transient = mp_opt.reset_transient;
1167                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTRX);
1168                 }
1169
1170                 if (!(mp_opt.suboptions & OPTION_MPTCP_DSS))
1171                         return true;
1172         }
1173
1174         /* we can't wait for recvmsg() to update the ack_seq, otherwise
1175          * monodirectional flows will stuck
1176          */
1177         if (mp_opt.use_ack)
1178                 ack_update_msk(msk, sk, &mp_opt);
1179
1180         /* Zero-data-length packets are dropped by the caller and not
1181          * propagated to the MPTCP layer, so the skb extension does not
1182          * need to be allocated or populated. DATA_FIN information, if
1183          * present, needs to be updated here before the skb is freed.
1184          */
1185         if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1186                 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1187                     mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64))
1188                         mptcp_schedule_work((struct sock *)msk);
1189
1190                 return true;
1191         }
1192
1193         mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1194         if (!mpext)
1195                 return true;
1196
1197         memset(mpext, 0, sizeof(*mpext));
1198
1199         if (likely(mp_opt.use_map)) {
1200                 if (mp_opt.mpc_map) {
1201                         /* this is an MP_CAPABLE carrying MPTCP data
1202                          * we know this map the first chunk of data
1203                          */
1204                         mptcp_crypto_key_sha(subflow->remote_key, NULL,
1205                                              &mpext->data_seq);
1206                         mpext->data_seq++;
1207                         mpext->subflow_seq = 1;
1208                         mpext->dsn64 = 1;
1209                         mpext->mpc_map = 1;
1210                         mpext->data_fin = 0;
1211                 } else {
1212                         mpext->data_seq = mp_opt.data_seq;
1213                         mpext->subflow_seq = mp_opt.subflow_seq;
1214                         mpext->dsn64 = mp_opt.dsn64;
1215                         mpext->data_fin = mp_opt.data_fin;
1216                 }
1217                 mpext->data_len = mp_opt.data_len;
1218                 mpext->use_map = 1;
1219                 mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD);
1220
1221                 if (mpext->csum_reqd)
1222                         mpext->csum = mp_opt.csum;
1223         }
1224
1225         return true;
1226 }
1227
1228 static void mptcp_set_rwin(struct tcp_sock *tp, struct tcphdr *th)
1229 {
1230         const struct sock *ssk = (const struct sock *)tp;
1231         struct mptcp_subflow_context *subflow;
1232         u64 ack_seq, rcv_wnd_old, rcv_wnd_new;
1233         struct mptcp_sock *msk;
1234         u32 new_win;
1235         u64 win;
1236
1237         subflow = mptcp_subflow_ctx(ssk);
1238         msk = mptcp_sk(subflow->conn);
1239
1240         ack_seq = READ_ONCE(msk->ack_seq);
1241         rcv_wnd_new = ack_seq + tp->rcv_wnd;
1242
1243         rcv_wnd_old = atomic64_read(&msk->rcv_wnd_sent);
1244         if (after64(rcv_wnd_new, rcv_wnd_old)) {
1245                 u64 rcv_wnd;
1246
1247                 for (;;) {
1248                         rcv_wnd = atomic64_cmpxchg(&msk->rcv_wnd_sent, rcv_wnd_old, rcv_wnd_new);
1249
1250                         if (rcv_wnd == rcv_wnd_old)
1251                                 break;
1252
1253                         rcv_wnd_old = rcv_wnd;
1254                         if (before64(rcv_wnd_new, rcv_wnd_old)) {
1255                                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICTUPDATE);
1256                                 goto raise_win;
1257                         }
1258                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICT);
1259                 }
1260                 return;
1261         }
1262
1263         if (rcv_wnd_new != rcv_wnd_old) {
1264 raise_win:
1265                 win = rcv_wnd_old - ack_seq;
1266                 tp->rcv_wnd = min_t(u64, win, U32_MAX);
1267                 new_win = tp->rcv_wnd;
1268
1269                 /* Make sure we do not exceed the maximum possible
1270                  * scaled window.
1271                  */
1272                 if (unlikely(th->syn))
1273                         new_win = min(new_win, 65535U) << tp->rx_opt.rcv_wscale;
1274                 if (!tp->rx_opt.rcv_wscale &&
1275                     READ_ONCE(sock_net(ssk)->ipv4.sysctl_tcp_workaround_signed_windows))
1276                         new_win = min(new_win, MAX_TCP_WINDOW);
1277                 else
1278                         new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale));
1279
1280                 /* RFC1323 scaling applied */
1281                 new_win >>= tp->rx_opt.rcv_wscale;
1282                 th->window = htons(new_win);
1283                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDSHARED);
1284         }
1285 }
1286
1287 __sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum)
1288 {
1289         struct csum_pseudo_header header;
1290         __wsum csum;
1291
1292         /* cfr RFC 8684 3.3.1.:
1293          * the data sequence number used in the pseudo-header is
1294          * always the 64-bit value, irrespective of what length is used in the
1295          * DSS option itself.
1296          */
1297         header.data_seq = cpu_to_be64(data_seq);
1298         header.subflow_seq = htonl(subflow_seq);
1299         header.data_len = htons(data_len);
1300         header.csum = 0;
1301
1302         csum = csum_partial(&header, sizeof(header), sum);
1303         return csum_fold(csum);
1304 }
1305
1306 static __sum16 mptcp_make_csum(const struct mptcp_ext *mpext)
1307 {
1308         return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1309                                  ~csum_unfold(mpext->csum));
1310 }
1311
1312 static void put_len_csum(u16 len, __sum16 csum, void *data)
1313 {
1314         __sum16 *sumptr = data + 2;
1315         __be16 *ptr = data;
1316
1317         put_unaligned_be16(len, ptr);
1318
1319         put_unaligned(csum, sumptr);
1320 }
1321
1322 void mptcp_write_options(struct tcphdr *th, __be32 *ptr, struct tcp_sock *tp,
1323                          struct mptcp_out_options *opts)
1324 {
1325         const struct sock *ssk = (const struct sock *)tp;
1326         struct mptcp_subflow_context *subflow;
1327
1328         /* Which options can be used together?
1329          *
1330          * X: mutually exclusive
1331          * O: often used together
1332          * C: can be used together in some cases
1333          * P: could be used together but we prefer not to (optimisations)
1334          *
1335          *  Opt: | MPC  | MPJ  | DSS  | ADD  |  RM  | PRIO | FAIL |  FC  |
1336          * ------|------|------|------|------|------|------|------|------|
1337          *  MPC  |------|------|------|------|------|------|------|------|
1338          *  MPJ  |  X   |------|------|------|------|------|------|------|
1339          *  DSS  |  X   |  X   |------|------|------|------|------|------|
1340          *  ADD  |  X   |  X   |  P   |------|------|------|------|------|
1341          *  RM   |  C   |  C   |  C   |  P   |------|------|------|------|
1342          *  PRIO |  X   |  C   |  C   |  C   |  C   |------|------|------|
1343          *  FAIL |  X   |  X   |  C   |  X   |  X   |  X   |------|------|
1344          *  FC   |  X   |  X   |  X   |  X   |  X   |  X   |  X   |------|
1345          *  RST  |  X   |  X   |  X   |  X   |  X   |  X   |  O   |  O   |
1346          * ------|------|------|------|------|------|------|------|------|
1347          *
1348          * The same applies in mptcp_established_options() function.
1349          */
1350         if (likely(OPTION_MPTCP_DSS & opts->suboptions)) {
1351                 struct mptcp_ext *mpext = &opts->ext_copy;
1352                 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1353                 u8 flags = 0;
1354
1355                 if (mpext->use_ack) {
1356                         flags = MPTCP_DSS_HAS_ACK;
1357                         if (mpext->ack64) {
1358                                 len += TCPOLEN_MPTCP_DSS_ACK64;
1359                                 flags |= MPTCP_DSS_ACK64;
1360                         } else {
1361                                 len += TCPOLEN_MPTCP_DSS_ACK32;
1362                         }
1363                 }
1364
1365                 if (mpext->use_map) {
1366                         len += TCPOLEN_MPTCP_DSS_MAP64;
1367
1368                         /* Use only 64-bit mapping flags for now, add
1369                          * support for optional 32-bit mappings later.
1370                          */
1371                         flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1372                         if (mpext->data_fin)
1373                                 flags |= MPTCP_DSS_DATA_FIN;
1374
1375                         if (opts->csum_reqd)
1376                                 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1377                 }
1378
1379                 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1380
1381                 if (mpext->use_ack) {
1382                         if (mpext->ack64) {
1383                                 put_unaligned_be64(mpext->data_ack, ptr);
1384                                 ptr += 2;
1385                         } else {
1386                                 put_unaligned_be32(mpext->data_ack32, ptr);
1387                                 ptr += 1;
1388                         }
1389                 }
1390
1391                 if (mpext->use_map) {
1392                         put_unaligned_be64(mpext->data_seq, ptr);
1393                         ptr += 2;
1394                         put_unaligned_be32(mpext->subflow_seq, ptr);
1395                         ptr += 1;
1396                         if (opts->csum_reqd) {
1397                                 /* data_len == 0 is reserved for the infinite mapping,
1398                                  * the checksum will also be set to 0.
1399                                  */
1400                                 put_len_csum(mpext->data_len,
1401                                              (mpext->data_len ? mptcp_make_csum(mpext) : 0),
1402                                              ptr);
1403                         } else {
1404                                 put_unaligned_be32(mpext->data_len << 16 |
1405                                                    TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1406                         }
1407                         ptr += 1;
1408                 }
1409
1410                 /* We might need to add MP_FAIL options in rare cases */
1411                 if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions))
1412                         goto mp_fail;
1413         } else if (OPTIONS_MPTCP_MPC & opts->suboptions) {
1414                 u8 len, flag = MPTCP_CAP_HMAC_SHA256;
1415
1416                 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) {
1417                         len = TCPOLEN_MPTCP_MPC_SYN;
1418                 } else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) {
1419                         len = TCPOLEN_MPTCP_MPC_SYNACK;
1420                 } else if (opts->data_len) {
1421                         len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1422                         if (opts->csum_reqd)
1423                                 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1424                 } else {
1425                         len = TCPOLEN_MPTCP_MPC_ACK;
1426                 }
1427
1428                 if (opts->csum_reqd)
1429                         flag |= MPTCP_CAP_CHECKSUM_REQD;
1430
1431                 if (!opts->allow_join_id0)
1432                         flag |= MPTCP_CAP_DENY_JOIN_ID0;
1433
1434                 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1435                                       MPTCP_SUPPORTED_VERSION,
1436                                       flag);
1437
1438                 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1439                     opts->suboptions))
1440                         goto mp_capable_done;
1441
1442                 put_unaligned_be64(opts->sndr_key, ptr);
1443                 ptr += 2;
1444                 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1445                         goto mp_capable_done;
1446
1447                 put_unaligned_be64(opts->rcvr_key, ptr);
1448                 ptr += 2;
1449                 if (!opts->data_len)
1450                         goto mp_capable_done;
1451
1452                 if (opts->csum_reqd) {
1453                         put_len_csum(opts->data_len,
1454                                      __mptcp_make_csum(opts->data_seq,
1455                                                        opts->subflow_seq,
1456                                                        opts->data_len,
1457                                                        ~csum_unfold(opts->csum)),
1458                                      ptr);
1459                 } else {
1460                         put_unaligned_be32(opts->data_len << 16 |
1461                                            TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1462                 }
1463                 ptr += 1;
1464
1465                 /* MPC is additionally mutually exclusive with MP_PRIO */
1466                 goto mp_capable_done;
1467         } else if (OPTIONS_MPTCP_MPJ & opts->suboptions) {
1468                 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1469                         *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1470                                               TCPOLEN_MPTCP_MPJ_SYN,
1471                                               opts->backup, opts->join_id);
1472                         put_unaligned_be32(opts->token, ptr);
1473                         ptr += 1;
1474                         put_unaligned_be32(opts->nonce, ptr);
1475                         ptr += 1;
1476                 } else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1477                         *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1478                                               TCPOLEN_MPTCP_MPJ_SYNACK,
1479                                               opts->backup, opts->join_id);
1480                         put_unaligned_be64(opts->thmac, ptr);
1481                         ptr += 2;
1482                         put_unaligned_be32(opts->nonce, ptr);
1483                         ptr += 1;
1484                 } else {
1485                         *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1486                                               TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1487                         memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1488                         ptr += 5;
1489                 }
1490         } else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1491                 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1492                 u8 echo = MPTCP_ADDR_ECHO;
1493
1494 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1495                 if (opts->addr.family == AF_INET6)
1496                         len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1497 #endif
1498
1499                 if (opts->addr.port)
1500                         len += TCPOLEN_MPTCP_PORT_LEN;
1501
1502                 if (opts->ahmac) {
1503                         len += sizeof(opts->ahmac);
1504                         echo = 0;
1505                 }
1506
1507                 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1508                                       len, echo, opts->addr.id);
1509                 if (opts->addr.family == AF_INET) {
1510                         memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1511                         ptr += 1;
1512                 }
1513 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1514                 else if (opts->addr.family == AF_INET6) {
1515                         memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1516                         ptr += 4;
1517                 }
1518 #endif
1519
1520                 if (!opts->addr.port) {
1521                         if (opts->ahmac) {
1522                                 put_unaligned_be64(opts->ahmac, ptr);
1523                                 ptr += 2;
1524                         }
1525                 } else {
1526                         u16 port = ntohs(opts->addr.port);
1527
1528                         if (opts->ahmac) {
1529                                 u8 *bptr = (u8 *)ptr;
1530
1531                                 put_unaligned_be16(port, bptr);
1532                                 bptr += 2;
1533                                 put_unaligned_be64(opts->ahmac, bptr);
1534                                 bptr += 8;
1535                                 put_unaligned_be16(TCPOPT_NOP << 8 |
1536                                                    TCPOPT_NOP, bptr);
1537
1538                                 ptr += 3;
1539                         } else {
1540                                 put_unaligned_be32(port << 16 |
1541                                                    TCPOPT_NOP << 8 |
1542                                                    TCPOPT_NOP, ptr);
1543                                 ptr += 1;
1544                         }
1545                 }
1546         } else if (unlikely(OPTION_MPTCP_FASTCLOSE & opts->suboptions)) {
1547                 /* FASTCLOSE is mutually exclusive with others except RST */
1548                 *ptr++ = mptcp_option(MPTCPOPT_MP_FASTCLOSE,
1549                                       TCPOLEN_MPTCP_FASTCLOSE,
1550                                       0, 0);
1551                 put_unaligned_be64(opts->rcvr_key, ptr);
1552                 ptr += 2;
1553
1554                 if (OPTION_MPTCP_RST & opts->suboptions)
1555                         goto mp_rst;
1556                 return;
1557         } else if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) {
1558 mp_fail:
1559                 /* MP_FAIL is mutually exclusive with others except RST */
1560                 subflow = mptcp_subflow_ctx(ssk);
1561                 subflow->send_mp_fail = 0;
1562
1563                 *ptr++ = mptcp_option(MPTCPOPT_MP_FAIL,
1564                                       TCPOLEN_MPTCP_FAIL,
1565                                       0, 0);
1566                 put_unaligned_be64(opts->fail_seq, ptr);
1567                 ptr += 2;
1568
1569                 if (OPTION_MPTCP_RST & opts->suboptions)
1570                         goto mp_rst;
1571                 return;
1572         } else if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) {
1573 mp_rst:
1574                 *ptr++ = mptcp_option(MPTCPOPT_RST,
1575                                       TCPOLEN_MPTCP_RST,
1576                                       opts->reset_transient,
1577                                       opts->reset_reason);
1578                 return;
1579         }
1580
1581         if (OPTION_MPTCP_PRIO & opts->suboptions) {
1582                 subflow = mptcp_subflow_ctx(ssk);
1583                 subflow->send_mp_prio = 0;
1584
1585                 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1586                                       TCPOLEN_MPTCP_PRIO,
1587                                       opts->backup, TCPOPT_NOP);
1588
1589                 MPTCP_INC_STATS(sock_net((const struct sock *)tp),
1590                                 MPTCP_MIB_MPPRIOTX);
1591         }
1592
1593 mp_capable_done:
1594         if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1595                 u8 i = 1;
1596
1597                 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1598                                       TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1599                                       0, opts->rm_list.ids[0]);
1600
1601                 while (i < opts->rm_list.nr) {
1602                         u8 id1, id2, id3, id4;
1603
1604                         id1 = opts->rm_list.ids[i];
1605                         id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1606                         id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1607                         id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1608                         put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1609                         ptr += 1;
1610                         i += 4;
1611                 }
1612         }
1613
1614         if (tp)
1615                 mptcp_set_rwin(tp, th);
1616 }
1617
1618 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1619 {
1620         const struct mptcp_ext *ext = mptcp_get_ext(skb);
1621         u8 flags, reason;
1622
1623         if (ext) {
1624                 flags = ext->reset_transient;
1625                 reason = ext->reset_reason;
1626
1627                 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1628                                     flags, reason);
1629         }
1630
1631         return htonl(0u);
1632 }
1633 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);