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 static bool mptcp_cap_flag_sha256(u8 flags)
  17 {
  18         return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256;
  19 }
  20
  21 static void mptcp_parse_option(const struct sk_buff *skb,
  22                                const unsigned char *ptr, int opsize,
  23                                struct mptcp_options_received *mp_opt)
  24 {
  25         u8 subtype = *ptr >> 4;
  26         int expected_opsize;
  27         u8 version;
  28         u8 flags;
  29
  30         switch (subtype) {
  31         case MPTCPOPT_MP_CAPABLE:
  32                 /* strict size checking */
  33                 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
  34                         if (skb->len > tcp_hdr(skb)->doff << 2)
  35                                 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA;
  36                         else
  37                                 expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
  38                 } else {
  39                         if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)
  40                                 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
  41                         else
  42                                 expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
  43                 }
  44                 if (opsize != expected_opsize)
  45                         break;
  46
  47                 /* try to be gentle vs future versions on the initial syn */
  48                 version = *ptr++ & MPTCP_VERSION_MASK;
  49                 if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
  50                         if (version != MPTCP_SUPPORTED_VERSION)
  51                                 break;
  52                 } else if (version < MPTCP_SUPPORTED_VERSION) {
  53                         break;
  54                 }
  55
  56                 flags = *ptr++;
  57                 if (!mptcp_cap_flag_sha256(flags) ||
  58                     (flags & MPTCP_CAP_EXTENSIBILITY))
  59                         break;
  60
  61                 /* RFC 6824, Section 3.1:
  62                  * "For the Checksum Required bit (labeled "A"), if either
  63                  * host requires the use of checksums, checksums MUST be used.
  64                  * In other words, the only way for checksums not to be used
  65                  * is if both hosts in their SYNs set A=0."
  66                  *
  67                  * Section 3.3.0:
  68                  * "If a checksum is not present when its use has been
  69                  * negotiated, the receiver MUST close the subflow with a RST as
  70                  * it is considered broken."
  71                  *
  72                  * We don't implement DSS checksum - fall back to TCP.
  73                  */
  74                 if (flags & MPTCP_CAP_CHECKSUM_REQD)
  75                         break;
  76
  77                 mp_opt->mp_capable = 1;
  78                 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
  79                         mp_opt->sndr_key = get_unaligned_be64(ptr);
  80                         ptr += 8;
  81                 }
  82                 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
  83                         mp_opt->rcvr_key = get_unaligned_be64(ptr);
  84                         ptr += 8;
  85                 }
  86                 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA) {
  87                         /* Section 3.1.:
  88                          * "the data parameters in a MP_CAPABLE are semantically
  89                          * equivalent to those in a DSS option and can be used
  90                          * interchangeably."
  91                          */
  92                         mp_opt->dss = 1;
  93                         mp_opt->use_map = 1;
  94                         mp_opt->mpc_map = 1;
  95                         mp_opt->data_len = get_unaligned_be16(ptr);
  96                         ptr += 2;
  97                 }
  98                 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d",
  99                          version, flags, opsize, mp_opt->sndr_key,
 100                          mp_opt->rcvr_key, mp_opt->data_len);
 101                 break;
 102
 103         case MPTCPOPT_MP_JOIN:
 104                 mp_opt->mp_join = 1;
 105                 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
 106                         mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
 107                         mp_opt->join_id = *ptr++;
 108                         mp_opt->token = get_unaligned_be32(ptr);
 109                         ptr += 4;
 110                         mp_opt->nonce = get_unaligned_be32(ptr);
 111                         ptr += 4;
 112                         pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
 113                                  mp_opt->backup, mp_opt->join_id,
 114                                  mp_opt->token, mp_opt->nonce);
 115                 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
 116                         mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
 117                         mp_opt->join_id = *ptr++;
 118                         mp_opt->thmac = get_unaligned_be64(ptr);
 119                         ptr += 8;
 120                         mp_opt->nonce = get_unaligned_be32(ptr);
 121                         ptr += 4;
 122                         pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
 123                                  mp_opt->backup, mp_opt->join_id,
 124                                  mp_opt->thmac, mp_opt->nonce);
 125                 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
 126                         ptr += 2;
 127                         memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
 128                         pr_debug("MP_JOIN hmac");
 129                 } else {
 130                         pr_warn("MP_JOIN bad option size");
 131                         mp_opt->mp_join = 0;
 132                 }
 133                 break;
 134
 135         case MPTCPOPT_DSS:
 136                 pr_debug("DSS");
 137                 ptr++;
 138
 139                 /* we must clear 'mpc_map' be able to detect MP_CAPABLE
 140                  * map vs DSS map in mptcp_incoming_options(), and reconstruct
 141                  * map info accordingly
 142                  */
 143                 mp_opt->mpc_map = 0;
 144                 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
 145                 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
 146                 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
 147                 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
 148                 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
 149                 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
 150
 151                 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
 152                          mp_opt->data_fin, mp_opt->dsn64,
 153                          mp_opt->use_map, mp_opt->ack64,
 154                          mp_opt->use_ack);
 155
 156                 expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
 157
 158                 if (mp_opt->use_ack) {
 159                         if (mp_opt->ack64)
 160                                 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
 161                         else
 162                                 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
 163                 }
 164
 165                 if (mp_opt->use_map) {
 166                         if (mp_opt->dsn64)
 167                                 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
 168                         else
 169                                 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
 170                 }
 171
 172                 /* RFC 6824, Section 3.3:
 173                  * If a checksum is present, but its use had
 174                  * not been negotiated in the MP_CAPABLE handshake,
 175                  * the checksum field MUST be ignored.
 176                  */
 177                 if (opsize != expected_opsize &&
 178                     opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
 179                         break;
 180
 181                 mp_opt->dss = 1;
 182
 183                 if (mp_opt->use_ack) {
 184                         if (mp_opt->ack64) {
 185                                 mp_opt->data_ack = get_unaligned_be64(ptr);
 186                                 ptr += 8;
 187                         } else {
 188                                 mp_opt->data_ack = get_unaligned_be32(ptr);
 189                                 ptr += 4;
 190                         }
 191
 192                         pr_debug("data_ack=%llu", mp_opt->data_ack);
 193                 }
 194
 195                 if (mp_opt->use_map) {
 196                         if (mp_opt->dsn64) {
 197                                 mp_opt->data_seq = get_unaligned_be64(ptr);
 198                                 ptr += 8;
 199                         } else {
 200                                 mp_opt->data_seq = get_unaligned_be32(ptr);
 201                                 ptr += 4;
 202                         }
 203
 204                         mp_opt->subflow_seq = get_unaligned_be32(ptr);
 205                         ptr += 4;
 206
 207                         mp_opt->data_len = get_unaligned_be16(ptr);
 208                         ptr += 2;
 209
 210                         pr_debug("data_seq=%llu subflow_seq=%u data_len=%u",
 211                                  mp_opt->data_seq, mp_opt->subflow_seq,
 212                                  mp_opt->data_len);
 213                 }
 214
 215                 break;
 216
 217         case MPTCPOPT_ADD_ADDR:
 218                 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
 219                 if (!mp_opt->echo) {
 220                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
 221                             opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
 222                                 mp_opt->family = MPTCP_ADDR_IPVERSION_4;
 223 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 224                         else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
 225                                  opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
 226                                 mp_opt->family = MPTCP_ADDR_IPVERSION_6;
 227 #endif
 228                         else
 229                                 break;
 230                 } else {
 231                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
 232                             opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
 233                                 mp_opt->family = MPTCP_ADDR_IPVERSION_4;
 234 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 235                         else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
 236                                  opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
 237                                 mp_opt->family = MPTCP_ADDR_IPVERSION_6;
 238 #endif
 239                         else
 240                                 break;
 241                 }
 242
 243                 mp_opt->add_addr = 1;
 244                 mp_opt->addr_id = *ptr++;
 245                 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4) {
 246                         memcpy((u8 *)&mp_opt->addr.s_addr, (u8 *)ptr, 4);
 247                         ptr += 4;
 248                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
 249                             opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
 250                                 mp_opt->port = get_unaligned_be16(ptr);
 251                                 ptr += 2;
 252                         }
 253                 }
 254 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 255                 else {
 256                         memcpy(mp_opt->addr6.s6_addr, (u8 *)ptr, 16);
 257                         ptr += 16;
 258                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
 259                             opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
 260                                 mp_opt->port = get_unaligned_be16(ptr);
 261                                 ptr += 2;
 262                         }
 263                 }
 264 #endif
 265                 if (!mp_opt->echo) {
 266                         mp_opt->ahmac = get_unaligned_be64(ptr);
 267                         ptr += 8;
 268                 }
 269                 pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d",
 270                          (mp_opt->family == MPTCP_ADDR_IPVERSION_6) ? "6" : "",
 271                          mp_opt->addr_id, mp_opt->ahmac, mp_opt->echo, mp_opt->port);
 272                 break;
 273
 274         case MPTCPOPT_RM_ADDR:
 275                 if (opsize != TCPOLEN_MPTCP_RM_ADDR_BASE)
 276                         break;
 277
 278                 ptr++;
 279
 280                 mp_opt->rm_addr = 1;
 281                 mp_opt->rm_id = *ptr++;
 282                 pr_debug("RM_ADDR: id=%d", mp_opt->rm_id);
 283                 break;
 284
 285         case MPTCPOPT_MP_FASTCLOSE:
 286                 if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
 287                         break;
 288
 289                 ptr += 2;
 290                 mp_opt->rcvr_key = get_unaligned_be64(ptr);
 291                 ptr += 8;
 292                 mp_opt->fastclose = 1;
 293                 break;
 294
 295         default:
 296                 break;
 297         }
 298 }
 299
 300 void mptcp_get_options(const struct sk_buff *skb,
 301                        struct mptcp_options_received *mp_opt)
 302 {
 303         const struct tcphdr *th = tcp_hdr(skb);
 304         const unsigned char *ptr;
 305         int length;
 306
 307         /* initialize option status */
 308         mp_opt->mp_capable = 0;
 309         mp_opt->mp_join = 0;
 310         mp_opt->add_addr = 0;
 311         mp_opt->ahmac = 0;
 312         mp_opt->fastclose = 0;
 313         mp_opt->port = 0;
 314         mp_opt->rm_addr = 0;
 315         mp_opt->dss = 0;
 316
 317         length = (th->doff * 4) - sizeof(struct tcphdr);
 318         ptr = (const unsigned char *)(th + 1);
 319
 320         while (length > 0) {
 321                 int opcode = *ptr++;
 322                 int opsize;
 323
 324                 switch (opcode) {
 325                 case TCPOPT_EOL:
 326                         return;
 327                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 328                         length--;
 329                         continue;
 330                 default:
 331                         opsize = *ptr++;
 332                         if (opsize < 2) /* "silly options" */
 333                                 return;
 334                         if (opsize > length)
 335                                 return; /* don't parse partial options */
 336                         if (opcode == TCPOPT_MPTCP)
 337                                 mptcp_parse_option(skb, ptr, opsize, mp_opt);
 338                         ptr += opsize - 2;
 339                         length -= opsize;
 340                 }
 341         }
 342 }
 343
 344 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
 345                        unsigned int *size, struct mptcp_out_options *opts)
 346 {
 347         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 348
 349         /* we will use snd_isn to detect first pkt [re]transmission
 350          * in mptcp_established_options_mp()
 351          */
 352         subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
 353         if (subflow->request_mptcp) {
 354                 opts->suboptions = OPTION_MPTCP_MPC_SYN;
 355                 *size = TCPOLEN_MPTCP_MPC_SYN;
 356                 return true;
 357         } else if (subflow->request_join) {
 358                 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
 359                          subflow->local_nonce);
 360                 opts->suboptions = OPTION_MPTCP_MPJ_SYN;
 361                 opts->join_id = subflow->local_id;
 362                 opts->token = subflow->remote_token;
 363                 opts->nonce = subflow->local_nonce;
 364                 opts->backup = subflow->request_bkup;
 365                 *size = TCPOLEN_MPTCP_MPJ_SYN;
 366                 return true;
 367         }
 368         return false;
 369 }
 370
 371 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
 372  * TCP can't schedule delack timer before the subflow is fully established.
 373  * MPTCP uses the delack timer to do 3rd ack retransmissions
 374  */
 375 static void schedule_3rdack_retransmission(struct sock *sk)
 376 {
 377         struct inet_connection_sock *icsk = inet_csk(sk);
 378         struct tcp_sock *tp = tcp_sk(sk);
 379         unsigned long timeout;
 380
 381         /* reschedule with a timeout above RTT, as we must look only for drop */
 382         if (tp->srtt_us)
 383                 timeout = tp->srtt_us << 1;
 384         else
 385                 timeout = TCP_TIMEOUT_INIT;
 386
 387         WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
 388         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 389         icsk->icsk_ack.timeout = timeout;
 390         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 391 }
 392
 393 static void clear_3rdack_retransmission(struct sock *sk)
 394 {
 395         struct inet_connection_sock *icsk = inet_csk(sk);
 396
 397         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 398         icsk->icsk_ack.timeout = 0;
 399         icsk->icsk_ack.ato = 0;
 400         icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
 401 }
 402
 403 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
 404                                          unsigned int *size,
 405                                          unsigned int remaining,
 406                                          struct mptcp_out_options *opts)
 407 {
 408         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 409         struct mptcp_ext *mpext;
 410         unsigned int data_len;
 411
 412         /* When skb is not available, we better over-estimate the emitted
 413          * options len. A full DSS option (28 bytes) is longer than
 414          * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
 415          * tell the caller to defer the estimate to
 416          * mptcp_established_options_dss(), which will reserve enough space.
 417          */
 418         if (!skb)
 419                 return false;
 420
 421         /* MPC/MPJ needed only on 3rd ack packet */
 422         if (subflow->fully_established ||
 423             subflow->snd_isn != TCP_SKB_CB(skb)->seq)
 424                 return false;
 425
 426         if (subflow->mp_capable) {
 427                 mpext = mptcp_get_ext(skb);
 428                 data_len = mpext ? mpext->data_len : 0;
 429
 430                 /* we will check ext_copy.data_len in mptcp_write_options() to
 431                  * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
 432                  * TCPOLEN_MPTCP_MPC_ACK
 433                  */
 434                 opts->ext_copy.data_len = data_len;
 435                 opts->suboptions = OPTION_MPTCP_MPC_ACK;
 436                 opts->sndr_key = subflow->local_key;
 437                 opts->rcvr_key = subflow->remote_key;
 438
 439                 /* Section 3.1.
 440                  * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
 441                  * packets that start the first subflow of an MPTCP connection,
 442                  * as well as the first packet that carries data
 443                  */
 444                 if (data_len > 0)
 445                         *size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4);
 446                 else
 447                         *size = TCPOLEN_MPTCP_MPC_ACK;
 448
 449                 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
 450                          subflow, subflow->local_key, subflow->remote_key,
 451                          data_len);
 452
 453                 return true;
 454         } else if (subflow->mp_join) {
 455                 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
 456                 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
 457                 *size = TCPOLEN_MPTCP_MPJ_ACK;
 458                 pr_debug("subflow=%p", subflow);
 459
 460                 schedule_3rdack_retransmission(sk);
 461                 return true;
 462         }
 463         return false;
 464 }
 465
 466 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
 467                                  struct sk_buff *skb, struct mptcp_ext *ext)
 468 {
 469         /* The write_seq value has already been incremented, so the actual
 470          * sequence number for the DATA_FIN is one less.
 471          */
 472         u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
 473
 474         if (!ext->use_map || !skb->len) {
 475                 /* RFC6824 requires a DSS mapping with specific values
 476                  * if DATA_FIN is set but no data payload is mapped
 477                  */
 478                 ext->data_fin = 1;
 479                 ext->use_map = 1;
 480                 ext->dsn64 = 1;
 481                 ext->data_seq = data_fin_tx_seq;
 482                 ext->subflow_seq = 0;
 483                 ext->data_len = 1;
 484         } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
 485                 /* If there's an existing DSS mapping and it is the
 486                  * final mapping, DATA_FIN consumes 1 additional byte of
 487                  * mapping space.
 488                  */
 489                 ext->data_fin = 1;
 490                 ext->data_len++;
 491         }
 492 }
 493
 494 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
 495                                           unsigned int *size,
 496                                           unsigned int remaining,
 497                                           struct mptcp_out_options *opts)
 498 {
 499         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 500         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 501         u64 snd_data_fin_enable, ack_seq;
 502         unsigned int dss_size = 0;
 503         struct mptcp_ext *mpext;
 504         unsigned int ack_size;
 505         bool ret = false;
 506
 507         mpext = skb ? mptcp_get_ext(skb) : NULL;
 508         snd_data_fin_enable = mptcp_data_fin_enabled(msk);
 509
 510         if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
 511                 unsigned int map_size;
 512
 513                 map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
 514
 515                 remaining -= map_size;
 516                 dss_size = map_size;
 517                 if (mpext)
 518                         opts->ext_copy = *mpext;
 519
 520                 if (skb && snd_data_fin_enable)
 521                         mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
 522                 ret = true;
 523         }
 524
 525         /* passive sockets msk will set the 'can_ack' after accept(), even
 526          * if the first subflow may have the already the remote key handy
 527          */
 528         opts->ext_copy.use_ack = 0;
 529         if (!READ_ONCE(msk->can_ack)) {
 530                 *size = ALIGN(dss_size, 4);
 531                 return ret;
 532         }
 533
 534         ack_seq = READ_ONCE(msk->ack_seq);
 535         if (READ_ONCE(msk->use_64bit_ack)) {
 536                 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
 537                 opts->ext_copy.data_ack = ack_seq;
 538                 opts->ext_copy.ack64 = 1;
 539         } else {
 540                 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
 541                 opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
 542                 opts->ext_copy.ack64 = 0;
 543         }
 544         opts->ext_copy.use_ack = 1;
 545         WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
 546
 547         /* Add kind/length/subtype/flag overhead if mapping is not populated */
 548         if (dss_size == 0)
 549                 ack_size += TCPOLEN_MPTCP_DSS_BASE;
 550
 551         dss_size += ack_size;
 552
 553         *size = ALIGN(dss_size, 4);
 554         return true;
 555 }
 556
 557 static u64 add_addr_generate_hmac(u64 key1, u64 key2, u8 addr_id,
 558                                   struct in_addr *addr)
 559 {
 560         u8 hmac[SHA256_DIGEST_SIZE];
 561         u8 msg[7];
 562
 563         msg[0] = addr_id;
 564         memcpy(&msg[1], &addr->s_addr, 4);
 565         msg[5] = 0;
 566         msg[6] = 0;
 567
 568         mptcp_crypto_hmac_sha(key1, key2, msg, 7, hmac);
 569
 570         return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
 571 }
 572
 573 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 574 static u64 add_addr6_generate_hmac(u64 key1, u64 key2, u8 addr_id,
 575                                    struct in6_addr *addr)
 576 {
 577         u8 hmac[SHA256_DIGEST_SIZE];
 578         u8 msg[19];
 579
 580         msg[0] = addr_id;
 581         memcpy(&msg[1], &addr->s6_addr, 16);
 582         msg[17] = 0;
 583         msg[18] = 0;
 584
 585         mptcp_crypto_hmac_sha(key1, key2, msg, 19, hmac);
 586
 587         return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
 588 }
 589 #endif
 590
 591 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
 592                                                unsigned int *size,
 593                                                unsigned int remaining,
 594                                                struct mptcp_out_options *opts)
 595 {
 596         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 597         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 598         bool drop_other_suboptions = false;
 599         unsigned int opt_size = *size;
 600         struct mptcp_addr_info saddr;
 601         bool echo;
 602         bool port;
 603         int len;
 604
 605         if ((mptcp_pm_should_add_signal_ipv6(msk) ||
 606              mptcp_pm_should_add_signal_port(msk)) &&
 607             skb && skb_is_tcp_pure_ack(skb)) {
 608                 pr_debug("drop other suboptions");
 609                 opts->suboptions = 0;
 610                 opts->ext_copy.use_ack = 0;
 611                 opts->ext_copy.use_map = 0;
 612                 remaining += opt_size;
 613                 drop_other_suboptions = true;
 614         }
 615
 616         if (!mptcp_pm_should_add_signal(msk) ||
 617             !(mptcp_pm_add_addr_signal(msk, remaining, &saddr, &echo, &port)))
 618                 return false;
 619
 620         len = mptcp_add_addr_len(saddr.family, echo, port);
 621         if (remaining < len)
 622                 return false;
 623
 624         *size = len;
 625         if (drop_other_suboptions)
 626                 *size -= opt_size;
 627         opts->addr_id = saddr.id;
 628         if (port)
 629                 opts->port = ntohs(saddr.port);
 630         if (saddr.family == AF_INET) {
 631                 opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
 632                 opts->addr = saddr.addr;
 633                 if (!echo) {
 634                         opts->ahmac = add_addr_generate_hmac(msk->local_key,
 635                                                              msk->remote_key,
 636                                                              opts->addr_id,
 637                                                              &opts->addr);
 638                 }
 639         }
 640 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 641         else if (saddr.family == AF_INET6) {
 642                 opts->suboptions |= OPTION_MPTCP_ADD_ADDR6;
 643                 opts->addr6 = saddr.addr6;
 644                 if (!echo) {
 645                         opts->ahmac = add_addr6_generate_hmac(msk->local_key,
 646                                                               msk->remote_key,
 647                                                               opts->addr_id,
 648                                                               &opts->addr6);
 649                 }
 650         }
 651 #endif
 652         pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
 653                  opts->addr_id, opts->ahmac, echo, opts->port);
 654
 655         return true;
 656 }
 657
 658 static bool mptcp_established_options_rm_addr(struct sock *sk,
 659                                               unsigned int *size,
 660                                               unsigned int remaining,
 661                                               struct mptcp_out_options *opts)
 662 {
 663         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 664         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 665         u8 rm_id;
 666
 667         if (!mptcp_pm_should_rm_signal(msk) ||
 668             !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_id)))
 669                 return false;
 670
 671         if (remaining < TCPOLEN_MPTCP_RM_ADDR_BASE)
 672                 return false;
 673
 674         *size = TCPOLEN_MPTCP_RM_ADDR_BASE;
 675         opts->suboptions |= OPTION_MPTCP_RM_ADDR;
 676         opts->rm_id = rm_id;
 677
 678         pr_debug("rm_id=%d", opts->rm_id);
 679
 680         return true;
 681 }
 682
 683 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
 684                                unsigned int *size, unsigned int remaining,
 685                                struct mptcp_out_options *opts)
 686 {
 687         unsigned int opt_size = 0;
 688         bool ret = false;
 689
 690         opts->suboptions = 0;
 691
 692         if (unlikely(mptcp_check_fallback(sk)))
 693                 return false;
 694
 695         /* prevent adding of any MPTCP related options on reset packet
 696          * until we support MP_TCPRST/MP_FASTCLOSE
 697          */
 698         if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
 699                 return false;
 700
 701         if (mptcp_established_options_mp(sk, skb, &opt_size, remaining, opts))
 702                 ret = true;
 703         else if (mptcp_established_options_dss(sk, skb, &opt_size, remaining,
 704                                                opts))
 705                 ret = true;
 706
 707         /* we reserved enough space for the above options, and exceeding the
 708          * TCP option space would be fatal
 709          */
 710         if (WARN_ON_ONCE(opt_size > remaining))
 711                 return false;
 712
 713         *size += opt_size;
 714         remaining -= opt_size;
 715         if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
 716                 *size += opt_size;
 717                 remaining -= opt_size;
 718                 ret = true;
 719         } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
 720                 *size += opt_size;
 721                 remaining -= opt_size;
 722                 ret = true;
 723         }
 724
 725         return ret;
 726 }
 727
 728 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
 729                           struct mptcp_out_options *opts)
 730 {
 731         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 732
 733         if (subflow_req->mp_capable) {
 734                 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
 735                 opts->sndr_key = subflow_req->local_key;
 736                 *size = TCPOLEN_MPTCP_MPC_SYNACK;
 737                 pr_debug("subflow_req=%p, local_key=%llu",
 738                          subflow_req, subflow_req->local_key);
 739                 return true;
 740         } else if (subflow_req->mp_join) {
 741                 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
 742                 opts->backup = subflow_req->backup;
 743                 opts->join_id = subflow_req->local_id;
 744                 opts->thmac = subflow_req->thmac;
 745                 opts->nonce = subflow_req->local_nonce;
 746                 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
 747                          subflow_req, opts->backup, opts->join_id,
 748                          opts->thmac, opts->nonce);
 749                 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
 750                 return true;
 751         }
 752         return false;
 753 }
 754
 755 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
 756                                     struct mptcp_subflow_context *subflow,
 757                                     struct sk_buff *skb,
 758                                     struct mptcp_options_received *mp_opt)
 759 {
 760         /* here we can process OoO, in-window pkts, only in-sequence 4th ack
 761          * will make the subflow fully established
 762          */
 763         if (likely(subflow->fully_established)) {
 764                 /* on passive sockets, check for 3rd ack retransmission
 765                  * note that msk is always set by subflow_syn_recv_sock()
 766                  * for mp_join subflows
 767                  */
 768                 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
 769                     TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
 770                     subflow->mp_join && mp_opt->mp_join &&
 771                     READ_ONCE(msk->pm.server_side))
 772                         tcp_send_ack(ssk);
 773                 goto fully_established;
 774         }
 775
 776         /* we must process OoO packets before the first subflow is fully
 777          * established. OoO packets are instead a protocol violation
 778          * for MP_JOIN subflows as the peer must not send any data
 779          * before receiving the forth ack - cfr. RFC 8684 section 3.2.
 780          */
 781         if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
 782                 if (subflow->mp_join)
 783                         goto reset;
 784                 return subflow->mp_capable;
 785         }
 786
 787         if (mp_opt->dss && mp_opt->use_ack) {
 788                 /* subflows are fully established as soon as we get any
 789                  * additional ack.
 790                  */
 791                 subflow->fully_established = 1;
 792                 WRITE_ONCE(msk->fully_established, true);
 793                 goto fully_established;
 794         }
 795
 796         if (mp_opt->add_addr) {
 797                 WRITE_ONCE(msk->fully_established, true);
 798                 return true;
 799         }
 800
 801         /* If the first established packet does not contain MP_CAPABLE + data
 802          * then fallback to TCP. Fallback scenarios requires a reset for
 803          * MP_JOIN subflows.
 804          */
 805         if (!mp_opt->mp_capable) {
 806                 if (subflow->mp_join)
 807                         goto reset;
 808                 subflow->mp_capable = 0;
 809                 pr_fallback(msk);
 810                 __mptcp_do_fallback(msk);
 811                 return false;
 812         }
 813
 814         if (unlikely(!READ_ONCE(msk->pm.server_side)))
 815                 pr_warn_once("bogus mpc option on established client sk");
 816         mptcp_subflow_fully_established(subflow, mp_opt);
 817
 818 fully_established:
 819         /* if the subflow is not already linked into the conn_list, we can't
 820          * notify the PM: this subflow is still on the listener queue
 821          * and the PM possibly acquiring the subflow lock could race with
 822          * the listener close
 823          */
 824         if (likely(subflow->pm_notified) || list_empty(&subflow->node))
 825                 return true;
 826
 827         subflow->pm_notified = 1;
 828         if (subflow->mp_join) {
 829                 clear_3rdack_retransmission(ssk);
 830                 mptcp_pm_subflow_established(msk, subflow);
 831         } else {
 832                 mptcp_pm_fully_established(msk);
 833         }
 834         return true;
 835
 836 reset:
 837         mptcp_subflow_reset(ssk);
 838         return false;
 839 }
 840
 841 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit)
 842 {
 843         u32 old_ack32, cur_ack32;
 844
 845         if (use_64bit)
 846                 return cur_ack;
 847
 848         old_ack32 = (u32)old_ack;
 849         cur_ack32 = (u32)cur_ack;
 850         cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32;
 851         if (unlikely(before(cur_ack32, old_ack32)))
 852                 return cur_ack + (1LL << 32);
 853         return cur_ack;
 854 }
 855
 856 static void ack_update_msk(struct mptcp_sock *msk,
 857                            struct sock *ssk,
 858                            struct mptcp_options_received *mp_opt)
 859 {
 860         u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
 861         struct sock *sk = (struct sock *)msk;
 862         u64 old_snd_una;
 863
 864         mptcp_data_lock(sk);
 865
 866         /* avoid ack expansion on update conflict, to reduce the risk of
 867          * wrongly expanding to a future ack sequence number, which is way
 868          * more dangerous than missing an ack
 869          */
 870         old_snd_una = msk->snd_una;
 871         new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
 872
 873         /* ACK for data not even sent yet? Ignore. */
 874         if (after64(new_snd_una, snd_nxt))
 875                 new_snd_una = old_snd_una;
 876
 877         new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
 878
 879         if (after64(new_wnd_end, msk->wnd_end))
 880                 msk->wnd_end = new_wnd_end;
 881
 882         /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
 883         if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)) &&
 884             sk_stream_memory_free(ssk))
 885                 __mptcp_check_push(sk, ssk);
 886
 887         if (after64(new_snd_una, old_snd_una)) {
 888                 msk->snd_una = new_snd_una;
 889                 __mptcp_data_acked(sk);
 890         }
 891         mptcp_data_unlock(sk);
 892 }
 893
 894 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
 895 {
 896         /* Skip if DATA_FIN was already received.
 897          * If updating simultaneously with the recvmsg loop, values
 898          * should match. If they mismatch, the peer is misbehaving and
 899          * we will prefer the most recent information.
 900          */
 901         if (READ_ONCE(msk->rcv_data_fin) || !READ_ONCE(msk->first))
 902                 return false;
 903
 904         WRITE_ONCE(msk->rcv_data_fin_seq,
 905                    expand_ack(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
 906         WRITE_ONCE(msk->rcv_data_fin, 1);
 907
 908         return true;
 909 }
 910
 911 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
 912                                 struct mptcp_options_received *mp_opt)
 913 {
 914         u64 hmac = 0;
 915
 916         if (mp_opt->echo)
 917                 return true;
 918
 919         if (mp_opt->family == MPTCP_ADDR_IPVERSION_4)
 920                 hmac = add_addr_generate_hmac(msk->remote_key,
 921                                               msk->local_key,
 922                                               mp_opt->addr_id, &mp_opt->addr);
 923 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 924         else
 925                 hmac = add_addr6_generate_hmac(msk->remote_key,
 926                                                msk->local_key,
 927                                                mp_opt->addr_id, &mp_opt->addr6);
 928 #endif
 929
 930         pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
 931                  msk, (unsigned long long)hmac,
 932                  (unsigned long long)mp_opt->ahmac);
 933
 934         return hmac == mp_opt->ahmac;
 935 }
 936
 937 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
 938 {
 939         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 940         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 941         struct mptcp_options_received mp_opt;
 942         struct mptcp_ext *mpext;
 943
 944         if (__mptcp_check_fallback(msk)) {
 945                 /* Keep it simple and unconditionally trigger send data cleanup and
 946                  * pending queue spooling. We will need to acquire the data lock
 947                  * for more accurate checks, and once the lock is acquired, such
 948                  * helpers are cheap.
 949                  */
 950                 mptcp_data_lock(subflow->conn);
 951                 if (sk_stream_memory_free(sk))
 952                         __mptcp_check_push(subflow->conn, sk);
 953                 __mptcp_data_acked(subflow->conn);
 954                 mptcp_data_unlock(subflow->conn);
 955                 return;
 956         }
 957
 958         mptcp_get_options(skb, &mp_opt);
 959         if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
 960                 return;
 961
 962         if (mp_opt.fastclose &&
 963             msk->local_key == mp_opt.rcvr_key) {
 964                 WRITE_ONCE(msk->rcv_fastclose, true);
 965                 mptcp_schedule_work((struct sock *)msk);
 966         }
 967
 968         if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) {
 969                 struct mptcp_addr_info addr;
 970
 971                 addr.port = htons(mp_opt.port);
 972                 addr.id = mp_opt.addr_id;
 973                 if (mp_opt.family == MPTCP_ADDR_IPVERSION_4) {
 974                         addr.family = AF_INET;
 975                         addr.addr = mp_opt.addr;
 976                 }
 977 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 978                 else if (mp_opt.family == MPTCP_ADDR_IPVERSION_6) {
 979                         addr.family = AF_INET6;
 980                         addr.addr6 = mp_opt.addr6;
 981                 }
 982 #endif
 983                 if (!mp_opt.echo) {
 984                         mptcp_pm_add_addr_received(msk, &addr);
 985                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
 986                 } else {
 987                         mptcp_pm_del_add_timer(msk, &addr);
 988                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
 989                 }
 990                 mp_opt.add_addr = 0;
 991         }
 992
 993         if (mp_opt.rm_addr) {
 994                 mptcp_pm_rm_addr_received(msk, mp_opt.rm_id);
 995                 mp_opt.rm_addr = 0;
 996         }
 997
 998         if (!mp_opt.dss)
 999                 return;
1000
1001         /* we can't wait for recvmsg() to update the ack_seq, otherwise
1002          * monodirectional flows will stuck
1003          */
1004         if (mp_opt.use_ack)
1005                 ack_update_msk(msk, sk, &mp_opt);
1006
1007         /* Zero-data-length packets are dropped by the caller and not
1008          * propagated to the MPTCP layer, so the skb extension does not
1009          * need to be allocated or populated. DATA_FIN information, if
1010          * present, needs to be updated here before the skb is freed.
1011          */
1012         if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1013                 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1014                     mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
1015                     schedule_work(&msk->work))
1016                         sock_hold(subflow->conn);
1017
1018                 return;
1019         }
1020
1021         mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1022         if (!mpext)
1023                 return;
1024
1025         memset(mpext, 0, sizeof(*mpext));
1026
1027         if (mp_opt.use_map) {
1028                 if (mp_opt.mpc_map) {
1029                         /* this is an MP_CAPABLE carrying MPTCP data
1030                          * we know this map the first chunk of data
1031                          */
1032                         mptcp_crypto_key_sha(subflow->remote_key, NULL,
1033                                              &mpext->data_seq);
1034                         mpext->data_seq++;
1035                         mpext->subflow_seq = 1;
1036                         mpext->dsn64 = 1;
1037                         mpext->mpc_map = 1;
1038                         mpext->data_fin = 0;
1039                 } else {
1040                         mpext->data_seq = mp_opt.data_seq;
1041                         mpext->subflow_seq = mp_opt.subflow_seq;
1042                         mpext->dsn64 = mp_opt.dsn64;
1043                         mpext->data_fin = mp_opt.data_fin;
1044                 }
1045                 mpext->data_len = mp_opt.data_len;
1046                 mpext->use_map = 1;
1047         }
1048 }
1049
1050 static void mptcp_set_rwin(const struct tcp_sock *tp)
1051 {
1052         const struct sock *ssk = (const struct sock *)tp;
1053         const struct mptcp_subflow_context *subflow;
1054         struct mptcp_sock *msk;
1055         u64 ack_seq;
1056
1057         subflow = mptcp_subflow_ctx(ssk);
1058         msk = mptcp_sk(subflow->conn);
1059
1060         ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd;
1061
1062         if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent)))
1063                 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
1064 }
1065
1066 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,
1067                          struct mptcp_out_options *opts)
1068 {
1069         if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
1070              OPTION_MPTCP_MPC_ACK) & opts->suboptions) {
1071                 u8 len;
1072
1073                 if (OPTION_MPTCP_MPC_SYN & opts->suboptions)
1074                         len = TCPOLEN_MPTCP_MPC_SYN;
1075                 else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions)
1076                         len = TCPOLEN_MPTCP_MPC_SYNACK;
1077                 else if (opts->ext_copy.data_len)
1078                         len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1079                 else
1080                         len = TCPOLEN_MPTCP_MPC_ACK;
1081
1082                 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1083                                       MPTCP_SUPPORTED_VERSION,
1084                                       MPTCP_CAP_HMAC_SHA256);
1085
1086                 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1087                     opts->suboptions))
1088                         goto mp_capable_done;
1089
1090                 put_unaligned_be64(opts->sndr_key, ptr);
1091                 ptr += 2;
1092                 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1093                         goto mp_capable_done;
1094
1095                 put_unaligned_be64(opts->rcvr_key, ptr);
1096                 ptr += 2;
1097                 if (!opts->ext_copy.data_len)
1098                         goto mp_capable_done;
1099
1100                 put_unaligned_be32(opts->ext_copy.data_len << 16 |
1101                                    TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1102                 ptr += 1;
1103         }
1104
1105 mp_capable_done:
1106         if ((OPTION_MPTCP_ADD_ADDR
1107 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1108              | OPTION_MPTCP_ADD_ADDR6
1109 #endif
1110             ) & opts->suboptions) {
1111                 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1112                 u8 echo = MPTCP_ADDR_ECHO;
1113
1114 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1115                 if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions)
1116                         len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1117 #endif
1118
1119                 if (opts->port)
1120                         len += TCPOLEN_MPTCP_PORT_LEN;
1121
1122                 if (opts->ahmac) {
1123                         len += sizeof(opts->ahmac);
1124                         echo = 0;
1125                 }
1126
1127                 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1128                                       len, echo, opts->addr_id);
1129                 if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1130                         memcpy((u8 *)ptr, (u8 *)&opts->addr.s_addr, 4);
1131                         ptr += 1;
1132                 }
1133 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1134                 else if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions) {
1135                         memcpy((u8 *)ptr, opts->addr6.s6_addr, 16);
1136                         ptr += 4;
1137                 }
1138 #endif
1139
1140                 if (!opts->port) {
1141                         if (opts->ahmac) {
1142                                 put_unaligned_be64(opts->ahmac, ptr);
1143                                 ptr += 2;
1144                         }
1145                 } else {
1146                         if (opts->ahmac) {
1147                                 u8 *bptr = (u8 *)ptr;
1148
1149                                 put_unaligned_be16(opts->port, bptr);
1150                                 bptr += 2;
1151                                 put_unaligned_be64(opts->ahmac, bptr);
1152                                 bptr += 8;
1153                                 put_unaligned_be16(TCPOPT_NOP << 8 |
1154                                                    TCPOPT_NOP, bptr);
1155
1156                                 ptr += 3;
1157                         } else {
1158                                 put_unaligned_be32(opts->port << 16 |
1159                                                    TCPOPT_NOP << 8 |
1160                                                    TCPOPT_NOP, ptr);
1161                                 ptr += 1;
1162                         }
1163                 }
1164         }
1165
1166         if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1167                 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1168                                       TCPOLEN_MPTCP_RM_ADDR_BASE,
1169                                       0, opts->rm_id);
1170         }
1171
1172         if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1173                 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1174                                       TCPOLEN_MPTCP_MPJ_SYN,
1175                                       opts->backup, opts->join_id);
1176                 put_unaligned_be32(opts->token, ptr);
1177                 ptr += 1;
1178                 put_unaligned_be32(opts->nonce, ptr);
1179                 ptr += 1;
1180         }
1181
1182         if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1183                 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1184                                       TCPOLEN_MPTCP_MPJ_SYNACK,
1185                                       opts->backup, opts->join_id);
1186                 put_unaligned_be64(opts->thmac, ptr);
1187                 ptr += 2;
1188                 put_unaligned_be32(opts->nonce, ptr);
1189                 ptr += 1;
1190         }
1191
1192         if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1193                 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1194                                       TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1195                 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1196                 ptr += 5;
1197         }
1198
1199         if (opts->ext_copy.use_ack || opts->ext_copy.use_map) {
1200                 struct mptcp_ext *mpext = &opts->ext_copy;
1201                 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1202                 u8 flags = 0;
1203
1204                 if (mpext->use_ack) {
1205                         flags = MPTCP_DSS_HAS_ACK;
1206                         if (mpext->ack64) {
1207                                 len += TCPOLEN_MPTCP_DSS_ACK64;
1208                                 flags |= MPTCP_DSS_ACK64;
1209                         } else {
1210                                 len += TCPOLEN_MPTCP_DSS_ACK32;
1211                         }
1212                 }
1213
1214                 if (mpext->use_map) {
1215                         len += TCPOLEN_MPTCP_DSS_MAP64;
1216
1217                         /* Use only 64-bit mapping flags for now, add
1218                          * support for optional 32-bit mappings later.
1219                          */
1220                         flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1221                         if (mpext->data_fin)
1222                                 flags |= MPTCP_DSS_DATA_FIN;
1223                 }
1224
1225                 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1226
1227                 if (mpext->use_ack) {
1228                         if (mpext->ack64) {
1229                                 put_unaligned_be64(mpext->data_ack, ptr);
1230                                 ptr += 2;
1231                         } else {
1232                                 put_unaligned_be32(mpext->data_ack32, ptr);
1233                                 ptr += 1;
1234                         }
1235                 }
1236
1237                 if (mpext->use_map) {
1238                         put_unaligned_be64(mpext->data_seq, ptr);
1239                         ptr += 2;
1240                         put_unaligned_be32(mpext->subflow_seq, ptr);
1241                         ptr += 1;
1242                         put_unaligned_be32(mpext->data_len << 16 |
1243                                            TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1244                 }
1245         }
1246
1247         if (tp)
1248                 mptcp_set_rwin(tp);
1249 }