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                 if (opsize != expected_opsize)
  48                         break;
  49
  50                 /* try to be gentle vs future versions on the initial syn */
  51                 version = *ptr++ & MPTCP_VERSION_MASK;
  52                 if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
  53                         if (version != MPTCP_SUPPORTED_VERSION)
  54                                 break;
  55                 } else if (version < MPTCP_SUPPORTED_VERSION) {
  56                         break;
  57                 }
  58
  59                 flags = *ptr++;
  60                 if (!mptcp_cap_flag_sha256(flags) ||
  61                     (flags & MPTCP_CAP_EXTENSIBILITY))
  62                         break;
  63
  64                 /* RFC 6824, Section 3.1:
  65                  * "For the Checksum Required bit (labeled "A"), if either
  66                  * host requires the use of checksums, checksums MUST be used.
  67                  * In other words, the only way for checksums not to be used
  68                  * is if both hosts in their SYNs set A=0."
  69                  *
  70                  * Section 3.3.0:
  71                  * "If a checksum is not present when its use has been
  72                  * negotiated, the receiver MUST close the subflow with a RST as
  73                  * it is considered broken."
  74                  *
  75                  * We don't implement DSS checksum - fall back to TCP.
  76                  */
  77                 if (flags & MPTCP_CAP_CHECKSUM_REQD)
  78                         break;
  79
  80                 mp_opt->mp_capable = 1;
  81                 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
  82                         mp_opt->sndr_key = get_unaligned_be64(ptr);
  83                         ptr += 8;
  84                 }
  85                 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
  86                         mp_opt->rcvr_key = get_unaligned_be64(ptr);
  87                         ptr += 8;
  88                 }
  89                 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA) {
  90                         /* Section 3.1.:
  91                          * "the data parameters in a MP_CAPABLE are semantically
  92                          * equivalent to those in a DSS option and can be used
  93                          * interchangeably."
  94                          */
  95                         mp_opt->dss = 1;
  96                         mp_opt->use_map = 1;
  97                         mp_opt->mpc_map = 1;
  98                         mp_opt->data_len = get_unaligned_be16(ptr);
  99                         ptr += 2;
 100                 }
 101                 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d",
 102                          version, flags, opsize, mp_opt->sndr_key,
 103                          mp_opt->rcvr_key, mp_opt->data_len);
 104                 break;
 105
 106         case MPTCPOPT_MP_JOIN:
 107                 mp_opt->mp_join = 1;
 108                 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
 109                         mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
 110                         mp_opt->join_id = *ptr++;
 111                         mp_opt->token = get_unaligned_be32(ptr);
 112                         ptr += 4;
 113                         mp_opt->nonce = get_unaligned_be32(ptr);
 114                         ptr += 4;
 115                         pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
 116                                  mp_opt->backup, mp_opt->join_id,
 117                                  mp_opt->token, mp_opt->nonce);
 118                 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
 119                         mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
 120                         mp_opt->join_id = *ptr++;
 121                         mp_opt->thmac = get_unaligned_be64(ptr);
 122                         ptr += 8;
 123                         mp_opt->nonce = get_unaligned_be32(ptr);
 124                         ptr += 4;
 125                         pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
 126                                  mp_opt->backup, mp_opt->join_id,
 127                                  mp_opt->thmac, mp_opt->nonce);
 128                 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
 129                         ptr += 2;
 130                         memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
 131                         pr_debug("MP_JOIN hmac");
 132                 } else {
 133                         mp_opt->mp_join = 0;
 134                 }
 135                 break;
 136
 137         case MPTCPOPT_DSS:
 138                 pr_debug("DSS");
 139                 ptr++;
 140
 141                 /* we must clear 'mpc_map' be able to detect MP_CAPABLE
 142                  * map vs DSS map in mptcp_incoming_options(), and reconstruct
 143                  * map info accordingly
 144                  */
 145                 mp_opt->mpc_map = 0;
 146                 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
 147                 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
 148                 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
 149                 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
 150                 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
 151                 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
 152
 153                 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
 154                          mp_opt->data_fin, mp_opt->dsn64,
 155                          mp_opt->use_map, mp_opt->ack64,
 156                          mp_opt->use_ack);
 157
 158                 expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
 159
 160                 if (mp_opt->use_ack) {
 161                         if (mp_opt->ack64)
 162                                 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
 163                         else
 164                                 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
 165                 }
 166
 167                 if (mp_opt->use_map) {
 168                         if (mp_opt->dsn64)
 169                                 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
 170                         else
 171                                 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
 172                 }
 173
 174                 /* RFC 6824, Section 3.3:
 175                  * If a checksum is present, but its use had
 176                  * not been negotiated in the MP_CAPABLE handshake,
 177                  * the checksum field MUST be ignored.
 178                  */
 179                 if (opsize != expected_opsize &&
 180                     opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
 181                         break;
 182
 183                 mp_opt->dss = 1;
 184
 185                 if (mp_opt->use_ack) {
 186                         if (mp_opt->ack64) {
 187                                 mp_opt->data_ack = get_unaligned_be64(ptr);
 188                                 ptr += 8;
 189                         } else {
 190                                 mp_opt->data_ack = get_unaligned_be32(ptr);
 191                                 ptr += 4;
 192                         }
 193
 194                         pr_debug("data_ack=%llu", mp_opt->data_ack);
 195                 }
 196
 197                 if (mp_opt->use_map) {
 198                         if (mp_opt->dsn64) {
 199                                 mp_opt->data_seq = get_unaligned_be64(ptr);
 200                                 ptr += 8;
 201                         } else {
 202                                 mp_opt->data_seq = get_unaligned_be32(ptr);
 203                                 ptr += 4;
 204                         }
 205
 206                         mp_opt->subflow_seq = get_unaligned_be32(ptr);
 207                         ptr += 4;
 208
 209                         mp_opt->data_len = get_unaligned_be16(ptr);
 210                         ptr += 2;
 211
 212                         pr_debug("data_seq=%llu subflow_seq=%u data_len=%u",
 213                                  mp_opt->data_seq, mp_opt->subflow_seq,
 214                                  mp_opt->data_len);
 215                 }
 216
 217                 break;
 218
 219         case MPTCPOPT_ADD_ADDR:
 220                 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
 221                 if (!mp_opt->echo) {
 222                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
 223                             opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
 224                                 mp_opt->addr.family = AF_INET;
 225 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 226                         else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
 227                                  opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
 228                                 mp_opt->addr.family = AF_INET6;
 229 #endif
 230                         else
 231                                 break;
 232                 } else {
 233                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
 234                             opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
 235                                 mp_opt->addr.family = AF_INET;
 236 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 237                         else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
 238                                  opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
 239                                 mp_opt->addr.family = AF_INET6;
 240 #endif
 241                         else
 242                                 break;
 243                 }
 244
 245                 mp_opt->add_addr = 1;
 246                 mp_opt->addr.id = *ptr++;
 247                 if (mp_opt->addr.family == AF_INET) {
 248                         memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4);
 249                         ptr += 4;
 250                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
 251                             opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
 252                                 mp_opt->addr.port = htons(get_unaligned_be16(ptr));
 253                                 ptr += 2;
 254                         }
 255                 }
 256 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 257                 else {
 258                         memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16);
 259                         ptr += 16;
 260                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
 261                             opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
 262                                 mp_opt->addr.port = htons(get_unaligned_be16(ptr));
 263                                 ptr += 2;
 264                         }
 265                 }
 266 #endif
 267                 if (!mp_opt->echo) {
 268                         mp_opt->ahmac = get_unaligned_be64(ptr);
 269                         ptr += 8;
 270                 }
 271                 pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d",
 272                          (mp_opt->addr.family == AF_INET6) ? "6" : "",
 273                          mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port));
 274                 break;
 275
 276         case MPTCPOPT_RM_ADDR:
 277                 if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 ||
 278                     opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX)
 279                         break;
 280
 281                 ptr++;
 282
 283                 mp_opt->rm_addr = 1;
 284                 mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE;
 285                 for (i = 0; i < mp_opt->rm_list.nr; i++)
 286                         mp_opt->rm_list.ids[i] = *ptr++;
 287                 pr_debug("RM_ADDR: rm_list_nr=%d", mp_opt->rm_list.nr);
 288                 break;
 289
 290         case MPTCPOPT_MP_PRIO:
 291                 if (opsize != TCPOLEN_MPTCP_PRIO)
 292                         break;
 293
 294                 mp_opt->mp_prio = 1;
 295                 mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP;
 296                 pr_debug("MP_PRIO: prio=%d", mp_opt->backup);
 297                 break;
 298
 299         case MPTCPOPT_MP_FASTCLOSE:
 300                 if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
 301                         break;
 302
 303                 ptr += 2;
 304                 mp_opt->rcvr_key = get_unaligned_be64(ptr);
 305                 ptr += 8;
 306                 mp_opt->fastclose = 1;
 307                 break;
 308
 309         case MPTCPOPT_RST:
 310                 if (opsize != TCPOLEN_MPTCP_RST)
 311                         break;
 312
 313                 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
 314                         break;
 315                 mp_opt->reset = 1;
 316                 flags = *ptr++;
 317                 mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT;
 318                 mp_opt->reset_reason = *ptr;
 319                 break;
 320
 321         default:
 322                 break;
 323         }
 324 }
 325
 326 void mptcp_get_options(const struct sk_buff *skb,
 327                        struct mptcp_options_received *mp_opt)
 328 {
 329         const struct tcphdr *th = tcp_hdr(skb);
 330         const unsigned char *ptr;
 331         int length;
 332
 333         /* initialize option status */
 334         mp_opt->mp_capable = 0;
 335         mp_opt->mp_join = 0;
 336         mp_opt->add_addr = 0;
 337         mp_opt->ahmac = 0;
 338         mp_opt->fastclose = 0;
 339         mp_opt->addr.port = 0;
 340         mp_opt->rm_addr = 0;
 341         mp_opt->dss = 0;
 342         mp_opt->mp_prio = 0;
 343         mp_opt->reset = 0;
 344
 345         length = (th->doff * 4) - sizeof(struct tcphdr);
 346         ptr = (const unsigned char *)(th + 1);
 347
 348         while (length > 0) {
 349                 int opcode = *ptr++;
 350                 int opsize;
 351
 352                 switch (opcode) {
 353                 case TCPOPT_EOL:
 354                         return;
 355                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 356                         length--;
 357                         continue;
 358                 default:
 359                         if (length < 2)
 360                                 return;
 361                         opsize = *ptr++;
 362                         if (opsize < 2) /* "silly options" */
 363                                 return;
 364                         if (opsize > length)
 365                                 return; /* don't parse partial options */
 366                         if (opcode == TCPOPT_MPTCP)
 367                                 mptcp_parse_option(skb, ptr, opsize, mp_opt);
 368                         ptr += opsize - 2;
 369                         length -= opsize;
 370                 }
 371         }
 372 }
 373
 374 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
 375                        unsigned int *size, struct mptcp_out_options *opts)
 376 {
 377         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 378
 379         /* we will use snd_isn to detect first pkt [re]transmission
 380          * in mptcp_established_options_mp()
 381          */
 382         subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
 383         if (subflow->request_mptcp) {
 384                 opts->suboptions = OPTION_MPTCP_MPC_SYN;
 385                 *size = TCPOLEN_MPTCP_MPC_SYN;
 386                 return true;
 387         } else if (subflow->request_join) {
 388                 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
 389                          subflow->local_nonce);
 390                 opts->suboptions = OPTION_MPTCP_MPJ_SYN;
 391                 opts->join_id = subflow->local_id;
 392                 opts->token = subflow->remote_token;
 393                 opts->nonce = subflow->local_nonce;
 394                 opts->backup = subflow->request_bkup;
 395                 *size = TCPOLEN_MPTCP_MPJ_SYN;
 396                 return true;
 397         }
 398         return false;
 399 }
 400
 401 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
 402  * TCP can't schedule delack timer before the subflow is fully established.
 403  * MPTCP uses the delack timer to do 3rd ack retransmissions
 404  */
 405 static void schedule_3rdack_retransmission(struct sock *sk)
 406 {
 407         struct inet_connection_sock *icsk = inet_csk(sk);
 408         struct tcp_sock *tp = tcp_sk(sk);
 409         unsigned long timeout;
 410
 411         /* reschedule with a timeout above RTT, as we must look only for drop */
 412         if (tp->srtt_us)
 413                 timeout = tp->srtt_us << 1;
 414         else
 415                 timeout = TCP_TIMEOUT_INIT;
 416
 417         WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
 418         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 419         icsk->icsk_ack.timeout = timeout;
 420         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 421 }
 422
 423 static void clear_3rdack_retransmission(struct sock *sk)
 424 {
 425         struct inet_connection_sock *icsk = inet_csk(sk);
 426
 427         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 428         icsk->icsk_ack.timeout = 0;
 429         icsk->icsk_ack.ato = 0;
 430         icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
 431 }
 432
 433 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
 434                                          bool snd_data_fin_enable,
 435                                          unsigned int *size,
 436                                          unsigned int remaining,
 437                                          struct mptcp_out_options *opts)
 438 {
 439         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 440         struct mptcp_ext *mpext;
 441         unsigned int data_len;
 442
 443         /* When skb is not available, we better over-estimate the emitted
 444          * options len. A full DSS option (28 bytes) is longer than
 445          * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
 446          * tell the caller to defer the estimate to
 447          * mptcp_established_options_dss(), which will reserve enough space.
 448          */
 449         if (!skb)
 450                 return false;
 451
 452         /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
 453         if (subflow->fully_established || snd_data_fin_enable ||
 454             subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
 455             sk->sk_state != TCP_ESTABLISHED)
 456                 return false;
 457
 458         if (subflow->mp_capable) {
 459                 mpext = mptcp_get_ext(skb);
 460                 data_len = mpext ? mpext->data_len : 0;
 461
 462                 /* we will check ext_copy.data_len in mptcp_write_options() to
 463                  * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
 464                  * TCPOLEN_MPTCP_MPC_ACK
 465                  */
 466                 opts->ext_copy.data_len = data_len;
 467                 opts->suboptions = OPTION_MPTCP_MPC_ACK;
 468                 opts->sndr_key = subflow->local_key;
 469                 opts->rcvr_key = subflow->remote_key;
 470
 471                 /* Section 3.1.
 472                  * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
 473                  * packets that start the first subflow of an MPTCP connection,
 474                  * as well as the first packet that carries data
 475                  */
 476                 if (data_len > 0)
 477                         *size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4);
 478                 else
 479                         *size = TCPOLEN_MPTCP_MPC_ACK;
 480
 481                 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
 482                          subflow, subflow->local_key, subflow->remote_key,
 483                          data_len);
 484
 485                 return true;
 486         } else if (subflow->mp_join) {
 487                 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
 488                 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
 489                 *size = TCPOLEN_MPTCP_MPJ_ACK;
 490                 pr_debug("subflow=%p", subflow);
 491
 492                 schedule_3rdack_retransmission(sk);
 493                 return true;
 494         }
 495         return false;
 496 }
 497
 498 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
 499                                  struct sk_buff *skb, struct mptcp_ext *ext)
 500 {
 501         /* The write_seq value has already been incremented, so the actual
 502          * sequence number for the DATA_FIN is one less.
 503          */
 504         u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
 505
 506         if (!ext->use_map || !skb->len) {
 507                 /* RFC6824 requires a DSS mapping with specific values
 508                  * if DATA_FIN is set but no data payload is mapped
 509                  */
 510                 ext->data_fin = 1;
 511                 ext->use_map = 1;
 512                 ext->dsn64 = 1;
 513                 ext->data_seq = data_fin_tx_seq;
 514                 ext->subflow_seq = 0;
 515                 ext->data_len = 1;
 516         } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
 517                 /* If there's an existing DSS mapping and it is the
 518                  * final mapping, DATA_FIN consumes 1 additional byte of
 519                  * mapping space.
 520                  */
 521                 ext->data_fin = 1;
 522                 ext->data_len++;
 523         }
 524 }
 525
 526 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
 527                                           bool snd_data_fin_enable,
 528                                           unsigned int *size,
 529                                           unsigned int remaining,
 530                                           struct mptcp_out_options *opts)
 531 {
 532         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 533         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 534         unsigned int dss_size = 0;
 535         struct mptcp_ext *mpext;
 536         unsigned int ack_size;
 537         bool ret = false;
 538         u64 ack_seq;
 539
 540         mpext = skb ? mptcp_get_ext(skb) : NULL;
 541
 542         if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
 543                 unsigned int map_size;
 544
 545                 map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
 546
 547                 remaining -= map_size;
 548                 dss_size = map_size;
 549                 if (mpext)
 550                         opts->ext_copy = *mpext;
 551
 552                 if (skb && snd_data_fin_enable)
 553                         mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
 554                 ret = true;
 555         }
 556
 557         /* passive sockets msk will set the 'can_ack' after accept(), even
 558          * if the first subflow may have the already the remote key handy
 559          */
 560         opts->ext_copy.use_ack = 0;
 561         if (!READ_ONCE(msk->can_ack)) {
 562                 *size = ALIGN(dss_size, 4);
 563                 return ret;
 564         }
 565
 566         ack_seq = READ_ONCE(msk->ack_seq);
 567         if (READ_ONCE(msk->use_64bit_ack)) {
 568                 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
 569                 opts->ext_copy.data_ack = ack_seq;
 570                 opts->ext_copy.ack64 = 1;
 571         } else {
 572                 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
 573                 opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
 574                 opts->ext_copy.ack64 = 0;
 575         }
 576         opts->ext_copy.use_ack = 1;
 577         WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
 578
 579         /* Add kind/length/subtype/flag overhead if mapping is not populated */
 580         if (dss_size == 0)
 581                 ack_size += TCPOLEN_MPTCP_DSS_BASE;
 582
 583         dss_size += ack_size;
 584
 585         *size = ALIGN(dss_size, 4);
 586         return true;
 587 }
 588
 589 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
 590                                   struct mptcp_addr_info *addr)
 591 {
 592         u16 port = ntohs(addr->port);
 593         u8 hmac[SHA256_DIGEST_SIZE];
 594         u8 msg[19];
 595         int i = 0;
 596
 597         msg[i++] = addr->id;
 598         if (addr->family == AF_INET) {
 599                 memcpy(&msg[i], &addr->addr.s_addr, 4);
 600                 i += 4;
 601         }
 602 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 603         else if (addr->family == AF_INET6) {
 604                 memcpy(&msg[i], &addr->addr6.s6_addr, 16);
 605                 i += 16;
 606         }
 607 #endif
 608         msg[i++] = port >> 8;
 609         msg[i++] = port & 0xFF;
 610
 611         mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
 612
 613         return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
 614 }
 615
 616 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
 617                                                unsigned int *size,
 618                                                unsigned int remaining,
 619                                                struct mptcp_out_options *opts)
 620 {
 621         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 622         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 623         bool drop_other_suboptions = false;
 624         unsigned int opt_size = *size;
 625         bool echo;
 626         bool port;
 627         int len;
 628
 629         if ((mptcp_pm_should_add_signal_ipv6(msk) ||
 630              mptcp_pm_should_add_signal_port(msk) ||
 631              mptcp_pm_should_add_signal_echo(msk)) &&
 632             skb && skb_is_tcp_pure_ack(skb)) {
 633                 pr_debug("drop other suboptions");
 634                 opts->suboptions = 0;
 635                 opts->ext_copy.use_ack = 0;
 636                 opts->ext_copy.use_map = 0;
 637                 remaining += opt_size;
 638                 drop_other_suboptions = true;
 639         }
 640
 641         if (!mptcp_pm_should_add_signal(msk) ||
 642             !(mptcp_pm_add_addr_signal(msk, remaining, &opts->addr, &echo, &port)))
 643                 return false;
 644
 645         len = mptcp_add_addr_len(opts->addr.family, echo, port);
 646         if (remaining < len)
 647                 return false;
 648
 649         *size = len;
 650         if (drop_other_suboptions)
 651                 *size -= opt_size;
 652         opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
 653         if (!echo) {
 654                 opts->ahmac = add_addr_generate_hmac(msk->local_key,
 655                                                      msk->remote_key,
 656                                                      &opts->addr);
 657         }
 658         pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
 659                  opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
 660
 661         return true;
 662 }
 663
 664 static bool mptcp_established_options_rm_addr(struct sock *sk,
 665                                               unsigned int *size,
 666                                               unsigned int remaining,
 667                                               struct mptcp_out_options *opts)
 668 {
 669         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 670         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 671         struct mptcp_rm_list rm_list;
 672         int i, len;
 673
 674         if (!mptcp_pm_should_rm_signal(msk) ||
 675             !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
 676                 return false;
 677
 678         len = mptcp_rm_addr_len(&rm_list);
 679         if (len < 0)
 680                 return false;
 681         if (remaining < len)
 682                 return false;
 683
 684         *size = len;
 685         opts->suboptions |= OPTION_MPTCP_RM_ADDR;
 686         opts->rm_list = rm_list;
 687
 688         for (i = 0; i < opts->rm_list.nr; i++)
 689                 pr_debug("rm_list_ids[%d]=%d", i, opts->rm_list.ids[i]);
 690
 691         return true;
 692 }
 693
 694 static bool mptcp_established_options_mp_prio(struct sock *sk,
 695                                               unsigned int *size,
 696                                               unsigned int remaining,
 697                                               struct mptcp_out_options *opts)
 698 {
 699         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 700
 701         if (!subflow->send_mp_prio)
 702                 return false;
 703
 704         /* account for the trailing 'nop' option */
 705         if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
 706                 return false;
 707
 708         *size = TCPOLEN_MPTCP_PRIO_ALIGN;
 709         opts->suboptions |= OPTION_MPTCP_PRIO;
 710         opts->backup = subflow->request_bkup;
 711
 712         pr_debug("prio=%d", opts->backup);
 713
 714         return true;
 715 }
 716
 717 static noinline void mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
 718                                                    unsigned int *size,
 719                                                    unsigned int remaining,
 720                                                    struct mptcp_out_options *opts)
 721 {
 722         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 723
 724         if (remaining < TCPOLEN_MPTCP_RST)
 725                 return;
 726
 727         *size = TCPOLEN_MPTCP_RST;
 728         opts->suboptions |= OPTION_MPTCP_RST;
 729         opts->reset_transient = subflow->reset_transient;
 730         opts->reset_reason = subflow->reset_reason;
 731 }
 732
 733 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
 734                                unsigned int *size, unsigned int remaining,
 735                                struct mptcp_out_options *opts)
 736 {
 737         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 738         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 739         unsigned int opt_size = 0;
 740         bool snd_data_fin;
 741         bool ret = false;
 742
 743         opts->suboptions = 0;
 744
 745         if (unlikely(__mptcp_check_fallback(msk)))
 746                 return false;
 747
 748         if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
 749                 mptcp_established_options_rst(sk, skb, size, remaining, opts);
 750                 return true;
 751         }
 752
 753         snd_data_fin = mptcp_data_fin_enabled(msk);
 754         if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, remaining, opts))
 755                 ret = true;
 756         else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, remaining, opts))
 757                 ret = true;
 758
 759         /* we reserved enough space for the above options, and exceeding the
 760          * TCP option space would be fatal
 761          */
 762         if (WARN_ON_ONCE(opt_size > remaining))
 763                 return false;
 764
 765         *size += opt_size;
 766         remaining -= opt_size;
 767         if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
 768                 *size += opt_size;
 769                 remaining -= opt_size;
 770                 ret = true;
 771         } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
 772                 *size += opt_size;
 773                 remaining -= opt_size;
 774                 ret = true;
 775         }
 776
 777         if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
 778                 *size += opt_size;
 779                 remaining -= opt_size;
 780                 ret = true;
 781         }
 782
 783         return ret;
 784 }
 785
 786 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
 787                           struct mptcp_out_options *opts)
 788 {
 789         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 790
 791         if (subflow_req->mp_capable) {
 792                 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
 793                 opts->sndr_key = subflow_req->local_key;
 794                 *size = TCPOLEN_MPTCP_MPC_SYNACK;
 795                 pr_debug("subflow_req=%p, local_key=%llu",
 796                          subflow_req, subflow_req->local_key);
 797                 return true;
 798         } else if (subflow_req->mp_join) {
 799                 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
 800                 opts->backup = subflow_req->backup;
 801                 opts->join_id = subflow_req->local_id;
 802                 opts->thmac = subflow_req->thmac;
 803                 opts->nonce = subflow_req->local_nonce;
 804                 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
 805                          subflow_req, opts->backup, opts->join_id,
 806                          opts->thmac, opts->nonce);
 807                 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
 808                 return true;
 809         }
 810         return false;
 811 }
 812
 813 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
 814                                     struct mptcp_subflow_context *subflow,
 815                                     struct sk_buff *skb,
 816                                     struct mptcp_options_received *mp_opt)
 817 {
 818         /* here we can process OoO, in-window pkts, only in-sequence 4th ack
 819          * will make the subflow fully established
 820          */
 821         if (likely(subflow->fully_established)) {
 822                 /* on passive sockets, check for 3rd ack retransmission
 823                  * note that msk is always set by subflow_syn_recv_sock()
 824                  * for mp_join subflows
 825                  */
 826                 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
 827                     TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
 828                     subflow->mp_join && mp_opt->mp_join &&
 829                     READ_ONCE(msk->pm.server_side))
 830                         tcp_send_ack(ssk);
 831                 goto fully_established;
 832         }
 833
 834         /* we must process OoO packets before the first subflow is fully
 835          * established. OoO packets are instead a protocol violation
 836          * for MP_JOIN subflows as the peer must not send any data
 837          * before receiving the forth ack - cfr. RFC 8684 section 3.2.
 838          */
 839         if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
 840                 if (subflow->mp_join)
 841                         goto reset;
 842                 return subflow->mp_capable;
 843         }
 844
 845         if (mp_opt->dss && mp_opt->use_ack) {
 846                 /* subflows are fully established as soon as we get any
 847                  * additional ack.
 848                  */
 849                 subflow->fully_established = 1;
 850                 WRITE_ONCE(msk->fully_established, true);
 851                 goto fully_established;
 852         }
 853
 854         if (mp_opt->add_addr) {
 855                 WRITE_ONCE(msk->fully_established, true);
 856                 return true;
 857         }
 858
 859         /* If the first established packet does not contain MP_CAPABLE + data
 860          * then fallback to TCP. Fallback scenarios requires a reset for
 861          * MP_JOIN subflows.
 862          */
 863         if (!mp_opt->mp_capable) {
 864                 if (subflow->mp_join)
 865                         goto reset;
 866                 subflow->mp_capable = 0;
 867                 pr_fallback(msk);
 868                 __mptcp_do_fallback(msk);
 869                 return false;
 870         }
 871
 872         if (unlikely(!READ_ONCE(msk->pm.server_side)))
 873                 pr_warn_once("bogus mpc option on established client sk");
 874         mptcp_subflow_fully_established(subflow, mp_opt);
 875
 876 fully_established:
 877         /* if the subflow is not already linked into the conn_list, we can't
 878          * notify the PM: this subflow is still on the listener queue
 879          * and the PM possibly acquiring the subflow lock could race with
 880          * the listener close
 881          */
 882         if (likely(subflow->pm_notified) || list_empty(&subflow->node))
 883                 return true;
 884
 885         subflow->pm_notified = 1;
 886         if (subflow->mp_join) {
 887                 clear_3rdack_retransmission(ssk);
 888                 mptcp_pm_subflow_established(msk);
 889         } else {
 890                 mptcp_pm_fully_established(msk, ssk, GFP_ATOMIC);
 891         }
 892         return true;
 893
 894 reset:
 895         mptcp_subflow_reset(ssk);
 896         return false;
 897 }
 898
 899 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit)
 900 {
 901         u32 old_ack32, cur_ack32;
 902
 903         if (use_64bit)
 904                 return cur_ack;
 905
 906         old_ack32 = (u32)old_ack;
 907         cur_ack32 = (u32)cur_ack;
 908         cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32;
 909         if (unlikely(before(cur_ack32, old_ack32)))
 910                 return cur_ack + (1LL << 32);
 911         return cur_ack;
 912 }
 913
 914 static void ack_update_msk(struct mptcp_sock *msk,
 915                            struct sock *ssk,
 916                            struct mptcp_options_received *mp_opt)
 917 {
 918         u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
 919         struct sock *sk = (struct sock *)msk;
 920         u64 old_snd_una;
 921
 922         mptcp_data_lock(sk);
 923
 924         /* avoid ack expansion on update conflict, to reduce the risk of
 925          * wrongly expanding to a future ack sequence number, which is way
 926          * more dangerous than missing an ack
 927          */
 928         old_snd_una = msk->snd_una;
 929         new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
 930
 931         /* ACK for data not even sent yet? Ignore. */
 932         if (after64(new_snd_una, snd_nxt))
 933                 new_snd_una = old_snd_una;
 934
 935         new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
 936
 937         if (after64(new_wnd_end, msk->wnd_end))
 938                 msk->wnd_end = new_wnd_end;
 939
 940         /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
 941         if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
 942                 __mptcp_check_push(sk, ssk);
 943
 944         if (after64(new_snd_una, old_snd_una)) {
 945                 msk->snd_una = new_snd_una;
 946                 __mptcp_data_acked(sk);
 947         }
 948         mptcp_data_unlock(sk);
 949
 950         trace_ack_update_msk(mp_opt->data_ack,
 951                              old_snd_una, new_snd_una,
 952                              new_wnd_end, msk->wnd_end);
 953 }
 954
 955 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
 956 {
 957         /* Skip if DATA_FIN was already received.
 958          * If updating simultaneously with the recvmsg loop, values
 959          * should match. If they mismatch, the peer is misbehaving and
 960          * we will prefer the most recent information.
 961          */
 962         if (READ_ONCE(msk->rcv_data_fin))
 963                 return false;
 964
 965         WRITE_ONCE(msk->rcv_data_fin_seq,
 966                    expand_ack(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
 967         WRITE_ONCE(msk->rcv_data_fin, 1);
 968
 969         return true;
 970 }
 971
 972 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
 973                                 struct mptcp_options_received *mp_opt)
 974 {
 975         u64 hmac = 0;
 976
 977         if (mp_opt->echo)
 978                 return true;
 979
 980         hmac = add_addr_generate_hmac(msk->remote_key,
 981                                       msk->local_key,
 982                                       &mp_opt->addr);
 983
 984         pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
 985                  msk, (unsigned long long)hmac,
 986                  (unsigned long long)mp_opt->ahmac);
 987
 988         return hmac == mp_opt->ahmac;
 989 }
 990
 991 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
 992 {
 993         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 994         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 995         struct mptcp_options_received mp_opt;
 996         struct mptcp_ext *mpext;
 997
 998         if (__mptcp_check_fallback(msk)) {
 999                 /* Keep it simple and unconditionally trigger send data cleanup and
1000                  * pending queue spooling. We will need to acquire the data lock
1001                  * for more accurate checks, and once the lock is acquired, such
1002                  * helpers are cheap.
1003                  */
1004                 mptcp_data_lock(subflow->conn);
1005                 if (sk_stream_memory_free(sk))
1006                         __mptcp_check_push(subflow->conn, sk);
1007                 __mptcp_data_acked(subflow->conn);
1008                 mptcp_data_unlock(subflow->conn);
1009                 return;
1010         }
1011
1012         mptcp_get_options(skb, &mp_opt);
1013         if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1014                 return;
1015
1016         if (mp_opt.fastclose &&
1017             msk->local_key == mp_opt.rcvr_key) {
1018                 WRITE_ONCE(msk->rcv_fastclose, true);
1019                 mptcp_schedule_work((struct sock *)msk);
1020         }
1021
1022         if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) {
1023                 if (!mp_opt.echo) {
1024                         mptcp_pm_add_addr_received(msk, &mp_opt.addr);
1025                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1026                 } else {
1027                         mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1028                         mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1029                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1030                 }
1031
1032                 if (mp_opt.addr.port)
1033                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1034
1035                 mp_opt.add_addr = 0;
1036         }
1037
1038         if (mp_opt.rm_addr) {
1039                 mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1040                 mp_opt.rm_addr = 0;
1041         }
1042
1043         if (mp_opt.mp_prio) {
1044                 mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1045                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1046                 mp_opt.mp_prio = 0;
1047         }
1048
1049         if (mp_opt.reset) {
1050                 subflow->reset_seen = 1;
1051                 subflow->reset_reason = mp_opt.reset_reason;
1052                 subflow->reset_transient = mp_opt.reset_transient;
1053         }
1054
1055         if (!mp_opt.dss)
1056                 return;
1057
1058         /* we can't wait for recvmsg() to update the ack_seq, otherwise
1059          * monodirectional flows will stuck
1060          */
1061         if (mp_opt.use_ack)
1062                 ack_update_msk(msk, sk, &mp_opt);
1063
1064         /* Zero-data-length packets are dropped by the caller and not
1065          * propagated to the MPTCP layer, so the skb extension does not
1066          * need to be allocated or populated. DATA_FIN information, if
1067          * present, needs to be updated here before the skb is freed.
1068          */
1069         if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1070                 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1071                     mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
1072                     schedule_work(&msk->work))
1073                         sock_hold(subflow->conn);
1074
1075                 return;
1076         }
1077
1078         mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1079         if (!mpext)
1080                 return;
1081
1082         memset(mpext, 0, sizeof(*mpext));
1083
1084         if (mp_opt.use_map) {
1085                 if (mp_opt.mpc_map) {
1086                         /* this is an MP_CAPABLE carrying MPTCP data
1087                          * we know this map the first chunk of data
1088                          */
1089                         mptcp_crypto_key_sha(subflow->remote_key, NULL,
1090                                              &mpext->data_seq);
1091                         mpext->data_seq++;
1092                         mpext->subflow_seq = 1;
1093                         mpext->dsn64 = 1;
1094                         mpext->mpc_map = 1;
1095                         mpext->data_fin = 0;
1096                 } else {
1097                         mpext->data_seq = mp_opt.data_seq;
1098                         mpext->subflow_seq = mp_opt.subflow_seq;
1099                         mpext->dsn64 = mp_opt.dsn64;
1100                         mpext->data_fin = mp_opt.data_fin;
1101                 }
1102                 mpext->data_len = mp_opt.data_len;
1103                 mpext->use_map = 1;
1104         }
1105 }
1106
1107 static void mptcp_set_rwin(const struct tcp_sock *tp)
1108 {
1109         const struct sock *ssk = (const struct sock *)tp;
1110         const struct mptcp_subflow_context *subflow;
1111         struct mptcp_sock *msk;
1112         u64 ack_seq;
1113
1114         subflow = mptcp_subflow_ctx(ssk);
1115         msk = mptcp_sk(subflow->conn);
1116
1117         ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd;
1118
1119         if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent)))
1120                 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
1121 }
1122
1123 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,
1124                          struct mptcp_out_options *opts)
1125 {
1126         if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
1127              OPTION_MPTCP_MPC_ACK) & opts->suboptions) {
1128                 u8 len;
1129
1130                 if (OPTION_MPTCP_MPC_SYN & opts->suboptions)
1131                         len = TCPOLEN_MPTCP_MPC_SYN;
1132                 else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions)
1133                         len = TCPOLEN_MPTCP_MPC_SYNACK;
1134                 else if (opts->ext_copy.data_len)
1135                         len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1136                 else
1137                         len = TCPOLEN_MPTCP_MPC_ACK;
1138
1139                 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1140                                       MPTCP_SUPPORTED_VERSION,
1141                                       MPTCP_CAP_HMAC_SHA256);
1142
1143                 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1144                     opts->suboptions))
1145                         goto mp_capable_done;
1146
1147                 put_unaligned_be64(opts->sndr_key, ptr);
1148                 ptr += 2;
1149                 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1150                         goto mp_capable_done;
1151
1152                 put_unaligned_be64(opts->rcvr_key, ptr);
1153                 ptr += 2;
1154                 if (!opts->ext_copy.data_len)
1155                         goto mp_capable_done;
1156
1157                 put_unaligned_be32(opts->ext_copy.data_len << 16 |
1158                                    TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1159                 ptr += 1;
1160         }
1161
1162 mp_capable_done:
1163         if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1164                 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1165                 u8 echo = MPTCP_ADDR_ECHO;
1166
1167 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1168                 if (opts->addr.family == AF_INET6)
1169                         len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1170 #endif
1171
1172                 if (opts->addr.port)
1173                         len += TCPOLEN_MPTCP_PORT_LEN;
1174
1175                 if (opts->ahmac) {
1176                         len += sizeof(opts->ahmac);
1177                         echo = 0;
1178                 }
1179
1180                 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1181                                       len, echo, opts->addr.id);
1182                 if (opts->addr.family == AF_INET) {
1183                         memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1184                         ptr += 1;
1185                 }
1186 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1187                 else if (opts->addr.family == AF_INET6) {
1188                         memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1189                         ptr += 4;
1190                 }
1191 #endif
1192
1193                 if (!opts->addr.port) {
1194                         if (opts->ahmac) {
1195                                 put_unaligned_be64(opts->ahmac, ptr);
1196                                 ptr += 2;
1197                         }
1198                 } else {
1199                         u16 port = ntohs(opts->addr.port);
1200
1201                         if (opts->ahmac) {
1202                                 u8 *bptr = (u8 *)ptr;
1203
1204                                 put_unaligned_be16(port, bptr);
1205                                 bptr += 2;
1206                                 put_unaligned_be64(opts->ahmac, bptr);
1207                                 bptr += 8;
1208                                 put_unaligned_be16(TCPOPT_NOP << 8 |
1209                                                    TCPOPT_NOP, bptr);
1210
1211                                 ptr += 3;
1212                         } else {
1213                                 put_unaligned_be32(port << 16 |
1214                                                    TCPOPT_NOP << 8 |
1215                                                    TCPOPT_NOP, ptr);
1216                                 ptr += 1;
1217                         }
1218                 }
1219         }
1220
1221         if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1222                 u8 i = 1;
1223
1224                 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1225                                       TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1226                                       0, opts->rm_list.ids[0]);
1227
1228                 while (i < opts->rm_list.nr) {
1229                         u8 id1, id2, id3, id4;
1230
1231                         id1 = opts->rm_list.ids[i];
1232                         id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1233                         id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1234                         id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1235                         put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1236                         ptr += 1;
1237                         i += 4;
1238                 }
1239         }
1240
1241         if (OPTION_MPTCP_PRIO & opts->suboptions) {
1242                 const struct sock *ssk = (const struct sock *)tp;
1243                 struct mptcp_subflow_context *subflow;
1244
1245                 subflow = mptcp_subflow_ctx(ssk);
1246                 subflow->send_mp_prio = 0;
1247
1248                 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1249                                       TCPOLEN_MPTCP_PRIO,
1250                                       opts->backup, TCPOPT_NOP);
1251         }
1252
1253         if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1254                 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1255                                       TCPOLEN_MPTCP_MPJ_SYN,
1256                                       opts->backup, opts->join_id);
1257                 put_unaligned_be32(opts->token, ptr);
1258                 ptr += 1;
1259                 put_unaligned_be32(opts->nonce, ptr);
1260                 ptr += 1;
1261         }
1262
1263         if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1264                 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1265                                       TCPOLEN_MPTCP_MPJ_SYNACK,
1266                                       opts->backup, opts->join_id);
1267                 put_unaligned_be64(opts->thmac, ptr);
1268                 ptr += 2;
1269                 put_unaligned_be32(opts->nonce, ptr);
1270                 ptr += 1;
1271         }
1272
1273         if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1274                 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1275                                       TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1276                 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1277                 ptr += 5;
1278         }
1279
1280         if (OPTION_MPTCP_RST & opts->suboptions)
1281                 *ptr++ = mptcp_option(MPTCPOPT_RST,
1282                                       TCPOLEN_MPTCP_RST,
1283                                       opts->reset_transient,
1284                                       opts->reset_reason);
1285
1286         if (opts->ext_copy.use_ack || opts->ext_copy.use_map) {
1287                 struct mptcp_ext *mpext = &opts->ext_copy;
1288                 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1289                 u8 flags = 0;
1290
1291                 if (mpext->use_ack) {
1292                         flags = MPTCP_DSS_HAS_ACK;
1293                         if (mpext->ack64) {
1294                                 len += TCPOLEN_MPTCP_DSS_ACK64;
1295                                 flags |= MPTCP_DSS_ACK64;
1296                         } else {
1297                                 len += TCPOLEN_MPTCP_DSS_ACK32;
1298                         }
1299                 }
1300
1301                 if (mpext->use_map) {
1302                         len += TCPOLEN_MPTCP_DSS_MAP64;
1303
1304                         /* Use only 64-bit mapping flags for now, add
1305                          * support for optional 32-bit mappings later.
1306                          */
1307                         flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1308                         if (mpext->data_fin)
1309                                 flags |= MPTCP_DSS_DATA_FIN;
1310                 }
1311
1312                 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1313
1314                 if (mpext->use_ack) {
1315                         if (mpext->ack64) {
1316                                 put_unaligned_be64(mpext->data_ack, ptr);
1317                                 ptr += 2;
1318                         } else {
1319                                 put_unaligned_be32(mpext->data_ack32, ptr);
1320                                 ptr += 1;
1321                         }
1322                 }
1323
1324                 if (mpext->use_map) {
1325                         put_unaligned_be64(mpext->data_seq, ptr);
1326                         ptr += 2;
1327                         put_unaligned_be32(mpext->subflow_seq, ptr);
1328                         ptr += 1;
1329                         put_unaligned_be32(mpext->data_len << 16 |
1330                                            TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1331                 }
1332         }
1333
1334         if (tp)
1335                 mptcp_set_rwin(tp);
1336 }
1337
1338 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1339 {
1340         const struct mptcp_ext *ext = mptcp_get_ext(skb);
1341         u8 flags, reason;
1342
1343         if (ext) {
1344                 flags = ext->reset_transient;
1345                 reason = ext->reset_reason;
1346
1347                 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1348                                     flags, reason);
1349         }
1350
1351         return htonl(0u);
1352 }
1353 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);