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 sock *sk,
 327                        const struct sk_buff *skb,
 328                        struct mptcp_options_received *mp_opt)
 329 {
 330         const struct tcphdr *th = tcp_hdr(skb);
 331         const unsigned char *ptr;
 332         int length;
 333
 334         /* initialize option status */
 335         mp_opt->mp_capable = 0;
 336         mp_opt->mp_join = 0;
 337         mp_opt->add_addr = 0;
 338         mp_opt->ahmac = 0;
 339         mp_opt->fastclose = 0;
 340         mp_opt->addr.port = 0;
 341         mp_opt->rm_addr = 0;
 342         mp_opt->dss = 0;
 343         mp_opt->mp_prio = 0;
 344         mp_opt->reset = 0;
 345
 346         length = (th->doff * 4) - sizeof(struct tcphdr);
 347         ptr = (const unsigned char *)(th + 1);
 348
 349         while (length > 0) {
 350                 int opcode = *ptr++;
 351                 int opsize;
 352
 353                 switch (opcode) {
 354                 case TCPOPT_EOL:
 355                         return;
 356                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 357                         length--;
 358                         continue;
 359                 default:
 360                         if (length < 2)
 361                                 return;
 362                         opsize = *ptr++;
 363                         if (opsize < 2) /* "silly options" */
 364                                 return;
 365                         if (opsize > length)
 366                                 return; /* don't parse partial options */
 367                         if (opcode == TCPOPT_MPTCP)
 368                                 mptcp_parse_option(skb, ptr, opsize, mp_opt);
 369                         ptr += opsize - 2;
 370                         length -= opsize;
 371                 }
 372         }
 373 }
 374
 375 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
 376                        unsigned int *size, struct mptcp_out_options *opts)
 377 {
 378         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 379
 380         /* we will use snd_isn to detect first pkt [re]transmission
 381          * in mptcp_established_options_mp()
 382          */
 383         subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
 384         if (subflow->request_mptcp) {
 385                 opts->suboptions = OPTION_MPTCP_MPC_SYN;
 386                 *size = TCPOLEN_MPTCP_MPC_SYN;
 387                 return true;
 388         } else if (subflow->request_join) {
 389                 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
 390                          subflow->local_nonce);
 391                 opts->suboptions = OPTION_MPTCP_MPJ_SYN;
 392                 opts->join_id = subflow->local_id;
 393                 opts->token = subflow->remote_token;
 394                 opts->nonce = subflow->local_nonce;
 395                 opts->backup = subflow->request_bkup;
 396                 *size = TCPOLEN_MPTCP_MPJ_SYN;
 397                 return true;
 398         }
 399         return false;
 400 }
 401
 402 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
 403  * TCP can't schedule delack timer before the subflow is fully established.
 404  * MPTCP uses the delack timer to do 3rd ack retransmissions
 405  */
 406 static void schedule_3rdack_retransmission(struct sock *sk)
 407 {
 408         struct inet_connection_sock *icsk = inet_csk(sk);
 409         struct tcp_sock *tp = tcp_sk(sk);
 410         unsigned long timeout;
 411
 412         /* reschedule with a timeout above RTT, as we must look only for drop */
 413         if (tp->srtt_us)
 414                 timeout = tp->srtt_us << 1;
 415         else
 416                 timeout = TCP_TIMEOUT_INIT;
 417
 418         WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
 419         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 420         icsk->icsk_ack.timeout = timeout;
 421         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 422 }
 423
 424 static void clear_3rdack_retransmission(struct sock *sk)
 425 {
 426         struct inet_connection_sock *icsk = inet_csk(sk);
 427
 428         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 429         icsk->icsk_ack.timeout = 0;
 430         icsk->icsk_ack.ato = 0;
 431         icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
 432 }
 433
 434 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
 435                                          bool snd_data_fin_enable,
 436                                          unsigned int *size,
 437                                          unsigned int remaining,
 438                                          struct mptcp_out_options *opts)
 439 {
 440         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 441         struct mptcp_ext *mpext;
 442         unsigned int data_len;
 443
 444         /* When skb is not available, we better over-estimate the emitted
 445          * options len. A full DSS option (28 bytes) is longer than
 446          * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
 447          * tell the caller to defer the estimate to
 448          * mptcp_established_options_dss(), which will reserve enough space.
 449          */
 450         if (!skb)
 451                 return false;
 452
 453         /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
 454         if (subflow->fully_established || snd_data_fin_enable ||
 455             subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
 456             sk->sk_state != TCP_ESTABLISHED)
 457                 return false;
 458
 459         if (subflow->mp_capable) {
 460                 mpext = mptcp_get_ext(skb);
 461                 data_len = mpext ? mpext->data_len : 0;
 462
 463                 /* we will check ext_copy.data_len in mptcp_write_options() to
 464                  * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
 465                  * TCPOLEN_MPTCP_MPC_ACK
 466                  */
 467                 opts->ext_copy.data_len = data_len;
 468                 opts->suboptions = OPTION_MPTCP_MPC_ACK;
 469                 opts->sndr_key = subflow->local_key;
 470                 opts->rcvr_key = subflow->remote_key;
 471
 472                 /* Section 3.1.
 473                  * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
 474                  * packets that start the first subflow of an MPTCP connection,
 475                  * as well as the first packet that carries data
 476                  */
 477                 if (data_len > 0)
 478                         *size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4);
 479                 else
 480                         *size = TCPOLEN_MPTCP_MPC_ACK;
 481
 482                 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
 483                          subflow, subflow->local_key, subflow->remote_key,
 484                          data_len);
 485
 486                 return true;
 487         } else if (subflow->mp_join) {
 488                 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
 489                 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
 490                 *size = TCPOLEN_MPTCP_MPJ_ACK;
 491                 pr_debug("subflow=%p", subflow);
 492
 493                 schedule_3rdack_retransmission(sk);
 494                 return true;
 495         }
 496         return false;
 497 }
 498
 499 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
 500                                  struct sk_buff *skb, struct mptcp_ext *ext)
 501 {
 502         /* The write_seq value has already been incremented, so the actual
 503          * sequence number for the DATA_FIN is one less.
 504          */
 505         u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
 506
 507         if (!ext->use_map || !skb->len) {
 508                 /* RFC6824 requires a DSS mapping with specific values
 509                  * if DATA_FIN is set but no data payload is mapped
 510                  */
 511                 ext->data_fin = 1;
 512                 ext->use_map = 1;
 513                 ext->dsn64 = 1;
 514                 ext->data_seq = data_fin_tx_seq;
 515                 ext->subflow_seq = 0;
 516                 ext->data_len = 1;
 517         } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
 518                 /* If there's an existing DSS mapping and it is the
 519                  * final mapping, DATA_FIN consumes 1 additional byte of
 520                  * mapping space.
 521                  */
 522                 ext->data_fin = 1;
 523                 ext->data_len++;
 524         }
 525 }
 526
 527 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
 528                                           bool snd_data_fin_enable,
 529                                           unsigned int *size,
 530                                           unsigned int remaining,
 531                                           struct mptcp_out_options *opts)
 532 {
 533         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 534         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 535         unsigned int dss_size = 0;
 536         struct mptcp_ext *mpext;
 537         unsigned int ack_size;
 538         bool ret = false;
 539         u64 ack_seq;
 540
 541         mpext = skb ? mptcp_get_ext(skb) : NULL;
 542
 543         if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
 544                 unsigned int map_size;
 545
 546                 map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
 547
 548                 remaining -= map_size;
 549                 dss_size = map_size;
 550                 if (mpext)
 551                         opts->ext_copy = *mpext;
 552
 553                 if (skb && snd_data_fin_enable)
 554                         mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
 555                 ret = true;
 556         }
 557
 558         /* passive sockets msk will set the 'can_ack' after accept(), even
 559          * if the first subflow may have the already the remote key handy
 560          */
 561         opts->ext_copy.use_ack = 0;
 562         if (!READ_ONCE(msk->can_ack)) {
 563                 *size = ALIGN(dss_size, 4);
 564                 return ret;
 565         }
 566
 567         ack_seq = READ_ONCE(msk->ack_seq);
 568         if (READ_ONCE(msk->use_64bit_ack)) {
 569                 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
 570                 opts->ext_copy.data_ack = ack_seq;
 571                 opts->ext_copy.ack64 = 1;
 572         } else {
 573                 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
 574                 opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
 575                 opts->ext_copy.ack64 = 0;
 576         }
 577         opts->ext_copy.use_ack = 1;
 578         WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
 579
 580         /* Add kind/length/subtype/flag overhead if mapping is not populated */
 581         if (dss_size == 0)
 582                 ack_size += TCPOLEN_MPTCP_DSS_BASE;
 583
 584         dss_size += ack_size;
 585
 586         *size = ALIGN(dss_size, 4);
 587         return true;
 588 }
 589
 590 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
 591                                   struct mptcp_addr_info *addr)
 592 {
 593         u16 port = ntohs(addr->port);
 594         u8 hmac[SHA256_DIGEST_SIZE];
 595         u8 msg[19];
 596         int i = 0;
 597
 598         msg[i++] = addr->id;
 599         if (addr->family == AF_INET) {
 600                 memcpy(&msg[i], &addr->addr.s_addr, 4);
 601                 i += 4;
 602         }
 603 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 604         else if (addr->family == AF_INET6) {
 605                 memcpy(&msg[i], &addr->addr6.s6_addr, 16);
 606                 i += 16;
 607         }
 608 #endif
 609         msg[i++] = port >> 8;
 610         msg[i++] = port & 0xFF;
 611
 612         mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
 613
 614         return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
 615 }
 616
 617 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
 618                                                unsigned int *size,
 619                                                unsigned int remaining,
 620                                                struct mptcp_out_options *opts)
 621 {
 622         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 623         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 624         bool drop_other_suboptions = false;
 625         unsigned int opt_size = *size;
 626         bool echo;
 627         bool port;
 628         int len;
 629
 630         if ((mptcp_pm_should_add_signal_ipv6(msk) ||
 631              mptcp_pm_should_add_signal_port(msk) ||
 632              mptcp_pm_should_add_signal_echo(msk)) &&
 633             skb && skb_is_tcp_pure_ack(skb)) {
 634                 pr_debug("drop other suboptions");
 635                 opts->suboptions = 0;
 636                 opts->ext_copy.use_ack = 0;
 637                 opts->ext_copy.use_map = 0;
 638                 remaining += opt_size;
 639                 drop_other_suboptions = true;
 640         }
 641
 642         if (!mptcp_pm_should_add_signal(msk) ||
 643             !(mptcp_pm_add_addr_signal(msk, remaining, &opts->addr, &echo, &port)))
 644                 return false;
 645
 646         len = mptcp_add_addr_len(opts->addr.family, echo, port);
 647         if (remaining < len)
 648                 return false;
 649
 650         *size = len;
 651         if (drop_other_suboptions)
 652                 *size -= opt_size;
 653         opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
 654         if (!echo) {
 655                 opts->ahmac = add_addr_generate_hmac(msk->local_key,
 656                                                      msk->remote_key,
 657                                                      &opts->addr);
 658         }
 659         pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
 660                  opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
 661
 662         return true;
 663 }
 664
 665 static bool mptcp_established_options_rm_addr(struct sock *sk,
 666                                               unsigned int *size,
 667                                               unsigned int remaining,
 668                                               struct mptcp_out_options *opts)
 669 {
 670         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 671         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 672         struct mptcp_rm_list rm_list;
 673         int i, len;
 674
 675         if (!mptcp_pm_should_rm_signal(msk) ||
 676             !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
 677                 return false;
 678
 679         len = mptcp_rm_addr_len(&rm_list);
 680         if (len < 0)
 681                 return false;
 682         if (remaining < len)
 683                 return false;
 684
 685         *size = len;
 686         opts->suboptions |= OPTION_MPTCP_RM_ADDR;
 687         opts->rm_list = rm_list;
 688
 689         for (i = 0; i < opts->rm_list.nr; i++)
 690                 pr_debug("rm_list_ids[%d]=%d", i, opts->rm_list.ids[i]);
 691
 692         return true;
 693 }
 694
 695 static bool mptcp_established_options_mp_prio(struct sock *sk,
 696                                               unsigned int *size,
 697                                               unsigned int remaining,
 698                                               struct mptcp_out_options *opts)
 699 {
 700         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 701
 702         if (!subflow->send_mp_prio)
 703                 return false;
 704
 705         /* account for the trailing 'nop' option */
 706         if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
 707                 return false;
 708
 709         *size = TCPOLEN_MPTCP_PRIO_ALIGN;
 710         opts->suboptions |= OPTION_MPTCP_PRIO;
 711         opts->backup = subflow->request_bkup;
 712
 713         pr_debug("prio=%d", opts->backup);
 714
 715         return true;
 716 }
 717
 718 static noinline void mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
 719                                                    unsigned int *size,
 720                                                    unsigned int remaining,
 721                                                    struct mptcp_out_options *opts)
 722 {
 723         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 724
 725         if (remaining < TCPOLEN_MPTCP_RST)
 726                 return;
 727
 728         *size = TCPOLEN_MPTCP_RST;
 729         opts->suboptions |= OPTION_MPTCP_RST;
 730         opts->reset_transient = subflow->reset_transient;
 731         opts->reset_reason = subflow->reset_reason;
 732 }
 733
 734 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
 735                                unsigned int *size, unsigned int remaining,
 736                                struct mptcp_out_options *opts)
 737 {
 738         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 739         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 740         unsigned int opt_size = 0;
 741         bool snd_data_fin;
 742         bool ret = false;
 743
 744         opts->suboptions = 0;
 745
 746         if (unlikely(__mptcp_check_fallback(msk)))
 747                 return false;
 748
 749         if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
 750                 mptcp_established_options_rst(sk, skb, size, remaining, opts);
 751                 return true;
 752         }
 753
 754         snd_data_fin = mptcp_data_fin_enabled(msk);
 755         if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, remaining, opts))
 756                 ret = true;
 757         else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, remaining, opts))
 758                 ret = true;
 759
 760         /* we reserved enough space for the above options, and exceeding the
 761          * TCP option space would be fatal
 762          */
 763         if (WARN_ON_ONCE(opt_size > remaining))
 764                 return false;
 765
 766         *size += opt_size;
 767         remaining -= opt_size;
 768         if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
 769                 *size += opt_size;
 770                 remaining -= opt_size;
 771                 ret = true;
 772         } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
 773                 *size += opt_size;
 774                 remaining -= opt_size;
 775                 ret = true;
 776         }
 777
 778         if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
 779                 *size += opt_size;
 780                 remaining -= opt_size;
 781                 ret = true;
 782         }
 783
 784         return ret;
 785 }
 786
 787 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
 788                           struct mptcp_out_options *opts)
 789 {
 790         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 791
 792         if (subflow_req->mp_capable) {
 793                 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
 794                 opts->sndr_key = subflow_req->local_key;
 795                 *size = TCPOLEN_MPTCP_MPC_SYNACK;
 796                 pr_debug("subflow_req=%p, local_key=%llu",
 797                          subflow_req, subflow_req->local_key);
 798                 return true;
 799         } else if (subflow_req->mp_join) {
 800                 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
 801                 opts->backup = subflow_req->backup;
 802                 opts->join_id = subflow_req->local_id;
 803                 opts->thmac = subflow_req->thmac;
 804                 opts->nonce = subflow_req->local_nonce;
 805                 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
 806                          subflow_req, opts->backup, opts->join_id,
 807                          opts->thmac, opts->nonce);
 808                 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
 809                 return true;
 810         }
 811         return false;
 812 }
 813
 814 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
 815                                     struct mptcp_subflow_context *subflow,
 816                                     struct sk_buff *skb,
 817                                     struct mptcp_options_received *mp_opt)
 818 {
 819         /* here we can process OoO, in-window pkts, only in-sequence 4th ack
 820          * will make the subflow fully established
 821          */
 822         if (likely(subflow->fully_established)) {
 823                 /* on passive sockets, check for 3rd ack retransmission
 824                  * note that msk is always set by subflow_syn_recv_sock()
 825                  * for mp_join subflows
 826                  */
 827                 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
 828                     TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
 829                     subflow->mp_join && mp_opt->mp_join &&
 830                     READ_ONCE(msk->pm.server_side))
 831                         tcp_send_ack(ssk);
 832                 goto fully_established;
 833         }
 834
 835         /* we must process OoO packets before the first subflow is fully
 836          * established. OoO packets are instead a protocol violation
 837          * for MP_JOIN subflows as the peer must not send any data
 838          * before receiving the forth ack - cfr. RFC 8684 section 3.2.
 839          */
 840         if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
 841                 if (subflow->mp_join)
 842                         goto reset;
 843                 return subflow->mp_capable;
 844         }
 845
 846         if ((mp_opt->dss && mp_opt->use_ack) ||
 847             (mp_opt->add_addr && !mp_opt->echo)) {
 848                 /* subflows are fully established as soon as we get any
 849                  * additional ack, including ADD_ADDR.
 850                  */
 851                 subflow->fully_established = 1;
 852                 WRITE_ONCE(msk->fully_established, true);
 853                 goto fully_established;
 854         }
 855
 856         /* If the first established packet does not contain MP_CAPABLE + data
 857          * then fallback to TCP. Fallback scenarios requires a reset for
 858          * MP_JOIN subflows.
 859          */
 860         if (!mp_opt->mp_capable) {
 861                 if (subflow->mp_join)
 862                         goto reset;
 863                 subflow->mp_capable = 0;
 864                 pr_fallback(msk);
 865                 __mptcp_do_fallback(msk);
 866                 return false;
 867         }
 868
 869         if (unlikely(!READ_ONCE(msk->pm.server_side)))
 870                 pr_warn_once("bogus mpc option on established client sk");
 871         mptcp_subflow_fully_established(subflow, mp_opt);
 872
 873 fully_established:
 874         /* if the subflow is not already linked into the conn_list, we can't
 875          * notify the PM: this subflow is still on the listener queue
 876          * and the PM possibly acquiring the subflow lock could race with
 877          * the listener close
 878          */
 879         if (likely(subflow->pm_notified) || list_empty(&subflow->node))
 880                 return true;
 881
 882         subflow->pm_notified = 1;
 883         if (subflow->mp_join) {
 884                 clear_3rdack_retransmission(ssk);
 885                 mptcp_pm_subflow_established(msk);
 886         } else {
 887                 mptcp_pm_fully_established(msk, ssk, GFP_ATOMIC);
 888         }
 889         return true;
 890
 891 reset:
 892         mptcp_subflow_reset(ssk);
 893         return false;
 894 }
 895
 896 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq)
 897 {
 898         u32 old_seq32, cur_seq32;
 899
 900         old_seq32 = (u32)old_seq;
 901         cur_seq32 = (u32)cur_seq;
 902         cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32;
 903         if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32)))
 904                 return cur_seq + (1LL << 32);
 905
 906         /* reverse wrap could happen, too */
 907         if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32)))
 908                 return cur_seq - (1LL << 32);
 909         return cur_seq;
 910 }
 911
 912 static void ack_update_msk(struct mptcp_sock *msk,
 913                            struct sock *ssk,
 914                            struct mptcp_options_received *mp_opt)
 915 {
 916         u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
 917         struct sock *sk = (struct sock *)msk;
 918         u64 old_snd_una;
 919
 920         mptcp_data_lock(sk);
 921
 922         /* avoid ack expansion on update conflict, to reduce the risk of
 923          * wrongly expanding to a future ack sequence number, which is way
 924          * more dangerous than missing an ack
 925          */
 926         old_snd_una = msk->snd_una;
 927         new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
 928
 929         /* ACK for data not even sent yet? Ignore. */
 930         if (after64(new_snd_una, snd_nxt))
 931                 new_snd_una = old_snd_una;
 932
 933         new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
 934
 935         if (after64(new_wnd_end, msk->wnd_end))
 936                 msk->wnd_end = new_wnd_end;
 937
 938         /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
 939         if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
 940                 __mptcp_check_push(sk, ssk);
 941
 942         if (after64(new_snd_una, old_snd_una)) {
 943                 msk->snd_una = new_snd_una;
 944                 __mptcp_data_acked(sk);
 945         }
 946         mptcp_data_unlock(sk);
 947
 948         trace_ack_update_msk(mp_opt->data_ack,
 949                              old_snd_una, new_snd_una,
 950                              new_wnd_end, msk->wnd_end);
 951 }
 952
 953 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
 954 {
 955         /* Skip if DATA_FIN was already received.
 956          * If updating simultaneously with the recvmsg loop, values
 957          * should match. If they mismatch, the peer is misbehaving and
 958          * we will prefer the most recent information.
 959          */
 960         if (READ_ONCE(msk->rcv_data_fin))
 961                 return false;
 962
 963         WRITE_ONCE(msk->rcv_data_fin_seq,
 964                    mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
 965         WRITE_ONCE(msk->rcv_data_fin, 1);
 966
 967         return true;
 968 }
 969
 970 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
 971                                 struct mptcp_options_received *mp_opt)
 972 {
 973         u64 hmac = 0;
 974
 975         if (mp_opt->echo)
 976                 return true;
 977
 978         hmac = add_addr_generate_hmac(msk->remote_key,
 979                                       msk->local_key,
 980                                       &mp_opt->addr);
 981
 982         pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
 983                  msk, (unsigned long long)hmac,
 984                  (unsigned long long)mp_opt->ahmac);
 985
 986         return hmac == mp_opt->ahmac;
 987 }
 988
 989 /* Return false if a subflow has been reset, else return true */
 990 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
 991 {
 992         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 993         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 994         struct mptcp_options_received mp_opt;
 995         struct mptcp_ext *mpext;
 996
 997         if (__mptcp_check_fallback(msk)) {
 998                 /* Keep it simple and unconditionally trigger send data cleanup and
 999                  * pending queue spooling. We will need to acquire the data lock
1000                  * for more accurate checks, and once the lock is acquired, such
1001                  * helpers are cheap.
1002                  */
1003                 mptcp_data_lock(subflow->conn);
1004                 if (sk_stream_memory_free(sk))
1005                         __mptcp_check_push(subflow->conn, sk);
1006                 __mptcp_data_acked(subflow->conn);
1007                 mptcp_data_unlock(subflow->conn);
1008                 return true;
1009         }
1010
1011         mptcp_get_options(sk, skb, &mp_opt);
1012
1013         /* The subflow can be in close state only if check_fully_established()
1014          * just sent a reset. If so, tell the caller to ignore the current packet.
1015          */
1016         if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1017                 return sk->sk_state != TCP_CLOSE;
1018
1019         if (mp_opt.fastclose &&
1020             msk->local_key == mp_opt.rcvr_key) {
1021                 WRITE_ONCE(msk->rcv_fastclose, true);
1022                 mptcp_schedule_work((struct sock *)msk);
1023         }
1024
1025         if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) {
1026                 if (!mp_opt.echo) {
1027                         mptcp_pm_add_addr_received(msk, &mp_opt.addr);
1028                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1029                 } else {
1030                         mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1031                         mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1032                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1033                 }
1034
1035                 if (mp_opt.addr.port)
1036                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1037
1038                 mp_opt.add_addr = 0;
1039         }
1040
1041         if (mp_opt.rm_addr) {
1042                 mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1043                 mp_opt.rm_addr = 0;
1044         }
1045
1046         if (mp_opt.mp_prio) {
1047                 mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1048                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1049                 mp_opt.mp_prio = 0;
1050         }
1051
1052         if (mp_opt.reset) {
1053                 subflow->reset_seen = 1;
1054                 subflow->reset_reason = mp_opt.reset_reason;
1055                 subflow->reset_transient = mp_opt.reset_transient;
1056         }
1057
1058         if (!mp_opt.dss)
1059                 return true;
1060
1061         /* we can't wait for recvmsg() to update the ack_seq, otherwise
1062          * monodirectional flows will stuck
1063          */
1064         if (mp_opt.use_ack)
1065                 ack_update_msk(msk, sk, &mp_opt);
1066
1067         /* Zero-data-length packets are dropped by the caller and not
1068          * propagated to the MPTCP layer, so the skb extension does not
1069          * need to be allocated or populated. DATA_FIN information, if
1070          * present, needs to be updated here before the skb is freed.
1071          */
1072         if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1073                 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1074                     mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
1075                     schedule_work(&msk->work))
1076                         sock_hold(subflow->conn);
1077
1078                 return true;
1079         }
1080
1081         mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1082         if (!mpext)
1083                 return true;
1084
1085         memset(mpext, 0, sizeof(*mpext));
1086
1087         if (mp_opt.use_map) {
1088                 if (mp_opt.mpc_map) {
1089                         /* this is an MP_CAPABLE carrying MPTCP data
1090                          * we know this map the first chunk of data
1091                          */
1092                         mptcp_crypto_key_sha(subflow->remote_key, NULL,
1093                                              &mpext->data_seq);
1094                         mpext->data_seq++;
1095                         mpext->subflow_seq = 1;
1096                         mpext->dsn64 = 1;
1097                         mpext->mpc_map = 1;
1098                         mpext->data_fin = 0;
1099                 } else {
1100                         mpext->data_seq = mp_opt.data_seq;
1101                         mpext->subflow_seq = mp_opt.subflow_seq;
1102                         mpext->dsn64 = mp_opt.dsn64;
1103                         mpext->data_fin = mp_opt.data_fin;
1104                 }
1105                 mpext->data_len = mp_opt.data_len;
1106                 mpext->use_map = 1;
1107         }
1108
1109         return true;
1110 }
1111
1112 static void mptcp_set_rwin(const struct tcp_sock *tp)
1113 {
1114         const struct sock *ssk = (const struct sock *)tp;
1115         const struct mptcp_subflow_context *subflow;
1116         struct mptcp_sock *msk;
1117         u64 ack_seq;
1118
1119         subflow = mptcp_subflow_ctx(ssk);
1120         msk = mptcp_sk(subflow->conn);
1121
1122         ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd;
1123
1124         if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent)))
1125                 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
1126 }
1127
1128 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,
1129                          struct mptcp_out_options *opts)
1130 {
1131         if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
1132              OPTION_MPTCP_MPC_ACK) & opts->suboptions) {
1133                 u8 len;
1134
1135                 if (OPTION_MPTCP_MPC_SYN & opts->suboptions)
1136                         len = TCPOLEN_MPTCP_MPC_SYN;
1137                 else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions)
1138                         len = TCPOLEN_MPTCP_MPC_SYNACK;
1139                 else if (opts->ext_copy.data_len)
1140                         len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1141                 else
1142                         len = TCPOLEN_MPTCP_MPC_ACK;
1143
1144                 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1145                                       MPTCP_SUPPORTED_VERSION,
1146                                       MPTCP_CAP_HMAC_SHA256);
1147
1148                 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1149                     opts->suboptions))
1150                         goto mp_capable_done;
1151
1152                 put_unaligned_be64(opts->sndr_key, ptr);
1153                 ptr += 2;
1154                 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1155                         goto mp_capable_done;
1156
1157                 put_unaligned_be64(opts->rcvr_key, ptr);
1158                 ptr += 2;
1159                 if (!opts->ext_copy.data_len)
1160                         goto mp_capable_done;
1161
1162                 put_unaligned_be32(opts->ext_copy.data_len << 16 |
1163                                    TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1164                 ptr += 1;
1165         }
1166
1167 mp_capable_done:
1168         if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1169                 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1170                 u8 echo = MPTCP_ADDR_ECHO;
1171
1172 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1173                 if (opts->addr.family == AF_INET6)
1174                         len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1175 #endif
1176
1177                 if (opts->addr.port)
1178                         len += TCPOLEN_MPTCP_PORT_LEN;
1179
1180                 if (opts->ahmac) {
1181                         len += sizeof(opts->ahmac);
1182                         echo = 0;
1183                 }
1184
1185                 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1186                                       len, echo, opts->addr.id);
1187                 if (opts->addr.family == AF_INET) {
1188                         memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1189                         ptr += 1;
1190                 }
1191 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1192                 else if (opts->addr.family == AF_INET6) {
1193                         memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1194                         ptr += 4;
1195                 }
1196 #endif
1197
1198                 if (!opts->addr.port) {
1199                         if (opts->ahmac) {
1200                                 put_unaligned_be64(opts->ahmac, ptr);
1201                                 ptr += 2;
1202                         }
1203                 } else {
1204                         u16 port = ntohs(opts->addr.port);
1205
1206                         if (opts->ahmac) {
1207                                 u8 *bptr = (u8 *)ptr;
1208
1209                                 put_unaligned_be16(port, bptr);
1210                                 bptr += 2;
1211                                 put_unaligned_be64(opts->ahmac, bptr);
1212                                 bptr += 8;
1213                                 put_unaligned_be16(TCPOPT_NOP << 8 |
1214                                                    TCPOPT_NOP, bptr);
1215
1216                                 ptr += 3;
1217                         } else {
1218                                 put_unaligned_be32(port << 16 |
1219                                                    TCPOPT_NOP << 8 |
1220                                                    TCPOPT_NOP, ptr);
1221                                 ptr += 1;
1222                         }
1223                 }
1224         }
1225
1226         if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1227                 u8 i = 1;
1228
1229                 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1230                                       TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1231                                       0, opts->rm_list.ids[0]);
1232
1233                 while (i < opts->rm_list.nr) {
1234                         u8 id1, id2, id3, id4;
1235
1236                         id1 = opts->rm_list.ids[i];
1237                         id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1238                         id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1239                         id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1240                         put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1241                         ptr += 1;
1242                         i += 4;
1243                 }
1244         }
1245
1246         if (OPTION_MPTCP_PRIO & opts->suboptions) {
1247                 const struct sock *ssk = (const struct sock *)tp;
1248                 struct mptcp_subflow_context *subflow;
1249
1250                 subflow = mptcp_subflow_ctx(ssk);
1251                 subflow->send_mp_prio = 0;
1252
1253                 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1254                                       TCPOLEN_MPTCP_PRIO,
1255                                       opts->backup, TCPOPT_NOP);
1256         }
1257
1258         if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1259                 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1260                                       TCPOLEN_MPTCP_MPJ_SYN,
1261                                       opts->backup, opts->join_id);
1262                 put_unaligned_be32(opts->token, ptr);
1263                 ptr += 1;
1264                 put_unaligned_be32(opts->nonce, ptr);
1265                 ptr += 1;
1266         }
1267
1268         if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1269                 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1270                                       TCPOLEN_MPTCP_MPJ_SYNACK,
1271                                       opts->backup, opts->join_id);
1272                 put_unaligned_be64(opts->thmac, ptr);
1273                 ptr += 2;
1274                 put_unaligned_be32(opts->nonce, ptr);
1275                 ptr += 1;
1276         }
1277
1278         if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1279                 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1280                                       TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1281                 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1282                 ptr += 5;
1283         }
1284
1285         if (OPTION_MPTCP_RST & opts->suboptions)
1286                 *ptr++ = mptcp_option(MPTCPOPT_RST,
1287                                       TCPOLEN_MPTCP_RST,
1288                                       opts->reset_transient,
1289                                       opts->reset_reason);
1290
1291         if (opts->ext_copy.use_ack || opts->ext_copy.use_map) {
1292                 struct mptcp_ext *mpext = &opts->ext_copy;
1293                 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1294                 u8 flags = 0;
1295
1296                 if (mpext->use_ack) {
1297                         flags = MPTCP_DSS_HAS_ACK;
1298                         if (mpext->ack64) {
1299                                 len += TCPOLEN_MPTCP_DSS_ACK64;
1300                                 flags |= MPTCP_DSS_ACK64;
1301                         } else {
1302                                 len += TCPOLEN_MPTCP_DSS_ACK32;
1303                         }
1304                 }
1305
1306                 if (mpext->use_map) {
1307                         len += TCPOLEN_MPTCP_DSS_MAP64;
1308
1309                         /* Use only 64-bit mapping flags for now, add
1310                          * support for optional 32-bit mappings later.
1311                          */
1312                         flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1313                         if (mpext->data_fin)
1314                                 flags |= MPTCP_DSS_DATA_FIN;
1315                 }
1316
1317                 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1318
1319                 if (mpext->use_ack) {
1320                         if (mpext->ack64) {
1321                                 put_unaligned_be64(mpext->data_ack, ptr);
1322                                 ptr += 2;
1323                         } else {
1324                                 put_unaligned_be32(mpext->data_ack32, ptr);
1325                                 ptr += 1;
1326                         }
1327                 }
1328
1329                 if (mpext->use_map) {
1330                         put_unaligned_be64(mpext->data_seq, ptr);
1331                         ptr += 2;
1332                         put_unaligned_be32(mpext->subflow_seq, ptr);
1333                         ptr += 1;
1334                         put_unaligned_be32(mpext->data_len << 16 |
1335                                            TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1336                 }
1337         }
1338
1339         if (tp)
1340                 mptcp_set_rwin(tp);
1341 }
1342
1343 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1344 {
1345         const struct mptcp_ext *ext = mptcp_get_ext(skb);
1346         u8 flags, reason;
1347
1348         if (ext) {
1349                 flags = ext->reset_transient;
1350                 reason = ext->reset_reason;
1351
1352                 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1353                                     flags, reason);
1354         }
1355
1356         return htonl(0u);
1357 }
1358 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);