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         u16 subopt;
  30         u8 version;
  31         u8 flags;
  32         u8 i;
  33
  34         switch (subtype) {
  35         case MPTCPOPT_MP_CAPABLE:
  36                 /* strict size checking */
  37                 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
  38                         if (skb->len > tcp_hdr(skb)->doff << 2)
  39                                 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA;
  40                         else
  41                                 expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
  42                         subopt = OPTION_MPTCP_MPC_ACK;
  43                 } else {
  44                         if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK) {
  45                                 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
  46                                 subopt = OPTION_MPTCP_MPC_SYNACK;
  47                         } else {
  48                                 expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
  49                                 subopt = OPTION_MPTCP_MPC_SYN;
  50                         }
  51                 }
  52
  53                 /* Cfr RFC 8684 Section 3.3.0:
  54                  * If a checksum is present but its use had
  55                  * not been negotiated in the MP_CAPABLE handshake, the receiver MUST
  56                  * close the subflow with a RST, as it is not behaving as negotiated.
  57                  * If a checksum is not present when its use has been negotiated, the
  58                  * receiver MUST close the subflow with a RST, as it is considered
  59                  * broken
  60                  * We parse even option with mismatching csum presence, so that
  61                  * later in subflow_data_ready we can trigger the reset.
  62                  */
  63                 if (opsize != expected_opsize &&
  64                     (expected_opsize != TCPOLEN_MPTCP_MPC_ACK_DATA ||
  65                      opsize != TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM))
  66                         break;
  67
  68                 /* try to be gentle vs future versions on the initial syn */
  69                 version = *ptr++ & MPTCP_VERSION_MASK;
  70                 if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
  71                         if (version != MPTCP_SUPPORTED_VERSION)
  72                                 break;
  73                 } else if (version < MPTCP_SUPPORTED_VERSION) {
  74                         break;
  75                 }
  76
  77                 flags = *ptr++;
  78                 if (!mptcp_cap_flag_sha256(flags) ||
  79                     (flags & MPTCP_CAP_EXTENSIBILITY))
  80                         break;
  81
  82                 /* RFC 6824, Section 3.1:
  83                  * "For the Checksum Required bit (labeled "A"), if either
  84                  * host requires the use of checksums, checksums MUST be used.
  85                  * In other words, the only way for checksums not to be used
  86                  * is if both hosts in their SYNs set A=0."
  87                  */
  88                 if (flags & MPTCP_CAP_CHECKSUM_REQD)
  89                         mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
  90
  91                 mp_opt->deny_join_id0 = !!(flags & MPTCP_CAP_DENY_JOIN_ID0);
  92
  93                 mp_opt->suboptions |= subopt;
  94                 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
  95                         mp_opt->sndr_key = get_unaligned_be64(ptr);
  96                         ptr += 8;
  97                 }
  98                 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
  99                         mp_opt->rcvr_key = get_unaligned_be64(ptr);
 100                         ptr += 8;
 101                 }
 102                 if (opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA) {
 103                         /* Section 3.1.:
 104                          * "the data parameters in a MP_CAPABLE are semantically
 105                          * equivalent to those in a DSS option and can be used
 106                          * interchangeably."
 107                          */
 108                         mp_opt->suboptions |= OPTION_MPTCP_DSS;
 109                         mp_opt->use_map = 1;
 110                         mp_opt->mpc_map = 1;
 111                         mp_opt->use_ack = 0;
 112                         mp_opt->data_len = get_unaligned_be16(ptr);
 113                         ptr += 2;
 114                 }
 115                 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM) {
 116                         mp_opt->csum = get_unaligned((__force __sum16 *)ptr);
 117                         mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
 118                         ptr += 2;
 119                 }
 120                 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d csum=%u\n",
 121                          version, flags, opsize, mp_opt->sndr_key,
 122                          mp_opt->rcvr_key, mp_opt->data_len, mp_opt->csum);
 123                 break;
 124
 125         case MPTCPOPT_MP_JOIN:
 126                 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
 127                         mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYN;
 128                         mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
 129                         mp_opt->join_id = *ptr++;
 130                         mp_opt->token = get_unaligned_be32(ptr);
 131                         ptr += 4;
 132                         mp_opt->nonce = get_unaligned_be32(ptr);
 133                         ptr += 4;
 134                         pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u\n",
 135                                  mp_opt->backup, mp_opt->join_id,
 136                                  mp_opt->token, mp_opt->nonce);
 137                 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
 138                         mp_opt->suboptions |= OPTION_MPTCP_MPJ_SYNACK;
 139                         mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
 140                         mp_opt->join_id = *ptr++;
 141                         mp_opt->thmac = get_unaligned_be64(ptr);
 142                         ptr += 8;
 143                         mp_opt->nonce = get_unaligned_be32(ptr);
 144                         ptr += 4;
 145                         pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u\n",
 146                                  mp_opt->backup, mp_opt->join_id,
 147                                  mp_opt->thmac, mp_opt->nonce);
 148                 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
 149                         mp_opt->suboptions |= OPTION_MPTCP_MPJ_ACK;
 150                         ptr += 2;
 151                         memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
 152                         pr_debug("MP_JOIN hmac\n");
 153                 }
 154                 break;
 155
 156         case MPTCPOPT_DSS:
 157                 pr_debug("DSS\n");
 158                 ptr++;
 159
 160                 /* we must clear 'mpc_map' be able to detect MP_CAPABLE
 161                  * map vs DSS map in mptcp_incoming_options(), and reconstruct
 162                  * map info accordingly
 163                  */
 164                 mp_opt->mpc_map = 0;
 165                 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
 166                 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
 167                 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
 168                 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
 169                 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
 170                 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
 171
 172                 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d\n",
 173                          mp_opt->data_fin, mp_opt->dsn64,
 174                          mp_opt->use_map, mp_opt->ack64,
 175                          mp_opt->use_ack);
 176
 177                 expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
 178
 179                 if (mp_opt->use_ack) {
 180                         if (mp_opt->ack64)
 181                                 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
 182                         else
 183                                 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
 184                 }
 185
 186                 if (mp_opt->use_map) {
 187                         if (mp_opt->dsn64)
 188                                 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
 189                         else
 190                                 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
 191                 }
 192
 193                 /* Always parse any csum presence combination, we will enforce
 194                  * RFC 8684 Section 3.3.0 checks later in subflow_data_ready
 195                  */
 196                 if (opsize != expected_opsize &&
 197                     opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
 198                         break;
 199
 200                 mp_opt->suboptions |= OPTION_MPTCP_DSS;
 201                 if (mp_opt->use_ack) {
 202                         if (mp_opt->ack64) {
 203                                 mp_opt->data_ack = get_unaligned_be64(ptr);
 204                                 ptr += 8;
 205                         } else {
 206                                 mp_opt->data_ack = get_unaligned_be32(ptr);
 207                                 ptr += 4;
 208                         }
 209
 210                         pr_debug("data_ack=%llu\n", mp_opt->data_ack);
 211                 }
 212
 213                 if (mp_opt->use_map) {
 214                         if (mp_opt->dsn64) {
 215                                 mp_opt->data_seq = get_unaligned_be64(ptr);
 216                                 ptr += 8;
 217                         } else {
 218                                 mp_opt->data_seq = get_unaligned_be32(ptr);
 219                                 ptr += 4;
 220                         }
 221
 222                         mp_opt->subflow_seq = get_unaligned_be32(ptr);
 223                         ptr += 4;
 224
 225                         mp_opt->data_len = get_unaligned_be16(ptr);
 226                         ptr += 2;
 227
 228                         if (opsize == expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) {
 229                                 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
 230                                 mp_opt->csum = get_unaligned((__force __sum16 *)ptr);
 231                                 ptr += 2;
 232                         }
 233
 234                         pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u\n",
 235                                  mp_opt->data_seq, mp_opt->subflow_seq,
 236                                  mp_opt->data_len, !!(mp_opt->suboptions & OPTION_MPTCP_CSUMREQD),
 237                                  mp_opt->csum);
 238                 }
 239
 240                 break;
 241
 242         case MPTCPOPT_ADD_ADDR:
 243                 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
 244                 if (!mp_opt->echo) {
 245                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
 246                             opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
 247                                 mp_opt->addr.family = AF_INET;
 248 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 249                         else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
 250                                  opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
 251                                 mp_opt->addr.family = AF_INET6;
 252 #endif
 253                         else
 254                                 break;
 255                 } else {
 256                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
 257                             opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
 258                                 mp_opt->addr.family = AF_INET;
 259 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 260                         else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
 261                                  opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
 262                                 mp_opt->addr.family = AF_INET6;
 263 #endif
 264                         else
 265                                 break;
 266                 }
 267
 268                 mp_opt->suboptions |= OPTION_MPTCP_ADD_ADDR;
 269                 mp_opt->addr.id = *ptr++;
 270                 mp_opt->addr.port = 0;
 271                 mp_opt->ahmac = 0;
 272                 if (mp_opt->addr.family == AF_INET) {
 273                         memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4);
 274                         ptr += 4;
 275                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
 276                             opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
 277                                 mp_opt->addr.port = htons(get_unaligned_be16(ptr));
 278                                 ptr += 2;
 279                         }
 280                 }
 281 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 282                 else {
 283                         memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16);
 284                         ptr += 16;
 285                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
 286                             opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
 287                                 mp_opt->addr.port = htons(get_unaligned_be16(ptr));
 288                                 ptr += 2;
 289                         }
 290                 }
 291 #endif
 292                 if (!mp_opt->echo) {
 293                         mp_opt->ahmac = get_unaligned_be64(ptr);
 294                         ptr += 8;
 295                 }
 296                 pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d\n",
 297                          (mp_opt->addr.family == AF_INET6) ? "6" : "",
 298                          mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port));
 299                 break;
 300
 301         case MPTCPOPT_RM_ADDR:
 302                 if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 ||
 303                     opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX)
 304                         break;
 305
 306                 ptr++;
 307
 308                 mp_opt->suboptions |= OPTION_MPTCP_RM_ADDR;
 309                 mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE;
 310                 for (i = 0; i < mp_opt->rm_list.nr; i++)
 311                         mp_opt->rm_list.ids[i] = *ptr++;
 312                 pr_debug("RM_ADDR: rm_list_nr=%d\n", mp_opt->rm_list.nr);
 313                 break;
 314
 315         case MPTCPOPT_MP_PRIO:
 316                 if (opsize != TCPOLEN_MPTCP_PRIO)
 317                         break;
 318
 319                 mp_opt->suboptions |= OPTION_MPTCP_PRIO;
 320                 mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP;
 321                 pr_debug("MP_PRIO: prio=%d\n", mp_opt->backup);
 322                 break;
 323
 324         case MPTCPOPT_MP_FASTCLOSE:
 325                 if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
 326                         break;
 327
 328                 ptr += 2;
 329                 mp_opt->rcvr_key = get_unaligned_be64(ptr);
 330                 ptr += 8;
 331                 mp_opt->suboptions |= OPTION_MPTCP_FASTCLOSE;
 332                 pr_debug("MP_FASTCLOSE: recv_key=%llu\n", mp_opt->rcvr_key);
 333                 break;
 334
 335         case MPTCPOPT_RST:
 336                 if (opsize != TCPOLEN_MPTCP_RST)
 337                         break;
 338
 339                 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
 340                         break;
 341
 342                 mp_opt->suboptions |= OPTION_MPTCP_RST;
 343                 flags = *ptr++;
 344                 mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT;
 345                 mp_opt->reset_reason = *ptr;
 346                 pr_debug("MP_RST: transient=%u reason=%u\n",
 347                          mp_opt->reset_transient, mp_opt->reset_reason);
 348                 break;
 349
 350         case MPTCPOPT_MP_FAIL:
 351                 if (opsize != TCPOLEN_MPTCP_FAIL)
 352                         break;
 353
 354                 ptr += 2;
 355                 mp_opt->suboptions |= OPTION_MPTCP_FAIL;
 356                 mp_opt->fail_seq = get_unaligned_be64(ptr);
 357                 pr_debug("MP_FAIL: data_seq=%llu\n", mp_opt->fail_seq);
 358                 break;
 359
 360         default:
 361                 break;
 362         }
 363 }
 364
 365 void mptcp_get_options(const struct sk_buff *skb,
 366                        struct mptcp_options_received *mp_opt)
 367 {
 368         const struct tcphdr *th = tcp_hdr(skb);
 369         const unsigned char *ptr;
 370         int length;
 371
 372         /* initialize option status */
 373         mp_opt->suboptions = 0;
 374
 375         length = (th->doff * 4) - sizeof(struct tcphdr);
 376         ptr = (const unsigned char *)(th + 1);
 377
 378         while (length > 0) {
 379                 int opcode = *ptr++;
 380                 int opsize;
 381
 382                 switch (opcode) {
 383                 case TCPOPT_EOL:
 384                         return;
 385                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 386                         length--;
 387                         continue;
 388                 default:
 389                         if (length < 2)
 390                                 return;
 391                         opsize = *ptr++;
 392                         if (opsize < 2) /* "silly options" */
 393                                 return;
 394                         if (opsize > length)
 395                                 return; /* don't parse partial options */
 396                         if (opcode == TCPOPT_MPTCP)
 397                                 mptcp_parse_option(skb, ptr, opsize, mp_opt);
 398                         ptr += opsize - 2;
 399                         length -= opsize;
 400                 }
 401         }
 402 }
 403
 404 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
 405                        unsigned int *size, struct mptcp_out_options *opts)
 406 {
 407         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 408
 409         /* we will use snd_isn to detect first pkt [re]transmission
 410          * in mptcp_established_options_mp()
 411          */
 412         subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
 413         if (subflow->request_mptcp) {
 414                 opts->suboptions = OPTION_MPTCP_MPC_SYN;
 415                 opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk));
 416                 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
 417                 *size = TCPOLEN_MPTCP_MPC_SYN;
 418                 return true;
 419         } else if (subflow->request_join) {
 420                 pr_debug("remote_token=%u, nonce=%u\n", subflow->remote_token,
 421                          subflow->local_nonce);
 422                 opts->suboptions = OPTION_MPTCP_MPJ_SYN;
 423                 opts->join_id = subflow->local_id;
 424                 opts->token = subflow->remote_token;
 425                 opts->nonce = subflow->local_nonce;
 426                 opts->backup = subflow->request_bkup;
 427                 *size = TCPOLEN_MPTCP_MPJ_SYN;
 428                 return true;
 429         }
 430         return false;
 431 }
 432
 433 static void clear_3rdack_retransmission(struct sock *sk)
 434 {
 435         struct inet_connection_sock *icsk = inet_csk(sk);
 436
 437         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 438         icsk->icsk_ack.timeout = 0;
 439         icsk->icsk_ack.ato = 0;
 440         icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
 441 }
 442
 443 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
 444                                          bool snd_data_fin_enable,
 445                                          unsigned int *size,
 446                                          struct mptcp_out_options *opts)
 447 {
 448         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 449         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 450         struct mptcp_ext *mpext;
 451         unsigned int data_len;
 452         u8 len;
 453
 454         /* When skb is not available, we better over-estimate the emitted
 455          * options len. A full DSS option (28 bytes) is longer than
 456          * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
 457          * tell the caller to defer the estimate to
 458          * mptcp_established_options_dss(), which will reserve enough space.
 459          */
 460         if (!skb)
 461                 return false;
 462
 463         /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
 464         if (READ_ONCE(subflow->fully_established) || snd_data_fin_enable ||
 465             subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
 466             sk->sk_state != TCP_ESTABLISHED)
 467                 return false;
 468
 469         if (subflow->mp_capable) {
 470                 mpext = mptcp_get_ext(skb);
 471                 data_len = mpext ? mpext->data_len : 0;
 472
 473                 /* we will check ops->data_len in mptcp_write_options() to
 474                  * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
 475                  * TCPOLEN_MPTCP_MPC_ACK
 476                  */
 477                 opts->data_len = data_len;
 478                 opts->suboptions = OPTION_MPTCP_MPC_ACK;
 479                 opts->sndr_key = subflow->local_key;
 480                 opts->rcvr_key = subflow->remote_key;
 481                 opts->csum_reqd = READ_ONCE(msk->csum_enabled);
 482                 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
 483
 484                 /* Section 3.1.
 485                  * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
 486                  * packets that start the first subflow of an MPTCP connection,
 487                  * as well as the first packet that carries data
 488                  */
 489                 if (data_len > 0) {
 490                         len = TCPOLEN_MPTCP_MPC_ACK_DATA;
 491                         if (opts->csum_reqd) {
 492                                 /* we need to propagate more info to csum the pseudo hdr */
 493                                 opts->data_seq = mpext->data_seq;
 494                                 opts->subflow_seq = mpext->subflow_seq;
 495                                 opts->csum = mpext->csum;
 496                                 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
 497                         }
 498                         *size = ALIGN(len, 4);
 499                 } else {
 500                         *size = TCPOLEN_MPTCP_MPC_ACK;
 501                 }
 502
 503                 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d\n",
 504                          subflow, subflow->local_key, subflow->remote_key,
 505                          data_len);
 506
 507                 return true;
 508         } else if (subflow->mp_join) {
 509                 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
 510                 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
 511                 *size = TCPOLEN_MPTCP_MPJ_ACK;
 512                 pr_debug("subflow=%p\n", subflow);
 513
 514                 /* we can use the full delegate action helper only from BH context
 515                  * If we are in process context - sk is flushing the backlog at
 516                  * socket lock release time - just set the appropriate flag, will
 517                  * be handled by the release callback
 518                  */
 519                 if (sock_owned_by_user(sk))
 520                         set_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status);
 521                 else
 522                         mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_ACK);
 523                 return true;
 524         }
 525         return false;
 526 }
 527
 528 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
 529                                  struct sk_buff *skb, struct mptcp_ext *ext)
 530 {
 531         /* The write_seq value has already been incremented, so the actual
 532          * sequence number for the DATA_FIN is one less.
 533          */
 534         u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
 535
 536         if (!ext->use_map || !skb->len) {
 537                 /* RFC6824 requires a DSS mapping with specific values
 538                  * if DATA_FIN is set but no data payload is mapped
 539                  */
 540                 ext->data_fin = 1;
 541                 ext->use_map = 1;
 542                 ext->dsn64 = 1;
 543                 ext->data_seq = data_fin_tx_seq;
 544                 ext->subflow_seq = 0;
 545                 ext->data_len = 1;
 546         } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
 547                 /* If there's an existing DSS mapping and it is the
 548                  * final mapping, DATA_FIN consumes 1 additional byte of
 549                  * mapping space.
 550                  */
 551                 ext->data_fin = 1;
 552                 ext->data_len++;
 553         }
 554 }
 555
 556 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
 557                                           bool snd_data_fin_enable,
 558                                           unsigned int *size,
 559                                           struct mptcp_out_options *opts)
 560 {
 561         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 562         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 563         unsigned int dss_size = 0;
 564         struct mptcp_ext *mpext;
 565         unsigned int ack_size;
 566         bool ret = false;
 567         u64 ack_seq;
 568
 569         opts->csum_reqd = READ_ONCE(msk->csum_enabled);
 570         mpext = skb ? mptcp_get_ext(skb) : NULL;
 571
 572         if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
 573                 unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
 574
 575                 if (mpext) {
 576                         if (opts->csum_reqd)
 577                                 map_size += TCPOLEN_MPTCP_DSS_CHECKSUM;
 578
 579                         opts->ext_copy = *mpext;
 580                 }
 581
 582                 dss_size = map_size;
 583                 if (skb && snd_data_fin_enable)
 584                         mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
 585                 opts->suboptions = OPTION_MPTCP_DSS;
 586                 ret = true;
 587         }
 588
 589         /* passive sockets msk will set the 'can_ack' after accept(), even
 590          * if the first subflow may have the already the remote key handy
 591          */
 592         opts->ext_copy.use_ack = 0;
 593         if (!READ_ONCE(msk->can_ack)) {
 594                 *size = ALIGN(dss_size, 4);
 595                 return ret;
 596         }
 597
 598         ack_seq = READ_ONCE(msk->ack_seq);
 599         if (READ_ONCE(msk->use_64bit_ack)) {
 600                 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
 601                 opts->ext_copy.data_ack = ack_seq;
 602                 opts->ext_copy.ack64 = 1;
 603         } else {
 604                 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
 605                 opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
 606                 opts->ext_copy.ack64 = 0;
 607         }
 608         opts->ext_copy.use_ack = 1;
 609         opts->suboptions = OPTION_MPTCP_DSS;
 610         WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
 611
 612         /* Add kind/length/subtype/flag overhead if mapping is not populated */
 613         if (dss_size == 0)
 614                 ack_size += TCPOLEN_MPTCP_DSS_BASE;
 615
 616         dss_size += ack_size;
 617
 618         *size = ALIGN(dss_size, 4);
 619         return true;
 620 }
 621
 622 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
 623                                   struct mptcp_addr_info *addr)
 624 {
 625         u16 port = ntohs(addr->port);
 626         u8 hmac[SHA256_DIGEST_SIZE];
 627         u8 msg[19];
 628         int i = 0;
 629
 630         msg[i++] = addr->id;
 631         if (addr->family == AF_INET) {
 632                 memcpy(&msg[i], &addr->addr.s_addr, 4);
 633                 i += 4;
 634         }
 635 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 636         else if (addr->family == AF_INET6) {
 637                 memcpy(&msg[i], &addr->addr6.s6_addr, 16);
 638                 i += 16;
 639         }
 640 #endif
 641         msg[i++] = port >> 8;
 642         msg[i++] = port & 0xFF;
 643
 644         mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
 645
 646         return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
 647 }
 648
 649 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
 650                                                unsigned int *size,
 651                                                unsigned int remaining,
 652                                                struct mptcp_out_options *opts)
 653 {
 654         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 655         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 656         bool drop_other_suboptions = false;
 657         unsigned int opt_size = *size;
 658         bool echo;
 659         int len;
 660
 661         /* add addr will strip the existing options, be sure to avoid breaking
 662          * MPC/MPJ handshakes
 663          */
 664         if (!mptcp_pm_should_add_signal(msk) ||
 665             (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) ||
 666             !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, &opts->addr,
 667                     &echo, &drop_other_suboptions))
 668                 return false;
 669
 670         /*
 671          * Later on, mptcp_write_options() will enforce mutually exclusion with
 672          * DSS, bail out if such option is set and we can't drop it.
 673          */
 674         if (drop_other_suboptions)
 675                 remaining += opt_size;
 676         else if (opts->suboptions & OPTION_MPTCP_DSS)
 677                 return false;
 678
 679         len = mptcp_add_addr_len(opts->addr.family, echo, !!opts->addr.port);
 680         if (remaining < len)
 681                 return false;
 682
 683         *size = len;
 684         if (drop_other_suboptions) {
 685                 pr_debug("drop other suboptions\n");
 686                 opts->suboptions = 0;
 687
 688                 /* note that e.g. DSS could have written into the memory
 689                  * aliased by ahmac, we must reset the field here
 690                  * to avoid appending the hmac even for ADD_ADDR echo
 691                  * options
 692                  */
 693                 opts->ahmac = 0;
 694                 *size -= opt_size;
 695         }
 696         opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
 697         if (!echo) {
 698                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDRTX);
 699                 opts->ahmac = add_addr_generate_hmac(READ_ONCE(msk->local_key),
 700                                                      READ_ONCE(msk->remote_key),
 701                                                      &opts->addr);
 702         } else {
 703                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADDTX);
 704         }
 705         pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d\n",
 706                  opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
 707
 708         return true;
 709 }
 710
 711 static bool mptcp_established_options_rm_addr(struct sock *sk,
 712                                               unsigned int *size,
 713                                               unsigned int remaining,
 714                                               struct mptcp_out_options *opts)
 715 {
 716         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 717         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 718         struct mptcp_rm_list rm_list;
 719         int i, len;
 720
 721         if (!mptcp_pm_should_rm_signal(msk) ||
 722             !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
 723                 return false;
 724
 725         len = mptcp_rm_addr_len(&rm_list);
 726         if (len < 0)
 727                 return false;
 728         if (remaining < len)
 729                 return false;
 730
 731         *size = len;
 732         opts->suboptions |= OPTION_MPTCP_RM_ADDR;
 733         opts->rm_list = rm_list;
 734
 735         for (i = 0; i < opts->rm_list.nr; i++)
 736                 pr_debug("rm_list_ids[%d]=%d\n", i, opts->rm_list.ids[i]);
 737         MPTCP_ADD_STATS(sock_net(sk), MPTCP_MIB_RMADDRTX, opts->rm_list.nr);
 738         return true;
 739 }
 740
 741 static bool mptcp_established_options_mp_prio(struct sock *sk,
 742                                               unsigned int *size,
 743                                               unsigned int remaining,
 744                                               struct mptcp_out_options *opts)
 745 {
 746         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 747
 748         /* can't send MP_PRIO with MPC, as they share the same option space:
 749          * 'backup'. Also it makes no sense at all
 750          */
 751         if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC))
 752                 return false;
 753
 754         /* account for the trailing 'nop' option */
 755         if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
 756                 return false;
 757
 758         *size = TCPOLEN_MPTCP_PRIO_ALIGN;
 759         opts->suboptions |= OPTION_MPTCP_PRIO;
 760         opts->backup = subflow->request_bkup;
 761
 762         pr_debug("prio=%d\n", opts->backup);
 763
 764         return true;
 765 }
 766
 767 static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
 768                                                    unsigned int *size,
 769                                                    unsigned int remaining,
 770                                                    struct mptcp_out_options *opts)
 771 {
 772         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 773
 774         if (remaining < TCPOLEN_MPTCP_RST)
 775                 return false;
 776
 777         *size = TCPOLEN_MPTCP_RST;
 778         opts->suboptions |= OPTION_MPTCP_RST;
 779         opts->reset_transient = subflow->reset_transient;
 780         opts->reset_reason = subflow->reset_reason;
 781         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTTX);
 782
 783         return true;
 784 }
 785
 786 static bool mptcp_established_options_fastclose(struct sock *sk,
 787                                                 unsigned int *size,
 788                                                 unsigned int remaining,
 789                                                 struct mptcp_out_options *opts)
 790 {
 791         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 792         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 793
 794         if (likely(!subflow->send_fastclose))
 795                 return false;
 796
 797         if (remaining < TCPOLEN_MPTCP_FASTCLOSE)
 798                 return false;
 799
 800         *size = TCPOLEN_MPTCP_FASTCLOSE;
 801         opts->suboptions |= OPTION_MPTCP_FASTCLOSE;
 802         opts->rcvr_key = READ_ONCE(msk->remote_key);
 803
 804         pr_debug("FASTCLOSE key=%llu\n", opts->rcvr_key);
 805         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSETX);
 806         return true;
 807 }
 808
 809 static bool mptcp_established_options_mp_fail(struct sock *sk,
 810                                               unsigned int *size,
 811                                               unsigned int remaining,
 812                                               struct mptcp_out_options *opts)
 813 {
 814         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 815
 816         if (likely(!subflow->send_mp_fail))
 817                 return false;
 818
 819         if (remaining < TCPOLEN_MPTCP_FAIL)
 820                 return false;
 821
 822         *size = TCPOLEN_MPTCP_FAIL;
 823         opts->suboptions |= OPTION_MPTCP_FAIL;
 824         opts->fail_seq = subflow->map_seq;
 825
 826         pr_debug("MP_FAIL fail_seq=%llu\n", opts->fail_seq);
 827         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILTX);
 828
 829         return true;
 830 }
 831
 832 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
 833                                unsigned int *size, unsigned int remaining,
 834                                struct mptcp_out_options *opts)
 835 {
 836         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 837         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 838         unsigned int opt_size = 0;
 839         bool snd_data_fin;
 840         bool ret = false;
 841
 842         opts->suboptions = 0;
 843
 844         if (unlikely(__mptcp_check_fallback(msk) && !mptcp_check_infinite_map(skb)))
 845                 return false;
 846
 847         if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
 848                 if (mptcp_established_options_fastclose(sk, &opt_size, remaining, opts) ||
 849                     mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) {
 850                         *size += opt_size;
 851                         remaining -= opt_size;
 852                 }
 853                 /* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */
 854                 if (mptcp_established_options_rst(sk, skb, &opt_size, remaining, opts)) {
 855                         *size += opt_size;
 856                         remaining -= opt_size;
 857                 }
 858                 return true;
 859         }
 860
 861         snd_data_fin = mptcp_data_fin_enabled(msk);
 862         if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, opts))
 863                 ret = true;
 864         else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, opts)) {
 865                 unsigned int mp_fail_size;
 866
 867                 ret = true;
 868                 if (mptcp_established_options_mp_fail(sk, &mp_fail_size,
 869                                                       remaining - opt_size, opts)) {
 870                         *size += opt_size + mp_fail_size;
 871                         remaining -= opt_size - mp_fail_size;
 872                         return true;
 873                 }
 874         }
 875
 876         /* we reserved enough space for the above options, and exceeding the
 877          * TCP option space would be fatal
 878          */
 879         if (WARN_ON_ONCE(opt_size > remaining))
 880                 return false;
 881
 882         *size += opt_size;
 883         remaining -= opt_size;
 884         if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
 885                 *size += opt_size;
 886                 remaining -= opt_size;
 887                 ret = true;
 888         } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
 889                 *size += opt_size;
 890                 remaining -= opt_size;
 891                 ret = true;
 892         }
 893
 894         if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
 895                 *size += opt_size;
 896                 remaining -= opt_size;
 897                 ret = true;
 898         }
 899
 900         return ret;
 901 }
 902
 903 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
 904                           struct mptcp_out_options *opts)
 905 {
 906         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 907
 908         if (subflow_req->mp_capable) {
 909                 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
 910                 opts->sndr_key = subflow_req->local_key;
 911                 opts->csum_reqd = subflow_req->csum_reqd;
 912                 opts->allow_join_id0 = subflow_req->allow_join_id0;
 913                 *size = TCPOLEN_MPTCP_MPC_SYNACK;
 914                 pr_debug("subflow_req=%p, local_key=%llu\n",
 915                          subflow_req, subflow_req->local_key);
 916                 return true;
 917         } else if (subflow_req->mp_join) {
 918                 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
 919                 opts->backup = subflow_req->request_bkup;
 920                 opts->join_id = subflow_req->local_id;
 921                 opts->thmac = subflow_req->thmac;
 922                 opts->nonce = subflow_req->local_nonce;
 923                 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u\n",
 924                          subflow_req, opts->backup, opts->join_id,
 925                          opts->thmac, opts->nonce);
 926                 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
 927                 return true;
 928         }
 929         return false;
 930 }
 931
 932 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
 933                                     struct mptcp_subflow_context *subflow,
 934                                     struct sk_buff *skb,
 935                                     struct mptcp_options_received *mp_opt)
 936 {
 937         /* here we can process OoO, in-window pkts, only in-sequence 4th ack
 938          * will make the subflow fully established
 939          */
 940         if (likely(READ_ONCE(subflow->fully_established))) {
 941                 /* on passive sockets, check for 3rd ack retransmission
 942                  * note that msk is always set by subflow_syn_recv_sock()
 943                  * for mp_join subflows
 944                  */
 945                 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
 946                     TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
 947                     subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) &&
 948                     !subflow->request_join)
 949                         tcp_send_ack(ssk);
 950                 goto check_notify;
 951         }
 952
 953         /* we must process OoO packets before the first subflow is fully
 954          * established. OoO packets are instead a protocol violation
 955          * for MP_JOIN subflows as the peer must not send any data
 956          * before receiving the forth ack - cfr. RFC 8684 section 3.2.
 957          */
 958         if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
 959                 if (subflow->mp_join)
 960                         goto reset;
 961                 if (subflow->is_mptfo && mp_opt->suboptions & OPTION_MPTCP_MPC_ACK)
 962                         goto set_fully_established;
 963                 return subflow->mp_capable;
 964         }
 965
 966         if (subflow->remote_key_valid &&
 967             (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) ||
 968              ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) &&
 969               (!mp_opt->echo || subflow->mp_join)))) {
 970                 /* subflows are fully established as soon as we get any
 971                  * additional ack, including ADD_ADDR.
 972                  */
 973                 goto set_fully_established;
 974         }
 975
 976         /* If the first established packet does not contain MP_CAPABLE + data
 977          * then fallback to TCP. Fallback scenarios requires a reset for
 978          * MP_JOIN subflows.
 979          */
 980         if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) {
 981                 if (subflow->mp_join)
 982                         goto reset;
 983                 subflow->mp_capable = 0;
 984                 pr_fallback(msk);
 985                 mptcp_do_fallback(ssk);
 986                 return false;
 987         }
 988
 989         if (mp_opt->deny_join_id0)
 990                 WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
 991
 992         if (unlikely(!READ_ONCE(msk->pm.server_side)))
 993                 pr_warn_once("bogus mpc option on established client sk");
 994
 995 set_fully_established:
 996         mptcp_data_lock((struct sock *)msk);
 997         __mptcp_subflow_fully_established(msk, subflow, mp_opt);
 998         mptcp_data_unlock((struct sock *)msk);
 999
1000 check_notify:
1001         /* if the subflow is not already linked into the conn_list, we can't
1002          * notify the PM: this subflow is still on the listener queue
1003          * and the PM possibly acquiring the subflow lock could race with
1004          * the listener close
1005          */
1006         if (likely(subflow->pm_notified) || list_empty(&subflow->node))
1007                 return true;
1008
1009         subflow->pm_notified = 1;
1010         if (subflow->mp_join) {
1011                 clear_3rdack_retransmission(ssk);
1012                 mptcp_pm_subflow_established(msk);
1013         } else {
1014                 mptcp_pm_fully_established(msk, ssk);
1015         }
1016         return true;
1017
1018 reset:
1019         mptcp_subflow_reset(ssk);
1020         return false;
1021 }
1022
1023 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq)
1024 {
1025         u32 old_seq32, cur_seq32;
1026
1027         old_seq32 = (u32)old_seq;
1028         cur_seq32 = (u32)cur_seq;
1029         cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32;
1030         if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32)))
1031                 return cur_seq + (1LL << 32);
1032
1033         /* reverse wrap could happen, too */
1034         if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32)))
1035                 return cur_seq - (1LL << 32);
1036         return cur_seq;
1037 }
1038
1039 static void __mptcp_snd_una_update(struct mptcp_sock *msk, u64 new_snd_una)
1040 {
1041         msk->bytes_acked += new_snd_una - msk->snd_una;
1042         WRITE_ONCE(msk->snd_una, new_snd_una);
1043 }
1044
1045 static void ack_update_msk(struct mptcp_sock *msk,
1046                            struct sock *ssk,
1047                            struct mptcp_options_received *mp_opt)
1048 {
1049         u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
1050         struct sock *sk = (struct sock *)msk;
1051         u64 old_snd_una;
1052
1053         mptcp_data_lock(sk);
1054
1055         /* avoid ack expansion on update conflict, to reduce the risk of
1056          * wrongly expanding to a future ack sequence number, which is way
1057          * more dangerous than missing an ack
1058          */
1059         old_snd_una = msk->snd_una;
1060         new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
1061
1062         /* ACK for data not even sent yet? Ignore.*/
1063         if (unlikely(after64(new_snd_una, snd_nxt)))
1064                 new_snd_una = old_snd_una;
1065
1066         new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
1067
1068         if (after64(new_wnd_end, msk->wnd_end))
1069                 WRITE_ONCE(msk->wnd_end, new_wnd_end);
1070
1071         /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
1072         if (after64(msk->wnd_end, snd_nxt))
1073                 __mptcp_check_push(sk, ssk);
1074
1075         if (after64(new_snd_una, old_snd_una)) {
1076                 __mptcp_snd_una_update(msk, new_snd_una);
1077                 __mptcp_data_acked(sk);
1078         }
1079         msk->last_ack_recv = tcp_jiffies32;
1080         mptcp_data_unlock(sk);
1081
1082         trace_ack_update_msk(mp_opt->data_ack,
1083                              old_snd_una, new_snd_una,
1084                              new_wnd_end, READ_ONCE(msk->wnd_end));
1085 }
1086
1087 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
1088 {
1089         /* Skip if DATA_FIN was already received.
1090          * If updating simultaneously with the recvmsg loop, values
1091          * should match. If they mismatch, the peer is misbehaving and
1092          * we will prefer the most recent information.
1093          */
1094         if (READ_ONCE(msk->rcv_data_fin))
1095                 return false;
1096
1097         WRITE_ONCE(msk->rcv_data_fin_seq,
1098                    mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
1099         WRITE_ONCE(msk->rcv_data_fin, 1);
1100
1101         return true;
1102 }
1103
1104 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
1105                                 struct mptcp_options_received *mp_opt)
1106 {
1107         u64 hmac = 0;
1108
1109         if (mp_opt->echo)
1110                 return true;
1111
1112         hmac = add_addr_generate_hmac(READ_ONCE(msk->remote_key),
1113                                       READ_ONCE(msk->local_key),
1114                                       &mp_opt->addr);
1115
1116         pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
1117                  msk, hmac, mp_opt->ahmac);
1118
1119         return hmac == mp_opt->ahmac;
1120 }
1121
1122 /* Return false if a subflow has been reset, else return true */
1123 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
1124 {
1125         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1126         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1127         struct mptcp_options_received mp_opt;
1128         struct mptcp_ext *mpext;
1129
1130         if (__mptcp_check_fallback(msk)) {
1131                 /* Keep it simple and unconditionally trigger send data cleanup and
1132                  * pending queue spooling. We will need to acquire the data lock
1133                  * for more accurate checks, and once the lock is acquired, such
1134                  * helpers are cheap.
1135                  */
1136                 mptcp_data_lock(subflow->conn);
1137                 if (sk_stream_memory_free(sk))
1138                         __mptcp_check_push(subflow->conn, sk);
1139
1140                 /* on fallback we just need to ignore the msk-level snd_una, as
1141                  * this is really plain TCP
1142                  */
1143                 __mptcp_snd_una_update(msk, READ_ONCE(msk->snd_nxt));
1144
1145                 __mptcp_data_acked(subflow->conn);
1146                 mptcp_data_unlock(subflow->conn);
1147                 return true;
1148         }
1149
1150         mptcp_get_options(skb, &mp_opt);
1151
1152         /* The subflow can be in close state only if check_fully_established()
1153          * just sent a reset. If so, tell the caller to ignore the current packet.
1154          */
1155         if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1156                 return sk->sk_state != TCP_CLOSE;
1157
1158         if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) {
1159                 if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) &&
1160                     READ_ONCE(msk->local_key) == mp_opt.rcvr_key) {
1161                         WRITE_ONCE(msk->rcv_fastclose, true);
1162                         mptcp_schedule_work((struct sock *)msk);
1163                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFASTCLOSERX);
1164                 }
1165
1166                 if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) &&
1167                     add_addr_hmac_valid(msk, &mp_opt)) {
1168                         if (!mp_opt.echo) {
1169                                 mptcp_pm_add_addr_received(sk, &mp_opt.addr);
1170                                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1171                         } else {
1172                                 mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1173                                 mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1174                                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1175                         }
1176
1177                         if (mp_opt.addr.port)
1178                                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1179                 }
1180
1181                 if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR)
1182                         mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1183
1184                 if (mp_opt.suboptions & OPTION_MPTCP_PRIO) {
1185                         mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1186                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1187                 }
1188
1189                 if (mp_opt.suboptions & OPTION_MPTCP_FAIL) {
1190                         mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq);
1191                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX);
1192                 }
1193
1194                 if (mp_opt.suboptions & OPTION_MPTCP_RST) {
1195                         subflow->reset_seen = 1;
1196                         subflow->reset_reason = mp_opt.reset_reason;
1197                         subflow->reset_transient = mp_opt.reset_transient;
1198                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPRSTRX);
1199                 }
1200
1201                 if (!(mp_opt.suboptions & OPTION_MPTCP_DSS))
1202                         return true;
1203         }
1204
1205         /* we can't wait for recvmsg() to update the ack_seq, otherwise
1206          * monodirectional flows will stuck
1207          */
1208         if (mp_opt.use_ack)
1209                 ack_update_msk(msk, sk, &mp_opt);
1210
1211         /* Zero-data-length packets are dropped by the caller and not
1212          * propagated to the MPTCP layer, so the skb extension does not
1213          * need to be allocated or populated. DATA_FIN information, if
1214          * present, needs to be updated here before the skb is freed.
1215          */
1216         if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1217                 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1218                     mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64))
1219                         mptcp_schedule_work((struct sock *)msk);
1220
1221                 return true;
1222         }
1223
1224         mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1225         if (!mpext)
1226                 return true;
1227
1228         memset(mpext, 0, sizeof(*mpext));
1229
1230         if (likely(mp_opt.use_map)) {
1231                 if (mp_opt.mpc_map) {
1232                         /* this is an MP_CAPABLE carrying MPTCP data
1233                          * we know this map the first chunk of data
1234                          */
1235                         mptcp_crypto_key_sha(subflow->remote_key, NULL,
1236                                              &mpext->data_seq);
1237                         mpext->data_seq++;
1238                         mpext->subflow_seq = 1;
1239                         mpext->dsn64 = 1;
1240                         mpext->mpc_map = 1;
1241                         mpext->data_fin = 0;
1242                 } else {
1243                         mpext->data_seq = mp_opt.data_seq;
1244                         mpext->subflow_seq = mp_opt.subflow_seq;
1245                         mpext->dsn64 = mp_opt.dsn64;
1246                         mpext->data_fin = mp_opt.data_fin;
1247                 }
1248                 mpext->data_len = mp_opt.data_len;
1249                 mpext->use_map = 1;
1250                 mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD);
1251
1252                 if (mpext->csum_reqd)
1253                         mpext->csum = mp_opt.csum;
1254         }
1255
1256         return true;
1257 }
1258
1259 static void mptcp_set_rwin(struct tcp_sock *tp, struct tcphdr *th)
1260 {
1261         const struct sock *ssk = (const struct sock *)tp;
1262         struct mptcp_subflow_context *subflow;
1263         u64 ack_seq, rcv_wnd_old, rcv_wnd_new;
1264         struct mptcp_sock *msk;
1265         u32 new_win;
1266         u64 win;
1267
1268         subflow = mptcp_subflow_ctx(ssk);
1269         msk = mptcp_sk(subflow->conn);
1270
1271         ack_seq = READ_ONCE(msk->ack_seq);
1272         rcv_wnd_new = ack_seq + tp->rcv_wnd;
1273
1274         rcv_wnd_old = atomic64_read(&msk->rcv_wnd_sent);
1275         if (after64(rcv_wnd_new, rcv_wnd_old)) {
1276                 u64 rcv_wnd;
1277
1278                 for (;;) {
1279                         rcv_wnd = atomic64_cmpxchg(&msk->rcv_wnd_sent, rcv_wnd_old, rcv_wnd_new);
1280
1281                         if (rcv_wnd == rcv_wnd_old)
1282                                 break;
1283
1284                         rcv_wnd_old = rcv_wnd;
1285                         if (before64(rcv_wnd_new, rcv_wnd_old)) {
1286                                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICTUPDATE);
1287                                 goto raise_win;
1288                         }
1289                         MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDCONFLICT);
1290                 }
1291                 return;
1292         }
1293
1294         if (rcv_wnd_new != rcv_wnd_old) {
1295 raise_win:
1296                 win = rcv_wnd_old - ack_seq;
1297                 tp->rcv_wnd = min_t(u64, win, U32_MAX);
1298                 new_win = tp->rcv_wnd;
1299
1300                 /* Make sure we do not exceed the maximum possible
1301                  * scaled window.
1302                  */
1303                 if (unlikely(th->syn))
1304                         new_win = min(new_win, 65535U) << tp->rx_opt.rcv_wscale;
1305                 if (!tp->rx_opt.rcv_wscale &&
1306                     READ_ONCE(sock_net(ssk)->ipv4.sysctl_tcp_workaround_signed_windows))
1307                         new_win = min(new_win, MAX_TCP_WINDOW);
1308                 else
1309                         new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale));
1310
1311                 /* RFC1323 scaling applied */
1312                 new_win >>= tp->rx_opt.rcv_wscale;
1313                 th->window = htons(new_win);
1314                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_RCVWNDSHARED);
1315         }
1316 }
1317
1318 __sum16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum)
1319 {
1320         struct csum_pseudo_header header;
1321         __wsum csum;
1322
1323         /* cfr RFC 8684 3.3.1.:
1324          * the data sequence number used in the pseudo-header is
1325          * always the 64-bit value, irrespective of what length is used in the
1326          * DSS option itself.
1327          */
1328         header.data_seq = cpu_to_be64(data_seq);
1329         header.subflow_seq = htonl(subflow_seq);
1330         header.data_len = htons(data_len);
1331         header.csum = 0;
1332
1333         csum = csum_partial(&header, sizeof(header), sum);
1334         return csum_fold(csum);
1335 }
1336
1337 static __sum16 mptcp_make_csum(const struct mptcp_ext *mpext)
1338 {
1339         return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1340                                  ~csum_unfold(mpext->csum));
1341 }
1342
1343 static void put_len_csum(u16 len, __sum16 csum, void *data)
1344 {
1345         __sum16 *sumptr = data + 2;
1346         __be16 *ptr = data;
1347
1348         put_unaligned_be16(len, ptr);
1349
1350         put_unaligned(csum, sumptr);
1351 }
1352
1353 void mptcp_write_options(struct tcphdr *th, __be32 *ptr, struct tcp_sock *tp,
1354                          struct mptcp_out_options *opts)
1355 {
1356         const struct sock *ssk = (const struct sock *)tp;
1357         struct mptcp_subflow_context *subflow;
1358
1359         /* Which options can be used together?
1360          *
1361          * X: mutually exclusive
1362          * O: often used together
1363          * C: can be used together in some cases
1364          * P: could be used together but we prefer not to (optimisations)
1365          *
1366          *  Opt: | MPC  | MPJ  | DSS  | ADD  |  RM  | PRIO | FAIL |  FC  |
1367          * ------|------|------|------|------|------|------|------|------|
1368          *  MPC  |------|------|------|------|------|------|------|------|
1369          *  MPJ  |  X   |------|------|------|------|------|------|------|
1370          *  DSS  |  X   |  X   |------|------|------|------|------|------|
1371          *  ADD  |  X   |  X   |  P   |------|------|------|------|------|
1372          *  RM   |  C   |  C   |  C   |  P   |------|------|------|------|
1373          *  PRIO |  X   |  C   |  C   |  C   |  C   |------|------|------|
1374          *  FAIL |  X   |  X   |  C   |  X   |  X   |  X   |------|------|
1375          *  FC   |  X   |  X   |  X   |  X   |  X   |  X   |  X   |------|
1376          *  RST  |  X   |  X   |  X   |  X   |  X   |  X   |  O   |  O   |
1377          * ------|------|------|------|------|------|------|------|------|
1378          *
1379          * The same applies in mptcp_established_options() function.
1380          */
1381         if (likely(OPTION_MPTCP_DSS & opts->suboptions)) {
1382                 struct mptcp_ext *mpext = &opts->ext_copy;
1383                 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1384                 u8 flags = 0;
1385
1386                 if (mpext->use_ack) {
1387                         flags = MPTCP_DSS_HAS_ACK;
1388                         if (mpext->ack64) {
1389                                 len += TCPOLEN_MPTCP_DSS_ACK64;
1390                                 flags |= MPTCP_DSS_ACK64;
1391                         } else {
1392                                 len += TCPOLEN_MPTCP_DSS_ACK32;
1393                         }
1394                 }
1395
1396                 if (mpext->use_map) {
1397                         len += TCPOLEN_MPTCP_DSS_MAP64;
1398
1399                         /* Use only 64-bit mapping flags for now, add
1400                          * support for optional 32-bit mappings later.
1401                          */
1402                         flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1403                         if (mpext->data_fin)
1404                                 flags |= MPTCP_DSS_DATA_FIN;
1405
1406                         if (opts->csum_reqd)
1407                                 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1408                 }
1409
1410                 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1411
1412                 if (mpext->use_ack) {
1413                         if (mpext->ack64) {
1414                                 put_unaligned_be64(mpext->data_ack, ptr);
1415                                 ptr += 2;
1416                         } else {
1417                                 put_unaligned_be32(mpext->data_ack32, ptr);
1418                                 ptr += 1;
1419                         }
1420                 }
1421
1422                 if (mpext->use_map) {
1423                         put_unaligned_be64(mpext->data_seq, ptr);
1424                         ptr += 2;
1425                         put_unaligned_be32(mpext->subflow_seq, ptr);
1426                         ptr += 1;
1427                         if (opts->csum_reqd) {
1428                                 /* data_len == 0 is reserved for the infinite mapping,
1429                                  * the checksum will also be set to 0.
1430                                  */
1431                                 put_len_csum(mpext->data_len,
1432                                              (mpext->data_len ? mptcp_make_csum(mpext) : 0),
1433                                              ptr);
1434                         } else {
1435                                 put_unaligned_be32(mpext->data_len << 16 |
1436                                                    TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1437                         }
1438                         ptr += 1;
1439                 }
1440
1441                 /* We might need to add MP_FAIL options in rare cases */
1442                 if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions))
1443                         goto mp_fail;
1444         } else if (OPTIONS_MPTCP_MPC & opts->suboptions) {
1445                 u8 len, flag = MPTCP_CAP_HMAC_SHA256;
1446
1447                 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) {
1448                         len = TCPOLEN_MPTCP_MPC_SYN;
1449                 } else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) {
1450                         len = TCPOLEN_MPTCP_MPC_SYNACK;
1451                 } else if (opts->data_len) {
1452                         len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1453                         if (opts->csum_reqd)
1454                                 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1455                 } else {
1456                         len = TCPOLEN_MPTCP_MPC_ACK;
1457                 }
1458
1459                 if (opts->csum_reqd)
1460                         flag |= MPTCP_CAP_CHECKSUM_REQD;
1461
1462                 if (!opts->allow_join_id0)
1463                         flag |= MPTCP_CAP_DENY_JOIN_ID0;
1464
1465                 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1466                                       MPTCP_SUPPORTED_VERSION,
1467                                       flag);
1468
1469                 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1470                     opts->suboptions))
1471                         goto mp_capable_done;
1472
1473                 put_unaligned_be64(opts->sndr_key, ptr);
1474                 ptr += 2;
1475                 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1476                         goto mp_capable_done;
1477
1478                 put_unaligned_be64(opts->rcvr_key, ptr);
1479                 ptr += 2;
1480                 if (!opts->data_len)
1481                         goto mp_capable_done;
1482
1483                 if (opts->csum_reqd) {
1484                         put_len_csum(opts->data_len,
1485                                      __mptcp_make_csum(opts->data_seq,
1486                                                        opts->subflow_seq,
1487                                                        opts->data_len,
1488                                                        ~csum_unfold(opts->csum)),
1489                                      ptr);
1490                 } else {
1491                         put_unaligned_be32(opts->data_len << 16 |
1492                                            TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1493                 }
1494                 ptr += 1;
1495
1496                 /* MPC is additionally mutually exclusive with MP_PRIO */
1497                 goto mp_capable_done;
1498         } else if (OPTIONS_MPTCP_MPJ & opts->suboptions) {
1499                 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1500                         *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1501                                               TCPOLEN_MPTCP_MPJ_SYN,
1502                                               opts->backup, opts->join_id);
1503                         put_unaligned_be32(opts->token, ptr);
1504                         ptr += 1;
1505                         put_unaligned_be32(opts->nonce, ptr);
1506                         ptr += 1;
1507                 } else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1508                         *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1509                                               TCPOLEN_MPTCP_MPJ_SYNACK,
1510                                               opts->backup, opts->join_id);
1511                         put_unaligned_be64(opts->thmac, ptr);
1512                         ptr += 2;
1513                         put_unaligned_be32(opts->nonce, ptr);
1514                         ptr += 1;
1515                 } else {
1516                         *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1517                                               TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1518                         memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1519                         ptr += 5;
1520                 }
1521         } else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1522                 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1523                 u8 echo = MPTCP_ADDR_ECHO;
1524
1525 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1526                 if (opts->addr.family == AF_INET6)
1527                         len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1528 #endif
1529
1530                 if (opts->addr.port)
1531                         len += TCPOLEN_MPTCP_PORT_LEN;
1532
1533                 if (opts->ahmac) {
1534                         len += sizeof(opts->ahmac);
1535                         echo = 0;
1536                 }
1537
1538                 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1539                                       len, echo, opts->addr.id);
1540                 if (opts->addr.family == AF_INET) {
1541                         memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1542                         ptr += 1;
1543                 }
1544 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1545                 else if (opts->addr.family == AF_INET6) {
1546                         memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1547                         ptr += 4;
1548                 }
1549 #endif
1550
1551                 if (!opts->addr.port) {
1552                         if (opts->ahmac) {
1553                                 put_unaligned_be64(opts->ahmac, ptr);
1554                                 ptr += 2;
1555                         }
1556                 } else {
1557                         u16 port = ntohs(opts->addr.port);
1558
1559                         if (opts->ahmac) {
1560                                 u8 *bptr = (u8 *)ptr;
1561
1562                                 put_unaligned_be16(port, bptr);
1563                                 bptr += 2;
1564                                 put_unaligned_be64(opts->ahmac, bptr);
1565                                 bptr += 8;
1566                                 put_unaligned_be16(TCPOPT_NOP << 8 |
1567                                                    TCPOPT_NOP, bptr);
1568
1569                                 ptr += 3;
1570                         } else {
1571                                 put_unaligned_be32(port << 16 |
1572                                                    TCPOPT_NOP << 8 |
1573                                                    TCPOPT_NOP, ptr);
1574                                 ptr += 1;
1575                         }
1576                 }
1577         } else if (unlikely(OPTION_MPTCP_FASTCLOSE & opts->suboptions)) {
1578                 /* FASTCLOSE is mutually exclusive with others except RST */
1579                 *ptr++ = mptcp_option(MPTCPOPT_MP_FASTCLOSE,
1580                                       TCPOLEN_MPTCP_FASTCLOSE,
1581                                       0, 0);
1582                 put_unaligned_be64(opts->rcvr_key, ptr);
1583                 ptr += 2;
1584
1585                 if (OPTION_MPTCP_RST & opts->suboptions)
1586                         goto mp_rst;
1587                 return;
1588         } else if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) {
1589 mp_fail:
1590                 /* MP_FAIL is mutually exclusive with others except RST */
1591                 subflow = mptcp_subflow_ctx(ssk);
1592                 subflow->send_mp_fail = 0;
1593
1594                 *ptr++ = mptcp_option(MPTCPOPT_MP_FAIL,
1595                                       TCPOLEN_MPTCP_FAIL,
1596                                       0, 0);
1597                 put_unaligned_be64(opts->fail_seq, ptr);
1598                 ptr += 2;
1599
1600                 if (OPTION_MPTCP_RST & opts->suboptions)
1601                         goto mp_rst;
1602                 return;
1603         } else if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) {
1604 mp_rst:
1605                 *ptr++ = mptcp_option(MPTCPOPT_RST,
1606                                       TCPOLEN_MPTCP_RST,
1607                                       opts->reset_transient,
1608                                       opts->reset_reason);
1609                 return;
1610         }
1611
1612         if (OPTION_MPTCP_PRIO & opts->suboptions) {
1613                 subflow = mptcp_subflow_ctx(ssk);
1614                 subflow->send_mp_prio = 0;
1615
1616                 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1617                                       TCPOLEN_MPTCP_PRIO,
1618                                       opts->backup, TCPOPT_NOP);
1619
1620                 MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MPPRIOTX);
1621         }
1622
1623 mp_capable_done:
1624         if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1625                 u8 i = 1;
1626
1627                 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1628                                       TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1629                                       0, opts->rm_list.ids[0]);
1630
1631                 while (i < opts->rm_list.nr) {
1632                         u8 id1, id2, id3, id4;
1633
1634                         id1 = opts->rm_list.ids[i];
1635                         id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1636                         id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1637                         id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1638                         put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1639                         ptr += 1;
1640                         i += 4;
1641                 }
1642         }
1643
1644         if (tp)
1645                 mptcp_set_rwin(tp, th);
1646 }
1647
1648 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1649 {
1650         const struct mptcp_ext *ext = mptcp_get_ext(skb);
1651         u8 flags, reason;
1652
1653         if (ext) {
1654                 flags = ext->reset_transient;
1655                 reason = ext->reset_reason;
1656
1657                 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1658                                     flags, reason);
1659         }
1660
1661         return htonl(0u);
1662 }
1663 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);