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/sha.h>
  11 #include <net/tcp.h>
  12 #include <net/mptcp.h>
  13 #include "protocol.h"
  14
  15 static bool mptcp_cap_flag_sha256(u8 flags)
  16 {
  17         return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256;
  18 }
  19
  20 static void mptcp_parse_option(const struct sk_buff *skb,
  21                                const unsigned char *ptr, int opsize,
  22                                struct mptcp_options_received *mp_opt)
  23 {
  24         u8 subtype = *ptr >> 4;
  25         int expected_opsize;
  26         u8 version;
  27         u8 flags;
  28
  29         switch (subtype) {
  30         case MPTCPOPT_MP_CAPABLE:
  31                 /* strict size checking */
  32                 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
  33                         if (skb->len > tcp_hdr(skb)->doff << 2)
  34                                 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA;
  35                         else
  36                                 expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
  37                 } else {
  38                         if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)
  39                                 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
  40                         else
  41                                 expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
  42                 }
  43                 if (opsize != expected_opsize)
  44                         break;
  45
  46                 /* try to be gentle vs future versions on the initial syn */
  47                 version = *ptr++ & MPTCP_VERSION_MASK;
  48                 if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
  49                         if (version != MPTCP_SUPPORTED_VERSION)
  50                                 break;
  51                 } else if (version < MPTCP_SUPPORTED_VERSION) {
  52                         break;
  53                 }
  54
  55                 flags = *ptr++;
  56                 if (!mptcp_cap_flag_sha256(flags) ||
  57                     (flags & MPTCP_CAP_EXTENSIBILITY))
  58                         break;
  59
  60                 /* RFC 6824, Section 3.1:
  61                  * "For the Checksum Required bit (labeled "A"), if either
  62                  * host requires the use of checksums, checksums MUST be used.
  63                  * In other words, the only way for checksums not to be used
  64                  * is if both hosts in their SYNs set A=0."
  65                  *
  66                  * Section 3.3.0:
  67                  * "If a checksum is not present when its use has been
  68                  * negotiated, the receiver MUST close the subflow with a RST as
  69                  * it is considered broken."
  70                  *
  71                  * We don't implement DSS checksum - fall back to TCP.
  72                  */
  73                 if (flags & MPTCP_CAP_CHECKSUM_REQD)
  74                         break;
  75
  76                 mp_opt->mp_capable = 1;
  77                 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
  78                         mp_opt->sndr_key = get_unaligned_be64(ptr);
  79                         ptr += 8;
  80                 }
  81                 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
  82                         mp_opt->rcvr_key = get_unaligned_be64(ptr);
  83                         ptr += 8;
  84                 }
  85                 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA) {
  86                         /* Section 3.1.:
  87                          * "the data parameters in a MP_CAPABLE are semantically
  88                          * equivalent to those in a DSS option and can be used
  89                          * interchangeably."
  90                          */
  91                         mp_opt->dss = 1;
  92                         mp_opt->use_map = 1;
  93                         mp_opt->mpc_map = 1;
  94                         mp_opt->data_len = get_unaligned_be16(ptr);
  95                         ptr += 2;
  96                 }
  97                 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d",
  98                          version, flags, opsize, mp_opt->sndr_key,
  99                          mp_opt->rcvr_key, mp_opt->data_len);
 100                 break;
 101
 102         case MPTCPOPT_MP_JOIN:
 103                 mp_opt->mp_join = 1;
 104                 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
 105                         mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
 106                         mp_opt->join_id = *ptr++;
 107                         mp_opt->token = get_unaligned_be32(ptr);
 108                         ptr += 4;
 109                         mp_opt->nonce = get_unaligned_be32(ptr);
 110                         ptr += 4;
 111                         pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
 112                                  mp_opt->backup, mp_opt->join_id,
 113                                  mp_opt->token, mp_opt->nonce);
 114                 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
 115                         mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
 116                         mp_opt->join_id = *ptr++;
 117                         mp_opt->thmac = get_unaligned_be64(ptr);
 118                         ptr += 8;
 119                         mp_opt->nonce = get_unaligned_be32(ptr);
 120                         ptr += 4;
 121                         pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
 122                                  mp_opt->backup, mp_opt->join_id,
 123                                  mp_opt->thmac, mp_opt->nonce);
 124                 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
 125                         ptr += 2;
 126                         memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
 127                         pr_debug("MP_JOIN hmac");
 128                 } else {
 129                         pr_warn("MP_JOIN bad option size");
 130                         mp_opt->mp_join = 0;
 131                 }
 132                 break;
 133
 134         case MPTCPOPT_DSS:
 135                 pr_debug("DSS");
 136                 ptr++;
 137
 138                 /* we must clear 'mpc_map' be able to detect MP_CAPABLE
 139                  * map vs DSS map in mptcp_incoming_options(), and reconstruct
 140                  * map info accordingly
 141                  */
 142                 mp_opt->mpc_map = 0;
 143                 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
 144                 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
 145                 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
 146                 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
 147                 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
 148                 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
 149
 150                 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
 151                          mp_opt->data_fin, mp_opt->dsn64,
 152                          mp_opt->use_map, mp_opt->ack64,
 153                          mp_opt->use_ack);
 154
 155                 expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
 156
 157                 if (mp_opt->use_ack) {
 158                         if (mp_opt->ack64)
 159                                 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
 160                         else
 161                                 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
 162                 }
 163
 164                 if (mp_opt->use_map) {
 165                         if (mp_opt->dsn64)
 166                                 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
 167                         else
 168                                 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
 169                 }
 170
 171                 /* RFC 6824, Section 3.3:
 172                  * If a checksum is present, but its use had
 173                  * not been negotiated in the MP_CAPABLE handshake,
 174                  * the checksum field MUST be ignored.
 175                  */
 176                 if (opsize != expected_opsize &&
 177                     opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
 178                         break;
 179
 180                 mp_opt->dss = 1;
 181
 182                 if (mp_opt->use_ack) {
 183                         if (mp_opt->ack64) {
 184                                 mp_opt->data_ack = get_unaligned_be64(ptr);
 185                                 ptr += 8;
 186                         } else {
 187                                 mp_opt->data_ack = get_unaligned_be32(ptr);
 188                                 ptr += 4;
 189                         }
 190
 191                         pr_debug("data_ack=%llu", mp_opt->data_ack);
 192                 }
 193
 194                 if (mp_opt->use_map) {
 195                         if (mp_opt->dsn64) {
 196                                 mp_opt->data_seq = get_unaligned_be64(ptr);
 197                                 ptr += 8;
 198                         } else {
 199                                 mp_opt->data_seq = get_unaligned_be32(ptr);
 200                                 ptr += 4;
 201                         }
 202
 203                         mp_opt->subflow_seq = get_unaligned_be32(ptr);
 204                         ptr += 4;
 205
 206                         mp_opt->data_len = get_unaligned_be16(ptr);
 207                         ptr += 2;
 208
 209                         pr_debug("data_seq=%llu subflow_seq=%u data_len=%u",
 210                                  mp_opt->data_seq, mp_opt->subflow_seq,
 211                                  mp_opt->data_len);
 212                 }
 213
 214                 break;
 215
 216         case MPTCPOPT_ADD_ADDR:
 217                 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
 218                 if (!mp_opt->echo) {
 219                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
 220                             opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
 221                                 mp_opt->family = MPTCP_ADDR_IPVERSION_4;
 222 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 223                         else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
 224                                  opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
 225                                 mp_opt->family = MPTCP_ADDR_IPVERSION_6;
 226 #endif
 227                         else
 228                                 break;
 229                 } else {
 230                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
 231                             opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
 232                                 mp_opt->family = MPTCP_ADDR_IPVERSION_4;
 233 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 234                         else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
 235                                  opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
 236                                 mp_opt->family = MPTCP_ADDR_IPVERSION_6;
 237 #endif
 238                         else
 239                                 break;
 240                 }
 241
 242                 mp_opt->add_addr = 1;
 243                 mp_opt->port = 0;
 244                 mp_opt->addr_id = *ptr++;
 245                 pr_debug("ADD_ADDR: id=%d", mp_opt->addr_id);
 246                 if (mp_opt->family == MPTCP_ADDR_IPVERSION_4) {
 247                         memcpy((u8 *)&mp_opt->addr.s_addr, (u8 *)ptr, 4);
 248                         ptr += 4;
 249                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
 250                             opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
 251                                 mp_opt->port = get_unaligned_be16(ptr);
 252                                 ptr += 2;
 253                         }
 254                 }
 255 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 256                 else {
 257                         memcpy(mp_opt->addr6.s6_addr, (u8 *)ptr, 16);
 258                         ptr += 16;
 259                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
 260                             opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
 261                                 mp_opt->port = get_unaligned_be16(ptr);
 262                                 ptr += 2;
 263                         }
 264                 }
 265 #endif
 266                 if (!mp_opt->echo) {
 267                         mp_opt->ahmac = get_unaligned_be64(ptr);
 268                         ptr += 8;
 269                 }
 270                 break;
 271
 272         case MPTCPOPT_RM_ADDR:
 273                 if (opsize != TCPOLEN_MPTCP_RM_ADDR_BASE)
 274                         break;
 275
 276                 ptr++;
 277
 278                 mp_opt->rm_addr = 1;
 279                 mp_opt->rm_id = *ptr++;
 280                 pr_debug("RM_ADDR: id=%d", mp_opt->rm_id);
 281                 break;
 282
 283         default:
 284                 break;
 285         }
 286 }
 287
 288 void mptcp_get_options(const struct sk_buff *skb,
 289                        struct mptcp_options_received *mp_opt)
 290 {
 291         const struct tcphdr *th = tcp_hdr(skb);
 292         const unsigned char *ptr;
 293         int length;
 294
 295         /* initialize option status */
 296         mp_opt->mp_capable = 0;
 297         mp_opt->mp_join = 0;
 298         mp_opt->add_addr = 0;
 299         mp_opt->rm_addr = 0;
 300         mp_opt->dss = 0;
 301
 302         length = (th->doff * 4) - sizeof(struct tcphdr);
 303         ptr = (const unsigned char *)(th + 1);
 304
 305         while (length > 0) {
 306                 int opcode = *ptr++;
 307                 int opsize;
 308
 309                 switch (opcode) {
 310                 case TCPOPT_EOL:
 311                         return;
 312                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 313                         length--;
 314                         continue;
 315                 default:
 316                         opsize = *ptr++;
 317                         if (opsize < 2) /* "silly options" */
 318                                 return;
 319                         if (opsize > length)
 320                                 return; /* don't parse partial options */
 321                         if (opcode == TCPOPT_MPTCP)
 322                                 mptcp_parse_option(skb, ptr, opsize, mp_opt);
 323                         ptr += opsize - 2;
 324                         length -= opsize;
 325                 }
 326         }
 327 }
 328
 329 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
 330                        unsigned int *size, struct mptcp_out_options *opts)
 331 {
 332         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 333
 334         /* we will use snd_isn to detect first pkt [re]transmission
 335          * in mptcp_established_options_mp()
 336          */
 337         subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
 338         if (subflow->request_mptcp) {
 339                 opts->suboptions = OPTION_MPTCP_MPC_SYN;
 340                 *size = TCPOLEN_MPTCP_MPC_SYN;
 341                 return true;
 342         } else if (subflow->request_join) {
 343                 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
 344                          subflow->local_nonce);
 345                 opts->suboptions = OPTION_MPTCP_MPJ_SYN;
 346                 opts->join_id = subflow->local_id;
 347                 opts->token = subflow->remote_token;
 348                 opts->nonce = subflow->local_nonce;
 349                 opts->backup = subflow->request_bkup;
 350                 *size = TCPOLEN_MPTCP_MPJ_SYN;
 351                 return true;
 352         }
 353         return false;
 354 }
 355
 356 /* MP_JOIN client subflow must wait for 4th ack before sending any data:
 357  * TCP can't schedule delack timer before the subflow is fully established.
 358  * MPTCP uses the delack timer to do 3rd ack retransmissions
 359  */
 360 static void schedule_3rdack_retransmission(struct sock *sk)
 361 {
 362         struct inet_connection_sock *icsk = inet_csk(sk);
 363         struct tcp_sock *tp = tcp_sk(sk);
 364         unsigned long timeout;
 365
 366         /* reschedule with a timeout above RTT, as we must look only for drop */
 367         if (tp->srtt_us)
 368                 timeout = tp->srtt_us << 1;
 369         else
 370                 timeout = TCP_TIMEOUT_INIT;
 371
 372         WARN_ON_ONCE(icsk->icsk_ack.pending & ICSK_ACK_TIMER);
 373         icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
 374         icsk->icsk_ack.timeout = timeout;
 375         sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
 376 }
 377
 378 static void clear_3rdack_retransmission(struct sock *sk)
 379 {
 380         struct inet_connection_sock *icsk = inet_csk(sk);
 381
 382         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 383         icsk->icsk_ack.timeout = 0;
 384         icsk->icsk_ack.ato = 0;
 385         icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
 386 }
 387
 388 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
 389                                          unsigned int *size,
 390                                          unsigned int remaining,
 391                                          struct mptcp_out_options *opts)
 392 {
 393         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 394         struct mptcp_ext *mpext;
 395         unsigned int data_len;
 396
 397         /* When skb is not available, we better over-estimate the emitted
 398          * options len. A full DSS option (28 bytes) is longer than
 399          * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
 400          * tell the caller to defer the estimate to
 401          * mptcp_established_options_dss(), which will reserve enough space.
 402          */
 403         if (!skb)
 404                 return false;
 405
 406         /* MPC/MPJ needed only on 3rd ack packet */
 407         if (subflow->fully_established ||
 408             subflow->snd_isn != TCP_SKB_CB(skb)->seq)
 409                 return false;
 410
 411         if (subflow->mp_capable) {
 412                 mpext = mptcp_get_ext(skb);
 413                 data_len = mpext ? mpext->data_len : 0;
 414
 415                 /* we will check ext_copy.data_len in mptcp_write_options() to
 416                  * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
 417                  * TCPOLEN_MPTCP_MPC_ACK
 418                  */
 419                 opts->ext_copy.data_len = data_len;
 420                 opts->suboptions = OPTION_MPTCP_MPC_ACK;
 421                 opts->sndr_key = subflow->local_key;
 422                 opts->rcvr_key = subflow->remote_key;
 423
 424                 /* Section 3.1.
 425                  * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
 426                  * packets that start the first subflow of an MPTCP connection,
 427                  * as well as the first packet that carries data
 428                  */
 429                 if (data_len > 0)
 430                         *size = ALIGN(TCPOLEN_MPTCP_MPC_ACK_DATA, 4);
 431                 else
 432                         *size = TCPOLEN_MPTCP_MPC_ACK;
 433
 434                 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
 435                          subflow, subflow->local_key, subflow->remote_key,
 436                          data_len);
 437
 438                 return true;
 439         } else if (subflow->mp_join) {
 440                 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
 441                 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
 442                 *size = TCPOLEN_MPTCP_MPJ_ACK;
 443                 pr_debug("subflow=%p", subflow);
 444
 445                 schedule_3rdack_retransmission(sk);
 446                 return true;
 447         }
 448         return false;
 449 }
 450
 451 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
 452                                  struct sk_buff *skb, struct mptcp_ext *ext)
 453 {
 454         /* The write_seq value has already been incremented, so the actual
 455          * sequence number for the DATA_FIN is one less.
 456          */
 457         u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
 458
 459         if (!ext->use_map || !skb->len) {
 460                 /* RFC6824 requires a DSS mapping with specific values
 461                  * if DATA_FIN is set but no data payload is mapped
 462                  */
 463                 ext->data_fin = 1;
 464                 ext->use_map = 1;
 465                 ext->dsn64 = 1;
 466                 ext->data_seq = data_fin_tx_seq;
 467                 ext->subflow_seq = 0;
 468                 ext->data_len = 1;
 469         } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
 470                 /* If there's an existing DSS mapping and it is the
 471                  * final mapping, DATA_FIN consumes 1 additional byte of
 472                  * mapping space.
 473                  */
 474                 ext->data_fin = 1;
 475                 ext->data_len++;
 476         }
 477 }
 478
 479 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
 480                                           unsigned int *size,
 481                                           unsigned int remaining,
 482                                           struct mptcp_out_options *opts)
 483 {
 484         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 485         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 486         unsigned int dss_size = 0;
 487         u64 snd_data_fin_enable;
 488         struct mptcp_ext *mpext;
 489         unsigned int ack_size;
 490         bool ret = false;
 491
 492         mpext = skb ? mptcp_get_ext(skb) : NULL;
 493         snd_data_fin_enable = READ_ONCE(msk->snd_data_fin_enable);
 494
 495         if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
 496                 unsigned int map_size;
 497
 498                 map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
 499
 500                 remaining -= map_size;
 501                 dss_size = map_size;
 502                 if (mpext)
 503                         opts->ext_copy = *mpext;
 504
 505                 if (skb && snd_data_fin_enable)
 506                         mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
 507                 ret = true;
 508         }
 509
 510         /* passive sockets msk will set the 'can_ack' after accept(), even
 511          * if the first subflow may have the already the remote key handy
 512          */
 513         opts->ext_copy.use_ack = 0;
 514         if (!READ_ONCE(msk->can_ack)) {
 515                 *size = ALIGN(dss_size, 4);
 516                 return ret;
 517         }
 518
 519         if (READ_ONCE(msk->use_64bit_ack)) {
 520                 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
 521                 opts->ext_copy.data_ack = READ_ONCE(msk->ack_seq);
 522                 opts->ext_copy.ack64 = 1;
 523         } else {
 524                 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
 525                 opts->ext_copy.data_ack32 = (uint32_t)READ_ONCE(msk->ack_seq);
 526                 opts->ext_copy.ack64 = 0;
 527         }
 528         opts->ext_copy.use_ack = 1;
 529
 530         /* Add kind/length/subtype/flag overhead if mapping is not populated */
 531         if (dss_size == 0)
 532                 ack_size += TCPOLEN_MPTCP_DSS_BASE;
 533
 534         dss_size += ack_size;
 535
 536         *size = ALIGN(dss_size, 4);
 537         return true;
 538 }
 539
 540 static u64 add_addr_generate_hmac(u64 key1, u64 key2, u8 addr_id,
 541                                   struct in_addr *addr)
 542 {
 543         u8 hmac[SHA256_DIGEST_SIZE];
 544         u8 msg[7];
 545
 546         msg[0] = addr_id;
 547         memcpy(&msg[1], &addr->s_addr, 4);
 548         msg[5] = 0;
 549         msg[6] = 0;
 550
 551         mptcp_crypto_hmac_sha(key1, key2, msg, 7, hmac);
 552
 553         return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
 554 }
 555
 556 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 557 static u64 add_addr6_generate_hmac(u64 key1, u64 key2, u8 addr_id,
 558                                    struct in6_addr *addr)
 559 {
 560         u8 hmac[SHA256_DIGEST_SIZE];
 561         u8 msg[19];
 562
 563         msg[0] = addr_id;
 564         memcpy(&msg[1], &addr->s6_addr, 16);
 565         msg[17] = 0;
 566         msg[18] = 0;
 567
 568         mptcp_crypto_hmac_sha(key1, key2, msg, 19, hmac);
 569
 570         return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
 571 }
 572 #endif
 573
 574 static bool mptcp_established_options_addr(struct sock *sk,
 575                                            unsigned int *size,
 576                                            unsigned int remaining,
 577                                            struct mptcp_out_options *opts)
 578 {
 579         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 580         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 581         struct mptcp_addr_info saddr;
 582         int len;
 583
 584         if (!mptcp_pm_should_signal(msk) ||
 585             !(mptcp_pm_addr_signal(msk, remaining, &saddr)))
 586                 return false;
 587
 588         len = mptcp_add_addr_len(saddr.family);
 589         if (remaining < len)
 590                 return false;
 591
 592         *size = len;
 593         opts->addr_id = saddr.id;
 594         if (saddr.family == AF_INET) {
 595                 opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
 596                 opts->addr = saddr.addr;
 597                 opts->ahmac = add_addr_generate_hmac(msk->local_key,
 598                                                      msk->remote_key,
 599                                                      opts->addr_id,
 600                                                      &opts->addr);
 601         }
 602 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 603         else if (saddr.family == AF_INET6) {
 604                 opts->suboptions |= OPTION_MPTCP_ADD_ADDR6;
 605                 opts->addr6 = saddr.addr6;
 606                 opts->ahmac = add_addr6_generate_hmac(msk->local_key,
 607                                                       msk->remote_key,
 608                                                       opts->addr_id,
 609                                                       &opts->addr6);
 610         }
 611 #endif
 612         pr_debug("addr_id=%d, ahmac=%llu", opts->addr_id, opts->ahmac);
 613
 614         return true;
 615 }
 616
 617 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
 618                                unsigned int *size, unsigned int remaining,
 619                                struct mptcp_out_options *opts)
 620 {
 621         unsigned int opt_size = 0;
 622         bool ret = false;
 623
 624         opts->suboptions = 0;
 625
 626         if (unlikely(mptcp_check_fallback(sk)))
 627                 return false;
 628
 629         /* prevent adding of any MPTCP related options on reset packet
 630          * until we support MP_TCPRST/MP_FASTCLOSE
 631          */
 632         if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
 633                 return false;
 634
 635         if (mptcp_established_options_mp(sk, skb, &opt_size, remaining, opts))
 636                 ret = true;
 637         else if (mptcp_established_options_dss(sk, skb, &opt_size, remaining,
 638                                                opts))
 639                 ret = true;
 640
 641         /* we reserved enough space for the above options, and exceeding the
 642          * TCP option space would be fatal
 643          */
 644         if (WARN_ON_ONCE(opt_size > remaining))
 645                 return false;
 646
 647         *size += opt_size;
 648         remaining -= opt_size;
 649         if (mptcp_established_options_addr(sk, &opt_size, remaining, opts)) {
 650                 *size += opt_size;
 651                 remaining -= opt_size;
 652                 ret = true;
 653         }
 654
 655         return ret;
 656 }
 657
 658 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
 659                           struct mptcp_out_options *opts)
 660 {
 661         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 662
 663         if (subflow_req->mp_capable) {
 664                 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
 665                 opts->sndr_key = subflow_req->local_key;
 666                 *size = TCPOLEN_MPTCP_MPC_SYNACK;
 667                 pr_debug("subflow_req=%p, local_key=%llu",
 668                          subflow_req, subflow_req->local_key);
 669                 return true;
 670         } else if (subflow_req->mp_join) {
 671                 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
 672                 opts->backup = subflow_req->backup;
 673                 opts->join_id = subflow_req->local_id;
 674                 opts->thmac = subflow_req->thmac;
 675                 opts->nonce = subflow_req->local_nonce;
 676                 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
 677                          subflow_req, opts->backup, opts->join_id,
 678                          opts->thmac, opts->nonce);
 679                 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
 680                 return true;
 681         }
 682         return false;
 683 }
 684
 685 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
 686                                     struct mptcp_subflow_context *subflow,
 687                                     struct sk_buff *skb,
 688                                     struct mptcp_options_received *mp_opt)
 689 {
 690         /* here we can process OoO, in-window pkts, only in-sequence 4th ack
 691          * will make the subflow fully established
 692          */
 693         if (likely(subflow->fully_established)) {
 694                 /* on passive sockets, check for 3rd ack retransmission
 695                  * note that msk is always set by subflow_syn_recv_sock()
 696                  * for mp_join subflows
 697                  */
 698                 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
 699                     TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
 700                     subflow->mp_join && mp_opt->mp_join &&
 701                     READ_ONCE(msk->pm.server_side))
 702                         tcp_send_ack(ssk);
 703                 goto fully_established;
 704         }
 705
 706         /* we must process OoO packets before the first subflow is fully
 707          * established. OoO packets are instead a protocol violation
 708          * for MP_JOIN subflows as the peer must not send any data
 709          * before receiving the forth ack - cfr. RFC 8684 section 3.2.
 710          */
 711         if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
 712                 if (subflow->mp_join)
 713                         goto reset;
 714                 return subflow->mp_capable;
 715         }
 716
 717         if (mp_opt->dss && mp_opt->use_ack) {
 718                 /* subflows are fully established as soon as we get any
 719                  * additional ack.
 720                  */
 721                 subflow->fully_established = 1;
 722                 WRITE_ONCE(msk->fully_established, true);
 723                 goto fully_established;
 724         }
 725
 726         /* If the first established packet does not contain MP_CAPABLE + data
 727          * then fallback to TCP. Fallback scenarios requires a reset for
 728          * MP_JOIN subflows.
 729          */
 730         if (!mp_opt->mp_capable) {
 731                 if (subflow->mp_join)
 732                         goto reset;
 733                 subflow->mp_capable = 0;
 734                 pr_fallback(msk);
 735                 __mptcp_do_fallback(msk);
 736                 return false;
 737         }
 738
 739         if (unlikely(!READ_ONCE(msk->pm.server_side)))
 740                 pr_warn_once("bogus mpc option on established client sk");
 741         mptcp_subflow_fully_established(subflow, mp_opt);
 742
 743 fully_established:
 744         if (likely(subflow->pm_notified))
 745                 return true;
 746
 747         subflow->pm_notified = 1;
 748         if (subflow->mp_join) {
 749                 clear_3rdack_retransmission(ssk);
 750                 mptcp_pm_subflow_established(msk, subflow);
 751         } else {
 752                 mptcp_pm_fully_established(msk);
 753         }
 754         return true;
 755
 756 reset:
 757         mptcp_subflow_reset(ssk);
 758         return false;
 759 }
 760
 761 static u64 expand_ack(u64 old_ack, u64 cur_ack, bool use_64bit)
 762 {
 763         u32 old_ack32, cur_ack32;
 764
 765         if (use_64bit)
 766                 return cur_ack;
 767
 768         old_ack32 = (u32)old_ack;
 769         cur_ack32 = (u32)cur_ack;
 770         cur_ack = (old_ack & GENMASK_ULL(63, 32)) + cur_ack32;
 771         if (unlikely(before(cur_ack32, old_ack32)))
 772                 return cur_ack + (1LL << 32);
 773         return cur_ack;
 774 }
 775
 776 static void update_una(struct mptcp_sock *msk,
 777                        struct mptcp_options_received *mp_opt)
 778 {
 779         u64 new_snd_una, snd_una, old_snd_una = atomic64_read(&msk->snd_una);
 780         u64 write_seq = READ_ONCE(msk->write_seq);
 781
 782         /* avoid ack expansion on update conflict, to reduce the risk of
 783          * wrongly expanding to a future ack sequence number, which is way
 784          * more dangerous than missing an ack
 785          */
 786         new_snd_una = expand_ack(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
 787
 788         /* ACK for data not even sent yet? Ignore. */
 789         if (after64(new_snd_una, write_seq))
 790                 new_snd_una = old_snd_una;
 791
 792         while (after64(new_snd_una, old_snd_una)) {
 793                 snd_una = old_snd_una;
 794                 old_snd_una = atomic64_cmpxchg(&msk->snd_una, snd_una,
 795                                                new_snd_una);
 796                 if (old_snd_una == snd_una) {
 797                         mptcp_data_acked((struct sock *)msk);
 798                         break;
 799                 }
 800         }
 801 }
 802
 803 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
 804 {
 805         /* Skip if DATA_FIN was already received.
 806          * If updating simultaneously with the recvmsg loop, values
 807          * should match. If they mismatch, the peer is misbehaving and
 808          * we will prefer the most recent information.
 809          */
 810         if (READ_ONCE(msk->rcv_data_fin) || !READ_ONCE(msk->first))
 811                 return false;
 812
 813         WRITE_ONCE(msk->rcv_data_fin_seq,
 814                    expand_ack(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
 815         WRITE_ONCE(msk->rcv_data_fin, 1);
 816
 817         return true;
 818 }
 819
 820 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
 821                                 struct mptcp_options_received *mp_opt)
 822 {
 823         u64 hmac = 0;
 824
 825         if (mp_opt->echo)
 826                 return true;
 827
 828         if (mp_opt->family == MPTCP_ADDR_IPVERSION_4)
 829                 hmac = add_addr_generate_hmac(msk->remote_key,
 830                                               msk->local_key,
 831                                               mp_opt->addr_id, &mp_opt->addr);
 832 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 833         else
 834                 hmac = add_addr6_generate_hmac(msk->remote_key,
 835                                                msk->local_key,
 836                                                mp_opt->addr_id, &mp_opt->addr6);
 837 #endif
 838
 839         pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
 840                  msk, (unsigned long long)hmac,
 841                  (unsigned long long)mp_opt->ahmac);
 842
 843         return hmac == mp_opt->ahmac;
 844 }
 845
 846 void mptcp_incoming_options(struct sock *sk, struct sk_buff *skb,
 847                             struct tcp_options_received *opt_rx)
 848 {
 849         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 850         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 851         struct mptcp_options_received mp_opt;
 852         struct mptcp_ext *mpext;
 853
 854         if (__mptcp_check_fallback(msk))
 855                 return;
 856
 857         mptcp_get_options(skb, &mp_opt);
 858         if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
 859                 return;
 860
 861         if (mp_opt.add_addr && add_addr_hmac_valid(msk, &mp_opt)) {
 862                 struct mptcp_addr_info addr;
 863
 864                 addr.port = htons(mp_opt.port);
 865                 addr.id = mp_opt.addr_id;
 866                 if (mp_opt.family == MPTCP_ADDR_IPVERSION_4) {
 867                         addr.family = AF_INET;
 868                         addr.addr = mp_opt.addr;
 869                 }
 870 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 871                 else if (mp_opt.family == MPTCP_ADDR_IPVERSION_6) {
 872                         addr.family = AF_INET6;
 873                         addr.addr6 = mp_opt.addr6;
 874                 }
 875 #endif
 876                 if (!mp_opt.echo)
 877                         mptcp_pm_add_addr_received(msk, &addr);
 878                 mp_opt.add_addr = 0;
 879         }
 880
 881         if (!mp_opt.dss)
 882                 return;
 883
 884         /* we can't wait for recvmsg() to update the ack_seq, otherwise
 885          * monodirectional flows will stuck
 886          */
 887         if (mp_opt.use_ack)
 888                 update_una(msk, &mp_opt);
 889
 890         /* Zero-data-length packets are dropped by the caller and not
 891          * propagated to the MPTCP layer, so the skb extension does not
 892          * need to be allocated or populated. DATA_FIN information, if
 893          * present, needs to be updated here before the skb is freed.
 894          */
 895         if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
 896                 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
 897                     mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
 898                     schedule_work(&msk->work))
 899                         sock_hold(subflow->conn);
 900
 901                 return;
 902         }
 903
 904         mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
 905         if (!mpext)
 906                 return;
 907
 908         memset(mpext, 0, sizeof(*mpext));
 909
 910         if (mp_opt.use_map) {
 911                 if (mp_opt.mpc_map) {
 912                         /* this is an MP_CAPABLE carrying MPTCP data
 913                          * we know this map the first chunk of data
 914                          */
 915                         mptcp_crypto_key_sha(subflow->remote_key, NULL,
 916                                              &mpext->data_seq);
 917                         mpext->data_seq++;
 918                         mpext->subflow_seq = 1;
 919                         mpext->dsn64 = 1;
 920                         mpext->mpc_map = 1;
 921                         mpext->data_fin = 0;
 922                 } else {
 923                         mpext->data_seq = mp_opt.data_seq;
 924                         mpext->subflow_seq = mp_opt.subflow_seq;
 925                         mpext->dsn64 = mp_opt.dsn64;
 926                         mpext->data_fin = mp_opt.data_fin;
 927                 }
 928                 mpext->data_len = mp_opt.data_len;
 929                 mpext->use_map = 1;
 930         }
 931 }
 932
 933 void mptcp_write_options(__be32 *ptr, struct mptcp_out_options *opts)
 934 {
 935         if ((OPTION_MPTCP_MPC_SYN | OPTION_MPTCP_MPC_SYNACK |
 936              OPTION_MPTCP_MPC_ACK) & opts->suboptions) {
 937                 u8 len;
 938
 939                 if (OPTION_MPTCP_MPC_SYN & opts->suboptions)
 940                         len = TCPOLEN_MPTCP_MPC_SYN;
 941                 else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions)
 942                         len = TCPOLEN_MPTCP_MPC_SYNACK;
 943                 else if (opts->ext_copy.data_len)
 944                         len = TCPOLEN_MPTCP_MPC_ACK_DATA;
 945                 else
 946                         len = TCPOLEN_MPTCP_MPC_ACK;
 947
 948                 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
 949                                       MPTCP_SUPPORTED_VERSION,
 950                                       MPTCP_CAP_HMAC_SHA256);
 951
 952                 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
 953                     opts->suboptions))
 954                         goto mp_capable_done;
 955
 956                 put_unaligned_be64(opts->sndr_key, ptr);
 957                 ptr += 2;
 958                 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
 959                         goto mp_capable_done;
 960
 961                 put_unaligned_be64(opts->rcvr_key, ptr);
 962                 ptr += 2;
 963                 if (!opts->ext_copy.data_len)
 964                         goto mp_capable_done;
 965
 966                 put_unaligned_be32(opts->ext_copy.data_len << 16 |
 967                                    TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
 968                 ptr += 1;
 969         }
 970
 971 mp_capable_done:
 972         if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
 973                 if (opts->ahmac)
 974                         *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
 975                                               TCPOLEN_MPTCP_ADD_ADDR, 0,
 976                                               opts->addr_id);
 977                 else
 978                         *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
 979                                               TCPOLEN_MPTCP_ADD_ADDR_BASE,
 980                                               MPTCP_ADDR_ECHO,
 981                                               opts->addr_id);
 982                 memcpy((u8 *)ptr, (u8 *)&opts->addr.s_addr, 4);
 983                 ptr += 1;
 984                 if (opts->ahmac) {
 985                         put_unaligned_be64(opts->ahmac, ptr);
 986                         ptr += 2;
 987                 }
 988         }
 989
 990 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 991         if (OPTION_MPTCP_ADD_ADDR6 & opts->suboptions) {
 992                 if (opts->ahmac)
 993                         *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
 994                                               TCPOLEN_MPTCP_ADD_ADDR6, 0,
 995                                               opts->addr_id);
 996                 else
 997                         *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
 998                                               TCPOLEN_MPTCP_ADD_ADDR6_BASE,
 999                                               MPTCP_ADDR_ECHO,
1000                                               opts->addr_id);
1001                 memcpy((u8 *)ptr, opts->addr6.s6_addr, 16);
1002                 ptr += 4;
1003                 if (opts->ahmac) {
1004                         put_unaligned_be64(opts->ahmac, ptr);
1005                         ptr += 2;
1006                 }
1007         }
1008 #endif
1009
1010         if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1011                 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1012                                       TCPOLEN_MPTCP_RM_ADDR_BASE,
1013                                       0, opts->rm_id);
1014         }
1015
1016         if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1017                 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1018                                       TCPOLEN_MPTCP_MPJ_SYN,
1019                                       opts->backup, opts->join_id);
1020                 put_unaligned_be32(opts->token, ptr);
1021                 ptr += 1;
1022                 put_unaligned_be32(opts->nonce, ptr);
1023                 ptr += 1;
1024         }
1025
1026         if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1027                 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1028                                       TCPOLEN_MPTCP_MPJ_SYNACK,
1029                                       opts->backup, opts->join_id);
1030                 put_unaligned_be64(opts->thmac, ptr);
1031                 ptr += 2;
1032                 put_unaligned_be32(opts->nonce, ptr);
1033                 ptr += 1;
1034         }
1035
1036         if (OPTION_MPTCP_MPJ_ACK & opts->suboptions) {
1037                 *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1038                                       TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1039                 memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1040                 ptr += 5;
1041         }
1042
1043         if (opts->ext_copy.use_ack || opts->ext_copy.use_map) {
1044                 struct mptcp_ext *mpext = &opts->ext_copy;
1045                 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1046                 u8 flags = 0;
1047
1048                 if (mpext->use_ack) {
1049                         flags = MPTCP_DSS_HAS_ACK;
1050                         if (mpext->ack64) {
1051                                 len += TCPOLEN_MPTCP_DSS_ACK64;
1052                                 flags |= MPTCP_DSS_ACK64;
1053                         } else {
1054                                 len += TCPOLEN_MPTCP_DSS_ACK32;
1055                         }
1056                 }
1057
1058                 if (mpext->use_map) {
1059                         len += TCPOLEN_MPTCP_DSS_MAP64;
1060
1061                         /* Use only 64-bit mapping flags for now, add
1062                          * support for optional 32-bit mappings later.
1063                          */
1064                         flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1065                         if (mpext->data_fin)
1066                                 flags |= MPTCP_DSS_DATA_FIN;
1067                 }
1068
1069                 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1070
1071                 if (mpext->use_ack) {
1072                         if (mpext->ack64) {
1073                                 put_unaligned_be64(mpext->data_ack, ptr);
1074                                 ptr += 2;
1075                         } else {
1076                                 put_unaligned_be32(mpext->data_ack32, ptr);
1077                                 ptr += 1;
1078                         }
1079                 }
1080
1081                 if (mpext->use_map) {
1082                         put_unaligned_be64(mpext->data_seq, ptr);
1083                         ptr += 2;
1084                         put_unaligned_be32(mpext->subflow_seq, ptr);
1085                         ptr += 1;
1086                         put_unaligned_be32(mpext->data_len << 16 |
1087                                            TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1088                 }
1089         }
1090 }