net/mptcp/options.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Multipath TCP
   3  *
   4  * Copyright (c) 2017 - 2019, Intel Corporation.
   5  */
   6
   7 #define pr_fmt(fmt) "MPTCP: " fmt
   8
   9 #include <linux/kernel.h>
  10 #include <crypto/sha2.h>
  11 #include <net/tcp.h>
  12 #include <net/mptcp.h>
  13 #include "protocol.h"
  14 #include "mib.h"
  15
  16 #include <trace/events/mptcp.h>
  17
  18 static bool mptcp_cap_flag_sha256(u8 flags)
  19 {
  20         return (flags & MPTCP_CAP_FLAG_MASK) == MPTCP_CAP_HMAC_SHA256;
  21 }
  22
  23 static void mptcp_parse_option(const struct sk_buff *skb,
  24                                const unsigned char *ptr, int opsize,
  25                                struct mptcp_options_received *mp_opt)
  26 {
  27         u8 subtype = *ptr >> 4;
  28         int expected_opsize;
  29         u8 version;
  30         u8 flags;
  31         u8 i;
  32
  33         switch (subtype) {
  34         case MPTCPOPT_MP_CAPABLE:
  35                 /* strict size checking */
  36                 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
  37                         if (skb->len > tcp_hdr(skb)->doff << 2)
  38                                 expected_opsize = TCPOLEN_MPTCP_MPC_ACK_DATA;
  39                         else
  40                                 expected_opsize = TCPOLEN_MPTCP_MPC_ACK;
  41                 } else {
  42                         if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)
  43                                 expected_opsize = TCPOLEN_MPTCP_MPC_SYNACK;
  44                         else
  45                                 expected_opsize = TCPOLEN_MPTCP_MPC_SYN;
  46                 }
  47
  48                 /* Cfr RFC 8684 Section 3.3.0:
  49                  * If a checksum is present but its use had
  50                  * not been negotiated in the MP_CAPABLE handshake, the receiver MUST
  51                  * close the subflow with a RST, as it is not behaving as negotiated.
  52                  * If a checksum is not present when its use has been negotiated, the
  53                  * receiver MUST close the subflow with a RST, as it is considered
  54                  * broken
  55                  * We parse even option with mismatching csum presence, so that
  56                  * later in subflow_data_ready we can trigger the reset.
  57                  */
  58                 if (opsize != expected_opsize &&
  59                     (expected_opsize != TCPOLEN_MPTCP_MPC_ACK_DATA ||
  60                      opsize != TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM))
  61                         break;
  62
  63                 /* try to be gentle vs future versions on the initial syn */
  64                 version = *ptr++ & MPTCP_VERSION_MASK;
  65                 if (opsize != TCPOLEN_MPTCP_MPC_SYN) {
  66                         if (version != MPTCP_SUPPORTED_VERSION)
  67                                 break;
  68                 } else if (version < MPTCP_SUPPORTED_VERSION) {
  69                         break;
  70                 }
  71
  72                 flags = *ptr++;
  73                 if (!mptcp_cap_flag_sha256(flags) ||
  74                     (flags & MPTCP_CAP_EXTENSIBILITY))
  75                         break;
  76
  77                 /* RFC 6824, Section 3.1:
  78                  * "For the Checksum Required bit (labeled "A"), if either
  79                  * host requires the use of checksums, checksums MUST be used.
  80                  * In other words, the only way for checksums not to be used
  81                  * is if both hosts in their SYNs set A=0."
  82                  */
  83                 if (flags & MPTCP_CAP_CHECKSUM_REQD)
  84                         mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
  85
  86                 mp_opt->deny_join_id0 = !!(flags & MPTCP_CAP_DENY_JOIN_ID0);
  87
  88                 mp_opt->suboptions |= OPTIONS_MPTCP_MPC;
  89                 if (opsize >= TCPOLEN_MPTCP_MPC_SYNACK) {
  90                         mp_opt->sndr_key = get_unaligned_be64(ptr);
  91                         ptr += 8;
  92                 }
  93                 if (opsize >= TCPOLEN_MPTCP_MPC_ACK) {
  94                         mp_opt->rcvr_key = get_unaligned_be64(ptr);
  95                         ptr += 8;
  96                 }
  97                 if (opsize >= TCPOLEN_MPTCP_MPC_ACK_DATA) {
  98                         /* Section 3.1.:
  99                          * "the data parameters in a MP_CAPABLE are semantically
 100                          * equivalent to those in a DSS option and can be used
 101                          * interchangeably."
 102                          */
 103                         mp_opt->suboptions |= OPTION_MPTCP_DSS;
 104                         mp_opt->use_map = 1;
 105                         mp_opt->mpc_map = 1;
 106                         mp_opt->data_len = get_unaligned_be16(ptr);
 107                         ptr += 2;
 108                 }
 109                 if (opsize == TCPOLEN_MPTCP_MPC_ACK_DATA_CSUM) {
 110                         mp_opt->csum = (__force __sum16)get_unaligned_be16(ptr);
 111                         mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
 112                         ptr += 2;
 113                 }
 114                 pr_debug("MP_CAPABLE version=%x, flags=%x, optlen=%d sndr=%llu, rcvr=%llu len=%d csum=%u",
 115                          version, flags, opsize, mp_opt->sndr_key,
 116                          mp_opt->rcvr_key, mp_opt->data_len, mp_opt->csum);
 117                 break;
 118
 119         case MPTCPOPT_MP_JOIN:
 120                 mp_opt->suboptions |= OPTIONS_MPTCP_MPJ;
 121                 if (opsize == TCPOLEN_MPTCP_MPJ_SYN) {
 122                         mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
 123                         mp_opt->join_id = *ptr++;
 124                         mp_opt->token = get_unaligned_be32(ptr);
 125                         ptr += 4;
 126                         mp_opt->nonce = get_unaligned_be32(ptr);
 127                         ptr += 4;
 128                         pr_debug("MP_JOIN bkup=%u, id=%u, token=%u, nonce=%u",
 129                                  mp_opt->backup, mp_opt->join_id,
 130                                  mp_opt->token, mp_opt->nonce);
 131                 } else if (opsize == TCPOLEN_MPTCP_MPJ_SYNACK) {
 132                         mp_opt->backup = *ptr++ & MPTCPOPT_BACKUP;
 133                         mp_opt->join_id = *ptr++;
 134                         mp_opt->thmac = get_unaligned_be64(ptr);
 135                         ptr += 8;
 136                         mp_opt->nonce = get_unaligned_be32(ptr);
 137                         ptr += 4;
 138                         pr_debug("MP_JOIN bkup=%u, id=%u, thmac=%llu, nonce=%u",
 139                                  mp_opt->backup, mp_opt->join_id,
 140                                  mp_opt->thmac, mp_opt->nonce);
 141                 } else if (opsize == TCPOLEN_MPTCP_MPJ_ACK) {
 142                         ptr += 2;
 143                         memcpy(mp_opt->hmac, ptr, MPTCPOPT_HMAC_LEN);
 144                         pr_debug("MP_JOIN hmac");
 145                 } else {
 146                         mp_opt->suboptions &= ~OPTIONS_MPTCP_MPJ;
 147                 }
 148                 break;
 149
 150         case MPTCPOPT_DSS:
 151                 pr_debug("DSS");
 152                 ptr++;
 153
 154                 /* we must clear 'mpc_map' be able to detect MP_CAPABLE
 155                  * map vs DSS map in mptcp_incoming_options(), and reconstruct
 156                  * map info accordingly
 157                  */
 158                 mp_opt->mpc_map = 0;
 159                 flags = (*ptr++) & MPTCP_DSS_FLAG_MASK;
 160                 mp_opt->data_fin = (flags & MPTCP_DSS_DATA_FIN) != 0;
 161                 mp_opt->dsn64 = (flags & MPTCP_DSS_DSN64) != 0;
 162                 mp_opt->use_map = (flags & MPTCP_DSS_HAS_MAP) != 0;
 163                 mp_opt->ack64 = (flags & MPTCP_DSS_ACK64) != 0;
 164                 mp_opt->use_ack = (flags & MPTCP_DSS_HAS_ACK);
 165
 166                 pr_debug("data_fin=%d dsn64=%d use_map=%d ack64=%d use_ack=%d",
 167                          mp_opt->data_fin, mp_opt->dsn64,
 168                          mp_opt->use_map, mp_opt->ack64,
 169                          mp_opt->use_ack);
 170
 171                 expected_opsize = TCPOLEN_MPTCP_DSS_BASE;
 172
 173                 if (mp_opt->use_ack) {
 174                         if (mp_opt->ack64)
 175                                 expected_opsize += TCPOLEN_MPTCP_DSS_ACK64;
 176                         else
 177                                 expected_opsize += TCPOLEN_MPTCP_DSS_ACK32;
 178                 }
 179
 180                 if (mp_opt->use_map) {
 181                         if (mp_opt->dsn64)
 182                                 expected_opsize += TCPOLEN_MPTCP_DSS_MAP64;
 183                         else
 184                                 expected_opsize += TCPOLEN_MPTCP_DSS_MAP32;
 185                 }
 186
 187                 /* Always parse any csum presence combination, we will enforce
 188                  * RFC 8684 Section 3.3.0 checks later in subflow_data_ready
 189                  */
 190                 if (opsize != expected_opsize &&
 191                     opsize != expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM)
 192                         break;
 193
 194                 mp_opt->suboptions |= OPTION_MPTCP_DSS;
 195                 if (mp_opt->use_ack) {
 196                         if (mp_opt->ack64) {
 197                                 mp_opt->data_ack = get_unaligned_be64(ptr);
 198                                 ptr += 8;
 199                         } else {
 200                                 mp_opt->data_ack = get_unaligned_be32(ptr);
 201                                 ptr += 4;
 202                         }
 203
 204                         pr_debug("data_ack=%llu", mp_opt->data_ack);
 205                 }
 206
 207                 if (mp_opt->use_map) {
 208                         if (mp_opt->dsn64) {
 209                                 mp_opt->data_seq = get_unaligned_be64(ptr);
 210                                 ptr += 8;
 211                         } else {
 212                                 mp_opt->data_seq = get_unaligned_be32(ptr);
 213                                 ptr += 4;
 214                         }
 215
 216                         mp_opt->subflow_seq = get_unaligned_be32(ptr);
 217                         ptr += 4;
 218
 219                         mp_opt->data_len = get_unaligned_be16(ptr);
 220                         ptr += 2;
 221
 222                         if (opsize == expected_opsize + TCPOLEN_MPTCP_DSS_CHECKSUM) {
 223                                 mp_opt->suboptions |= OPTION_MPTCP_CSUMREQD;
 224                                 mp_opt->csum = (__force __sum16)get_unaligned_be16(ptr);
 225                                 ptr += 2;
 226                         }
 227
 228                         pr_debug("data_seq=%llu subflow_seq=%u data_len=%u csum=%d:%u",
 229                                  mp_opt->data_seq, mp_opt->subflow_seq,
 230                                  mp_opt->data_len, !!(mp_opt->suboptions & OPTION_MPTCP_CSUMREQD),
 231                                  mp_opt->csum);
 232                 }
 233
 234                 break;
 235
 236         case MPTCPOPT_ADD_ADDR:
 237                 mp_opt->echo = (*ptr++) & MPTCP_ADDR_ECHO;
 238                 if (!mp_opt->echo) {
 239                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR ||
 240                             opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT)
 241                                 mp_opt->addr.family = AF_INET;
 242 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 243                         else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6 ||
 244                                  opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT)
 245                                 mp_opt->addr.family = AF_INET6;
 246 #endif
 247                         else
 248                                 break;
 249                 } else {
 250                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE ||
 251                             opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT)
 252                                 mp_opt->addr.family = AF_INET;
 253 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 254                         else if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE ||
 255                                  opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT)
 256                                 mp_opt->addr.family = AF_INET6;
 257 #endif
 258                         else
 259                                 break;
 260                 }
 261
 262                 mp_opt->suboptions |= OPTION_MPTCP_ADD_ADDR;
 263                 mp_opt->addr.id = *ptr++;
 264                 mp_opt->addr.port = 0;
 265                 mp_opt->ahmac = 0;
 266                 if (mp_opt->addr.family == AF_INET) {
 267                         memcpy((u8 *)&mp_opt->addr.addr.s_addr, (u8 *)ptr, 4);
 268                         ptr += 4;
 269                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR_PORT ||
 270                             opsize == TCPOLEN_MPTCP_ADD_ADDR_BASE_PORT) {
 271                                 mp_opt->addr.port = htons(get_unaligned_be16(ptr));
 272                                 ptr += 2;
 273                         }
 274                 }
 275 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 276                 else {
 277                         memcpy(mp_opt->addr.addr6.s6_addr, (u8 *)ptr, 16);
 278                         ptr += 16;
 279                         if (opsize == TCPOLEN_MPTCP_ADD_ADDR6_PORT ||
 280                             opsize == TCPOLEN_MPTCP_ADD_ADDR6_BASE_PORT) {
 281                                 mp_opt->addr.port = htons(get_unaligned_be16(ptr));
 282                                 ptr += 2;
 283                         }
 284                 }
 285 #endif
 286                 if (!mp_opt->echo) {
 287                         mp_opt->ahmac = get_unaligned_be64(ptr);
 288                         ptr += 8;
 289                 }
 290                 pr_debug("ADD_ADDR%s: id=%d, ahmac=%llu, echo=%d, port=%d",
 291                          (mp_opt->addr.family == AF_INET6) ? "6" : "",
 292                          mp_opt->addr.id, mp_opt->ahmac, mp_opt->echo, ntohs(mp_opt->addr.port));
 293                 break;
 294
 295         case MPTCPOPT_RM_ADDR:
 296                 if (opsize < TCPOLEN_MPTCP_RM_ADDR_BASE + 1 ||
 297                     opsize > TCPOLEN_MPTCP_RM_ADDR_BASE + MPTCP_RM_IDS_MAX)
 298                         break;
 299
 300                 ptr++;
 301
 302                 mp_opt->suboptions |= OPTION_MPTCP_RM_ADDR;
 303                 mp_opt->rm_list.nr = opsize - TCPOLEN_MPTCP_RM_ADDR_BASE;
 304                 for (i = 0; i < mp_opt->rm_list.nr; i++)
 305                         mp_opt->rm_list.ids[i] = *ptr++;
 306                 pr_debug("RM_ADDR: rm_list_nr=%d", mp_opt->rm_list.nr);
 307                 break;
 308
 309         case MPTCPOPT_MP_PRIO:
 310                 if (opsize != TCPOLEN_MPTCP_PRIO)
 311                         break;
 312
 313                 mp_opt->suboptions |= OPTION_MPTCP_PRIO;
 314                 mp_opt->backup = *ptr++ & MPTCP_PRIO_BKUP;
 315                 pr_debug("MP_PRIO: prio=%d", mp_opt->backup);
 316                 break;
 317
 318         case MPTCPOPT_MP_FASTCLOSE:
 319                 if (opsize != TCPOLEN_MPTCP_FASTCLOSE)
 320                         break;
 321
 322                 ptr += 2;
 323                 mp_opt->rcvr_key = get_unaligned_be64(ptr);
 324                 ptr += 8;
 325                 mp_opt->suboptions |= OPTION_MPTCP_FASTCLOSE;
 326                 break;
 327
 328         case MPTCPOPT_RST:
 329                 if (opsize != TCPOLEN_MPTCP_RST)
 330                         break;
 331
 332                 if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST))
 333                         break;
 334
 335                 mp_opt->suboptions |= OPTION_MPTCP_RST;
 336                 flags = *ptr++;
 337                 mp_opt->reset_transient = flags & MPTCP_RST_TRANSIENT;
 338                 mp_opt->reset_reason = *ptr;
 339                 break;
 340
 341         case MPTCPOPT_MP_FAIL:
 342                 if (opsize != TCPOLEN_MPTCP_FAIL)
 343                         break;
 344
 345                 ptr += 2;
 346                 mp_opt->suboptions |= OPTION_MPTCP_FAIL;
 347                 mp_opt->fail_seq = get_unaligned_be64(ptr);
 348                 pr_debug("MP_FAIL: data_seq=%llu", mp_opt->fail_seq);
 349                 break;
 350
 351         default:
 352                 break;
 353         }
 354 }
 355
 356 void mptcp_get_options(const struct sock *sk,
 357                        const struct sk_buff *skb,
 358                        struct mptcp_options_received *mp_opt)
 359 {
 360         const struct tcphdr *th = tcp_hdr(skb);
 361         const unsigned char *ptr;
 362         int length;
 363
 364         /* initialize option status */
 365         mp_opt->suboptions = 0;
 366
 367         length = (th->doff * 4) - sizeof(struct tcphdr);
 368         ptr = (const unsigned char *)(th + 1);
 369
 370         while (length > 0) {
 371                 int opcode = *ptr++;
 372                 int opsize;
 373
 374                 switch (opcode) {
 375                 case TCPOPT_EOL:
 376                         return;
 377                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 378                         length--;
 379                         continue;
 380                 default:
 381                         if (length < 2)
 382                                 return;
 383                         opsize = *ptr++;
 384                         if (opsize < 2) /* "silly options" */
 385                                 return;
 386                         if (opsize > length)
 387                                 return; /* don't parse partial options */
 388                         if (opcode == TCPOPT_MPTCP)
 389                                 mptcp_parse_option(skb, ptr, opsize, mp_opt);
 390                         ptr += opsize - 2;
 391                         length -= opsize;
 392                 }
 393         }
 394 }
 395
 396 bool mptcp_syn_options(struct sock *sk, const struct sk_buff *skb,
 397                        unsigned int *size, struct mptcp_out_options *opts)
 398 {
 399         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 400
 401         /* we will use snd_isn to detect first pkt [re]transmission
 402          * in mptcp_established_options_mp()
 403          */
 404         subflow->snd_isn = TCP_SKB_CB(skb)->end_seq;
 405         if (subflow->request_mptcp) {
 406                 opts->suboptions = OPTION_MPTCP_MPC_SYN;
 407                 opts->csum_reqd = mptcp_is_checksum_enabled(sock_net(sk));
 408                 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
 409                 *size = TCPOLEN_MPTCP_MPC_SYN;
 410                 return true;
 411         } else if (subflow->request_join) {
 412                 pr_debug("remote_token=%u, nonce=%u", subflow->remote_token,
 413                          subflow->local_nonce);
 414                 opts->suboptions = OPTION_MPTCP_MPJ_SYN;
 415                 opts->join_id = subflow->local_id;
 416                 opts->token = subflow->remote_token;
 417                 opts->nonce = subflow->local_nonce;
 418                 opts->backup = subflow->request_bkup;
 419                 *size = TCPOLEN_MPTCP_MPJ_SYN;
 420                 return true;
 421         }
 422         return false;
 423 }
 424
 425 static void clear_3rdack_retransmission(struct sock *sk)
 426 {
 427         struct inet_connection_sock *icsk = inet_csk(sk);
 428
 429         sk_stop_timer(sk, &icsk->icsk_delack_timer);
 430         icsk->icsk_ack.timeout = 0;
 431         icsk->icsk_ack.ato = 0;
 432         icsk->icsk_ack.pending &= ~(ICSK_ACK_SCHED | ICSK_ACK_TIMER);
 433 }
 434
 435 static bool mptcp_established_options_mp(struct sock *sk, struct sk_buff *skb,
 436                                          bool snd_data_fin_enable,
 437                                          unsigned int *size,
 438                                          unsigned int remaining,
 439                                          struct mptcp_out_options *opts)
 440 {
 441         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 442         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 443         struct mptcp_ext *mpext;
 444         unsigned int data_len;
 445         u8 len;
 446
 447         /* When skb is not available, we better over-estimate the emitted
 448          * options len. A full DSS option (28 bytes) is longer than
 449          * TCPOLEN_MPTCP_MPC_ACK_DATA(22) or TCPOLEN_MPTCP_MPJ_ACK(24), so
 450          * tell the caller to defer the estimate to
 451          * mptcp_established_options_dss(), which will reserve enough space.
 452          */
 453         if (!skb)
 454                 return false;
 455
 456         /* MPC/MPJ needed only on 3rd ack packet, DATA_FIN and TCP shutdown take precedence */
 457         if (subflow->fully_established || snd_data_fin_enable ||
 458             subflow->snd_isn != TCP_SKB_CB(skb)->seq ||
 459             sk->sk_state != TCP_ESTABLISHED)
 460                 return false;
 461
 462         if (subflow->mp_capable) {
 463                 mpext = mptcp_get_ext(skb);
 464                 data_len = mpext ? mpext->data_len : 0;
 465
 466                 /* we will check ops->data_len in mptcp_write_options() to
 467                  * discriminate between TCPOLEN_MPTCP_MPC_ACK_DATA and
 468                  * TCPOLEN_MPTCP_MPC_ACK
 469                  */
 470                 opts->data_len = data_len;
 471                 opts->suboptions = OPTION_MPTCP_MPC_ACK;
 472                 opts->sndr_key = subflow->local_key;
 473                 opts->rcvr_key = subflow->remote_key;
 474                 opts->csum_reqd = READ_ONCE(msk->csum_enabled);
 475                 opts->allow_join_id0 = mptcp_allow_join_id0(sock_net(sk));
 476
 477                 /* Section 3.1.
 478                  * The MP_CAPABLE option is carried on the SYN, SYN/ACK, and ACK
 479                  * packets that start the first subflow of an MPTCP connection,
 480                  * as well as the first packet that carries data
 481                  */
 482                 if (data_len > 0) {
 483                         len = TCPOLEN_MPTCP_MPC_ACK_DATA;
 484                         if (opts->csum_reqd) {
 485                                 /* we need to propagate more info to csum the pseudo hdr */
 486                                 opts->data_seq = mpext->data_seq;
 487                                 opts->subflow_seq = mpext->subflow_seq;
 488                                 opts->csum = mpext->csum;
 489                                 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
 490                         }
 491                         *size = ALIGN(len, 4);
 492                 } else {
 493                         *size = TCPOLEN_MPTCP_MPC_ACK;
 494                 }
 495
 496                 pr_debug("subflow=%p, local_key=%llu, remote_key=%llu map_len=%d",
 497                          subflow, subflow->local_key, subflow->remote_key,
 498                          data_len);
 499
 500                 return true;
 501         } else if (subflow->mp_join) {
 502                 opts->suboptions = OPTION_MPTCP_MPJ_ACK;
 503                 memcpy(opts->hmac, subflow->hmac, MPTCPOPT_HMAC_LEN);
 504                 *size = TCPOLEN_MPTCP_MPJ_ACK;
 505                 pr_debug("subflow=%p", subflow);
 506
 507                 /* we can use the full delegate action helper only from BH context
 508                  * If we are in process context - sk is flushing the backlog at
 509                  * socket lock release time - just set the appropriate flag, will
 510                  * be handled by the release callback
 511                  */
 512                 if (sock_owned_by_user(sk))
 513                         set_bit(MPTCP_DELEGATE_ACK, &subflow->delegated_status);
 514                 else
 515                         mptcp_subflow_delegate(subflow, MPTCP_DELEGATE_ACK);
 516                 return true;
 517         }
 518         return false;
 519 }
 520
 521 static void mptcp_write_data_fin(struct mptcp_subflow_context *subflow,
 522                                  struct sk_buff *skb, struct mptcp_ext *ext)
 523 {
 524         /* The write_seq value has already been incremented, so the actual
 525          * sequence number for the DATA_FIN is one less.
 526          */
 527         u64 data_fin_tx_seq = READ_ONCE(mptcp_sk(subflow->conn)->write_seq) - 1;
 528
 529         if (!ext->use_map || !skb->len) {
 530                 /* RFC6824 requires a DSS mapping with specific values
 531                  * if DATA_FIN is set but no data payload is mapped
 532                  */
 533                 ext->data_fin = 1;
 534                 ext->use_map = 1;
 535                 ext->dsn64 = 1;
 536                 ext->data_seq = data_fin_tx_seq;
 537                 ext->subflow_seq = 0;
 538                 ext->data_len = 1;
 539         } else if (ext->data_seq + ext->data_len == data_fin_tx_seq) {
 540                 /* If there's an existing DSS mapping and it is the
 541                  * final mapping, DATA_FIN consumes 1 additional byte of
 542                  * mapping space.
 543                  */
 544                 ext->data_fin = 1;
 545                 ext->data_len++;
 546         }
 547 }
 548
 549 static bool mptcp_established_options_dss(struct sock *sk, struct sk_buff *skb,
 550                                           bool snd_data_fin_enable,
 551                                           unsigned int *size,
 552                                           unsigned int remaining,
 553                                           struct mptcp_out_options *opts)
 554 {
 555         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 556         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 557         unsigned int dss_size = 0;
 558         struct mptcp_ext *mpext;
 559         unsigned int ack_size;
 560         bool ret = false;
 561         u64 ack_seq;
 562
 563         opts->csum_reqd = READ_ONCE(msk->csum_enabled);
 564         mpext = skb ? mptcp_get_ext(skb) : NULL;
 565
 566         if (!skb || (mpext && mpext->use_map) || snd_data_fin_enable) {
 567                 unsigned int map_size = TCPOLEN_MPTCP_DSS_BASE + TCPOLEN_MPTCP_DSS_MAP64;
 568
 569                 if (mpext) {
 570                         if (opts->csum_reqd)
 571                                 map_size += TCPOLEN_MPTCP_DSS_CHECKSUM;
 572
 573                         opts->ext_copy = *mpext;
 574                 }
 575
 576                 remaining -= map_size;
 577                 dss_size = map_size;
 578                 if (skb && snd_data_fin_enable)
 579                         mptcp_write_data_fin(subflow, skb, &opts->ext_copy);
 580                 opts->suboptions = OPTION_MPTCP_DSS;
 581                 ret = true;
 582         }
 583
 584         /* passive sockets msk will set the 'can_ack' after accept(), even
 585          * if the first subflow may have the already the remote key handy
 586          */
 587         opts->ext_copy.use_ack = 0;
 588         if (!READ_ONCE(msk->can_ack)) {
 589                 *size = ALIGN(dss_size, 4);
 590                 return ret;
 591         }
 592
 593         ack_seq = READ_ONCE(msk->ack_seq);
 594         if (READ_ONCE(msk->use_64bit_ack)) {
 595                 ack_size = TCPOLEN_MPTCP_DSS_ACK64;
 596                 opts->ext_copy.data_ack = ack_seq;
 597                 opts->ext_copy.ack64 = 1;
 598         } else {
 599                 ack_size = TCPOLEN_MPTCP_DSS_ACK32;
 600                 opts->ext_copy.data_ack32 = (uint32_t)ack_seq;
 601                 opts->ext_copy.ack64 = 0;
 602         }
 603         opts->ext_copy.use_ack = 1;
 604         opts->suboptions = OPTION_MPTCP_DSS;
 605         WRITE_ONCE(msk->old_wspace, __mptcp_space((struct sock *)msk));
 606
 607         /* Add kind/length/subtype/flag overhead if mapping is not populated */
 608         if (dss_size == 0)
 609                 ack_size += TCPOLEN_MPTCP_DSS_BASE;
 610
 611         dss_size += ack_size;
 612
 613         *size = ALIGN(dss_size, 4);
 614         return true;
 615 }
 616
 617 static u64 add_addr_generate_hmac(u64 key1, u64 key2,
 618                                   struct mptcp_addr_info *addr)
 619 {
 620         u16 port = ntohs(addr->port);
 621         u8 hmac[SHA256_DIGEST_SIZE];
 622         u8 msg[19];
 623         int i = 0;
 624
 625         msg[i++] = addr->id;
 626         if (addr->family == AF_INET) {
 627                 memcpy(&msg[i], &addr->addr.s_addr, 4);
 628                 i += 4;
 629         }
 630 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
 631         else if (addr->family == AF_INET6) {
 632                 memcpy(&msg[i], &addr->addr6.s6_addr, 16);
 633                 i += 16;
 634         }
 635 #endif
 636         msg[i++] = port >> 8;
 637         msg[i++] = port & 0xFF;
 638
 639         mptcp_crypto_hmac_sha(key1, key2, msg, i, hmac);
 640
 641         return get_unaligned_be64(&hmac[SHA256_DIGEST_SIZE - sizeof(u64)]);
 642 }
 643
 644 static bool mptcp_established_options_add_addr(struct sock *sk, struct sk_buff *skb,
 645                                                unsigned int *size,
 646                                                unsigned int remaining,
 647                                                struct mptcp_out_options *opts)
 648 {
 649         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 650         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 651         bool drop_other_suboptions = false;
 652         unsigned int opt_size = *size;
 653         bool echo;
 654         bool port;
 655         int len;
 656
 657         /* add addr will strip the existing options, be sure to avoid breaking
 658          * MPC/MPJ handshakes
 659          */
 660         if (!mptcp_pm_should_add_signal(msk) ||
 661             (opts->suboptions & (OPTION_MPTCP_MPJ_ACK | OPTION_MPTCP_MPC_ACK)) ||
 662             !mptcp_pm_add_addr_signal(msk, skb, opt_size, remaining, &opts->addr,
 663                     &echo, &port, &drop_other_suboptions))
 664                 return false;
 665
 666         if (drop_other_suboptions)
 667                 remaining += opt_size;
 668         len = mptcp_add_addr_len(opts->addr.family, echo, port);
 669         if (remaining < len)
 670                 return false;
 671
 672         *size = len;
 673         if (drop_other_suboptions) {
 674                 pr_debug("drop other suboptions");
 675                 opts->suboptions = 0;
 676
 677                 /* note that e.g. DSS could have written into the memory
 678                  * aliased by ahmac, we must reset the field here
 679                  * to avoid appending the hmac even for ADD_ADDR echo
 680                  * options
 681                  */
 682                 opts->ahmac = 0;
 683                 *size -= opt_size;
 684         }
 685         opts->suboptions |= OPTION_MPTCP_ADD_ADDR;
 686         if (!echo) {
 687                 opts->ahmac = add_addr_generate_hmac(msk->local_key,
 688                                                      msk->remote_key,
 689                                                      &opts->addr);
 690         }
 691         pr_debug("addr_id=%d, ahmac=%llu, echo=%d, port=%d",
 692                  opts->addr.id, opts->ahmac, echo, ntohs(opts->addr.port));
 693
 694         return true;
 695 }
 696
 697 static bool mptcp_established_options_rm_addr(struct sock *sk,
 698                                               unsigned int *size,
 699                                               unsigned int remaining,
 700                                               struct mptcp_out_options *opts)
 701 {
 702         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 703         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 704         struct mptcp_rm_list rm_list;
 705         int i, len;
 706
 707         if (!mptcp_pm_should_rm_signal(msk) ||
 708             !(mptcp_pm_rm_addr_signal(msk, remaining, &rm_list)))
 709                 return false;
 710
 711         len = mptcp_rm_addr_len(&rm_list);
 712         if (len < 0)
 713                 return false;
 714         if (remaining < len)
 715                 return false;
 716
 717         *size = len;
 718         opts->suboptions |= OPTION_MPTCP_RM_ADDR;
 719         opts->rm_list = rm_list;
 720
 721         for (i = 0; i < opts->rm_list.nr; i++)
 722                 pr_debug("rm_list_ids[%d]=%d", i, opts->rm_list.ids[i]);
 723
 724         return true;
 725 }
 726
 727 static bool mptcp_established_options_mp_prio(struct sock *sk,
 728                                               unsigned int *size,
 729                                               unsigned int remaining,
 730                                               struct mptcp_out_options *opts)
 731 {
 732         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 733
 734         /* can't send MP_PRIO with MPC, as they share the same option space:
 735          * 'backup'. Also it makes no sense at all
 736          */
 737         if (!subflow->send_mp_prio || (opts->suboptions & OPTIONS_MPTCP_MPC))
 738                 return false;
 739
 740         /* account for the trailing 'nop' option */
 741         if (remaining < TCPOLEN_MPTCP_PRIO_ALIGN)
 742                 return false;
 743
 744         *size = TCPOLEN_MPTCP_PRIO_ALIGN;
 745         opts->suboptions |= OPTION_MPTCP_PRIO;
 746         opts->backup = subflow->request_bkup;
 747
 748         pr_debug("prio=%d", opts->backup);
 749
 750         return true;
 751 }
 752
 753 static noinline bool mptcp_established_options_rst(struct sock *sk, struct sk_buff *skb,
 754                                                    unsigned int *size,
 755                                                    unsigned int remaining,
 756                                                    struct mptcp_out_options *opts)
 757 {
 758         const struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 759
 760         if (remaining < TCPOLEN_MPTCP_RST)
 761                 return false;
 762
 763         *size = TCPOLEN_MPTCP_RST;
 764         opts->suboptions |= OPTION_MPTCP_RST;
 765         opts->reset_transient = subflow->reset_transient;
 766         opts->reset_reason = subflow->reset_reason;
 767
 768         return true;
 769 }
 770
 771 static bool mptcp_established_options_fastclose(struct sock *sk,
 772                                                 unsigned int *size,
 773                                                 unsigned int remaining,
 774                                                 struct mptcp_out_options *opts)
 775 {
 776         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 777         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 778
 779         if (likely(!subflow->send_fastclose))
 780                 return false;
 781
 782         if (remaining < TCPOLEN_MPTCP_FASTCLOSE)
 783                 return false;
 784
 785         *size = TCPOLEN_MPTCP_FASTCLOSE;
 786         opts->suboptions |= OPTION_MPTCP_FASTCLOSE;
 787         opts->rcvr_key = msk->remote_key;
 788
 789         pr_debug("FASTCLOSE key=%llu", opts->rcvr_key);
 790         return true;
 791 }
 792
 793 static bool mptcp_established_options_mp_fail(struct sock *sk,
 794                                               unsigned int *size,
 795                                               unsigned int remaining,
 796                                               struct mptcp_out_options *opts)
 797 {
 798         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 799
 800         if (likely(!subflow->send_mp_fail))
 801                 return false;
 802
 803         if (remaining < TCPOLEN_MPTCP_FAIL)
 804                 return false;
 805
 806         *size = TCPOLEN_MPTCP_FAIL;
 807         opts->suboptions |= OPTION_MPTCP_FAIL;
 808         opts->fail_seq = subflow->map_seq;
 809
 810         pr_debug("MP_FAIL fail_seq=%llu", opts->fail_seq);
 811
 812         return true;
 813 }
 814
 815 bool mptcp_established_options(struct sock *sk, struct sk_buff *skb,
 816                                unsigned int *size, unsigned int remaining,
 817                                struct mptcp_out_options *opts)
 818 {
 819         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
 820         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
 821         unsigned int opt_size = 0;
 822         bool snd_data_fin;
 823         bool ret = false;
 824
 825         opts->suboptions = 0;
 826
 827         if (unlikely(__mptcp_check_fallback(msk)))
 828                 return false;
 829
 830         if (unlikely(skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_RST)) {
 831                 if (mptcp_established_options_fastclose(sk, &opt_size, remaining, opts) ||
 832                     mptcp_established_options_mp_fail(sk, &opt_size, remaining, opts)) {
 833                         *size += opt_size;
 834                         remaining -= opt_size;
 835                 }
 836                 /* MP_RST can be used with MP_FASTCLOSE and MP_FAIL if there is room */
 837                 if (mptcp_established_options_rst(sk, skb, &opt_size, remaining, opts)) {
 838                         *size += opt_size;
 839                         remaining -= opt_size;
 840                 }
 841                 return true;
 842         }
 843
 844         snd_data_fin = mptcp_data_fin_enabled(msk);
 845         if (mptcp_established_options_mp(sk, skb, snd_data_fin, &opt_size, remaining, opts))
 846                 ret = true;
 847         else if (mptcp_established_options_dss(sk, skb, snd_data_fin, &opt_size, remaining, opts)) {
 848                 unsigned int mp_fail_size;
 849
 850                 ret = true;
 851                 if (mptcp_established_options_mp_fail(sk, &mp_fail_size,
 852                                                       remaining - opt_size, opts)) {
 853                         *size += opt_size + mp_fail_size;
 854                         remaining -= opt_size - mp_fail_size;
 855                         return true;
 856                 }
 857         }
 858
 859         /* we reserved enough space for the above options, and exceeding the
 860          * TCP option space would be fatal
 861          */
 862         if (WARN_ON_ONCE(opt_size > remaining))
 863                 return false;
 864
 865         *size += opt_size;
 866         remaining -= opt_size;
 867         if (mptcp_established_options_add_addr(sk, skb, &opt_size, remaining, opts)) {
 868                 *size += opt_size;
 869                 remaining -= opt_size;
 870                 ret = true;
 871         } else if (mptcp_established_options_rm_addr(sk, &opt_size, remaining, opts)) {
 872                 *size += opt_size;
 873                 remaining -= opt_size;
 874                 ret = true;
 875         }
 876
 877         if (mptcp_established_options_mp_prio(sk, &opt_size, remaining, opts)) {
 878                 *size += opt_size;
 879                 remaining -= opt_size;
 880                 ret = true;
 881         }
 882
 883         return ret;
 884 }
 885
 886 bool mptcp_synack_options(const struct request_sock *req, unsigned int *size,
 887                           struct mptcp_out_options *opts)
 888 {
 889         struct mptcp_subflow_request_sock *subflow_req = mptcp_subflow_rsk(req);
 890
 891         if (subflow_req->mp_capable) {
 892                 opts->suboptions = OPTION_MPTCP_MPC_SYNACK;
 893                 opts->sndr_key = subflow_req->local_key;
 894                 opts->csum_reqd = subflow_req->csum_reqd;
 895                 opts->allow_join_id0 = subflow_req->allow_join_id0;
 896                 *size = TCPOLEN_MPTCP_MPC_SYNACK;
 897                 pr_debug("subflow_req=%p, local_key=%llu",
 898                          subflow_req, subflow_req->local_key);
 899                 return true;
 900         } else if (subflow_req->mp_join) {
 901                 opts->suboptions = OPTION_MPTCP_MPJ_SYNACK;
 902                 opts->backup = subflow_req->backup;
 903                 opts->join_id = subflow_req->local_id;
 904                 opts->thmac = subflow_req->thmac;
 905                 opts->nonce = subflow_req->local_nonce;
 906                 pr_debug("req=%p, bkup=%u, id=%u, thmac=%llu, nonce=%u",
 907                          subflow_req, opts->backup, opts->join_id,
 908                          opts->thmac, opts->nonce);
 909                 *size = TCPOLEN_MPTCP_MPJ_SYNACK;
 910                 return true;
 911         }
 912         return false;
 913 }
 914
 915 static bool check_fully_established(struct mptcp_sock *msk, struct sock *ssk,
 916                                     struct mptcp_subflow_context *subflow,
 917                                     struct sk_buff *skb,
 918                                     struct mptcp_options_received *mp_opt)
 919 {
 920         /* here we can process OoO, in-window pkts, only in-sequence 4th ack
 921          * will make the subflow fully established
 922          */
 923         if (likely(subflow->fully_established)) {
 924                 /* on passive sockets, check for 3rd ack retransmission
 925                  * note that msk is always set by subflow_syn_recv_sock()
 926                  * for mp_join subflows
 927                  */
 928                 if (TCP_SKB_CB(skb)->seq == subflow->ssn_offset + 1 &&
 929                     TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq &&
 930                     subflow->mp_join && (mp_opt->suboptions & OPTIONS_MPTCP_MPJ) &&
 931                     READ_ONCE(msk->pm.server_side))
 932                         tcp_send_ack(ssk);
 933                 goto fully_established;
 934         }
 935
 936         /* we must process OoO packets before the first subflow is fully
 937          * established. OoO packets are instead a protocol violation
 938          * for MP_JOIN subflows as the peer must not send any data
 939          * before receiving the forth ack - cfr. RFC 8684 section 3.2.
 940          */
 941         if (TCP_SKB_CB(skb)->seq != subflow->ssn_offset + 1) {
 942                 if (subflow->mp_join)
 943                         goto reset;
 944                 return subflow->mp_capable;
 945         }
 946
 947         if (((mp_opt->suboptions & OPTION_MPTCP_DSS) && mp_opt->use_ack) ||
 948             ((mp_opt->suboptions & OPTION_MPTCP_ADD_ADDR) && !mp_opt->echo)) {
 949                 /* subflows are fully established as soon as we get any
 950                  * additional ack, including ADD_ADDR.
 951                  */
 952                 subflow->fully_established = 1;
 953                 WRITE_ONCE(msk->fully_established, true);
 954                 goto fully_established;
 955         }
 956
 957         /* If the first established packet does not contain MP_CAPABLE + data
 958          * then fallback to TCP. Fallback scenarios requires a reset for
 959          * MP_JOIN subflows.
 960          */
 961         if (!(mp_opt->suboptions & OPTIONS_MPTCP_MPC)) {
 962                 if (subflow->mp_join)
 963                         goto reset;
 964                 subflow->mp_capable = 0;
 965                 pr_fallback(msk);
 966                 __mptcp_do_fallback(msk);
 967                 return false;
 968         }
 969
 970         if (mp_opt->deny_join_id0)
 971                 WRITE_ONCE(msk->pm.remote_deny_join_id0, true);
 972
 973         if (unlikely(!READ_ONCE(msk->pm.server_side)))
 974                 pr_warn_once("bogus mpc option on established client sk");
 975         mptcp_subflow_fully_established(subflow, mp_opt);
 976
 977 fully_established:
 978         /* if the subflow is not already linked into the conn_list, we can't
 979          * notify the PM: this subflow is still on the listener queue
 980          * and the PM possibly acquiring the subflow lock could race with
 981          * the listener close
 982          */
 983         if (likely(subflow->pm_notified) || list_empty(&subflow->node))
 984                 return true;
 985
 986         subflow->pm_notified = 1;
 987         if (subflow->mp_join) {
 988                 clear_3rdack_retransmission(ssk);
 989                 mptcp_pm_subflow_established(msk);
 990         } else {
 991                 mptcp_pm_fully_established(msk, ssk, GFP_ATOMIC);
 992         }
 993         return true;
 994
 995 reset:
 996         mptcp_subflow_reset(ssk);
 997         return false;
 998 }
 999
1000 u64 __mptcp_expand_seq(u64 old_seq, u64 cur_seq)
1001 {
1002         u32 old_seq32, cur_seq32;
1003
1004         old_seq32 = (u32)old_seq;
1005         cur_seq32 = (u32)cur_seq;
1006         cur_seq = (old_seq & GENMASK_ULL(63, 32)) + cur_seq32;
1007         if (unlikely(cur_seq32 < old_seq32 && before(old_seq32, cur_seq32)))
1008                 return cur_seq + (1LL << 32);
1009
1010         /* reverse wrap could happen, too */
1011         if (unlikely(cur_seq32 > old_seq32 && after(old_seq32, cur_seq32)))
1012                 return cur_seq - (1LL << 32);
1013         return cur_seq;
1014 }
1015
1016 static void ack_update_msk(struct mptcp_sock *msk,
1017                            struct sock *ssk,
1018                            struct mptcp_options_received *mp_opt)
1019 {
1020         u64 new_wnd_end, new_snd_una, snd_nxt = READ_ONCE(msk->snd_nxt);
1021         struct sock *sk = (struct sock *)msk;
1022         u64 old_snd_una;
1023
1024         mptcp_data_lock(sk);
1025
1026         /* avoid ack expansion on update conflict, to reduce the risk of
1027          * wrongly expanding to a future ack sequence number, which is way
1028          * more dangerous than missing an ack
1029          */
1030         old_snd_una = msk->snd_una;
1031         new_snd_una = mptcp_expand_seq(old_snd_una, mp_opt->data_ack, mp_opt->ack64);
1032
1033         /* ACK for data not even sent yet? Ignore.*/
1034         if (unlikely(after64(new_snd_una, snd_nxt)))
1035                 new_snd_una = old_snd_una;
1036
1037         new_wnd_end = new_snd_una + tcp_sk(ssk)->snd_wnd;
1038
1039         if (after64(new_wnd_end, msk->wnd_end))
1040                 msk->wnd_end = new_wnd_end;
1041
1042         /* this assumes mptcp_incoming_options() is invoked after tcp_ack() */
1043         if (after64(msk->wnd_end, READ_ONCE(msk->snd_nxt)))
1044                 __mptcp_check_push(sk, ssk);
1045
1046         if (after64(new_snd_una, old_snd_una)) {
1047                 msk->snd_una = new_snd_una;
1048                 __mptcp_data_acked(sk);
1049         }
1050         mptcp_data_unlock(sk);
1051
1052         trace_ack_update_msk(mp_opt->data_ack,
1053                              old_snd_una, new_snd_una,
1054                              new_wnd_end, msk->wnd_end);
1055 }
1056
1057 bool mptcp_update_rcv_data_fin(struct mptcp_sock *msk, u64 data_fin_seq, bool use_64bit)
1058 {
1059         /* Skip if DATA_FIN was already received.
1060          * If updating simultaneously with the recvmsg loop, values
1061          * should match. If they mismatch, the peer is misbehaving and
1062          * we will prefer the most recent information.
1063          */
1064         if (READ_ONCE(msk->rcv_data_fin))
1065                 return false;
1066
1067         WRITE_ONCE(msk->rcv_data_fin_seq,
1068                    mptcp_expand_seq(READ_ONCE(msk->ack_seq), data_fin_seq, use_64bit));
1069         WRITE_ONCE(msk->rcv_data_fin, 1);
1070
1071         return true;
1072 }
1073
1074 static bool add_addr_hmac_valid(struct mptcp_sock *msk,
1075                                 struct mptcp_options_received *mp_opt)
1076 {
1077         u64 hmac = 0;
1078
1079         if (mp_opt->echo)
1080                 return true;
1081
1082         hmac = add_addr_generate_hmac(msk->remote_key,
1083                                       msk->local_key,
1084                                       &mp_opt->addr);
1085
1086         pr_debug("msk=%p, ahmac=%llu, mp_opt->ahmac=%llu\n",
1087                  msk, (unsigned long long)hmac,
1088                  (unsigned long long)mp_opt->ahmac);
1089
1090         return hmac == mp_opt->ahmac;
1091 }
1092
1093 /* Return false if a subflow has been reset, else return true */
1094 bool mptcp_incoming_options(struct sock *sk, struct sk_buff *skb)
1095 {
1096         struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
1097         struct mptcp_sock *msk = mptcp_sk(subflow->conn);
1098         struct mptcp_options_received mp_opt;
1099         struct mptcp_ext *mpext;
1100
1101         if (__mptcp_check_fallback(msk)) {
1102                 /* Keep it simple and unconditionally trigger send data cleanup and
1103                  * pending queue spooling. We will need to acquire the data lock
1104                  * for more accurate checks, and once the lock is acquired, such
1105                  * helpers are cheap.
1106                  */
1107                 mptcp_data_lock(subflow->conn);
1108                 if (sk_stream_memory_free(sk))
1109                         __mptcp_check_push(subflow->conn, sk);
1110                 __mptcp_data_acked(subflow->conn);
1111                 mptcp_data_unlock(subflow->conn);
1112                 return true;
1113         }
1114
1115         mptcp_get_options(sk, skb, &mp_opt);
1116
1117         /* The subflow can be in close state only if check_fully_established()
1118          * just sent a reset. If so, tell the caller to ignore the current packet.
1119          */
1120         if (!check_fully_established(msk, sk, subflow, skb, &mp_opt))
1121                 return sk->sk_state != TCP_CLOSE;
1122
1123         if (unlikely(mp_opt.suboptions != OPTION_MPTCP_DSS)) {
1124                 if ((mp_opt.suboptions & OPTION_MPTCP_FASTCLOSE) &&
1125                     msk->local_key == mp_opt.rcvr_key) {
1126                         WRITE_ONCE(msk->rcv_fastclose, true);
1127                         mptcp_schedule_work((struct sock *)msk);
1128                 }
1129
1130                 if ((mp_opt.suboptions & OPTION_MPTCP_ADD_ADDR) &&
1131                     add_addr_hmac_valid(msk, &mp_opt)) {
1132                         if (!mp_opt.echo) {
1133                                 mptcp_pm_add_addr_received(msk, &mp_opt.addr);
1134                                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ADDADDR);
1135                         } else {
1136                                 mptcp_pm_add_addr_echoed(msk, &mp_opt.addr);
1137                                 mptcp_pm_del_add_timer(msk, &mp_opt.addr, true);
1138                                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_ECHOADD);
1139                         }
1140
1141                         if (mp_opt.addr.port)
1142                                 MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_PORTADD);
1143                 }
1144
1145                 if (mp_opt.suboptions & OPTION_MPTCP_RM_ADDR)
1146                         mptcp_pm_rm_addr_received(msk, &mp_opt.rm_list);
1147
1148                 if (mp_opt.suboptions & OPTION_MPTCP_PRIO) {
1149                         mptcp_pm_mp_prio_received(sk, mp_opt.backup);
1150                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPPRIORX);
1151                 }
1152
1153                 if (mp_opt.suboptions & OPTION_MPTCP_FAIL) {
1154                         mptcp_pm_mp_fail_received(sk, mp_opt.fail_seq);
1155                         MPTCP_INC_STATS(sock_net(sk), MPTCP_MIB_MPFAILRX);
1156                 }
1157
1158                 if (mp_opt.suboptions & OPTION_MPTCP_RST) {
1159                         subflow->reset_seen = 1;
1160                         subflow->reset_reason = mp_opt.reset_reason;
1161                         subflow->reset_transient = mp_opt.reset_transient;
1162                 }
1163
1164                 if (!(mp_opt.suboptions & OPTION_MPTCP_DSS))
1165                         return true;
1166         }
1167
1168         /* we can't wait for recvmsg() to update the ack_seq, otherwise
1169          * monodirectional flows will stuck
1170          */
1171         if (mp_opt.use_ack)
1172                 ack_update_msk(msk, sk, &mp_opt);
1173
1174         /* Zero-data-length packets are dropped by the caller and not
1175          * propagated to the MPTCP layer, so the skb extension does not
1176          * need to be allocated or populated. DATA_FIN information, if
1177          * present, needs to be updated here before the skb is freed.
1178          */
1179         if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {
1180                 if (mp_opt.data_fin && mp_opt.data_len == 1 &&
1181                     mptcp_update_rcv_data_fin(msk, mp_opt.data_seq, mp_opt.dsn64) &&
1182                     schedule_work(&msk->work))
1183                         sock_hold(subflow->conn);
1184
1185                 return true;
1186         }
1187
1188         mpext = skb_ext_add(skb, SKB_EXT_MPTCP);
1189         if (!mpext)
1190                 return true;
1191
1192         memset(mpext, 0, sizeof(*mpext));
1193
1194         if (likely(mp_opt.use_map)) {
1195                 if (mp_opt.mpc_map) {
1196                         /* this is an MP_CAPABLE carrying MPTCP data
1197                          * we know this map the first chunk of data
1198                          */
1199                         mptcp_crypto_key_sha(subflow->remote_key, NULL,
1200                                              &mpext->data_seq);
1201                         mpext->data_seq++;
1202                         mpext->subflow_seq = 1;
1203                         mpext->dsn64 = 1;
1204                         mpext->mpc_map = 1;
1205                         mpext->data_fin = 0;
1206                 } else {
1207                         mpext->data_seq = mp_opt.data_seq;
1208                         mpext->subflow_seq = mp_opt.subflow_seq;
1209                         mpext->dsn64 = mp_opt.dsn64;
1210                         mpext->data_fin = mp_opt.data_fin;
1211                 }
1212                 mpext->data_len = mp_opt.data_len;
1213                 mpext->use_map = 1;
1214                 mpext->csum_reqd = !!(mp_opt.suboptions & OPTION_MPTCP_CSUMREQD);
1215
1216                 if (mpext->csum_reqd)
1217                         mpext->csum = mp_opt.csum;
1218         }
1219
1220         return true;
1221 }
1222
1223 static void mptcp_set_rwin(const struct tcp_sock *tp)
1224 {
1225         const struct sock *ssk = (const struct sock *)tp;
1226         const struct mptcp_subflow_context *subflow;
1227         struct mptcp_sock *msk;
1228         u64 ack_seq;
1229
1230         subflow = mptcp_subflow_ctx(ssk);
1231         msk = mptcp_sk(subflow->conn);
1232
1233         ack_seq = READ_ONCE(msk->ack_seq) + tp->rcv_wnd;
1234
1235         if (after64(ack_seq, READ_ONCE(msk->rcv_wnd_sent)))
1236                 WRITE_ONCE(msk->rcv_wnd_sent, ack_seq);
1237 }
1238
1239 u16 __mptcp_make_csum(u64 data_seq, u32 subflow_seq, u16 data_len, __wsum sum)
1240 {
1241         struct csum_pseudo_header header;
1242         __wsum csum;
1243
1244         /* cfr RFC 8684 3.3.1.:
1245          * the data sequence number used in the pseudo-header is
1246          * always the 64-bit value, irrespective of what length is used in the
1247          * DSS option itself.
1248          */
1249         header.data_seq = cpu_to_be64(data_seq);
1250         header.subflow_seq = htonl(subflow_seq);
1251         header.data_len = htons(data_len);
1252         header.csum = 0;
1253
1254         csum = csum_partial(&header, sizeof(header), sum);
1255         return (__force u16)csum_fold(csum);
1256 }
1257
1258 static u16 mptcp_make_csum(const struct mptcp_ext *mpext)
1259 {
1260         return __mptcp_make_csum(mpext->data_seq, mpext->subflow_seq, mpext->data_len,
1261                                  ~csum_unfold(mpext->csum));
1262 }
1263
1264 void mptcp_write_options(__be32 *ptr, const struct tcp_sock *tp,
1265                          struct mptcp_out_options *opts)
1266 {
1267         if (unlikely(OPTION_MPTCP_FAIL & opts->suboptions)) {
1268                 const struct sock *ssk = (const struct sock *)tp;
1269                 struct mptcp_subflow_context *subflow;
1270
1271                 subflow = mptcp_subflow_ctx(ssk);
1272                 subflow->send_mp_fail = 0;
1273
1274                 *ptr++ = mptcp_option(MPTCPOPT_MP_FAIL,
1275                                       TCPOLEN_MPTCP_FAIL,
1276                                       0, 0);
1277                 put_unaligned_be64(opts->fail_seq, ptr);
1278                 ptr += 2;
1279         }
1280
1281         /* DSS, MPC, MPJ, ADD_ADDR, FASTCLOSE and RST are mutually exclusive,
1282          * see mptcp_established_options*()
1283          */
1284         if (likely(OPTION_MPTCP_DSS & opts->suboptions)) {
1285                 struct mptcp_ext *mpext = &opts->ext_copy;
1286                 u8 len = TCPOLEN_MPTCP_DSS_BASE;
1287                 u8 flags = 0;
1288
1289                 if (mpext->use_ack) {
1290                         flags = MPTCP_DSS_HAS_ACK;
1291                         if (mpext->ack64) {
1292                                 len += TCPOLEN_MPTCP_DSS_ACK64;
1293                                 flags |= MPTCP_DSS_ACK64;
1294                         } else {
1295                                 len += TCPOLEN_MPTCP_DSS_ACK32;
1296                         }
1297                 }
1298
1299                 if (mpext->use_map) {
1300                         len += TCPOLEN_MPTCP_DSS_MAP64;
1301
1302                         /* Use only 64-bit mapping flags for now, add
1303                          * support for optional 32-bit mappings later.
1304                          */
1305                         flags |= MPTCP_DSS_HAS_MAP | MPTCP_DSS_DSN64;
1306                         if (mpext->data_fin)
1307                                 flags |= MPTCP_DSS_DATA_FIN;
1308
1309                         if (opts->csum_reqd)
1310                                 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1311                 }
1312
1313                 *ptr++ = mptcp_option(MPTCPOPT_DSS, len, 0, flags);
1314
1315                 if (mpext->use_ack) {
1316                         if (mpext->ack64) {
1317                                 put_unaligned_be64(mpext->data_ack, ptr);
1318                                 ptr += 2;
1319                         } else {
1320                                 put_unaligned_be32(mpext->data_ack32, ptr);
1321                                 ptr += 1;
1322                         }
1323                 }
1324
1325                 if (mpext->use_map) {
1326                         put_unaligned_be64(mpext->data_seq, ptr);
1327                         ptr += 2;
1328                         put_unaligned_be32(mpext->subflow_seq, ptr);
1329                         ptr += 1;
1330                         if (opts->csum_reqd) {
1331                                 put_unaligned_be32(mpext->data_len << 16 |
1332                                                    mptcp_make_csum(mpext), ptr);
1333                         } else {
1334                                 put_unaligned_be32(mpext->data_len << 16 |
1335                                                    TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1336                         }
1337                         ptr += 1;
1338                 }
1339         } else if (OPTIONS_MPTCP_MPC & opts->suboptions) {
1340                 u8 len, flag = MPTCP_CAP_HMAC_SHA256;
1341
1342                 if (OPTION_MPTCP_MPC_SYN & opts->suboptions) {
1343                         len = TCPOLEN_MPTCP_MPC_SYN;
1344                 } else if (OPTION_MPTCP_MPC_SYNACK & opts->suboptions) {
1345                         len = TCPOLEN_MPTCP_MPC_SYNACK;
1346                 } else if (opts->data_len) {
1347                         len = TCPOLEN_MPTCP_MPC_ACK_DATA;
1348                         if (opts->csum_reqd)
1349                                 len += TCPOLEN_MPTCP_DSS_CHECKSUM;
1350                 } else {
1351                         len = TCPOLEN_MPTCP_MPC_ACK;
1352                 }
1353
1354                 if (opts->csum_reqd)
1355                         flag |= MPTCP_CAP_CHECKSUM_REQD;
1356
1357                 if (!opts->allow_join_id0)
1358                         flag |= MPTCP_CAP_DENY_JOIN_ID0;
1359
1360                 *ptr++ = mptcp_option(MPTCPOPT_MP_CAPABLE, len,
1361                                       MPTCP_SUPPORTED_VERSION,
1362                                       flag);
1363
1364                 if (!((OPTION_MPTCP_MPC_SYNACK | OPTION_MPTCP_MPC_ACK) &
1365                     opts->suboptions))
1366                         goto mp_capable_done;
1367
1368                 put_unaligned_be64(opts->sndr_key, ptr);
1369                 ptr += 2;
1370                 if (!((OPTION_MPTCP_MPC_ACK) & opts->suboptions))
1371                         goto mp_capable_done;
1372
1373                 put_unaligned_be64(opts->rcvr_key, ptr);
1374                 ptr += 2;
1375                 if (!opts->data_len)
1376                         goto mp_capable_done;
1377
1378                 if (opts->csum_reqd) {
1379                         put_unaligned_be32(opts->data_len << 16 |
1380                                            __mptcp_make_csum(opts->data_seq,
1381                                                              opts->subflow_seq,
1382                                                              opts->data_len,
1383                                                              ~csum_unfold(opts->csum)), ptr);
1384                 } else {
1385                         put_unaligned_be32(opts->data_len << 16 |
1386                                            TCPOPT_NOP << 8 | TCPOPT_NOP, ptr);
1387                 }
1388                 ptr += 1;
1389
1390                 /* MPC is additionally mutually exclusive with MP_PRIO */
1391                 goto mp_capable_done;
1392         } else if (OPTIONS_MPTCP_MPJ & opts->suboptions) {
1393                 if (OPTION_MPTCP_MPJ_SYN & opts->suboptions) {
1394                         *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1395                                               TCPOLEN_MPTCP_MPJ_SYN,
1396                                               opts->backup, opts->join_id);
1397                         put_unaligned_be32(opts->token, ptr);
1398                         ptr += 1;
1399                         put_unaligned_be32(opts->nonce, ptr);
1400                         ptr += 1;
1401                 } else if (OPTION_MPTCP_MPJ_SYNACK & opts->suboptions) {
1402                         *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1403                                               TCPOLEN_MPTCP_MPJ_SYNACK,
1404                                               opts->backup, opts->join_id);
1405                         put_unaligned_be64(opts->thmac, ptr);
1406                         ptr += 2;
1407                         put_unaligned_be32(opts->nonce, ptr);
1408                         ptr += 1;
1409                 } else {
1410                         *ptr++ = mptcp_option(MPTCPOPT_MP_JOIN,
1411                                               TCPOLEN_MPTCP_MPJ_ACK, 0, 0);
1412                         memcpy(ptr, opts->hmac, MPTCPOPT_HMAC_LEN);
1413                         ptr += 5;
1414                 }
1415         } else if (OPTION_MPTCP_ADD_ADDR & opts->suboptions) {
1416                 u8 len = TCPOLEN_MPTCP_ADD_ADDR_BASE;
1417                 u8 echo = MPTCP_ADDR_ECHO;
1418
1419 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1420                 if (opts->addr.family == AF_INET6)
1421                         len = TCPOLEN_MPTCP_ADD_ADDR6_BASE;
1422 #endif
1423
1424                 if (opts->addr.port)
1425                         len += TCPOLEN_MPTCP_PORT_LEN;
1426
1427                 if (opts->ahmac) {
1428                         len += sizeof(opts->ahmac);
1429                         echo = 0;
1430                 }
1431
1432                 *ptr++ = mptcp_option(MPTCPOPT_ADD_ADDR,
1433                                       len, echo, opts->addr.id);
1434                 if (opts->addr.family == AF_INET) {
1435                         memcpy((u8 *)ptr, (u8 *)&opts->addr.addr.s_addr, 4);
1436                         ptr += 1;
1437                 }
1438 #if IS_ENABLED(CONFIG_MPTCP_IPV6)
1439                 else if (opts->addr.family == AF_INET6) {
1440                         memcpy((u8 *)ptr, opts->addr.addr6.s6_addr, 16);
1441                         ptr += 4;
1442                 }
1443 #endif
1444
1445                 if (!opts->addr.port) {
1446                         if (opts->ahmac) {
1447                                 put_unaligned_be64(opts->ahmac, ptr);
1448                                 ptr += 2;
1449                         }
1450                 } else {
1451                         u16 port = ntohs(opts->addr.port);
1452
1453                         if (opts->ahmac) {
1454                                 u8 *bptr = (u8 *)ptr;
1455
1456                                 put_unaligned_be16(port, bptr);
1457                                 bptr += 2;
1458                                 put_unaligned_be64(opts->ahmac, bptr);
1459                                 bptr += 8;
1460                                 put_unaligned_be16(TCPOPT_NOP << 8 |
1461                                                    TCPOPT_NOP, bptr);
1462
1463                                 ptr += 3;
1464                         } else {
1465                                 put_unaligned_be32(port << 16 |
1466                                                    TCPOPT_NOP << 8 |
1467                                                    TCPOPT_NOP, ptr);
1468                                 ptr += 1;
1469                         }
1470                 }
1471         } else if (unlikely(OPTION_MPTCP_FASTCLOSE & opts->suboptions)) {
1472                 /* FASTCLOSE is mutually exclusive with others except RST */
1473                 *ptr++ = mptcp_option(MPTCPOPT_MP_FASTCLOSE,
1474                                       TCPOLEN_MPTCP_FASTCLOSE,
1475                                       0, 0);
1476                 put_unaligned_be64(opts->rcvr_key, ptr);
1477                 ptr += 2;
1478
1479                 if (OPTION_MPTCP_RST & opts->suboptions)
1480                         goto mp_rst;
1481                 return;
1482         } else if (unlikely(OPTION_MPTCP_RST & opts->suboptions)) {
1483 mp_rst:
1484                 *ptr++ = mptcp_option(MPTCPOPT_RST,
1485                                       TCPOLEN_MPTCP_RST,
1486                                       opts->reset_transient,
1487                                       opts->reset_reason);
1488                 return;
1489         }
1490
1491         if (OPTION_MPTCP_PRIO & opts->suboptions) {
1492                 const struct sock *ssk = (const struct sock *)tp;
1493                 struct mptcp_subflow_context *subflow;
1494
1495                 subflow = mptcp_subflow_ctx(ssk);
1496                 subflow->send_mp_prio = 0;
1497
1498                 *ptr++ = mptcp_option(MPTCPOPT_MP_PRIO,
1499                                       TCPOLEN_MPTCP_PRIO,
1500                                       opts->backup, TCPOPT_NOP);
1501         }
1502
1503 mp_capable_done:
1504         if (OPTION_MPTCP_RM_ADDR & opts->suboptions) {
1505                 u8 i = 1;
1506
1507                 *ptr++ = mptcp_option(MPTCPOPT_RM_ADDR,
1508                                       TCPOLEN_MPTCP_RM_ADDR_BASE + opts->rm_list.nr,
1509                                       0, opts->rm_list.ids[0]);
1510
1511                 while (i < opts->rm_list.nr) {
1512                         u8 id1, id2, id3, id4;
1513
1514                         id1 = opts->rm_list.ids[i];
1515                         id2 = i + 1 < opts->rm_list.nr ? opts->rm_list.ids[i + 1] : TCPOPT_NOP;
1516                         id3 = i + 2 < opts->rm_list.nr ? opts->rm_list.ids[i + 2] : TCPOPT_NOP;
1517                         id4 = i + 3 < opts->rm_list.nr ? opts->rm_list.ids[i + 3] : TCPOPT_NOP;
1518                         put_unaligned_be32(id1 << 24 | id2 << 16 | id3 << 8 | id4, ptr);
1519                         ptr += 1;
1520                         i += 4;
1521                 }
1522         }
1523
1524         if (tp)
1525                 mptcp_set_rwin(tp);
1526 }
1527
1528 __be32 mptcp_get_reset_option(const struct sk_buff *skb)
1529 {
1530         const struct mptcp_ext *ext = mptcp_get_ext(skb);
1531         u8 flags, reason;
1532
1533         if (ext) {
1534                 flags = ext->reset_transient;
1535                 reason = ext->reset_reason;
1536
1537                 return mptcp_option(MPTCPOPT_RST, TCPOLEN_MPTCP_RST,
1538                                     flags, reason);
1539         }
1540
1541         return htonl(0u);
1542 }
1543 EXPORT_SYMBOL_GPL(mptcp_get_reset_option);