net/sctp/output.c

   1 /* SCTP kernel implementation
   2  * (C) Copyright IBM Corp. 2001, 2004
   3  * Copyright (c) 1999-2000 Cisco, Inc.
   4  * Copyright (c) 1999-2001 Motorola, Inc.
   5  *
   6  * This file is part of the SCTP kernel implementation
   7  *
   8  * These functions handle output processing.
   9  *
  10  * This SCTP implementation is free software;
  11  * you can redistribute it and/or modify it under the terms of
  12  * the GNU General Public License as published by
  13  * the Free Software Foundation; either version 2, or (at your option)
  14  * any later version.
  15  *
  16  * This SCTP implementation is distributed in the hope that it
  17  * will be useful, but WITHOUT ANY WARRANTY; without even the implied
  18  *                 ************************
  19  * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  20  * See the GNU General Public License for more details.
  21  *
  22  * You should have received a copy of the GNU General Public License
  23  * along with GNU CC; see the file COPYING.  If not, see
  24  * <http://www.gnu.org/licenses/>.
  25  *
  26  * Please send any bug reports or fixes you make to the
  27  * email address(es):
  28  *    lksctp developers <linux-sctp@vger.kernel.org>
  29  *
  30  * Written or modified by:
  31  *    La Monte H.P. Yarroll <piggy@acm.org>
  32  *    Karl Knutson          <karl@athena.chicago.il.us>
  33  *    Jon Grimm             <jgrimm@austin.ibm.com>
  34  *    Sridhar Samudrala     <sri@us.ibm.com>
  35  */
  36
  37 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  38
  39 #include <linux/types.h>
  40 #include <linux/kernel.h>
  41 #include <linux/wait.h>
  42 #include <linux/time.h>
  43 #include <linux/ip.h>
  44 #include <linux/ipv6.h>
  45 #include <linux/init.h>
  46 #include <linux/slab.h>
  47 #include <net/inet_ecn.h>
  48 #include <net/ip.h>
  49 #include <net/icmp.h>
  50 #include <net/net_namespace.h>
  51
  52 #include <linux/socket.h> /* for sa_family_t */
  53 #include <net/sock.h>
  54
  55 #include <net/sctp/sctp.h>
  56 #include <net/sctp/sm.h>
  57 #include <net/sctp/checksum.h>
  58
  59 /* Forward declarations for private helpers. */
  60 static sctp_xmit_t __sctp_packet_append_chunk(struct sctp_packet *packet,
  61                                               struct sctp_chunk *chunk);
  62 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
  63                                            struct sctp_chunk *chunk);
  64 static void sctp_packet_append_data(struct sctp_packet *packet,
  65                                            struct sctp_chunk *chunk);
  66 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
  67                                         struct sctp_chunk *chunk,
  68                                         u16 chunk_len);
  69
  70 static void sctp_packet_reset(struct sctp_packet *packet)
  71 {
  72         packet->size = packet->overhead;
  73         packet->has_cookie_echo = 0;
  74         packet->has_sack = 0;
  75         packet->has_data = 0;
  76         packet->has_auth = 0;
  77         packet->ipfragok = 0;
  78         packet->auth = NULL;
  79 }
  80
  81 /* Config a packet.
  82  * This appears to be a followup set of initializations.
  83  */
  84 struct sctp_packet *sctp_packet_config(struct sctp_packet *packet,
  85                                        __u32 vtag, int ecn_capable)
  86 {
  87         struct sctp_transport *tp = packet->transport;
  88         struct sctp_association *asoc = tp->asoc;
  89
  90         pr_debug("%s: packet:%p vtag:0x%x\n", __func__, packet, vtag);
  91
  92         packet->vtag = vtag;
  93
  94         if (asoc && tp->dst) {
  95                 struct sock *sk = asoc->base.sk;
  96
  97                 rcu_read_lock();
  98                 if (__sk_dst_get(sk) != tp->dst) {
  99                         dst_hold(tp->dst);
 100                         sk_setup_caps(sk, tp->dst);
 101                 }
 102
 103                 if (sk_can_gso(sk)) {
 104                         struct net_device *dev = tp->dst->dev;
 105
 106                         packet->max_size = dev->gso_max_size;
 107                 } else {
 108                         packet->max_size = asoc->pathmtu;
 109                 }
 110                 rcu_read_unlock();
 111
 112         } else {
 113                 packet->max_size = tp->pathmtu;
 114         }
 115
 116         if (ecn_capable && sctp_packet_empty(packet)) {
 117                 struct sctp_chunk *chunk;
 118
 119                 /* If there a is a prepend chunk stick it on the list before
 120                  * any other chunks get appended.
 121                  */
 122                 chunk = sctp_get_ecne_prepend(asoc);
 123                 if (chunk)
 124                         sctp_packet_append_chunk(packet, chunk);
 125         }
 126
 127         return packet;
 128 }
 129
 130 /* Initialize the packet structure. */
 131 struct sctp_packet *sctp_packet_init(struct sctp_packet *packet,
 132                                      struct sctp_transport *transport,
 133                                      __u16 sport, __u16 dport)
 134 {
 135         struct sctp_association *asoc = transport->asoc;
 136         size_t overhead;
 137
 138         pr_debug("%s: packet:%p transport:%p\n", __func__, packet, transport);
 139
 140         packet->transport = transport;
 141         packet->source_port = sport;
 142         packet->destination_port = dport;
 143         INIT_LIST_HEAD(&packet->chunk_list);
 144         if (asoc) {
 145                 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
 146                 overhead = sp->pf->af->net_header_len;
 147         } else {
 148                 overhead = sizeof(struct ipv6hdr);
 149         }
 150         overhead += sizeof(struct sctphdr);
 151         packet->overhead = overhead;
 152         sctp_packet_reset(packet);
 153         packet->vtag = 0;
 154
 155         return packet;
 156 }
 157
 158 /* Free a packet.  */
 159 void sctp_packet_free(struct sctp_packet *packet)
 160 {
 161         struct sctp_chunk *chunk, *tmp;
 162
 163         pr_debug("%s: packet:%p\n", __func__, packet);
 164
 165         list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
 166                 list_del_init(&chunk->list);
 167                 sctp_chunk_free(chunk);
 168         }
 169 }
 170
 171 /* This routine tries to append the chunk to the offered packet. If adding
 172  * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
 173  * is not present in the packet, it transmits the input packet.
 174  * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
 175  * as it can fit in the packet, but any more data that does not fit in this
 176  * packet can be sent only after receiving the COOKIE_ACK.
 177  */
 178 sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
 179                                        struct sctp_chunk *chunk,
 180                                        int one_packet, gfp_t gfp)
 181 {
 182         sctp_xmit_t retval;
 183
 184         pr_debug("%s: packet:%p size:%Zu chunk:%p size:%d\n", __func__,
 185                  packet, packet->size, chunk, chunk->skb ? chunk->skb->len : -1);
 186
 187         switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
 188         case SCTP_XMIT_PMTU_FULL:
 189                 if (!packet->has_cookie_echo) {
 190                         int error = 0;
 191
 192                         error = sctp_packet_transmit(packet, gfp);
 193                         if (error < 0)
 194                                 chunk->skb->sk->sk_err = -error;
 195
 196                         /* If we have an empty packet, then we can NOT ever
 197                          * return PMTU_FULL.
 198                          */
 199                         if (!one_packet)
 200                                 retval = sctp_packet_append_chunk(packet,
 201                                                                   chunk);
 202                 }
 203                 break;
 204
 205         case SCTP_XMIT_RWND_FULL:
 206         case SCTP_XMIT_OK:
 207         case SCTP_XMIT_DELAY:
 208                 break;
 209         }
 210
 211         return retval;
 212 }
 213
 214 /* Try to bundle an auth chunk into the packet. */
 215 static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
 216                                            struct sctp_chunk *chunk)
 217 {
 218         struct sctp_association *asoc = pkt->transport->asoc;
 219         struct sctp_chunk *auth;
 220         sctp_xmit_t retval = SCTP_XMIT_OK;
 221
 222         /* if we don't have an association, we can't do authentication */
 223         if (!asoc)
 224                 return retval;
 225
 226         /* See if this is an auth chunk we are bundling or if
 227          * auth is already bundled.
 228          */
 229         if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth)
 230                 return retval;
 231
 232         /* if the peer did not request this chunk to be authenticated,
 233          * don't do it
 234          */
 235         if (!chunk->auth)
 236                 return retval;
 237
 238         auth = sctp_make_auth(asoc);
 239         if (!auth)
 240                 return retval;
 241
 242         retval = __sctp_packet_append_chunk(pkt, auth);
 243
 244         if (retval != SCTP_XMIT_OK)
 245                 sctp_chunk_free(auth);
 246
 247         return retval;
 248 }
 249
 250 /* Try to bundle a SACK with the packet. */
 251 static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
 252                                            struct sctp_chunk *chunk)
 253 {
 254         sctp_xmit_t retval = SCTP_XMIT_OK;
 255
 256         /* If sending DATA and haven't aleady bundled a SACK, try to
 257          * bundle one in to the packet.
 258          */
 259         if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
 260             !pkt->has_cookie_echo) {
 261                 struct sctp_association *asoc;
 262                 struct timer_list *timer;
 263                 asoc = pkt->transport->asoc;
 264                 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
 265
 266                 /* If the SACK timer is running, we have a pending SACK */
 267                 if (timer_pending(timer)) {
 268                         struct sctp_chunk *sack;
 269
 270                         if (pkt->transport->sack_generation !=
 271                             pkt->transport->asoc->peer.sack_generation)
 272                                 return retval;
 273
 274                         asoc->a_rwnd = asoc->rwnd;
 275                         sack = sctp_make_sack(asoc);
 276                         if (sack) {
 277                                 retval = __sctp_packet_append_chunk(pkt, sack);
 278                                 if (retval != SCTP_XMIT_OK) {
 279                                         sctp_chunk_free(sack);
 280                                         goto out;
 281                                 }
 282                                 asoc->peer.sack_needed = 0;
 283                                 if (del_timer(timer))
 284                                         sctp_association_put(asoc);
 285                         }
 286                 }
 287         }
 288 out:
 289         return retval;
 290 }
 291
 292
 293 /* Append a chunk to the offered packet reporting back any inability to do
 294  * so.
 295  */
 296 static sctp_xmit_t __sctp_packet_append_chunk(struct sctp_packet *packet,
 297                                               struct sctp_chunk *chunk)
 298 {
 299         sctp_xmit_t retval = SCTP_XMIT_OK;
 300         __u16 chunk_len = SCTP_PAD4(ntohs(chunk->chunk_hdr->length));
 301
 302         /* Check to see if this chunk will fit into the packet */
 303         retval = sctp_packet_will_fit(packet, chunk, chunk_len);
 304         if (retval != SCTP_XMIT_OK)
 305                 goto finish;
 306
 307         /* We believe that this chunk is OK to add to the packet */
 308         switch (chunk->chunk_hdr->type) {
 309         case SCTP_CID_DATA:
 310                 /* Account for the data being in the packet */
 311                 sctp_packet_append_data(packet, chunk);
 312                 /* Disallow SACK bundling after DATA. */
 313                 packet->has_sack = 1;
 314                 /* Disallow AUTH bundling after DATA */
 315                 packet->has_auth = 1;
 316                 /* Let it be knows that packet has DATA in it */
 317                 packet->has_data = 1;
 318                 /* timestamp the chunk for rtx purposes */
 319                 chunk->sent_at = jiffies;
 320                 /* Mainly used for prsctp RTX policy */
 321                 chunk->sent_count++;
 322                 break;
 323         case SCTP_CID_COOKIE_ECHO:
 324                 packet->has_cookie_echo = 1;
 325                 break;
 326
 327         case SCTP_CID_SACK:
 328                 packet->has_sack = 1;
 329                 if (chunk->asoc)
 330                         chunk->asoc->stats.osacks++;
 331                 break;
 332
 333         case SCTP_CID_AUTH:
 334                 packet->has_auth = 1;
 335                 packet->auth = chunk;
 336                 break;
 337         }
 338
 339         /* It is OK to send this chunk.  */
 340         list_add_tail(&chunk->list, &packet->chunk_list);
 341         packet->size += chunk_len;
 342         chunk->transport = packet->transport;
 343 finish:
 344         return retval;
 345 }
 346
 347 /* Append a chunk to the offered packet reporting back any inability to do
 348  * so.
 349  */
 350 sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
 351                                      struct sctp_chunk *chunk)
 352 {
 353         sctp_xmit_t retval = SCTP_XMIT_OK;
 354
 355         pr_debug("%s: packet:%p chunk:%p\n", __func__, packet, chunk);
 356
 357         /* Data chunks are special.  Before seeing what else we can
 358          * bundle into this packet, check to see if we are allowed to
 359          * send this DATA.
 360          */
 361         if (sctp_chunk_is_data(chunk)) {
 362                 retval = sctp_packet_can_append_data(packet, chunk);
 363                 if (retval != SCTP_XMIT_OK)
 364                         goto finish;
 365         }
 366
 367         /* Try to bundle AUTH chunk */
 368         retval = sctp_packet_bundle_auth(packet, chunk);
 369         if (retval != SCTP_XMIT_OK)
 370                 goto finish;
 371
 372         /* Try to bundle SACK chunk */
 373         retval = sctp_packet_bundle_sack(packet, chunk);
 374         if (retval != SCTP_XMIT_OK)
 375                 goto finish;
 376
 377         retval = __sctp_packet_append_chunk(packet, chunk);
 378
 379 finish:
 380         return retval;
 381 }
 382
 383 static void sctp_packet_release_owner(struct sk_buff *skb)
 384 {
 385         sk_free(skb->sk);
 386 }
 387
 388 static void sctp_packet_set_owner_w(struct sk_buff *skb, struct sock *sk)
 389 {
 390         skb_orphan(skb);
 391         skb->sk = sk;
 392         skb->destructor = sctp_packet_release_owner;
 393
 394         /*
 395          * The data chunks have already been accounted for in sctp_sendmsg(),
 396          * therefore only reserve a single byte to keep socket around until
 397          * the packet has been transmitted.
 398          */
 399         atomic_inc(&sk->sk_wmem_alloc);
 400 }
 401
 402 /* All packets are sent to the network through this function from
 403  * sctp_outq_tail().
 404  *
 405  * The return value is a normal kernel error return value.
 406  */
 407 int sctp_packet_transmit(struct sctp_packet *packet, gfp_t gfp)
 408 {
 409         struct sctp_transport *tp = packet->transport;
 410         struct sctp_association *asoc = tp->asoc;
 411         struct sctphdr *sh;
 412         struct sk_buff *nskb = NULL, *head = NULL;
 413         struct sctp_chunk *chunk, *tmp;
 414         struct sock *sk;
 415         int err = 0;
 416         int padding;            /* How much padding do we need?  */
 417         int pkt_size;
 418         __u8 has_data = 0;
 419         int gso = 0;
 420         int pktcount = 0;
 421         int auth_len = 0;
 422         struct dst_entry *dst;
 423         unsigned char *auth = NULL;     /* pointer to auth in skb data */
 424
 425         pr_debug("%s: packet:%p\n", __func__, packet);
 426
 427         /* Do NOT generate a chunkless packet. */
 428         if (list_empty(&packet->chunk_list))
 429                 return err;
 430
 431         /* Set up convenience variables... */
 432         chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
 433         sk = chunk->skb->sk;
 434
 435         /* Allocate the head skb, or main one if not in GSO */
 436         if (packet->size > tp->pathmtu && !packet->ipfragok) {
 437                 if (sk_can_gso(sk)) {
 438                         gso = 1;
 439                         pkt_size = packet->overhead;
 440                 } else {
 441                         /* If this happens, we trash this packet and try
 442                          * to build a new one, hopefully correct this
 443                          * time. Application may notice this error.
 444                          */
 445                         pr_err_once("Trying to GSO but underlying device doesn't support it.");
 446                         goto err;
 447                 }
 448         } else {
 449                 pkt_size = packet->size;
 450         }
 451         head = alloc_skb(pkt_size + MAX_HEADER, gfp);
 452         if (!head)
 453                 goto err;
 454         if (gso) {
 455                 NAPI_GRO_CB(head)->last = head;
 456                 skb_shinfo(head)->gso_type = sk->sk_gso_type;
 457         }
 458
 459         /* Make sure the outbound skb has enough header room reserved. */
 460         skb_reserve(head, packet->overhead + MAX_HEADER);
 461
 462         /* Set the owning socket so that we know where to get the
 463          * destination IP address.
 464          */
 465         sctp_packet_set_owner_w(head, sk);
 466
 467         if (!sctp_transport_dst_check(tp)) {
 468                 sctp_transport_route(tp, NULL, sctp_sk(sk));
 469                 if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) {
 470                         sctp_assoc_sync_pmtu(sk, asoc);
 471                 }
 472         }
 473         dst = dst_clone(tp->dst);
 474         if (!dst) {
 475                 if (asoc)
 476                         IP_INC_STATS(sock_net(asoc->base.sk),
 477                                      IPSTATS_MIB_OUTNOROUTES);
 478                 goto nodst;
 479         }
 480         skb_dst_set(head, dst);
 481
 482         /* Build the SCTP header.  */
 483         sh = (struct sctphdr *)skb_push(head, sizeof(struct sctphdr));
 484         skb_reset_transport_header(head);
 485         sh->source = htons(packet->source_port);
 486         sh->dest   = htons(packet->destination_port);
 487
 488         /* From 6.8 Adler-32 Checksum Calculation:
 489          * After the packet is constructed (containing the SCTP common
 490          * header and one or more control or DATA chunks), the
 491          * transmitter shall:
 492          *
 493          * 1) Fill in the proper Verification Tag in the SCTP common
 494          *    header and initialize the checksum field to 0's.
 495          */
 496         sh->vtag     = htonl(packet->vtag);
 497         sh->checksum = 0;
 498
 499         pr_debug("***sctp_transmit_packet***\n");
 500
 501         do {
 502                 /* Set up convenience variables... */
 503                 chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
 504                 pktcount++;
 505
 506                 /* Calculate packet size, so it fits in PMTU. Leave
 507                  * other chunks for the next packets.
 508                  */
 509                 if (gso) {
 510                         pkt_size = packet->overhead;
 511                         list_for_each_entry(chunk, &packet->chunk_list, list) {
 512                                 int padded = SCTP_PAD4(chunk->skb->len);
 513
 514                                 if (chunk == packet->auth)
 515                                         auth_len = padded;
 516                                 else if (auth_len + padded + packet->overhead >
 517                                          tp->pathmtu)
 518                                         goto nomem;
 519                                 else if (pkt_size + padded > tp->pathmtu)
 520                                         break;
 521                                 pkt_size += padded;
 522                         }
 523
 524                         /* Allocate a new skb. */
 525                         nskb = alloc_skb(pkt_size + MAX_HEADER, gfp);
 526                         if (!nskb)
 527                                 goto nomem;
 528
 529                         /* Make sure the outbound skb has enough header
 530                          * room reserved.
 531                          */
 532                         skb_reserve(nskb, packet->overhead + MAX_HEADER);
 533                 } else {
 534                         nskb = head;
 535                 }
 536
 537                 /**
 538                  * 3.2  Chunk Field Descriptions
 539                  *
 540                  * The total length of a chunk (including Type, Length and
 541                  * Value fields) MUST be a multiple of 4 bytes.  If the length
 542                  * of the chunk is not a multiple of 4 bytes, the sender MUST
 543                  * pad the chunk with all zero bytes and this padding is not
 544                  * included in the chunk length field.  The sender should
 545                  * never pad with more than 3 bytes.
 546                  *
 547                  * [This whole comment explains SCTP_PAD4() below.]
 548                  */
 549
 550                 pkt_size -= packet->overhead;
 551                 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
 552                         list_del_init(&chunk->list);
 553                         if (sctp_chunk_is_data(chunk)) {
 554                                 /* 6.3.1 C4) When data is in flight and when allowed
 555                                  * by rule C5, a new RTT measurement MUST be made each
 556                                  * round trip.  Furthermore, new RTT measurements
 557                                  * SHOULD be made no more than once per round-trip
 558                                  * for a given destination transport address.
 559                                  */
 560
 561                                 if (!chunk->resent && !tp->rto_pending) {
 562                                         chunk->rtt_in_progress = 1;
 563                                         tp->rto_pending = 1;
 564                                 }
 565
 566                                 has_data = 1;
 567                         }
 568
 569                         padding = SCTP_PAD4(chunk->skb->len) - chunk->skb->len;
 570                         if (padding)
 571                                 memset(skb_put(chunk->skb, padding), 0, padding);
 572
 573                         /* if this is the auth chunk that we are adding,
 574                          * store pointer where it will be added and put
 575                          * the auth into the packet.
 576                          */
 577                         if (chunk == packet->auth)
 578                                 auth = skb_tail_pointer(nskb);
 579
 580                         memcpy(skb_put(nskb, chunk->skb->len),
 581                                chunk->skb->data, chunk->skb->len);
 582
 583                         pr_debug("*** Chunk:%p[%s] %s 0x%x, length:%d, chunk->skb->len:%d, rtt_in_progress:%d\n",
 584                                  chunk,
 585                                  sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)),
 586                                  chunk->has_tsn ? "TSN" : "No TSN",
 587                                  chunk->has_tsn ? ntohl(chunk->subh.data_hdr->tsn) : 0,
 588                                  ntohs(chunk->chunk_hdr->length), chunk->skb->len,
 589                                  chunk->rtt_in_progress);
 590
 591                         /* If this is a control chunk, this is our last
 592                          * reference. Free data chunks after they've been
 593                          * acknowledged or have failed.
 594                          * Re-queue auth chunks if needed.
 595                          */
 596                         pkt_size -= SCTP_PAD4(chunk->skb->len);
 597
 598                         if (!sctp_chunk_is_data(chunk) && chunk != packet->auth)
 599                                 sctp_chunk_free(chunk);
 600
 601                         if (!pkt_size)
 602                                 break;
 603                 }
 604
 605                 /* SCTP-AUTH, Section 6.2
 606                  *    The sender MUST calculate the MAC as described in RFC2104 [2]
 607                  *    using the hash function H as described by the MAC Identifier and
 608                  *    the shared association key K based on the endpoint pair shared key
 609                  *    described by the shared key identifier.  The 'data' used for the
 610                  *    computation of the AUTH-chunk is given by the AUTH chunk with its
 611                  *    HMAC field set to zero (as shown in Figure 6) followed by all
 612                  *    chunks that are placed after the AUTH chunk in the SCTP packet.
 613                  */
 614                 if (auth)
 615                         sctp_auth_calculate_hmac(asoc, nskb,
 616                                                  (struct sctp_auth_chunk *)auth,
 617                                                  gfp);
 618
 619                 if (packet->auth) {
 620                         if (!list_empty(&packet->chunk_list)) {
 621                                 /* We will generate more packets, so re-queue
 622                                  * auth chunk.
 623                                  */
 624                                 list_add(&packet->auth->list,
 625                                          &packet->chunk_list);
 626                         } else {
 627                                 sctp_chunk_free(packet->auth);
 628                                 packet->auth = NULL;
 629                         }
 630                 }
 631
 632                 if (!gso)
 633                         break;
 634
 635                 if (skb_gro_receive(&head, nskb)) {
 636                         kfree_skb(nskb);
 637                         goto nomem;
 638                 }
 639                 nskb = NULL;
 640                 if (WARN_ON_ONCE(skb_shinfo(head)->gso_segs >=
 641                                  sk->sk_gso_max_segs))
 642                         goto nomem;
 643         } while (!list_empty(&packet->chunk_list));
 644
 645         /* 2) Calculate the Adler-32 checksum of the whole packet,
 646          *    including the SCTP common header and all the
 647          *    chunks.
 648          *
 649          * Note: Adler-32 is no longer applicable, as has been replaced
 650          * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
 651          *
 652          * If it's a GSO packet, it's postponed to sctp_skb_segment.
 653          */
 654         if (!sctp_checksum_disable || gso) {
 655                 if (!gso && (!(dst->dev->features & NETIF_F_SCTP_CRC) ||
 656                              dst_xfrm(dst) || packet->ipfragok)) {
 657                         sh->checksum = sctp_compute_cksum(head, 0);
 658                 } else {
 659                         /* no need to seed pseudo checksum for SCTP */
 660                         head->ip_summed = CHECKSUM_PARTIAL;
 661                         head->csum_start = skb_transport_header(head) - head->head;
 662                         head->csum_offset = offsetof(struct sctphdr, checksum);
 663                 }
 664         }
 665
 666         /* IP layer ECN support
 667          * From RFC 2481
 668          *  "The ECN-Capable Transport (ECT) bit would be set by the
 669          *   data sender to indicate that the end-points of the
 670          *   transport protocol are ECN-capable."
 671          *
 672          * Now setting the ECT bit all the time, as it should not cause
 673          * any problems protocol-wise even if our peer ignores it.
 674          *
 675          * Note: The works for IPv6 layer checks this bit too later
 676          * in transmission.  See IP6_ECN_flow_xmit().
 677          */
 678         tp->af_specific->ecn_capable(sk);
 679
 680         /* Set up the IP options.  */
 681         /* BUG: not implemented
 682          * For v4 this all lives somewhere in sk->sk_opt...
 683          */
 684
 685         /* Dump that on IP!  */
 686         if (asoc) {
 687                 asoc->stats.opackets += pktcount;
 688                 if (asoc->peer.last_sent_to != tp)
 689                         /* Considering the multiple CPU scenario, this is a
 690                          * "correcter" place for last_sent_to.  --xguo
 691                          */
 692                         asoc->peer.last_sent_to = tp;
 693         }
 694
 695         if (has_data) {
 696                 struct timer_list *timer;
 697                 unsigned long timeout;
 698
 699                 /* Restart the AUTOCLOSE timer when sending data. */
 700                 if (sctp_state(asoc, ESTABLISHED) &&
 701                     asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE]) {
 702                         timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
 703                         timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
 704
 705                         if (!mod_timer(timer, jiffies + timeout))
 706                                 sctp_association_hold(asoc);
 707                 }
 708         }
 709
 710         pr_debug("***sctp_transmit_packet*** skb->len:%d\n", head->len);
 711
 712         if (gso) {
 713                 /* Cleanup our debris for IP stacks */
 714                 memset(head->cb, 0, max(sizeof(struct inet_skb_parm),
 715                                         sizeof(struct inet6_skb_parm)));
 716
 717                 skb_shinfo(head)->gso_segs = pktcount;
 718                 skb_shinfo(head)->gso_size = GSO_BY_FRAGS;
 719
 720                 /* We have to refresh this in case we are xmiting to
 721                  * more than one transport at a time
 722                  */
 723                 rcu_read_lock();
 724                 if (__sk_dst_get(sk) != tp->dst) {
 725                         dst_hold(tp->dst);
 726                         sk_setup_caps(sk, tp->dst);
 727                 }
 728                 rcu_read_unlock();
 729         }
 730         head->ignore_df = packet->ipfragok;
 731         tp->af_specific->sctp_xmit(head, tp);
 732         goto out;
 733
 734 nomem:
 735         if (packet->auth && list_empty(&packet->auth->list))
 736                 sctp_chunk_free(packet->auth);
 737
 738 nodst:
 739         /* FIXME: Returning the 'err' will effect all the associations
 740          * associated with a socket, although only one of the paths of the
 741          * association is unreachable.
 742          * The real failure of a transport or association can be passed on
 743          * to the user via notifications. So setting this error may not be
 744          * required.
 745          */
 746          /* err = -EHOSTUNREACH; */
 747         kfree_skb(head);
 748
 749 err:
 750         list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
 751                 list_del_init(&chunk->list);
 752                 if (!sctp_chunk_is_data(chunk))
 753                         sctp_chunk_free(chunk);
 754         }
 755
 756 out:
 757         sctp_packet_reset(packet);
 758         return err;
 759 }
 760
 761 /********************************************************************
 762  * 2nd Level Abstractions
 763  ********************************************************************/
 764
 765 /* This private function check to see if a chunk can be added */
 766 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
 767                                            struct sctp_chunk *chunk)
 768 {
 769         size_t datasize, rwnd, inflight, flight_size;
 770         struct sctp_transport *transport = packet->transport;
 771         struct sctp_association *asoc = transport->asoc;
 772         struct sctp_outq *q = &asoc->outqueue;
 773
 774         /* RFC 2960 6.1  Transmission of DATA Chunks
 775          *
 776          * A) At any given time, the data sender MUST NOT transmit new data to
 777          * any destination transport address if its peer's rwnd indicates
 778          * that the peer has no buffer space (i.e. rwnd is 0, see Section
 779          * 6.2.1).  However, regardless of the value of rwnd (including if it
 780          * is 0), the data sender can always have one DATA chunk in flight to
 781          * the receiver if allowed by cwnd (see rule B below).  This rule
 782          * allows the sender to probe for a change in rwnd that the sender
 783          * missed due to the SACK having been lost in transit from the data
 784          * receiver to the data sender.
 785          */
 786
 787         rwnd = asoc->peer.rwnd;
 788         inflight = q->outstanding_bytes;
 789         flight_size = transport->flight_size;
 790
 791         datasize = sctp_data_size(chunk);
 792
 793         if (datasize > rwnd && inflight > 0)
 794                 /* We have (at least) one data chunk in flight,
 795                  * so we can't fall back to rule 6.1 B).
 796                  */
 797                 return SCTP_XMIT_RWND_FULL;
 798
 799         /* RFC 2960 6.1  Transmission of DATA Chunks
 800          *
 801          * B) At any given time, the sender MUST NOT transmit new data
 802          * to a given transport address if it has cwnd or more bytes
 803          * of data outstanding to that transport address.
 804          */
 805         /* RFC 7.2.4 & the Implementers Guide 2.8.
 806          *
 807          * 3) ...
 808          *    When a Fast Retransmit is being performed the sender SHOULD
 809          *    ignore the value of cwnd and SHOULD NOT delay retransmission.
 810          */
 811         if (chunk->fast_retransmit != SCTP_NEED_FRTX &&
 812             flight_size >= transport->cwnd)
 813                 return SCTP_XMIT_RWND_FULL;
 814
 815         /* Nagle's algorithm to solve small-packet problem:
 816          * Inhibit the sending of new chunks when new outgoing data arrives
 817          * if any previously transmitted data on the connection remains
 818          * unacknowledged.
 819          */
 820
 821         if (sctp_sk(asoc->base.sk)->nodelay)
 822                 /* Nagle disabled */
 823                 return SCTP_XMIT_OK;
 824
 825         if (!sctp_packet_empty(packet))
 826                 /* Append to packet */
 827                 return SCTP_XMIT_OK;
 828
 829         if (inflight == 0)
 830                 /* Nothing unacked */
 831                 return SCTP_XMIT_OK;
 832
 833         if (!sctp_state(asoc, ESTABLISHED))
 834                 return SCTP_XMIT_OK;
 835
 836         /* Check whether this chunk and all the rest of pending data will fit
 837          * or delay in hopes of bundling a full sized packet.
 838          */
 839         if (chunk->skb->len + q->out_qlen >
 840                 transport->pathmtu - packet->overhead - sizeof(sctp_data_chunk_t) - 4)
 841                 /* Enough data queued to fill a packet */
 842                 return SCTP_XMIT_OK;
 843
 844         /* Don't delay large message writes that may have been fragmented */
 845         if (!chunk->msg->can_delay)
 846                 return SCTP_XMIT_OK;
 847
 848         /* Defer until all data acked or packet full */
 849         return SCTP_XMIT_DELAY;
 850 }
 851
 852 /* This private function does management things when adding DATA chunk */
 853 static void sctp_packet_append_data(struct sctp_packet *packet,
 854                                 struct sctp_chunk *chunk)
 855 {
 856         struct sctp_transport *transport = packet->transport;
 857         size_t datasize = sctp_data_size(chunk);
 858         struct sctp_association *asoc = transport->asoc;
 859         u32 rwnd = asoc->peer.rwnd;
 860
 861         /* Keep track of how many bytes are in flight over this transport. */
 862         transport->flight_size += datasize;
 863
 864         /* Keep track of how many bytes are in flight to the receiver. */
 865         asoc->outqueue.outstanding_bytes += datasize;
 866
 867         /* Update our view of the receiver's rwnd. */
 868         if (datasize < rwnd)
 869                 rwnd -= datasize;
 870         else
 871                 rwnd = 0;
 872
 873         asoc->peer.rwnd = rwnd;
 874         /* Has been accepted for transmission. */
 875         if (!asoc->peer.prsctp_capable)
 876                 chunk->msg->can_abandon = 0;
 877         sctp_chunk_assign_tsn(chunk);
 878         sctp_chunk_assign_ssn(chunk);
 879 }
 880
 881 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
 882                                         struct sctp_chunk *chunk,
 883                                         u16 chunk_len)
 884 {
 885         size_t psize, pmtu, maxsize;
 886         sctp_xmit_t retval = SCTP_XMIT_OK;
 887
 888         psize = packet->size;
 889         if (packet->transport->asoc)
 890                 pmtu = packet->transport->asoc->pathmtu;
 891         else
 892                 pmtu = packet->transport->pathmtu;
 893
 894         /* Decide if we need to fragment or resubmit later. */
 895         if (psize + chunk_len > pmtu) {
 896                 /* It's OK to fragment at IP level if any one of the following
 897                  * is true:
 898                  *      1. The packet is empty (meaning this chunk is greater
 899                  *         the MTU)
 900                  *      2. The packet doesn't have any data in it yet and data
 901                  *         requires authentication.
 902                  */
 903                 if (sctp_packet_empty(packet) ||
 904                     (!packet->has_data && chunk->auth)) {
 905                         /* We no longer do re-fragmentation.
 906                          * Just fragment at the IP layer, if we
 907                          * actually hit this condition
 908                          */
 909                         packet->ipfragok = 1;
 910                         goto out;
 911                 }
 912
 913                 /* Similarly, if this chunk was built before a PMTU
 914                  * reduction, we have to fragment it at IP level now. So
 915                  * if the packet already contains something, we need to
 916                  * flush.
 917                  */
 918                 maxsize = pmtu - packet->overhead;
 919                 if (packet->auth)
 920                         maxsize -= SCTP_PAD4(packet->auth->skb->len);
 921                 if (chunk_len > maxsize)
 922                         retval = SCTP_XMIT_PMTU_FULL;
 923
 924                 /* It is also okay to fragment if the chunk we are
 925                  * adding is a control chunk, but only if current packet
 926                  * is not a GSO one otherwise it causes fragmentation of
 927                  * a large frame. So in this case we allow the
 928                  * fragmentation by forcing it to be in a new packet.
 929                  */
 930                 if (!sctp_chunk_is_data(chunk) && packet->has_data)
 931                         retval = SCTP_XMIT_PMTU_FULL;
 932
 933                 if (psize + chunk_len > packet->max_size)
 934                         /* Hit GSO/PMTU limit, gotta flush */
 935                         retval = SCTP_XMIT_PMTU_FULL;
 936
 937                 if (!packet->transport->burst_limited &&
 938                     psize + chunk_len > (packet->transport->cwnd >> 1))
 939                         /* Do not allow a single GSO packet to use more
 940                          * than half of cwnd.
 941                          */
 942                         retval = SCTP_XMIT_PMTU_FULL;
 943
 944                 if (packet->transport->burst_limited &&
 945                     psize + chunk_len > (packet->transport->burst_limited >> 1))
 946                         /* Do not allow a single GSO packet to use more
 947                          * than half of original cwnd.
 948                          */
 949                         retval = SCTP_XMIT_PMTU_FULL;
 950                 /* Otherwise it will fit in the GSO packet */
 951         }
 952
 953 out:
 954         return retval;
 955 }