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         int error = 0;
 184
 185         pr_debug("%s: packet:%p size:%Zu chunk:%p size:%d\n", __func__,
 186                  packet, packet->size, chunk, chunk->skb ? chunk->skb->len : -1);
 187
 188         switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
 189         case SCTP_XMIT_PMTU_FULL:
 190                 if (!packet->has_cookie_echo) {
 191                         error = sctp_packet_transmit(packet, gfp);
 192                         if (error < 0)
 193                                 chunk->skb->sk->sk_err = -error;
 194
 195                         /* If we have an empty packet, then we can NOT ever
 196                          * return PMTU_FULL.
 197                          */
 198                         if (!one_packet)
 199                                 retval = sctp_packet_append_chunk(packet,
 200                                                                   chunk);
 201                 }
 202                 break;
 203
 204         case SCTP_XMIT_RWND_FULL:
 205         case SCTP_XMIT_OK:
 206         case SCTP_XMIT_DELAY:
 207                 break;
 208         }
 209
 210         return retval;
 211 }
 212
 213 /* Try to bundle an auth chunk into the packet. */
 214 static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
 215                                            struct sctp_chunk *chunk)
 216 {
 217         struct sctp_association *asoc = pkt->transport->asoc;
 218         struct sctp_chunk *auth;
 219         sctp_xmit_t retval = SCTP_XMIT_OK;
 220
 221         /* if we don't have an association, we can't do authentication */
 222         if (!asoc)
 223                 return retval;
 224
 225         /* See if this is an auth chunk we are bundling or if
 226          * auth is already bundled.
 227          */
 228         if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth)
 229                 return retval;
 230
 231         /* if the peer did not request this chunk to be authenticated,
 232          * don't do it
 233          */
 234         if (!chunk->auth)
 235                 return retval;
 236
 237         auth = sctp_make_auth(asoc);
 238         if (!auth)
 239                 return retval;
 240
 241         retval = __sctp_packet_append_chunk(pkt, auth);
 242
 243         if (retval != SCTP_XMIT_OK)
 244                 sctp_chunk_free(auth);
 245
 246         return retval;
 247 }
 248
 249 /* Try to bundle a SACK with the packet. */
 250 static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
 251                                            struct sctp_chunk *chunk)
 252 {
 253         sctp_xmit_t retval = SCTP_XMIT_OK;
 254
 255         /* If sending DATA and haven't aleady bundled a SACK, try to
 256          * bundle one in to the packet.
 257          */
 258         if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
 259             !pkt->has_cookie_echo) {
 260                 struct sctp_association *asoc;
 261                 struct timer_list *timer;
 262                 asoc = pkt->transport->asoc;
 263                 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
 264
 265                 /* If the SACK timer is running, we have a pending SACK */
 266                 if (timer_pending(timer)) {
 267                         struct sctp_chunk *sack;
 268
 269                         if (pkt->transport->sack_generation !=
 270                             pkt->transport->asoc->peer.sack_generation)
 271                                 return retval;
 272
 273                         asoc->a_rwnd = asoc->rwnd;
 274                         sack = sctp_make_sack(asoc);
 275                         if (sack) {
 276                                 retval = __sctp_packet_append_chunk(pkt, sack);
 277                                 if (retval != SCTP_XMIT_OK) {
 278                                         sctp_chunk_free(sack);
 279                                         goto out;
 280                                 }
 281                                 asoc->peer.sack_needed = 0;
 282                                 if (del_timer(timer))
 283                                         sctp_association_put(asoc);
 284                         }
 285                 }
 286         }
 287 out:
 288         return retval;
 289 }
 290
 291
 292 /* Append a chunk to the offered packet reporting back any inability to do
 293  * so.
 294  */
 295 static sctp_xmit_t __sctp_packet_append_chunk(struct sctp_packet *packet,
 296                                               struct sctp_chunk *chunk)
 297 {
 298         sctp_xmit_t retval = SCTP_XMIT_OK;
 299         __u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length));
 300
 301         /* Check to see if this chunk will fit into the packet */
 302         retval = sctp_packet_will_fit(packet, chunk, chunk_len);
 303         if (retval != SCTP_XMIT_OK)
 304                 goto finish;
 305
 306         /* We believe that this chunk is OK to add to the packet */
 307         switch (chunk->chunk_hdr->type) {
 308         case SCTP_CID_DATA:
 309                 /* Account for the data being in the packet */
 310                 sctp_packet_append_data(packet, chunk);
 311                 /* Disallow SACK bundling after DATA. */
 312                 packet->has_sack = 1;
 313                 /* Disallow AUTH bundling after DATA */
 314                 packet->has_auth = 1;
 315                 /* Let it be knows that packet has DATA in it */
 316                 packet->has_data = 1;
 317                 /* timestamp the chunk for rtx purposes */
 318                 chunk->sent_at = jiffies;
 319                 break;
 320         case SCTP_CID_COOKIE_ECHO:
 321                 packet->has_cookie_echo = 1;
 322                 break;
 323
 324         case SCTP_CID_SACK:
 325                 packet->has_sack = 1;
 326                 if (chunk->asoc)
 327                         chunk->asoc->stats.osacks++;
 328                 break;
 329
 330         case SCTP_CID_AUTH:
 331                 packet->has_auth = 1;
 332                 packet->auth = chunk;
 333                 break;
 334         }
 335
 336         /* It is OK to send this chunk.  */
 337         list_add_tail(&chunk->list, &packet->chunk_list);
 338         packet->size += chunk_len;
 339         chunk->transport = packet->transport;
 340 finish:
 341         return retval;
 342 }
 343
 344 /* Append a chunk to the offered packet reporting back any inability to do
 345  * so.
 346  */
 347 sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
 348                                      struct sctp_chunk *chunk)
 349 {
 350         sctp_xmit_t retval = SCTP_XMIT_OK;
 351
 352         pr_debug("%s: packet:%p chunk:%p\n", __func__, packet, chunk);
 353
 354         /* Data chunks are special.  Before seeing what else we can
 355          * bundle into this packet, check to see if we are allowed to
 356          * send this DATA.
 357          */
 358         if (sctp_chunk_is_data(chunk)) {
 359                 retval = sctp_packet_can_append_data(packet, chunk);
 360                 if (retval != SCTP_XMIT_OK)
 361                         goto finish;
 362         }
 363
 364         /* Try to bundle AUTH chunk */
 365         retval = sctp_packet_bundle_auth(packet, chunk);
 366         if (retval != SCTP_XMIT_OK)
 367                 goto finish;
 368
 369         /* Try to bundle SACK chunk */
 370         retval = sctp_packet_bundle_sack(packet, chunk);
 371         if (retval != SCTP_XMIT_OK)
 372                 goto finish;
 373
 374         retval = __sctp_packet_append_chunk(packet, chunk);
 375
 376 finish:
 377         return retval;
 378 }
 379
 380 static void sctp_packet_release_owner(struct sk_buff *skb)
 381 {
 382         sk_free(skb->sk);
 383 }
 384
 385 static void sctp_packet_set_owner_w(struct sk_buff *skb, struct sock *sk)
 386 {
 387         skb_orphan(skb);
 388         skb->sk = sk;
 389         skb->destructor = sctp_packet_release_owner;
 390
 391         /*
 392          * The data chunks have already been accounted for in sctp_sendmsg(),
 393          * therefore only reserve a single byte to keep socket around until
 394          * the packet has been transmitted.
 395          */
 396         atomic_inc(&sk->sk_wmem_alloc);
 397 }
 398
 399 /* All packets are sent to the network through this function from
 400  * sctp_outq_tail().
 401  *
 402  * The return value is a normal kernel error return value.
 403  */
 404 int sctp_packet_transmit(struct sctp_packet *packet, gfp_t gfp)
 405 {
 406         struct sctp_transport *tp = packet->transport;
 407         struct sctp_association *asoc = tp->asoc;
 408         struct sctphdr *sh;
 409         struct sk_buff *nskb = NULL, *head = NULL;
 410         struct sctp_chunk *chunk, *tmp;
 411         struct sock *sk;
 412         int err = 0;
 413         int padding;            /* How much padding do we need?  */
 414         int pkt_size;
 415         __u8 has_data = 0;
 416         int gso = 0;
 417         int pktcount = 0;
 418         struct dst_entry *dst;
 419         unsigned char *auth = NULL;     /* pointer to auth in skb data */
 420
 421         pr_debug("%s: packet:%p\n", __func__, packet);
 422
 423         /* Do NOT generate a chunkless packet. */
 424         if (list_empty(&packet->chunk_list))
 425                 return err;
 426
 427         /* Set up convenience variables... */
 428         chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
 429         sk = chunk->skb->sk;
 430
 431         /* Allocate the head skb, or main one if not in GSO */
 432         if (packet->size > tp->pathmtu && !packet->ipfragok) {
 433                 if (sk_can_gso(sk)) {
 434                         gso = 1;
 435                         pkt_size = packet->overhead;
 436                 } else {
 437                         /* If this happens, we trash this packet and try
 438                          * to build a new one, hopefully correct this
 439                          * time. Application may notice this error.
 440                          */
 441                         pr_err_once("Trying to GSO but underlying device doesn't support it.");
 442                         goto nomem;
 443                 }
 444         } else {
 445                 pkt_size = packet->size;
 446         }
 447         head = alloc_skb(pkt_size + MAX_HEADER, gfp);
 448         if (!head)
 449                 goto nomem;
 450         if (gso) {
 451                 NAPI_GRO_CB(head)->last = head;
 452                 skb_shinfo(head)->gso_type = sk->sk_gso_type;
 453         }
 454
 455         /* Make sure the outbound skb has enough header room reserved. */
 456         skb_reserve(head, packet->overhead + MAX_HEADER);
 457
 458         /* Set the owning socket so that we know where to get the
 459          * destination IP address.
 460          */
 461         sctp_packet_set_owner_w(head, sk);
 462
 463         if (!sctp_transport_dst_check(tp)) {
 464                 sctp_transport_route(tp, NULL, sctp_sk(sk));
 465                 if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) {
 466                         sctp_assoc_sync_pmtu(sk, asoc);
 467                 }
 468         }
 469         dst = dst_clone(tp->dst);
 470         if (!dst)
 471                 goto no_route;
 472         skb_dst_set(head, dst);
 473
 474         /* Build the SCTP header.  */
 475         sh = (struct sctphdr *)skb_push(head, sizeof(struct sctphdr));
 476         skb_reset_transport_header(head);
 477         sh->source = htons(packet->source_port);
 478         sh->dest   = htons(packet->destination_port);
 479
 480         /* From 6.8 Adler-32 Checksum Calculation:
 481          * After the packet is constructed (containing the SCTP common
 482          * header and one or more control or DATA chunks), the
 483          * transmitter shall:
 484          *
 485          * 1) Fill in the proper Verification Tag in the SCTP common
 486          *    header and initialize the checksum field to 0's.
 487          */
 488         sh->vtag     = htonl(packet->vtag);
 489         sh->checksum = 0;
 490
 491         pr_debug("***sctp_transmit_packet***\n");
 492
 493         do {
 494                 /* Set up convenience variables... */
 495                 chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
 496                 pktcount++;
 497
 498                 /* Calculate packet size, so it fits in PMTU. Leave
 499                  * other chunks for the next packets.
 500                  */
 501                 if (gso) {
 502                         pkt_size = packet->overhead;
 503                         list_for_each_entry(chunk, &packet->chunk_list, list) {
 504                                 int padded = WORD_ROUND(chunk->skb->len);
 505
 506                                 if (pkt_size + padded > tp->pathmtu)
 507                                         break;
 508                                 pkt_size += padded;
 509                         }
 510
 511                         /* Allocate a new skb. */
 512                         nskb = alloc_skb(pkt_size + MAX_HEADER, gfp);
 513                         if (!nskb)
 514                                 goto nomem;
 515
 516                         /* Make sure the outbound skb has enough header
 517                          * room reserved.
 518                          */
 519                         skb_reserve(nskb, packet->overhead + MAX_HEADER);
 520                 } else {
 521                         nskb = head;
 522                 }
 523
 524                 /**
 525                  * 3.2  Chunk Field Descriptions
 526                  *
 527                  * The total length of a chunk (including Type, Length and
 528                  * Value fields) MUST be a multiple of 4 bytes.  If the length
 529                  * of the chunk is not a multiple of 4 bytes, the sender MUST
 530                  * pad the chunk with all zero bytes and this padding is not
 531                  * included in the chunk length field.  The sender should
 532                  * never pad with more than 3 bytes.
 533                  *
 534                  * [This whole comment explains WORD_ROUND() below.]
 535                  */
 536
 537                 pkt_size -= packet->overhead;
 538                 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
 539                         list_del_init(&chunk->list);
 540                         if (sctp_chunk_is_data(chunk)) {
 541                                 /* 6.3.1 C4) When data is in flight and when allowed
 542                                  * by rule C5, a new RTT measurement MUST be made each
 543                                  * round trip.  Furthermore, new RTT measurements
 544                                  * SHOULD be made no more than once per round-trip
 545                                  * for a given destination transport address.
 546                                  */
 547
 548                                 if (!chunk->resent && !tp->rto_pending) {
 549                                         chunk->rtt_in_progress = 1;
 550                                         tp->rto_pending = 1;
 551                                 }
 552
 553                                 has_data = 1;
 554                         }
 555
 556                         padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len;
 557                         if (padding)
 558                                 memset(skb_put(chunk->skb, padding), 0, padding);
 559
 560                         /* if this is the auth chunk that we are adding,
 561                          * store pointer where it will be added and put
 562                          * the auth into the packet.
 563                          */
 564                         if (chunk == packet->auth)
 565                                 auth = skb_tail_pointer(nskb);
 566
 567                         memcpy(skb_put(nskb, chunk->skb->len),
 568                                chunk->skb->data, chunk->skb->len);
 569
 570                         pr_debug("*** Chunk:%p[%s] %s 0x%x, length:%d, chunk->skb->len:%d, rtt_in_progress:%d\n",
 571                                  chunk,
 572                                  sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)),
 573                                  chunk->has_tsn ? "TSN" : "No TSN",
 574                                  chunk->has_tsn ? ntohl(chunk->subh.data_hdr->tsn) : 0,
 575                                  ntohs(chunk->chunk_hdr->length), chunk->skb->len,
 576                                  chunk->rtt_in_progress);
 577
 578                         /* If this is a control chunk, this is our last
 579                          * reference. Free data chunks after they've been
 580                          * acknowledged or have failed.
 581                          * Re-queue auth chunks if needed.
 582                          */
 583                         pkt_size -= WORD_ROUND(chunk->skb->len);
 584
 585                         if (chunk == packet->auth && !list_empty(&packet->chunk_list))
 586                                 list_add(&chunk->list, &packet->chunk_list);
 587                         else if (!sctp_chunk_is_data(chunk))
 588                                 sctp_chunk_free(chunk);
 589
 590                         if (!pkt_size)
 591                                 break;
 592                 }
 593
 594                 /* SCTP-AUTH, Section 6.2
 595                  *    The sender MUST calculate the MAC as described in RFC2104 [2]
 596                  *    using the hash function H as described by the MAC Identifier and
 597                  *    the shared association key K based on the endpoint pair shared key
 598                  *    described by the shared key identifier.  The 'data' used for the
 599                  *    computation of the AUTH-chunk is given by the AUTH chunk with its
 600                  *    HMAC field set to zero (as shown in Figure 6) followed by all
 601                  *    chunks that are placed after the AUTH chunk in the SCTP packet.
 602                  */
 603                 if (auth)
 604                         sctp_auth_calculate_hmac(asoc, nskb,
 605                                                  (struct sctp_auth_chunk *)auth,
 606                                                  gfp);
 607
 608                 if (!gso)
 609                         break;
 610
 611                 if (skb_gro_receive(&head, nskb))
 612                         goto nomem;
 613                 nskb = NULL;
 614                 if (WARN_ON_ONCE(skb_shinfo(head)->gso_segs >=
 615                                  sk->sk_gso_max_segs))
 616                         goto nomem;
 617         } while (!list_empty(&packet->chunk_list));
 618
 619         /* 2) Calculate the Adler-32 checksum of the whole packet,
 620          *    including the SCTP common header and all the
 621          *    chunks.
 622          *
 623          * Note: Adler-32 is no longer applicable, as has been replaced
 624          * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
 625          *
 626          * If it's a GSO packet, it's postponed to sctp_skb_segment.
 627          */
 628         if (!sctp_checksum_disable || gso) {
 629                 if (!gso && (!(dst->dev->features & NETIF_F_SCTP_CRC) ||
 630                              dst_xfrm(dst) || packet->ipfragok)) {
 631                         sh->checksum = sctp_compute_cksum(head, 0);
 632                 } else {
 633                         /* no need to seed pseudo checksum for SCTP */
 634                         head->ip_summed = CHECKSUM_PARTIAL;
 635                         head->csum_start = skb_transport_header(head) - head->head;
 636                         head->csum_offset = offsetof(struct sctphdr, checksum);
 637                 }
 638         }
 639
 640         /* IP layer ECN support
 641          * From RFC 2481
 642          *  "The ECN-Capable Transport (ECT) bit would be set by the
 643          *   data sender to indicate that the end-points of the
 644          *   transport protocol are ECN-capable."
 645          *
 646          * Now setting the ECT bit all the time, as it should not cause
 647          * any problems protocol-wise even if our peer ignores it.
 648          *
 649          * Note: The works for IPv6 layer checks this bit too later
 650          * in transmission.  See IP6_ECN_flow_xmit().
 651          */
 652         tp->af_specific->ecn_capable(sk);
 653
 654         /* Set up the IP options.  */
 655         /* BUG: not implemented
 656          * For v4 this all lives somewhere in sk->sk_opt...
 657          */
 658
 659         /* Dump that on IP!  */
 660         if (asoc) {
 661                 asoc->stats.opackets += pktcount;
 662                 if (asoc->peer.last_sent_to != tp)
 663                         /* Considering the multiple CPU scenario, this is a
 664                          * "correcter" place for last_sent_to.  --xguo
 665                          */
 666                         asoc->peer.last_sent_to = tp;
 667         }
 668
 669         if (has_data) {
 670                 struct timer_list *timer;
 671                 unsigned long timeout;
 672
 673                 /* Restart the AUTOCLOSE timer when sending data. */
 674                 if (sctp_state(asoc, ESTABLISHED) &&
 675                     asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE]) {
 676                         timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
 677                         timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
 678
 679                         if (!mod_timer(timer, jiffies + timeout))
 680                                 sctp_association_hold(asoc);
 681                 }
 682         }
 683
 684         pr_debug("***sctp_transmit_packet*** skb->len:%d\n", head->len);
 685
 686         if (gso) {
 687                 /* Cleanup our debris for IP stacks */
 688                 memset(head->cb, 0, max(sizeof(struct inet_skb_parm),
 689                                         sizeof(struct inet6_skb_parm)));
 690
 691                 skb_shinfo(head)->gso_segs = pktcount;
 692                 skb_shinfo(head)->gso_size = GSO_BY_FRAGS;
 693
 694                 /* We have to refresh this in case we are xmiting to
 695                  * more than one transport at a time
 696                  */
 697                 rcu_read_lock();
 698                 if (__sk_dst_get(sk) != tp->dst) {
 699                         dst_hold(tp->dst);
 700                         sk_setup_caps(sk, tp->dst);
 701                 }
 702                 rcu_read_unlock();
 703         }
 704         head->ignore_df = packet->ipfragok;
 705         tp->af_specific->sctp_xmit(head, tp);
 706
 707 out:
 708         sctp_packet_reset(packet);
 709         return err;
 710 no_route:
 711         kfree_skb(head);
 712         if (nskb != head)
 713                 kfree_skb(nskb);
 714
 715         if (asoc)
 716                 IP_INC_STATS(sock_net(asoc->base.sk), IPSTATS_MIB_OUTNOROUTES);
 717
 718         /* FIXME: Returning the 'err' will effect all the associations
 719          * associated with a socket, although only one of the paths of the
 720          * association is unreachable.
 721          * The real failure of a transport or association can be passed on
 722          * to the user via notifications. So setting this error may not be
 723          * required.
 724          */
 725          /* err = -EHOSTUNREACH; */
 726 err:
 727         /* Control chunks are unreliable so just drop them.  DATA chunks
 728          * will get resent or dropped later.
 729          */
 730
 731         list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
 732                 list_del_init(&chunk->list);
 733                 if (!sctp_chunk_is_data(chunk))
 734                         sctp_chunk_free(chunk);
 735         }
 736         goto out;
 737 nomem:
 738         err = -ENOMEM;
 739         goto err;
 740 }
 741
 742 /********************************************************************
 743  * 2nd Level Abstractions
 744  ********************************************************************/
 745
 746 /* This private function check to see if a chunk can be added */
 747 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
 748                                            struct sctp_chunk *chunk)
 749 {
 750         size_t datasize, rwnd, inflight, flight_size;
 751         struct sctp_transport *transport = packet->transport;
 752         struct sctp_association *asoc = transport->asoc;
 753         struct sctp_outq *q = &asoc->outqueue;
 754
 755         /* RFC 2960 6.1  Transmission of DATA Chunks
 756          *
 757          * A) At any given time, the data sender MUST NOT transmit new data to
 758          * any destination transport address if its peer's rwnd indicates
 759          * that the peer has no buffer space (i.e. rwnd is 0, see Section
 760          * 6.2.1).  However, regardless of the value of rwnd (including if it
 761          * is 0), the data sender can always have one DATA chunk in flight to
 762          * the receiver if allowed by cwnd (see rule B below).  This rule
 763          * allows the sender to probe for a change in rwnd that the sender
 764          * missed due to the SACK having been lost in transit from the data
 765          * receiver to the data sender.
 766          */
 767
 768         rwnd = asoc->peer.rwnd;
 769         inflight = q->outstanding_bytes;
 770         flight_size = transport->flight_size;
 771
 772         datasize = sctp_data_size(chunk);
 773
 774         if (datasize > rwnd && inflight > 0)
 775                 /* We have (at least) one data chunk in flight,
 776                  * so we can't fall back to rule 6.1 B).
 777                  */
 778                 return SCTP_XMIT_RWND_FULL;
 779
 780         /* RFC 2960 6.1  Transmission of DATA Chunks
 781          *
 782          * B) At any given time, the sender MUST NOT transmit new data
 783          * to a given transport address if it has cwnd or more bytes
 784          * of data outstanding to that transport address.
 785          */
 786         /* RFC 7.2.4 & the Implementers Guide 2.8.
 787          *
 788          * 3) ...
 789          *    When a Fast Retransmit is being performed the sender SHOULD
 790          *    ignore the value of cwnd and SHOULD NOT delay retransmission.
 791          */
 792         if (chunk->fast_retransmit != SCTP_NEED_FRTX &&
 793             flight_size >= transport->cwnd)
 794                 return SCTP_XMIT_RWND_FULL;
 795
 796         /* Nagle's algorithm to solve small-packet problem:
 797          * Inhibit the sending of new chunks when new outgoing data arrives
 798          * if any previously transmitted data on the connection remains
 799          * unacknowledged.
 800          */
 801
 802         if (sctp_sk(asoc->base.sk)->nodelay)
 803                 /* Nagle disabled */
 804                 return SCTP_XMIT_OK;
 805
 806         if (!sctp_packet_empty(packet))
 807                 /* Append to packet */
 808                 return SCTP_XMIT_OK;
 809
 810         if (inflight == 0)
 811                 /* Nothing unacked */
 812                 return SCTP_XMIT_OK;
 813
 814         if (!sctp_state(asoc, ESTABLISHED))
 815                 return SCTP_XMIT_OK;
 816
 817         /* Check whether this chunk and all the rest of pending data will fit
 818          * or delay in hopes of bundling a full sized packet.
 819          */
 820         if (chunk->skb->len + q->out_qlen >
 821                 transport->pathmtu - packet->overhead - sizeof(sctp_data_chunk_t) - 4)
 822                 /* Enough data queued to fill a packet */
 823                 return SCTP_XMIT_OK;
 824
 825         /* Don't delay large message writes that may have been fragmented */
 826         if (!chunk->msg->can_delay)
 827                 return SCTP_XMIT_OK;
 828
 829         /* Defer until all data acked or packet full */
 830         return SCTP_XMIT_DELAY;
 831 }
 832
 833 /* This private function does management things when adding DATA chunk */
 834 static void sctp_packet_append_data(struct sctp_packet *packet,
 835                                 struct sctp_chunk *chunk)
 836 {
 837         struct sctp_transport *transport = packet->transport;
 838         size_t datasize = sctp_data_size(chunk);
 839         struct sctp_association *asoc = transport->asoc;
 840         u32 rwnd = asoc->peer.rwnd;
 841
 842         /* Keep track of how many bytes are in flight over this transport. */
 843         transport->flight_size += datasize;
 844
 845         /* Keep track of how many bytes are in flight to the receiver. */
 846         asoc->outqueue.outstanding_bytes += datasize;
 847
 848         /* Update our view of the receiver's rwnd. */
 849         if (datasize < rwnd)
 850                 rwnd -= datasize;
 851         else
 852                 rwnd = 0;
 853
 854         asoc->peer.rwnd = rwnd;
 855         /* Has been accepted for transmission. */
 856         if (!asoc->peer.prsctp_capable)
 857                 chunk->msg->can_abandon = 0;
 858         sctp_chunk_assign_tsn(chunk);
 859         sctp_chunk_assign_ssn(chunk);
 860 }
 861
 862 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
 863                                         struct sctp_chunk *chunk,
 864                                         u16 chunk_len)
 865 {
 866         size_t psize, pmtu;
 867         sctp_xmit_t retval = SCTP_XMIT_OK;
 868
 869         psize = packet->size;
 870         if (packet->transport->asoc)
 871                 pmtu = packet->transport->asoc->pathmtu;
 872         else
 873                 pmtu = packet->transport->pathmtu;
 874
 875         /* Decide if we need to fragment or resubmit later. */
 876         if (psize + chunk_len > pmtu) {
 877                 /* It's OK to fragment at IP level if any one of the following
 878                  * is true:
 879                  *      1. The packet is empty (meaning this chunk is greater
 880                  *         the MTU)
 881                  *      2. The packet doesn't have any data in it yet and data
 882                  *         requires authentication.
 883                  */
 884                 if (sctp_packet_empty(packet) ||
 885                     (!packet->has_data && chunk->auth)) {
 886                         /* We no longer do re-fragmentation.
 887                          * Just fragment at the IP layer, if we
 888                          * actually hit this condition
 889                          */
 890                         packet->ipfragok = 1;
 891                         goto out;
 892                 }
 893
 894                 /* It is also okay to fragment if the chunk we are
 895                  * adding is a control chunk, but only if current packet
 896                  * is not a GSO one otherwise it causes fragmentation of
 897                  * a large frame. So in this case we allow the
 898                  * fragmentation by forcing it to be in a new packet.
 899                  */
 900                 if (!sctp_chunk_is_data(chunk) && packet->has_data)
 901                         retval = SCTP_XMIT_PMTU_FULL;
 902
 903                 if (psize + chunk_len > packet->max_size)
 904                         /* Hit GSO/PMTU limit, gotta flush */
 905                         retval = SCTP_XMIT_PMTU_FULL;
 906
 907                 if (!packet->transport->burst_limited &&
 908                     psize + chunk_len > (packet->transport->cwnd >> 1))
 909                         /* Do not allow a single GSO packet to use more
 910                          * than half of cwnd.
 911                          */
 912                         retval = SCTP_XMIT_PMTU_FULL;
 913
 914                 if (packet->transport->burst_limited &&
 915                     psize + chunk_len > (packet->transport->burst_limited >> 1))
 916                         /* Do not allow a single GSO packet to use more
 917                          * than half of original cwnd.
 918                          */
 919                         retval = SCTP_XMIT_PMTU_FULL;
 920                 /* Otherwise it will fit in the GSO packet */
 921         }
 922
 923 out:
 924         return retval;
 925 }