net/sctp/output.c

   1 /* SCTP kernel reference Implementation
   2  * Copyright (c) 1999-2000 Cisco, Inc.
   3  * Copyright (c) 1999-2001 Motorola, Inc.
   4  * Copyright (c) 2001 International Business Machines, Corp.
   5  *
   6  * This file is part of the SCTP kernel reference Implementation
   7  *
   8  * These functions handle output processing.
   9  *
  10  * The SCTP reference 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  * The SCTP reference 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, write to
  24  * the Free Software Foundation, 59 Temple Place - Suite 330,
  25  * Boston, MA 02111-1307, USA.
  26  *
  27  * Please send any bug reports or fixes you make to the
  28  * email address(es):
  29  *    lksctp developers <lksctp-developers@lists.sourceforge.net>
  30  *
  31  * Or submit a bug report through the following website:
  32  *    http://www.sf.net/projects/lksctp
  33  *
  34  * Written or modified by:
  35  *    La Monte H.P. Yarroll <piggy@acm.org>
  36  *    Karl Knutson          <karl@athena.chicago.il.us>
  37  *    Jon Grimm             <jgrimm@austin.ibm.com>
  38  *    Sridhar Samudrala     <sri@us.ibm.com>
  39  *
  40  * Any bugs reported given to us we will try to fix... any fixes shared will
  41  * be incorporated into the next SCTP release.
  42  */
  43
  44 #include <linux/types.h>
  45 #include <linux/kernel.h>
  46 #include <linux/wait.h>
  47 #include <linux/time.h>
  48 #include <linux/ip.h>
  49 #include <linux/ipv6.h>
  50 #include <linux/init.h>
  51 #include <net/inet_ecn.h>
  52 #include <net/icmp.h>
  53
  54 #ifndef TEST_FRAME
  55 #include <net/tcp.h>
  56 #endif /* TEST_FRAME (not defined) */
  57
  58 #include <linux/socket.h> /* for sa_family_t */
  59 #include <net/sock.h>
  60
  61 #include <net/sctp/sctp.h>
  62 #include <net/sctp/sm.h>
  63
  64 /* Forward declarations for private helpers. */
  65 __u32 count_crc(__u8 *ptr, __u16  count);
  66 static void sctp_packet_reset(sctp_packet_t *packet);
  67 static sctp_xmit_t sctp_packet_append_data(sctp_packet_t *packet,
  68                                            sctp_chunk_t *chunk);
  69
  70 /* Config a packet.
  71  * This appears to be a followup set of initializations.)
  72  */
  73 sctp_packet_t *sctp_packet_config(sctp_packet_t *packet,
  74                                   __u32 vtag,
  75                                   int ecn_capable,
  76                                   sctp_packet_phandler_t *prepend_handler)
  77 {
  78         int packet_empty = (packet->size == SCTP_IP_OVERHEAD);
  79
  80         packet->vtag = vtag;
  81         packet->ecn_capable = ecn_capable;
  82         packet->get_prepend_chunk = prepend_handler;
  83         packet->has_cookie_echo = 0;
  84
  85         /* We might need to call the prepend_handler right away.  */
  86         if (packet_empty)
  87                 sctp_packet_reset(packet);
  88         return packet;
  89 }
  90
  91 /* Initialize the packet structure. */
  92 sctp_packet_t *sctp_packet_init(sctp_packet_t *packet,
  93                                 sctp_transport_t *transport,
  94                                 __u16 sport,
  95                                 __u16 dport)
  96 {
  97         packet->transport = transport;
  98         packet->source_port = sport;
  99         packet->destination_port = dport;
 100         skb_queue_head_init(&packet->chunks);
 101         packet->vtag = 0;
 102         packet->ecn_capable = 0;
 103         packet->get_prepend_chunk = NULL;
 104         packet->has_cookie_echo = 0;
 105         packet->malloced = 0;
 106         sctp_packet_reset(packet);
 107         return packet;
 108 }
 109
 110 /* Free a packet.  */
 111 void sctp_packet_free(sctp_packet_t *packet)
 112 {
 113         sctp_chunk_t *chunk;
 114
 115         while (NULL !=
 116                (chunk = (sctp_chunk_t *)skb_dequeue(&packet->chunks))) {
 117                 sctp_free_chunk(chunk);
 118         }
 119
 120         if (packet->malloced)
 121                 kfree(packet);
 122 }
 123
 124 /* This routine tries to append the chunk to the offered packet. If adding
 125  * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
 126  * is not present in the packet, it transmits the input packet.
 127  * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
 128  * as it can fit in the packet, but any more data that does not fit in this
 129  * packet can be sent only after receiving the COOKIE_ACK.
 130  */
 131 sctp_xmit_t sctp_packet_transmit_chunk(sctp_packet_t *packet,
 132                                        sctp_chunk_t *chunk)
 133 {
 134         sctp_xmit_t retval;
 135         int error = 0;
 136
 137         switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
 138         case SCTP_XMIT_PMTU_FULL:
 139                 if (!packet->has_cookie_echo) {
 140                         error = sctp_packet_transmit(packet);
 141                         if (error < 0)
 142                                 chunk->skb->sk->err = -error;
 143
 144                         /* If we have an empty packet, then we can NOT ever
 145                          * return PMTU_FULL.
 146                          */
 147                         retval = sctp_packet_append_chunk(packet, chunk);
 148                 }
 149                 break;
 150
 151         case SCTP_XMIT_MUST_FRAG:
 152         case SCTP_XMIT_RWND_FULL:
 153         case SCTP_XMIT_OK:
 154                 break;
 155         };
 156
 157         return retval;
 158 }
 159
 160 /* Append a chunk to the offered packet reporting back any inability to do
 161  * so.
 162  */
 163 sctp_xmit_t sctp_packet_append_chunk(sctp_packet_t *packet, sctp_chunk_t *chunk)
 164 {
 165         sctp_xmit_t retval = SCTP_XMIT_OK;
 166         __u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length));
 167         size_t psize = packet->size;
 168         size_t pmtu;
 169         int too_big;
 170
 171         pmtu  = ((packet->transport->asoc) ?
 172                  (packet->transport->asoc->pmtu) :
 173                  (packet->transport->pmtu));
 174
 175         too_big = (psize + chunk_len > pmtu);
 176
 177         /* Decide if we need to fragment or resubmit later. */
 178         if (too_big) {
 179                 int packet_empty = (packet->size == SCTP_IP_OVERHEAD);
 180
 181                 /* Both control chunks and data chunks with TSNs are
 182                  * non-fragmentable.
 183                  */
 184                 int fragmentable = sctp_chunk_is_data(chunk) &&
 185                         (!chunk->has_tsn);
 186                 if (packet_empty) {
 187                         if (fragmentable) {
 188                                 retval = SCTP_XMIT_MUST_FRAG;
 189                                 goto finish;
 190                         } else {
 191                                 /* The packet is too big but we can
 192                                  * not fragment it--we have to just
 193                                  * transmit and rely on IP
 194                                  * fragmentation.
 195                                  */
 196                                 goto append;
 197                         }
 198                 } else { /* !packet_empty */
 199                         retval = SCTP_XMIT_PMTU_FULL;
 200                         goto finish;
 201                 }
 202         } else {
 203                 /* The chunk fits in the packet.  */
 204                 goto append;
 205         }
 206
 207 append:
 208         /* We believe that this chunk is OK to add to the packet (as
 209          * long as we have the cwnd for it).
 210          */
 211
 212         /* DATA is a special case since we must examine both rwnd and cwnd
 213          * before we send DATA.
 214          */
 215         if (sctp_chunk_is_data(chunk)) {
 216                 retval = sctp_packet_append_data(packet, chunk);
 217                 if (SCTP_XMIT_OK != retval)
 218                         goto finish;
 219         } else if (SCTP_CID_COOKIE_ECHO == chunk->chunk_hdr->type) {
 220                 packet->has_cookie_echo = 1;
 221         }
 222
 223         /* It is OK to send this chunk.  */
 224         skb_queue_tail(&packet->chunks, (struct sk_buff *)chunk);
 225         packet->size += chunk_len;
 226 finish:
 227         return retval;
 228 }
 229
 230 /* All packets are sent to the network through this function from
 231  * sctp_push_outqueue().
 232  *
 233  * The return value is a normal kernel error return value.
 234  */
 235 int sctp_packet_transmit(sctp_packet_t *packet)
 236 {
 237         sctp_transport_t *transport = packet->transport;
 238         sctp_association_t *asoc = transport->asoc;
 239         struct sctphdr *sh;
 240         __u32 crc32;
 241         struct sk_buff *nskb;
 242         sctp_chunk_t *chunk;
 243         struct sock *sk;
 244         int err = 0;
 245         int padding;            /* How much padding do we need?  */
 246         __u8 packet_has_data = 0;
 247         struct dst_entry *dst;
 248
 249         /* Do NOT generate a chunkless packet... */
 250         if (skb_queue_empty(&packet->chunks))
 251                 return err;
 252
 253         /* Set up convenience variables... */
 254         chunk = (sctp_chunk_t *) (packet->chunks.next);
 255         sk = chunk->skb->sk;
 256
 257         /* Allocate the new skb.  */
 258         nskb = dev_alloc_skb(packet->size);
 259         if (!nskb) {
 260                 err = -ENOMEM;
 261                 goto out;
 262         }
 263
 264         /* Make sure the outbound skb has enough header room reserved. */
 265         skb_reserve(nskb, SCTP_IP_OVERHEAD);
 266
 267         /* Set the owning socket so that we know where to get the
 268          * destination IP address.
 269          */
 270         skb_set_owner_w(nskb, sk);
 271
 272         /**
 273          * 6.10 Bundling
 274          *
 275          *    An endpoint bundles chunks by simply including multiple
 276          *    chunks in one outbound SCTP packet.  ...
 277          */
 278
 279         /**
 280          * 3.2  Chunk Field Descriptions
 281          *
 282          * The total length of a chunk (including Type, Length and
 283          * Value fields) MUST be a multiple of 4 bytes.  If the length
 284          * of the chunk is not a multiple of 4 bytes, the sender MUST
 285          * pad the chunk with all zero bytes and this padding is not
 286          * included in the chunk length field.  The sender should
 287          * never pad with more than 3 bytes.
 288          *
 289          * [This whole comment explains WORD_ROUND() below.]
 290          */
 291         SCTP_DEBUG_PRINTK("***sctp_transmit_packet***\n");
 292         while (NULL != (chunk = (sctp_chunk_t *)
 293                         skb_dequeue(&packet->chunks))) {
 294                 chunk->num_times_sent++;
 295                 chunk->sent_at = jiffies;
 296                 if (sctp_chunk_is_data(chunk)) {
 297                         sctp_chunk_assign_tsn(chunk);
 298
 299                         /* 6.3.1 C4) When data is in flight and when allowed
 300                          * by rule C5, a new RTT measurement MUST be made each
 301                          * round trip.  Furthermore, new RTT measurements
 302                          * SHOULD be made no more than once per round-trip
 303                          * for a given destination transport address.
 304                          */
 305                         if ((1 == chunk->num_times_sent) &&
 306                             (!transport->rto_pending)) {
 307                                 chunk->rtt_in_progress = 1;
 308                                 transport->rto_pending = 1;
 309                         }
 310                         packet_has_data = 1;
 311                 }
 312                 memcpy(skb_put(nskb, chunk->skb->len),
 313                        chunk->skb->data, chunk->skb->len);
 314                 padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len;
 315                 memset(skb_put(nskb, padding), 0, padding);
 316                 SCTP_DEBUG_PRINTK("%s %p[%s] %s 0x%x, %s %d, %s %d, %s %d, "
 317                                   "%s %d\n",
 318                                   "*** Chunk", chunk,
 319                                   sctp_cname(SCTP_ST_CHUNK(
 320                                           chunk->chunk_hdr->type)),
 321                                   chunk->has_tsn ? "TSN" : "No TSN",
 322                                   chunk->has_tsn ?
 323                                   ntohl(chunk->subh.data_hdr->tsn) : 0,
 324                                   "length", ntohs(chunk->chunk_hdr->length),
 325                                   "chunk->skb->len", chunk->skb->len,
 326                                   "num_times_sent", chunk->num_times_sent,
 327                                   "rtt_in_progress", chunk->rtt_in_progress);
 328
 329                 /*
 330                  * If this is a control chunk, this is our last
 331                  * reference. Free data chunks after they've been
 332                  * acknowledged or have failed.
 333                  */
 334                 if (!sctp_chunk_is_data(chunk))
 335                         sctp_free_chunk(chunk);
 336         }
 337
 338         /* Build the SCTP header.  */
 339         sh = (struct sctphdr *)skb_push(nskb, sizeof(struct sctphdr));
 340         sh->source = htons(packet->source_port);
 341         sh->dest   = htons(packet->destination_port);
 342
 343         /* From 6.8 Adler-32 Checksum Calculation:
 344          * After the packet is constructed (containing the SCTP common
 345          * header and one or more control or DATA chunks), the
 346          * transmitter shall:
 347          *
 348          * 1) Fill in the proper Verification Tag in the SCTP common
 349          *    header and initialize the checksum field to 0's.
 350          */
 351         sh->vtag     = htonl(packet->vtag);
 352         sh->checksum = 0;
 353
 354         /* 2) Calculate the Adler-32 checksum of the whole packet,
 355          *    including the SCTP common header and all the
 356          *    chunks.
 357          *
 358          * Note: Adler-32 is no longer applicable, as has been replaced
 359          * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
 360          */
 361         crc32 = count_crc((__u8 *)sh, nskb->len);
 362
 363         /* 3) Put the resultant value into the checksum field in the
 364          *    common header, and leave the rest of the bits unchanged.
 365          */
 366         sh->checksum = htonl(crc32);
 367
 368         /* FIXME:  Delete the rest of this switch statement once phase 2
 369          * of address selection (ipv6 support) drops in.
 370          */
 371         switch (transport->ipaddr.sa.sa_family) {
 372         case AF_INET6:
 373                 SCTP_V6(inet6_sk(sk)->daddr = transport->ipaddr.v6.sin6_addr;)
 374                 break;
 375         };
 376
 377         /* IP layer ECN support
 378          * From RFC 2481
 379          *  "The ECN-Capable Transport (ECT) bit would be set by the
 380          *   data sender to indicate that the end-points of the
 381          *   transport protocol are ECN-capable."
 382          *
 383          * If ECN capable && negotiated && it makes sense for
 384          * this packet to support it (e.g. post ECN negotiation)
 385          * then lets set the ECT bit
 386          *
 387          * FIXME:  Need to do something else for IPv6
 388          */
 389         if (packet->ecn_capable) {
 390                 INET_ECN_xmit(nskb->sk);
 391         } else {
 392                 INET_ECN_dontxmit(nskb->sk);
 393         }
 394
 395         /* Set up the IP options.  */
 396         /* BUG: not implemented
 397          * For v4 this all lives somewhere in sk->opt...
 398          */
 399
 400         /* Dump that on IP!  */
 401         if (asoc && asoc->peer.last_sent_to != transport) {
 402                 /* Considering the multiple CPU scenario, this is a
 403                  * "correcter" place for last_sent_to.  --xguo
 404                  */
 405                 asoc->peer.last_sent_to = transport;
 406         }
 407
 408         if (packet_has_data) {
 409                 struct timer_list *timer;
 410                 unsigned long timeout;
 411
 412                 transport->last_time_used = jiffies;
 413
 414                 /* Restart the AUTOCLOSE timer when sending data. */
 415                 if ((SCTP_STATE_ESTABLISHED == asoc->state) &&
 416                     (asoc->autoclose)) {
 417                         timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
 418                         timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
 419
 420                         if (!mod_timer(timer, jiffies + timeout))
 421                                 sctp_association_hold(asoc);
 422                 }
 423         }
 424
 425         dst = transport->dst;
 426         if (!dst || dst->obsolete) {
 427                 sctp_transport_route(transport, NULL, sctp_sk(sk));
 428         }
 429
 430         nskb->dst = dst_clone(transport->dst);
 431         if (!nskb->dst)
 432                 goto no_route;
 433
 434         SCTP_DEBUG_PRINTK("***sctp_transmit_packet*** skb length %d\n",
 435                           nskb->len);
 436         (*transport->af_specific->queue_xmit)(nskb);
 437 out:
 438         packet->size = SCTP_IP_OVERHEAD;
 439         return err;
 440 no_route:
 441         kfree_skb(nskb);
 442         IP_INC_STATS_BH(IpOutNoRoutes);
 443         err = -EHOSTUNREACH;
 444         goto out;
 445 }
 446
 447 /********************************************************************
 448  * 2nd Level Abstractions
 449  ********************************************************************/
 450
 451 /*
 452  * This private function resets the packet to a fresh state.
 453  */
 454 static void sctp_packet_reset(sctp_packet_t *packet)
 455 {
 456         sctp_chunk_t *chunk = NULL;
 457
 458         packet->size = SCTP_IP_OVERHEAD;
 459
 460         if (packet->get_prepend_chunk)
 461                 chunk = packet->get_prepend_chunk(packet->transport->asoc);
 462
 463         /* If there a is a prepend chunk stick it on the list before
 464          * any other chunks get appended.
 465          */
 466         if (chunk)
 467                 sctp_packet_append_chunk(packet, chunk);
 468 }
 469
 470 /* This private function handles the specifics of appending DATA chunks.  */
 471 static sctp_xmit_t sctp_packet_append_data(sctp_packet_t *packet,
 472                                            sctp_chunk_t *chunk)
 473 {
 474         sctp_xmit_t retval = SCTP_XMIT_OK;
 475         size_t datasize, rwnd, inflight;
 476         sctp_transport_t *transport = packet->transport;
 477         __u32 max_burst_bytes;
 478
 479         /* RFC 2960 6.1  Transmission of DATA Chunks
 480          *
 481          * A) At any given time, the data sender MUST NOT transmit new data to
 482          * any destination transport address if its peer's rwnd indicates
 483          * that the peer has no buffer space (i.e. rwnd is 0, see Section
 484          * 6.2.1).  However, regardless of the value of rwnd (including if it
 485          * is 0), the data sender can always have one DATA chunk in flight to
 486          * the receiver if allowed by cwnd (see rule B below).  This rule
 487          * allows the sender to probe for a change in rwnd that the sender
 488          * missed due to the SACK having been lost in transit from the data
 489          * receiver to the data sender.
 490          */
 491
 492         rwnd = transport->asoc->peer.rwnd;
 493         inflight = transport->asoc->outqueue.outstanding_bytes;
 494
 495         datasize = sctp_data_size(chunk);
 496
 497         if (datasize > rwnd) {
 498                 if (inflight > 0) {
 499                         /* We have (at least) one data chunk in flight,
 500                          * so we can't fall back to rule 6.1 B).
 501                          */
 502                         retval = SCTP_XMIT_RWND_FULL;
 503                         goto finish;
 504                 }
 505         }
 506
 507         /* sctpimpguide-05 2.14.2
 508          * D) When the time comes for the sender to
 509          * transmit new DATA chunks, the protocol parameter Max.Burst MUST
 510          * first be applied to limit how many new DATA chunks may be sent.
 511          * The limit is applied by adjusting cwnd as follows:
 512          *      if ((flightsize + Max.Burst * MTU) < cwnd)
 513          *              cwnd = flightsize + Max.Burst * MTU
 514          */
 515         max_burst_bytes = transport->asoc->max_burst * transport->asoc->pmtu;
 516         if ((transport->flight_size + max_burst_bytes) < transport->cwnd) {
 517                 transport->cwnd = transport->flight_size + max_burst_bytes;
 518                 SCTP_DEBUG_PRINTK("%s: cwnd limited by max_burst: "
 519                                   "transport: %p, cwnd: %d, "
 520                                   "ssthresh: %d, flight_size: %d, "
 521                                   "pba: %d\n",
 522                                   __FUNCTION__, transport,
 523                                   transport->cwnd,
 524                                   transport->ssthresh,
 525                                   transport->flight_size,
 526                                   transport->partial_bytes_acked);
 527         }
 528
 529         /* RFC 2960 6.1  Transmission of DATA Chunks
 530          *
 531          * B) At any given time, the sender MUST NOT transmit new data
 532          * to a given transport address if it has cwnd or more bytes
 533          * of data outstanding to that transport address.
 534          */
 535         /* RFC 7.2.4 & the Implementers Guide 2.8.
 536          *
 537          * 3) ...
 538          *    When a Fast Retransmit is being performed the sender SHOULD
 539          *    ignore the value of cwnd and SHOULD NOT delay retransmission.
 540          */
 541         if (!chunk->fast_retransmit) {
 542                 if (transport->flight_size >= transport->cwnd) {
 543                         retval = SCTP_XMIT_RWND_FULL;
 544                         goto finish;
 545                 }
 546         }
 547
 548         /* Keep track of how many bytes are in flight over this transport. */
 549         transport->flight_size += datasize;
 550
 551         /* Keep track of how many bytes are in flight to the receiver. */
 552         transport->asoc->outqueue.outstanding_bytes += datasize;
 553
 554         /* Update our view of the receiver's rwnd. */
 555         if (datasize < rwnd) {
 556                 rwnd -= datasize;
 557         } else {
 558                 rwnd = 0;
 559         }
 560
 561         transport->asoc->peer.rwnd = rwnd;
 562
 563 finish:
 564         return retval;
 565 }