net/sctp/outqueue.c

   1 /* SCTP kernel reference Implementation
   2  * (C) Copyright IBM Corp. 2001, 2004
   3  * Copyright (c) 1999-2000 Cisco, Inc.
   4  * Copyright (c) 1999-2001 Motorola, Inc.
   5  * Copyright (c) 2001-2003 Intel Corp.
   6  *
   7  * This file is part of the SCTP kernel reference Implementation
   8  *
   9  * These functions implement the sctp_outq class.   The outqueue handles
  10  * bundling and queueing of outgoing SCTP chunks.
  11  *
  12  * The SCTP reference implementation is free software;
  13  * you can redistribute it and/or modify it under the terms of
  14  * the GNU General Public License as published by
  15  * the Free Software Foundation; either version 2, or (at your option)
  16  * any later version.
  17  *
  18  * The SCTP reference implementation is distributed in the hope that it
  19  * will be useful, but WITHOUT ANY WARRANTY; without even the implied
  20  *                 ************************
  21  * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  22  * See the GNU General Public License for more details.
  23  *
  24  * You should have received a copy of the GNU General Public License
  25  * along with GNU CC; see the file COPYING.  If not, write to
  26  * the Free Software Foundation, 59 Temple Place - Suite 330,
  27  * Boston, MA 02111-1307, USA.
  28  *
  29  * Please send any bug reports or fixes you make to the
  30  * email address(es):
  31  *    lksctp developers <lksctp-developers@lists.sourceforge.net>
  32  *
  33  * Or submit a bug report through the following website:
  34  *    http://www.sf.net/projects/lksctp
  35  *
  36  * Written or modified by:
  37  *    La Monte H.P. Yarroll <piggy@acm.org>
  38  *    Karl Knutson          <karl@athena.chicago.il.us>
  39  *    Perry Melange         <pmelange@null.cc.uic.edu>
  40  *    Xingang Guo           <xingang.guo@intel.com>
  41  *    Hui Huang             <hui.huang@nokia.com>
  42  *    Sridhar Samudrala     <sri@us.ibm.com>
  43  *    Jon Grimm             <jgrimm@us.ibm.com>
  44  *
  45  * Any bugs reported given to us we will try to fix... any fixes shared will
  46  * be incorporated into the next SCTP release.
  47  */
  48
  49 #include <linux/types.h>
  50 #include <linux/list.h>   /* For struct list_head */
  51 #include <linux/socket.h>
  52 #include <linux/ip.h>
  53 #include <net/sock.h>     /* For skb_set_owner_w */
  54
  55 #include <net/sctp/sctp.h>
  56 #include <net/sctp/sm.h>
  57
  58 /* Declare internal functions here.  */
  59 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn);
  60 static void sctp_check_transmitted(struct sctp_outq *q,
  61                                    struct list_head *transmitted_queue,
  62                                    struct sctp_transport *transport,
  63                                    struct sctp_sackhdr *sack,
  64                                    __u32 highest_new_tsn);
  65
  66 static void sctp_mark_missing(struct sctp_outq *q,
  67                               struct list_head *transmitted_queue,
  68                               struct sctp_transport *transport,
  69                               __u32 highest_new_tsn,
  70                               int count_of_newacks);
  71
  72 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn);
  73
  74 /* Add data to the front of the queue. */
  75 static inline void sctp_outq_head_data(struct sctp_outq *q,
  76                                         struct sctp_chunk *ch)
  77 {
  78         __skb_queue_head(&q->out, (struct sk_buff *)ch);
  79         q->out_qlen += ch->skb->len;
  80         return;
  81 }
  82
  83 /* Take data from the front of the queue. */
  84 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
  85 {
  86         struct sctp_chunk *ch;
  87         ch = (struct sctp_chunk *)__skb_dequeue(&q->out);
  88         if (ch)
  89                 q->out_qlen -= ch->skb->len;
  90         return ch;
  91 }
  92 /* Add data chunk to the end of the queue. */
  93 static inline void sctp_outq_tail_data(struct sctp_outq *q,
  94                                        struct sctp_chunk *ch)
  95 {
  96         __skb_queue_tail(&q->out, (struct sk_buff *)ch);
  97         q->out_qlen += ch->skb->len;
  98         return;
  99 }
 100
 101 /* Insert a chunk behind chunk 'pos'. */
 102 static inline void sctp_outq_insert_data(struct sctp_outq *q,
 103                                          struct sctp_chunk *ch,
 104                                          struct sctp_chunk *pos)
 105 {
 106         __skb_insert((struct sk_buff *)ch, (struct sk_buff *)pos->prev,
 107                      (struct sk_buff *)pos, pos->list);
 108         q->out_qlen += ch->skb->len;
 109 }
 110
 111 /*
 112  * SFR-CACC algorithm:
 113  * D) If count_of_newacks is greater than or equal to 2
 114  * and t was not sent to the current primary then the
 115  * sender MUST NOT increment missing report count for t.
 116  */
 117 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary,
 118                                        struct sctp_transport *transport,
 119                                        int count_of_newacks)
 120 {
 121         if (count_of_newacks >=2 && transport != primary)
 122                 return 1;
 123         return 0;
 124 }
 125
 126 /*
 127  * SFR-CACC algorithm:
 128  * F) If count_of_newacks is less than 2, let d be the
 129  * destination to which t was sent. If cacc_saw_newack
 130  * is 0 for destination d, then the sender MUST NOT
 131  * increment missing report count for t.
 132  */
 133 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport,
 134                                        int count_of_newacks)
 135 {
 136         if (count_of_newacks < 2 && !transport->cacc.cacc_saw_newack)
 137                 return 1;
 138         return 0;
 139 }
 140
 141 /*
 142  * SFR-CACC algorithm:
 143  * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
 144  * execute steps C, D, F.
 145  *
 146  * C has been implemented in sctp_outq_sack
 147  */
 148 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
 149                                      struct sctp_transport *transport,
 150                                      int count_of_newacks)
 151 {
 152         if (!primary->cacc.cycling_changeover) {
 153                 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
 154                         return 1;
 155                 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
 156                         return 1;
 157                 return 0;
 158         }
 159         return 0;
 160 }
 161
 162 /*
 163  * SFR-CACC algorithm:
 164  * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
 165  * than next_tsn_at_change of the current primary, then
 166  * the sender MUST NOT increment missing report count
 167  * for t.
 168  */
 169 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
 170 {
 171         if (primary->cacc.cycling_changeover &&
 172             TSN_lt(tsn, primary->cacc.next_tsn_at_change))
 173                 return 1;
 174         return 0;
 175 }
 176
 177 /*
 178  * SFR-CACC algorithm:
 179  * 3) If the missing report count for TSN t is to be
 180  * incremented according to [RFC2960] and
 181  * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
 182  * then the sender MUST futher execute steps 3.1 and
 183  * 3.2 to determine if the missing report count for
 184  * TSN t SHOULD NOT be incremented.
 185  *
 186  * 3.3) If 3.1 and 3.2 do not dictate that the missing
 187  * report count for t should not be incremented, then
 188  * the sender SOULD increment missing report count for
 189  * t (according to [RFC2960] and [SCTP_STEWART_2002]).
 190  */
 191 static inline int sctp_cacc_skip(struct sctp_transport *primary,
 192                                  struct sctp_transport *transport,
 193                                  int count_of_newacks,
 194                                  __u32 tsn)
 195 {
 196         if (primary->cacc.changeover_active &&
 197             (sctp_cacc_skip_3_1(primary, transport, count_of_newacks)
 198              || sctp_cacc_skip_3_2(primary, tsn)))
 199                 return 1;
 200         return 0;
 201 }
 202
 203 /* Generate a new outqueue.  */
 204 struct sctp_outq *sctp_outq_new(struct sctp_association *asoc)
 205 {
 206         struct sctp_outq *q;
 207
 208         q = t_new(struct sctp_outq, GFP_KERNEL);
 209         if (q) {
 210                 sctp_outq_init(asoc, q);
 211                 q->malloced = 1;
 212         }
 213         return q;
 214 }
 215
 216 /* Initialize an existing sctp_outq.  This does the boring stuff.
 217  * You still need to define handlers if you really want to DO
 218  * something with this structure...
 219  */
 220 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
 221 {
 222         q->asoc = asoc;
 223         skb_queue_head_init(&q->out);
 224         skb_queue_head_init(&q->control);
 225         INIT_LIST_HEAD(&q->retransmit);
 226         INIT_LIST_HEAD(&q->sacked);
 227         INIT_LIST_HEAD(&q->abandoned);
 228
 229         q->outstanding_bytes = 0;
 230         q->empty = 1;
 231         q->cork  = 0;
 232
 233         q->malloced = 0;
 234         q->out_qlen = 0;
 235 }
 236
 237 /* Free the outqueue structure and any related pending chunks.
 238  */
 239 void sctp_outq_teardown(struct sctp_outq *q)
 240 {
 241         struct sctp_transport *transport;
 242         struct list_head *lchunk, *pos, *temp;
 243         struct sctp_chunk *chunk;
 244
 245         /* Throw away unacknowledged chunks. */
 246         list_for_each(pos, &q->asoc->peer.transport_addr_list) {
 247                 transport = list_entry(pos, struct sctp_transport, transports);
 248                 while ((lchunk = sctp_list_dequeue(&transport->transmitted))) {
 249                         chunk = list_entry(lchunk, struct sctp_chunk,
 250                                            transmitted_list);
 251                         /* Mark as part of a failed message. */
 252                         sctp_chunk_fail(chunk, q->error);
 253                         sctp_chunk_free(chunk);
 254                 }
 255         }
 256
 257         /* Throw away chunks that have been gap ACKed.  */
 258         list_for_each_safe(lchunk, temp, &q->sacked) {
 259                 list_del_init(lchunk);
 260                 chunk = list_entry(lchunk, struct sctp_chunk,
 261                                    transmitted_list);
 262                 sctp_chunk_fail(chunk, q->error);
 263                 sctp_chunk_free(chunk);
 264         }
 265
 266         /* Throw away any chunks in the retransmit queue. */
 267         list_for_each_safe(lchunk, temp, &q->retransmit) {
 268                 list_del_init(lchunk);
 269                 chunk = list_entry(lchunk, struct sctp_chunk,
 270                                    transmitted_list);
 271                 sctp_chunk_fail(chunk, q->error);
 272                 sctp_chunk_free(chunk);
 273         }
 274
 275         /* Throw away any chunks that are in the abandoned queue. */
 276         list_for_each_safe(lchunk, temp, &q->abandoned) {
 277                 list_del_init(lchunk);
 278                 chunk = list_entry(lchunk, struct sctp_chunk,
 279                                    transmitted_list);
 280                 sctp_chunk_fail(chunk, q->error);
 281                 sctp_chunk_free(chunk);
 282         }
 283
 284         /* Throw away any leftover data chunks. */
 285         while ((chunk = sctp_outq_dequeue_data(q))) {
 286
 287                 /* Mark as send failure. */
 288                 sctp_chunk_fail(chunk, q->error);
 289                 sctp_chunk_free(chunk);
 290         }
 291
 292         q->error = 0;
 293
 294         /* Throw away any leftover control chunks. */
 295         while ((chunk = (struct sctp_chunk *) skb_dequeue(&q->control)))
 296                 sctp_chunk_free(chunk);
 297 }
 298
 299 /* Free the outqueue structure and any related pending chunks.  */
 300 void sctp_outq_free(struct sctp_outq *q)
 301 {
 302         /* Throw away leftover chunks. */
 303         sctp_outq_teardown(q);
 304
 305         /* If we were kmalloc()'d, free the memory.  */
 306         if (q->malloced)
 307                 kfree(q);
 308 }
 309
 310 /* Put a new chunk in an sctp_outq.  */
 311 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk)
 312 {
 313         int error = 0;
 314
 315         SCTP_DEBUG_PRINTK("sctp_outq_tail(%p, %p[%s])\n",
 316                           q, chunk, chunk && chunk->chunk_hdr ?
 317                           sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
 318                           : "Illegal Chunk");
 319
 320         /* If it is data, queue it up, otherwise, send it
 321          * immediately.
 322          */
 323         if (SCTP_CID_DATA == chunk->chunk_hdr->type) {
 324                 /* Is it OK to queue data chunks?  */
 325                 /* From 9. Termination of Association
 326                  *
 327                  * When either endpoint performs a shutdown, the
 328                  * association on each peer will stop accepting new
 329                  * data from its user and only deliver data in queue
 330                  * at the time of sending or receiving the SHUTDOWN
 331                  * chunk.
 332                  */
 333                 switch (q->asoc->state) {
 334                 case SCTP_STATE_EMPTY:
 335                 case SCTP_STATE_CLOSED:
 336                 case SCTP_STATE_SHUTDOWN_PENDING:
 337                 case SCTP_STATE_SHUTDOWN_SENT:
 338                 case SCTP_STATE_SHUTDOWN_RECEIVED:
 339                 case SCTP_STATE_SHUTDOWN_ACK_SENT:
 340                         /* Cannot send after transport endpoint shutdown */
 341                         error = -ESHUTDOWN;
 342                         break;
 343
 344                 default:
 345                         SCTP_DEBUG_PRINTK("outqueueing (%p, %p[%s])\n",
 346                           q, chunk, chunk && chunk->chunk_hdr ?
 347                           sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
 348                           : "Illegal Chunk");
 349
 350                         sctp_outq_tail_data(q, chunk);
 351                         if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
 352                                 SCTP_INC_STATS(SctpOutUnorderChunks);
 353                         else
 354                                 SCTP_INC_STATS(SctpOutOrderChunks);
 355                         q->empty = 0;
 356                         break;
 357                 };
 358         } else {
 359                 __skb_queue_tail(&q->control, (struct sk_buff *) chunk);
 360                 SCTP_INC_STATS(SctpOutCtrlChunks);
 361         }
 362
 363         if (error < 0)
 364                 return error;
 365
 366         if (!q->cork)
 367                 error = sctp_outq_flush(q, 0);
 368
 369         return error;
 370 }
 371
 372 /* Insert a chunk into the sorted list based on the TSNs.  The retransmit list
 373  * and the abandoned list are in ascending order.
 374  */
 375 void sctp_insert_list(struct list_head *head, struct list_head *new)
 376 {
 377         struct list_head *pos;
 378         struct sctp_chunk *nchunk, *lchunk;
 379         __u32 ntsn, ltsn;
 380         int done = 0;
 381
 382         nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
 383         ntsn = ntohl(nchunk->subh.data_hdr->tsn);
 384
 385         list_for_each(pos, head) {
 386                 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
 387                 ltsn = ntohl(lchunk->subh.data_hdr->tsn);
 388                 if (TSN_lt(ntsn, ltsn)) {
 389                         list_add(new, pos->prev);
 390                         done = 1;
 391                         break;
 392                 }
 393         }
 394         if (!done)
 395                 list_add_tail(new, head);
 396 }
 397
 398 /* Mark all the eligible packets on a transport for retransmission.  */
 399 void sctp_retransmit_mark(struct sctp_outq *q,
 400                           struct sctp_transport *transport,
 401                           __u8 fast_retransmit)
 402 {
 403         struct list_head *lchunk, *ltemp;
 404         struct sctp_chunk *chunk;
 405
 406         /* Walk through the specified transmitted queue.  */
 407         list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
 408                 chunk = list_entry(lchunk, struct sctp_chunk,
 409                                    transmitted_list);
 410
 411                 /* If the chunk is abandoned, move it to abandoned list. */
 412                 if (sctp_chunk_abandoned(chunk)) {
 413                         list_del_init(lchunk);
 414                         sctp_insert_list(&q->abandoned, lchunk);
 415                         continue;
 416                 }
 417
 418                 /* If we are doing retransmission due to a fast retransmit,
 419                  * only the chunk's that are marked for fast retransmit
 420                  * should be added to the retransmit queue.  If we are doing
 421                  * retransmission due to a timeout or pmtu discovery, only the
 422                  * chunks that are not yet acked should be added to the
 423                  * retransmit queue.
 424                  */
 425                 if ((fast_retransmit && chunk->fast_retransmit) ||
 426                    (!fast_retransmit && !chunk->tsn_gap_acked)) {
 427                         /* RFC 2960 6.2.1 Processing a Received SACK
 428                          *
 429                          * C) Any time a DATA chunk is marked for
 430                          * retransmission (via either T3-rtx timer expiration
 431                          * (Section 6.3.3) or via fast retransmit
 432                          * (Section 7.2.4)), add the data size of those
 433                          * chunks to the rwnd.
 434                          */
 435                         q->asoc->peer.rwnd += sctp_data_size(chunk);
 436                         q->outstanding_bytes -= sctp_data_size(chunk);
 437                         transport->flight_size -= sctp_data_size(chunk);
 438
 439                         /* sctpimpguide-05 Section 2.8.2
 440                          * M5) If a T3-rtx timer expires, the
 441                          * 'TSN.Missing.Report' of all affected TSNs is set
 442                          * to 0.
 443                          */
 444                         chunk->tsn_missing_report = 0;
 445
 446                         /* If a chunk that is being used for RTT measurement
 447                          * has to be retransmitted, we cannot use this chunk
 448                          * anymore for RTT measurements. Reset rto_pending so
 449                          * that a new RTT measurement is started when a new
 450                          * data chunk is sent.
 451                          */
 452                         if (chunk->rtt_in_progress) {
 453                                 chunk->rtt_in_progress = 0;
 454                                 transport->rto_pending = 0;
 455                         }
 456
 457                         /* Move the chunk to the retransmit queue. The chunks
 458                          * on the retransmit queue are always kept in order.
 459                          */
 460                         list_del_init(lchunk);
 461                         sctp_insert_list(&q->retransmit, lchunk);
 462                 }
 463         }
 464
 465         SCTP_DEBUG_PRINTK("%s: transport: %p, fast_retransmit: %d, "
 466                           "cwnd: %d, ssthresh: %d, flight_size: %d, "
 467                           "pba: %d\n", __FUNCTION__,
 468                           transport, fast_retransmit,
 469                           transport->cwnd, transport->ssthresh,
 470                           transport->flight_size,
 471                           transport->partial_bytes_acked);
 472
 473 }
 474
 475 /* Mark all the eligible packets on a transport for retransmission and force
 476  * one packet out.
 477  */
 478 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
 479                      sctp_retransmit_reason_t reason)
 480 {
 481         int error = 0;
 482         __u8 fast_retransmit = 0;
 483
 484         switch(reason) {
 485         case SCTP_RTXR_T3_RTX:
 486                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
 487                 /* Update the retran path if the T3-rtx timer has expired for
 488                  * the current retran path.
 489                  */
 490                 if (transport == transport->asoc->peer.retran_path)
 491                         sctp_assoc_update_retran_path(transport->asoc);
 492                 break;
 493         case SCTP_RTXR_FAST_RTX:
 494                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
 495                 fast_retransmit = 1;
 496                 break;
 497         case SCTP_RTXR_PMTUD:
 498         default:
 499                 break;
 500         }
 501
 502         sctp_retransmit_mark(q, transport, fast_retransmit);
 503
 504         /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
 505          * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
 506          * following the procedures outlined in C1 - C5.
 507          */
 508         sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
 509
 510         error = sctp_outq_flush(q, /* rtx_timeout */ 1);
 511
 512         if (error)
 513                 q->asoc->base.sk->sk_err = -error;
 514 }
 515
 516 /*
 517  * Transmit DATA chunks on the retransmit queue.  Upon return from
 518  * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
 519  * need to be transmitted by the caller.
 520  * We assume that pkt->transport has already been set.
 521  *
 522  * The return value is a normal kernel error return value.
 523  */
 524 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
 525                                int rtx_timeout, int *start_timer)
 526 {
 527         struct list_head *lqueue;
 528         struct list_head *lchunk;
 529         struct sctp_transport *transport = pkt->transport;
 530         sctp_xmit_t status;
 531         struct sctp_chunk *chunk;
 532         struct sctp_association *asoc;
 533         int error = 0;
 534
 535         asoc = q->asoc;
 536         lqueue = &q->retransmit;
 537
 538         /* RFC 2960 6.3.3 Handle T3-rtx Expiration
 539          *
 540          * E3) Determine how many of the earliest (i.e., lowest TSN)
 541          * outstanding DATA chunks for the address for which the
 542          * T3-rtx has expired will fit into a single packet, subject
 543          * to the MTU constraint for the path corresponding to the
 544          * destination transport address to which the retransmission
 545          * is being sent (this may be different from the address for
 546          * which the timer expires [see Section 6.4]). Call this value
 547          * K. Bundle and retransmit those K DATA chunks in a single
 548          * packet to the destination endpoint.
 549          *
 550          * [Just to be painfully clear, if we are retransmitting
 551          * because a timeout just happened, we should send only ONE
 552          * packet of retransmitted data.]
 553          */
 554         lchunk = sctp_list_dequeue(lqueue);
 555
 556         while (lchunk) {
 557                 chunk = list_entry(lchunk, struct sctp_chunk,
 558                                    transmitted_list);
 559
 560                 /* Make sure that Gap Acked TSNs are not retransmitted.  A
 561                  * simple approach is just to move such TSNs out of the
 562                  * way and into a 'transmitted' queue and skip to the
 563                  * next chunk.
 564                  */
 565                 if (chunk->tsn_gap_acked) {
 566                         list_add_tail(lchunk, &transport->transmitted);
 567                         lchunk = sctp_list_dequeue(lqueue);
 568                         continue;
 569                 }
 570
 571                 /* Attempt to append this chunk to the packet. */
 572                 status = sctp_packet_append_chunk(pkt, chunk);
 573
 574                 switch (status) {
 575                 case SCTP_XMIT_PMTU_FULL:
 576                         /* Send this packet.  */
 577                         if ((error = sctp_packet_transmit(pkt)) == 0)
 578                                 *start_timer = 1;
 579
 580                         /* If we are retransmitting, we should only
 581                          * send a single packet.
 582                          */
 583                         if (rtx_timeout) {
 584                                 list_add(lchunk, lqueue);
 585                                 lchunk = NULL;
 586                         }
 587
 588                         /* Bundle lchunk in the next round.  */
 589                         break;
 590
 591                 case SCTP_XMIT_RWND_FULL:
 592                         /* Send this packet. */
 593                         if ((error = sctp_packet_transmit(pkt)) == 0)
 594                                 *start_timer = 1;
 595
 596                         /* Stop sending DATA as there is no more room
 597                          * at the receiver.
 598                          */
 599                         list_add(lchunk, lqueue);
 600                         lchunk = NULL;
 601                         break;
 602
 603                 case SCTP_XMIT_NAGLE_DELAY:
 604                         /* Send this packet. */
 605                         if ((error = sctp_packet_transmit(pkt)) == 0)
 606                                 *start_timer = 1;
 607
 608                         /* Stop sending DATA because of nagle delay. */
 609                         list_add(lchunk, lqueue);
 610                         lchunk = NULL;
 611                         break;
 612
 613                 default:
 614                         /* The append was successful, so add this chunk to
 615                          * the transmitted list.
 616                          */
 617                         list_add_tail(lchunk, &transport->transmitted);
 618                         *start_timer = 1;
 619                         q->empty = 0;
 620
 621                         /* Retrieve a new chunk to bundle. */
 622                         lchunk = sctp_list_dequeue(lqueue);
 623                         break;
 624                 };
 625         }
 626
 627         return error;
 628 }
 629
 630 /* Cork the outqueue so queued chunks are really queued. */
 631 int sctp_outq_uncork(struct sctp_outq *q)
 632 {
 633         int error = 0;
 634         if (q->cork) {
 635                 q->cork = 0;
 636                 error = sctp_outq_flush(q, 0);
 637         }
 638         return error;
 639 }
 640
 641 /*
 642  * Try to flush an outqueue.
 643  *
 644  * Description: Send everything in q which we legally can, subject to
 645  * congestion limitations.
 646  * * Note: This function can be called from multiple contexts so appropriate
 647  * locking concerns must be made.  Today we use the sock lock to protect
 648  * this function.
 649  */
 650 int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
 651 {
 652         struct sctp_packet *packet;
 653         struct sctp_packet singleton;
 654         struct sctp_association *asoc = q->asoc;
 655         __u16 sport = asoc->base.bind_addr.port;
 656         __u16 dport = asoc->peer.port;
 657         __u32 vtag = asoc->peer.i.init_tag;
 658         struct sk_buff_head *queue;
 659         struct sctp_transport *transport = NULL;
 660         struct sctp_transport *new_transport;
 661         struct sctp_chunk *chunk;
 662         sctp_xmit_t status;
 663         int error = 0;
 664         int start_timer = 0;
 665
 666         /* These transports have chunks to send. */
 667         struct list_head transport_list;
 668         struct list_head *ltransport;
 669
 670         INIT_LIST_HEAD(&transport_list);
 671         packet = NULL;
 672
 673         /*
 674          * 6.10 Bundling
 675          *   ...
 676          *   When bundling control chunks with DATA chunks, an
 677          *   endpoint MUST place control chunks first in the outbound
 678          *   SCTP packet.  The transmitter MUST transmit DATA chunks
 679          *   within a SCTP packet in increasing order of TSN.
 680          *   ...
 681          */
 682
 683         queue = &q->control;
 684         while ((chunk = (struct sctp_chunk *)skb_dequeue(queue))) {
 685                 /* Pick the right transport to use. */
 686                 new_transport = chunk->transport;
 687
 688                 if (!new_transport) {
 689                         new_transport = asoc->peer.active_path;
 690                 } else if (!new_transport->active) {
 691                         /* If the chunk is Heartbeat, send it to
 692                          * chunk->transport, even it's inactive.
 693                          */
 694                         if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT)
 695                                 new_transport = asoc->peer.active_path;
 696                 }
 697
 698                 /* Are we switching transports?
 699                  * Take care of transport locks.
 700                  */
 701                 if (new_transport != transport) {
 702                         transport = new_transport;
 703                         if (list_empty(&transport->send_ready)) {
 704                                 list_add_tail(&transport->send_ready,
 705                                               &transport_list);
 706                         }
 707                         packet = &transport->packet;
 708                         sctp_packet_config(packet, vtag,
 709                                            asoc->peer.ecn_capable);
 710                 }
 711
 712                 switch (chunk->chunk_hdr->type) {
 713                 /*
 714                  * 6.10 Bundling
 715                  *   ...
 716                  *   An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
 717                  *   COMPLETE with any other chunks.  [Send them immediately.]
 718                  */
 719                 case SCTP_CID_INIT:
 720                 case SCTP_CID_INIT_ACK:
 721                 case SCTP_CID_SHUTDOWN_COMPLETE:
 722                         sctp_packet_init(&singleton, transport, sport, dport);
 723                         sctp_packet_config(&singleton, vtag, 0);
 724                         sctp_packet_append_chunk(&singleton, chunk);
 725                         error = sctp_packet_transmit(&singleton);
 726                         if (error < 0)
 727                                 return error;
 728                         break;
 729
 730                 case SCTP_CID_ABORT:
 731                 case SCTP_CID_SACK:
 732                 case SCTP_CID_HEARTBEAT:
 733                 case SCTP_CID_HEARTBEAT_ACK:
 734                 case SCTP_CID_SHUTDOWN:
 735                 case SCTP_CID_SHUTDOWN_ACK:
 736                 case SCTP_CID_ERROR:
 737                 case SCTP_CID_COOKIE_ECHO:
 738                 case SCTP_CID_COOKIE_ACK:
 739                 case SCTP_CID_ECN_ECNE:
 740                 case SCTP_CID_ECN_CWR:
 741                 case SCTP_CID_ASCONF:
 742                 case SCTP_CID_ASCONF_ACK:
 743                 case SCTP_CID_FWD_TSN:
 744                         sctp_packet_transmit_chunk(packet, chunk);
 745                         break;
 746
 747                 default:
 748                         /* We built a chunk with an illegal type! */
 749                         BUG();
 750                 };
 751         }
 752
 753         /* Is it OK to send data chunks?  */
 754         switch (asoc->state) {
 755         case SCTP_STATE_COOKIE_ECHOED:
 756                 /* Only allow bundling when this packet has a COOKIE-ECHO
 757                  * chunk.
 758                  */
 759                 if (!packet || !packet->has_cookie_echo)
 760                         break;
 761
 762                 /* fallthru */
 763         case SCTP_STATE_ESTABLISHED:
 764         case SCTP_STATE_SHUTDOWN_PENDING:
 765         case SCTP_STATE_SHUTDOWN_RECEIVED:
 766                 /*
 767                  * RFC 2960 6.1  Transmission of DATA Chunks
 768                  *
 769                  * C) When the time comes for the sender to transmit,
 770                  * before sending new DATA chunks, the sender MUST
 771                  * first transmit any outstanding DATA chunks which
 772                  * are marked for retransmission (limited by the
 773                  * current cwnd).
 774                  */
 775                 if (!list_empty(&q->retransmit)) {
 776                         if (transport == asoc->peer.retran_path)
 777                                 goto retran;
 778
 779                         /* Switch transports & prepare the packet.  */
 780
 781                         transport = asoc->peer.retran_path;
 782
 783                         if (list_empty(&transport->send_ready)) {
 784                                 list_add_tail(&transport->send_ready,
 785                                               &transport_list);
 786                         }
 787
 788                         packet = &transport->packet;
 789                         sctp_packet_config(packet, vtag,
 790                                            asoc->peer.ecn_capable);
 791                 retran:
 792                         error = sctp_outq_flush_rtx(q, packet,
 793                                                     rtx_timeout, &start_timer);
 794
 795                         if (start_timer)
 796                                 sctp_transport_reset_timers(transport);
 797
 798                         /* This can happen on COOKIE-ECHO resend.  Only
 799                          * one chunk can get bundled with a COOKIE-ECHO.
 800                          */
 801                         if (packet->has_cookie_echo)
 802                                 goto sctp_flush_out;
 803
 804                         /* Don't send new data if there is still data
 805                          * waiting to retransmit.
 806                          */
 807                         if (!list_empty(&q->retransmit))
 808                                 goto sctp_flush_out;
 809                 }
 810
 811                 /* Finally, transmit new packets.  */
 812                 start_timer = 0;
 813                 queue = &q->out;
 814
 815                 while ((chunk = sctp_outq_dequeue_data(q))) {
 816                         /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
 817                          * stream identifier.
 818                          */
 819                         if (chunk->sinfo.sinfo_stream >=
 820                             asoc->c.sinit_num_ostreams) {
 821
 822                                 /* Mark as failed send. */
 823                                 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
 824                                 sctp_chunk_free(chunk);
 825                                 continue;
 826                         }
 827
 828                         /* Has this chunk expired? */
 829                         if (sctp_chunk_abandoned(chunk)) {
 830                                 sctp_chunk_fail(chunk, 0);
 831                                 sctp_chunk_free(chunk);
 832                                 continue;
 833                         }
 834
 835                         /* If there is a specified transport, use it.
 836                          * Otherwise, we want to use the active path.
 837                          */
 838                         new_transport = chunk->transport;
 839                         if (!new_transport || !new_transport->active)
 840                                 new_transport = asoc->peer.active_path;
 841
 842                         /* Change packets if necessary.  */
 843                         if (new_transport != transport) {
 844                                 transport = new_transport;
 845
 846                                 /* Schedule to have this transport's
 847                                  * packet flushed.
 848                                  */
 849                                 if (list_empty(&transport->send_ready)) {
 850                                         list_add_tail(&transport->send_ready,
 851                                                       &transport_list);
 852                                 }
 853
 854                                 packet = &transport->packet;
 855                                 sctp_packet_config(packet, vtag,
 856                                                    asoc->peer.ecn_capable);
 857                         }
 858
 859                         SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
 860                                           q, chunk,
 861                                           chunk && chunk->chunk_hdr ?
 862                                           sctp_cname(SCTP_ST_CHUNK(
 863                                                   chunk->chunk_hdr->type))
 864                                           : "Illegal Chunk");
 865
 866                         SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
 867                                         "%p skb->users %d.\n",
 868                                         ntohl(chunk->subh.data_hdr->tsn),
 869                                         chunk->skb ?chunk->skb->head : 0,
 870                                         chunk->skb ?
 871                                         atomic_read(&chunk->skb->users) : -1);
 872
 873                         /* Add the chunk to the packet.  */
 874                         status = sctp_packet_transmit_chunk(packet, chunk);
 875
 876                         switch (status) {
 877                         case SCTP_XMIT_PMTU_FULL:
 878                         case SCTP_XMIT_RWND_FULL:
 879                         case SCTP_XMIT_NAGLE_DELAY:
 880                                 /* We could not append this chunk, so put
 881                                  * the chunk back on the output queue.
 882                                  */
 883                                 SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
 884                                         "not transmit TSN: 0x%x, status: %d\n",
 885                                         ntohl(chunk->subh.data_hdr->tsn),
 886                                         status);
 887                                 sctp_outq_head_data(q, chunk);
 888                                 goto sctp_flush_out;
 889                                 break;
 890
 891                         case SCTP_XMIT_OK:
 892                                 break;
 893
 894                         default:
 895                                 BUG();
 896                         }
 897
 898                         /* BUG: We assume that the sctp_packet_transmit()
 899                          * call below will succeed all the time and add the
 900                          * chunk to the transmitted list and restart the
 901                          * timers.
 902                          * It is possible that the call can fail under OOM
 903                          * conditions.
 904                          *
 905                          * Is this really a problem?  Won't this behave
 906                          * like a lost TSN?
 907                          */
 908                         list_add_tail(&chunk->transmitted_list,
 909                                       &transport->transmitted);
 910
 911                         sctp_transport_reset_timers(transport);
 912
 913                         q->empty = 0;
 914
 915                         /* Only let one DATA chunk get bundled with a
 916                          * COOKIE-ECHO chunk.
 917                          */
 918                         if (packet->has_cookie_echo)
 919                                 goto sctp_flush_out;
 920                 }
 921                 break;
 922
 923         default:
 924                 /* Do nothing.  */
 925                 break;
 926         }
 927
 928 sctp_flush_out:
 929
 930         /* Before returning, examine all the transports touched in
 931          * this call.  Right now, we bluntly force clear all the
 932          * transports.  Things might change after we implement Nagle.
 933          * But such an examination is still required.
 934          *
 935          * --xguo
 936          */
 937         while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
 938                 struct sctp_transport *t = list_entry(ltransport,
 939                                                       struct sctp_transport,
 940                                                       send_ready);
 941                 packet = &t->packet;
 942                 if (!sctp_packet_empty(packet))
 943                         error = sctp_packet_transmit(packet);
 944         }
 945
 946         return error;
 947 }
 948
 949 /* Update unack_data based on the incoming SACK chunk */
 950 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
 951                                         struct sctp_sackhdr *sack)
 952 {
 953         sctp_sack_variable_t *frags;
 954         __u16 unack_data;
 955         int i;
 956
 957         unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
 958
 959         frags = sack->variable;
 960         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
 961                 unack_data -= ((ntohs(frags[i].gab.end) -
 962                                 ntohs(frags[i].gab.start) + 1));
 963         }
 964
 965         assoc->unack_data = unack_data;
 966 }
 967
 968 /* Return the highest new tsn that is acknowledged by the given SACK chunk. */
 969 static __u32 sctp_highest_new_tsn(struct sctp_sackhdr *sack,
 970                                   struct sctp_association *asoc)
 971 {
 972         struct list_head *ltransport, *lchunk;
 973         struct sctp_transport *transport;
 974         struct sctp_chunk *chunk;
 975         __u32 highest_new_tsn, tsn;
 976         struct list_head *transport_list = &asoc->peer.transport_addr_list;
 977
 978         highest_new_tsn = ntohl(sack->cum_tsn_ack);
 979
 980         list_for_each(ltransport, transport_list) {
 981                 transport = list_entry(ltransport, struct sctp_transport,
 982                                        transports);
 983                 list_for_each(lchunk, &transport->transmitted) {
 984                         chunk = list_entry(lchunk, struct sctp_chunk,
 985                                            transmitted_list);
 986                         tsn = ntohl(chunk->subh.data_hdr->tsn);
 987
 988                         if (!chunk->tsn_gap_acked &&
 989                             TSN_lt(highest_new_tsn, tsn) &&
 990                             sctp_acked(sack, tsn))
 991                                 highest_new_tsn = tsn;
 992                 }
 993         }
 994
 995         return highest_new_tsn;
 996 }
 997
 998 /* This is where we REALLY process a SACK.
 999  *
1000  * Process the SACK against the outqueue.  Mostly, this just frees
1001  * things off the transmitted queue.
1002  */
1003 int sctp_outq_sack(struct sctp_outq *q, struct sctp_sackhdr *sack)
1004 {
1005         struct sctp_association *asoc = q->asoc;
1006         struct sctp_transport *transport;
1007         struct sctp_chunk *tchunk = NULL;
1008         struct list_head *lchunk, *transport_list, *pos, *temp;
1009         sctp_sack_variable_t *frags = sack->variable;
1010         __u32 sack_ctsn, ctsn, tsn;
1011         __u32 highest_tsn, highest_new_tsn;
1012         __u32 sack_a_rwnd;
1013         unsigned outstanding;
1014         struct sctp_transport *primary = asoc->peer.primary_path;
1015         int count_of_newacks = 0;
1016
1017         /* Grab the association's destination address list. */
1018         transport_list = &asoc->peer.transport_addr_list;
1019
1020         sack_ctsn = ntohl(sack->cum_tsn_ack);
1021
1022         /*
1023          * SFR-CACC algorithm:
1024          * On receipt of a SACK the sender SHOULD execute the
1025          * following statements.
1026          *
1027          * 1) If the cumulative ack in the SACK passes next tsn_at_change
1028          * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1029          * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1030          * all destinations.
1031          */
1032         if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1033                 primary->cacc.changeover_active = 0;
1034                 list_for_each(pos, transport_list) {
1035                         transport = list_entry(pos, struct sctp_transport,
1036                                         transports);
1037                         transport->cacc.cycling_changeover = 0;
1038                 }
1039         }
1040
1041         /*
1042          * SFR-CACC algorithm:
1043          * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1044          * is set the receiver of the SACK MUST take the following actions:
1045          *
1046          * A) Initialize the cacc_saw_newack to 0 for all destination
1047          * addresses.
1048          */
1049         if (sack->num_gap_ack_blocks > 0 &&
1050             primary->cacc.changeover_active) {
1051                 list_for_each(pos, transport_list) {
1052                         transport = list_entry(pos, struct sctp_transport,
1053                                         transports);
1054                         transport->cacc.cacc_saw_newack = 0;
1055                 }
1056         }
1057
1058         /* Get the highest TSN in the sack. */
1059         highest_tsn = sack_ctsn;
1060         if (sack->num_gap_ack_blocks)
1061                 highest_tsn +=
1062                     ntohs(frags[ntohs(sack->num_gap_ack_blocks) - 1].gab.end);
1063
1064         if (TSN_lt(asoc->highest_sacked, highest_tsn)) {
1065                 highest_new_tsn = highest_tsn;
1066                 asoc->highest_sacked = highest_tsn;
1067         } else {
1068                 highest_new_tsn = sctp_highest_new_tsn(sack, asoc);
1069         }
1070
1071         /* Run through the retransmit queue.  Credit bytes received
1072          * and free those chunks that we can.
1073          */
1074         sctp_check_transmitted(q, &q->retransmit, NULL, sack, highest_new_tsn);
1075         sctp_mark_missing(q, &q->retransmit, NULL, highest_new_tsn, 0);
1076
1077         /* Run through the transmitted queue.
1078          * Credit bytes received and free those chunks which we can.
1079          *
1080          * This is a MASSIVE candidate for optimization.
1081          */
1082         list_for_each(pos, transport_list) {
1083                 transport  = list_entry(pos, struct sctp_transport,
1084                                         transports);
1085                 sctp_check_transmitted(q, &transport->transmitted,
1086                                        transport, sack, highest_new_tsn);
1087                 /*
1088                  * SFR-CACC algorithm:
1089                  * C) Let count_of_newacks be the number of
1090                  * destinations for which cacc_saw_newack is set.
1091                  */
1092                 if (transport->cacc.cacc_saw_newack)
1093                         count_of_newacks ++;
1094         }
1095
1096         list_for_each(pos, transport_list) {
1097                 transport  = list_entry(pos, struct sctp_transport,
1098                                         transports);
1099                 sctp_mark_missing(q, &transport->transmitted, transport,
1100                                   highest_new_tsn, count_of_newacks);
1101         }
1102
1103         /* Move the Cumulative TSN Ack Point if appropriate.  */
1104         if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn))
1105                 asoc->ctsn_ack_point = sack_ctsn;
1106
1107         /* Update unack_data field in the assoc. */
1108         sctp_sack_update_unack_data(asoc, sack);
1109
1110         ctsn = asoc->ctsn_ack_point;
1111
1112         /* Throw away stuff rotting on the sack queue.  */
1113         list_for_each_safe(lchunk, temp, &q->sacked) {
1114                 tchunk = list_entry(lchunk, struct sctp_chunk,
1115                                     transmitted_list);
1116                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1117                 if (TSN_lte(tsn, ctsn))
1118                         sctp_chunk_free(tchunk);
1119         }
1120
1121         /* ii) Set rwnd equal to the newly received a_rwnd minus the
1122          *     number of bytes still outstanding after processing the
1123          *     Cumulative TSN Ack and the Gap Ack Blocks.
1124          */
1125
1126         sack_a_rwnd = ntohl(sack->a_rwnd);
1127         outstanding = q->outstanding_bytes;
1128
1129         if (outstanding < sack_a_rwnd)
1130                 sack_a_rwnd -= outstanding;
1131         else
1132                 sack_a_rwnd = 0;
1133
1134         asoc->peer.rwnd = sack_a_rwnd;
1135
1136         sctp_generate_fwdtsn(q, sack_ctsn);
1137
1138         SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n",
1139                           __FUNCTION__, sack_ctsn);
1140         SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, "
1141                           "%p is 0x%x. Adv peer ack point: 0x%x\n",
1142                           __FUNCTION__, asoc, ctsn, asoc->adv_peer_ack_point);
1143
1144         /* See if all chunks are acked.
1145          * Make sure the empty queue handler will get run later.
1146          */
1147         q->empty = skb_queue_empty(&q->out) && skb_queue_empty(&q->control) &&
1148                         list_empty(&q->retransmit);
1149         if (!q->empty)
1150                 goto finish;
1151
1152         list_for_each(pos, transport_list) {
1153                 transport  = list_entry(pos, struct sctp_transport,
1154                                         transports);
1155                 q->empty = q->empty && list_empty(&transport->transmitted);
1156                 if (!q->empty)
1157                         goto finish;
1158         }
1159
1160         SCTP_DEBUG_PRINTK("sack queue is empty.\n");
1161 finish:
1162         return q->empty;
1163 }
1164
1165 /* Is the outqueue empty?  */
1166 int sctp_outq_is_empty(const struct sctp_outq *q)
1167 {
1168         return q->empty;
1169 }
1170
1171 /********************************************************************
1172  * 2nd Level Abstractions
1173  ********************************************************************/
1174
1175 /* Go through a transport's transmitted list or the association's retransmit
1176  * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1177  * The retransmit list will not have an associated transport.
1178  *
1179  * I added coherent debug information output.   --xguo
1180  *
1181  * Instead of printing 'sacked' or 'kept' for each TSN on the
1182  * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1183  * KEPT TSN6-TSN7, etc.
1184  */
1185 static void sctp_check_transmitted(struct sctp_outq *q,
1186                                    struct list_head *transmitted_queue,
1187                                    struct sctp_transport *transport,
1188                                    struct sctp_sackhdr *sack,
1189                                    __u32 highest_new_tsn_in_sack)
1190 {
1191         struct list_head *lchunk;
1192         struct sctp_chunk *tchunk;
1193         struct list_head tlist;
1194         __u32 tsn;
1195         __u32 sack_ctsn;
1196         __u32 rtt;
1197         __u8 restart_timer = 0;
1198         int bytes_acked = 0;
1199
1200         /* These state variables are for coherent debug output. --xguo */
1201
1202 #if SCTP_DEBUG
1203         __u32 dbg_ack_tsn = 0;  /* An ACKed TSN range starts here... */
1204         __u32 dbg_last_ack_tsn = 0;  /* ...and finishes here.        */
1205         __u32 dbg_kept_tsn = 0; /* An un-ACKed range starts here...  */
1206         __u32 dbg_last_kept_tsn = 0; /* ...and finishes here.        */
1207
1208         /* 0 : The last TSN was ACKed.
1209          * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1210          * -1: We need to initialize.
1211          */
1212         int dbg_prt_state = -1;
1213 #endif /* SCTP_DEBUG */
1214
1215         sack_ctsn = ntohl(sack->cum_tsn_ack);
1216
1217         INIT_LIST_HEAD(&tlist);
1218
1219         /* The while loop will skip empty transmitted queues. */
1220         while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1221                 tchunk = list_entry(lchunk, struct sctp_chunk,
1222                                     transmitted_list);
1223
1224                 if (sctp_chunk_abandoned(tchunk)) {
1225                         /* Move the chunk to abandoned list. */
1226                         sctp_insert_list(&q->abandoned, lchunk);
1227                         continue;
1228                 }
1229
1230                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1231                 if (sctp_acked(sack, tsn)) {
1232                         /* If this queue is the retransmit queue, the
1233                          * retransmit timer has already reclaimed
1234                          * the outstanding bytes for this chunk, so only
1235                          * count bytes associated with a transport.
1236                          */
1237                         if (transport) {
1238                                 /* If this chunk is being used for RTT
1239                                  * measurement, calculate the RTT and update
1240                                  * the RTO using this value.
1241                                  *
1242                                  * 6.3.1 C5) Karn's algorithm: RTT measurements
1243                                  * MUST NOT be made using packets that were
1244                                  * retransmitted (and thus for which it is
1245                                  * ambiguous whether the reply was for the
1246                                  * first instance of the packet or a later
1247                                  * instance).
1248                                  */
1249                                 if (!tchunk->tsn_gap_acked &&
1250                                     !tchunk->resent &&
1251                                     tchunk->rtt_in_progress) {
1252                                         rtt = jiffies - tchunk->sent_at;
1253                                         sctp_transport_update_rto(transport,
1254                                                                   rtt);
1255                                 }
1256                         }
1257                         if (TSN_lte(tsn, sack_ctsn)) {
1258                                 /* RFC 2960  6.3.2 Retransmission Timer Rules
1259                                  *
1260                                  * R3) Whenever a SACK is received
1261                                  * that acknowledges the DATA chunk
1262                                  * with the earliest outstanding TSN
1263                                  * for that address, restart T3-rtx
1264                                  * timer for that address with its
1265                                  * current RTO.
1266                                  */
1267                                 restart_timer = 1;
1268
1269                                 if (!tchunk->tsn_gap_acked) {
1270                                         tchunk->tsn_gap_acked = 1;
1271                                         bytes_acked += sctp_data_size(tchunk);
1272                                         /*
1273                                          * SFR-CACC algorithm:
1274                                          * 2) If the SACK contains gap acks
1275                                          * and the flag CHANGEOVER_ACTIVE is
1276                                          * set the receiver of the SACK MUST
1277                                          * take the following action:
1278                                          *
1279                                          * B) For each TSN t being acked that
1280                                          * has not been acked in any SACK so
1281                                          * far, set cacc_saw_newack to 1 for
1282                                          * the destination that the TSN was
1283                                          * sent to.
1284                                          */
1285                                         if (transport &&
1286                                             sack->num_gap_ack_blocks &&
1287                                             q->asoc->peer.primary_path->cacc.
1288                                             changeover_active)
1289                                                 transport->cacc.cacc_saw_newack
1290                                                         = 1;
1291                                 }
1292
1293                                 list_add_tail(&tchunk->transmitted_list,
1294                                               &q->sacked);
1295                         } else {
1296                                 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1297                                  * M2) Each time a SACK arrives reporting
1298                                  * 'Stray DATA chunk(s)' record the highest TSN
1299                                  * reported as newly acknowledged, call this
1300                                  * value 'HighestTSNinSack'. A newly
1301                                  * acknowledged DATA chunk is one not
1302                                  * previously acknowledged in a SACK.
1303                                  *
1304                                  * When the SCTP sender of data receives a SACK
1305                                  * chunk that acknowledges, for the first time,
1306                                  * the receipt of a DATA chunk, all the still
1307                                  * unacknowledged DATA chunks whose TSN is
1308                                  * older than that newly acknowledged DATA
1309                                  * chunk, are qualified as 'Stray DATA chunks'.
1310                                  */
1311                                 if (!tchunk->tsn_gap_acked) {
1312                                         tchunk->tsn_gap_acked = 1;
1313                                         bytes_acked += sctp_data_size(tchunk);
1314                                 }
1315                                 list_add_tail(lchunk, &tlist);
1316                         }
1317
1318 #if SCTP_DEBUG
1319                         switch (dbg_prt_state) {
1320                         case 0: /* last TSN was ACKed */
1321                                 if (dbg_last_ack_tsn + 1 == tsn) {
1322                                         /* This TSN belongs to the
1323                                          * current ACK range.
1324                                          */
1325                                         break;
1326                                 }
1327
1328                                 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1329                                         /* Display the end of the
1330                                          * current range.
1331                                          */
1332                                         SCTP_DEBUG_PRINTK("-%08x",
1333                                                           dbg_last_ack_tsn);
1334                                 }
1335
1336                                 /* Start a new range.  */
1337                                 SCTP_DEBUG_PRINTK(",%08x", tsn);
1338                                 dbg_ack_tsn = tsn;
1339                                 break;
1340
1341                         case 1: /* The last TSN was NOT ACKed. */
1342                                 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1343                                         /* Display the end of current range. */
1344                                         SCTP_DEBUG_PRINTK("-%08x",
1345                                                           dbg_last_kept_tsn);
1346                                 }
1347
1348                                 SCTP_DEBUG_PRINTK("\n");
1349
1350                                 /* FALL THROUGH... */
1351                         default:
1352                                 /* This is the first-ever TSN we examined.  */
1353                                 /* Start a new range of ACK-ed TSNs.  */
1354                                 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn);
1355                                 dbg_prt_state = 0;
1356                                 dbg_ack_tsn = tsn;
1357                         };
1358
1359                         dbg_last_ack_tsn = tsn;
1360 #endif /* SCTP_DEBUG */
1361
1362                 } else {
1363                         if (tchunk->tsn_gap_acked) {
1364                                 SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1365                                                   "data TSN: 0x%x\n",
1366                                                   __FUNCTION__,
1367                                                   tsn);
1368                                 tchunk->tsn_gap_acked = 0;
1369
1370                                 bytes_acked -= sctp_data_size(tchunk);
1371
1372                                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1373                                  *
1374                                  * R4) Whenever a SACK is received missing a
1375                                  * TSN that was previously acknowledged via a
1376                                  * Gap Ack Block, start T3-rtx for the
1377                                  * destination address to which the DATA
1378                                  * chunk was originally
1379                                  * transmitted if it is not already running.
1380                                  */
1381                                 restart_timer = 1;
1382                         }
1383
1384                         list_add_tail(lchunk, &tlist);
1385
1386 #if SCTP_DEBUG
1387                         /* See the above comments on ACK-ed TSNs. */
1388                         switch (dbg_prt_state) {
1389                         case 1:
1390                                 if (dbg_last_kept_tsn + 1 == tsn)
1391                                         break;
1392
1393                                 if (dbg_last_kept_tsn != dbg_kept_tsn)
1394                                         SCTP_DEBUG_PRINTK("-%08x",
1395                                                           dbg_last_kept_tsn);
1396
1397                                 SCTP_DEBUG_PRINTK(",%08x", tsn);
1398                                 dbg_kept_tsn = tsn;
1399                                 break;
1400
1401                         case 0:
1402                                 if (dbg_last_ack_tsn != dbg_ack_tsn)
1403                                         SCTP_DEBUG_PRINTK("-%08x",
1404                                                           dbg_last_ack_tsn);
1405                                 SCTP_DEBUG_PRINTK("\n");
1406
1407                                 /* FALL THROUGH... */
1408                         default:
1409                                 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn);
1410                                 dbg_prt_state = 1;
1411                                 dbg_kept_tsn = tsn;
1412                         };
1413
1414                         dbg_last_kept_tsn = tsn;
1415 #endif /* SCTP_DEBUG */
1416                 }
1417         }
1418
1419 #if SCTP_DEBUG
1420         /* Finish off the last range, displaying its ending TSN.  */
1421         switch (dbg_prt_state) {
1422         case 0:
1423                 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1424                         SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_ack_tsn);
1425                 } else {
1426                         SCTP_DEBUG_PRINTK("\n");
1427                 }
1428         break;
1429
1430         case 1:
1431                 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1432                         SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_kept_tsn);
1433                 } else {
1434                         SCTP_DEBUG_PRINTK("\n");
1435                 }
1436         };
1437 #endif /* SCTP_DEBUG */
1438         if (transport) {
1439                 if (bytes_acked) {
1440                         /* 8.2. When an outstanding TSN is acknowledged,
1441                          * the endpoint shall clear the error counter of
1442                          * the destination transport address to which the
1443                          * DATA chunk was last sent.
1444                          * The association's overall error counter is
1445                          * also cleared.
1446                          */
1447                         transport->error_count = 0;
1448                         transport->asoc->overall_error_count = 0;
1449
1450                         /* Mark the destination transport address as
1451                          * active if it is not so marked.
1452                          */
1453                         if (!transport->active) {
1454                                 sctp_assoc_control_transport(
1455                                         transport->asoc,
1456                                         transport,
1457                                         SCTP_TRANSPORT_UP,
1458                                         SCTP_RECEIVED_SACK);
1459                         }
1460
1461                         sctp_transport_raise_cwnd(transport, sack_ctsn,
1462                                                   bytes_acked);
1463
1464                         transport->flight_size -= bytes_acked;
1465                         q->outstanding_bytes -= bytes_acked;
1466                 } else {
1467                         /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1468                          * When a sender is doing zero window probing, it
1469                          * should not timeout the association if it continues
1470                          * to receive new packets from the receiver. The
1471                          * reason is that the receiver MAY keep its window
1472                          * closed for an indefinite time.
1473                          * A sender is doing zero window probing when the
1474                          * receiver's advertised window is zero, and there is
1475                          * only one data chunk in flight to the receiver.
1476                          */
1477                         if (!q->asoc->peer.rwnd &&
1478                             !list_empty(&tlist) &&
1479                             (sack_ctsn+2 == q->asoc->next_tsn)) {
1480                                 SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1481                                                   "window probe: %u\n",
1482                                                   __FUNCTION__, sack_ctsn);
1483                                 q->asoc->overall_error_count = 0;
1484                                 transport->error_count = 0;
1485                         }
1486                 }
1487
1488                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1489                  *
1490                  * R2) Whenever all outstanding data sent to an address have
1491                  * been acknowledged, turn off the T3-rtx timer of that
1492                  * address.
1493                  */
1494                 if (!transport->flight_size) {
1495                         if (timer_pending(&transport->T3_rtx_timer) &&
1496                             del_timer(&transport->T3_rtx_timer)) {
1497                                 sctp_transport_put(transport);
1498                         }
1499                 } else if (restart_timer) {
1500                         if (!mod_timer(&transport->T3_rtx_timer,
1501                                        jiffies + transport->rto))
1502                                 sctp_transport_hold(transport);
1503                 }
1504         }
1505
1506         list_splice(&tlist, transmitted_queue);
1507 }
1508
1509 /* Mark chunks as missing and consequently may get retransmitted. */
1510 static void sctp_mark_missing(struct sctp_outq *q,
1511                               struct list_head *transmitted_queue,
1512                               struct sctp_transport *transport,
1513                               __u32 highest_new_tsn_in_sack,
1514                               int count_of_newacks)
1515 {
1516         struct sctp_chunk *chunk;
1517         struct list_head *pos;
1518         __u32 tsn;
1519         char do_fast_retransmit = 0;
1520         struct sctp_transport *primary = q->asoc->peer.primary_path;
1521
1522         list_for_each(pos, transmitted_queue) {
1523
1524                 chunk = list_entry(pos, struct sctp_chunk, transmitted_list);
1525                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1526
1527                 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1528                  * 'Unacknowledged TSN's', if the TSN number of an
1529                  * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1530                  * value, increment the 'TSN.Missing.Report' count on that
1531                  * chunk if it has NOT been fast retransmitted or marked for
1532                  * fast retransmit already.
1533                  */
1534                 if (!chunk->fast_retransmit &&
1535                     !chunk->tsn_gap_acked &&
1536                     TSN_lt(tsn, highest_new_tsn_in_sack)) {
1537
1538                         /* SFR-CACC may require us to skip marking
1539                          * this chunk as missing.
1540                          */
1541                         if (!transport || !sctp_cacc_skip(primary, transport,
1542                                             count_of_newacks, tsn)) {
1543                                 chunk->tsn_missing_report++;
1544
1545                                 SCTP_DEBUG_PRINTK(
1546                                         "%s: TSN 0x%x missing counter: %d\n",
1547                                         __FUNCTION__, tsn,
1548                                         chunk->tsn_missing_report);
1549                         }
1550                 }
1551                 /*
1552                  * M4) If any DATA chunk is found to have a
1553                  * 'TSN.Missing.Report'
1554                  * value larger than or equal to 4, mark that chunk for
1555                  * retransmission and start the fast retransmit procedure.
1556                  */
1557
1558                 if (chunk->tsn_missing_report >= 4) {
1559                         chunk->fast_retransmit = 1;
1560                         do_fast_retransmit = 1;
1561                 }
1562         }
1563
1564         if (transport) {
1565                 if (do_fast_retransmit)
1566                         sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1567
1568                 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1569                                   "ssthresh: %d, flight_size: %d, pba: %d\n",
1570                                   __FUNCTION__, transport, transport->cwnd,
1571                                   transport->ssthresh, transport->flight_size,
1572                                   transport->partial_bytes_acked);
1573         }
1574 }
1575
1576 /* Is the given TSN acked by this packet?  */
1577 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1578 {
1579         int i;
1580         sctp_sack_variable_t *frags;
1581         __u16 gap;
1582         __u32 ctsn = ntohl(sack->cum_tsn_ack);
1583
1584         if (TSN_lte(tsn, ctsn))
1585                 goto pass;
1586
1587         /* 3.3.4 Selective Acknowledgement (SACK) (3):
1588          *
1589          * Gap Ack Blocks:
1590          *  These fields contain the Gap Ack Blocks. They are repeated
1591          *  for each Gap Ack Block up to the number of Gap Ack Blocks
1592          *  defined in the Number of Gap Ack Blocks field. All DATA
1593          *  chunks with TSNs greater than or equal to (Cumulative TSN
1594          *  Ack + Gap Ack Block Start) and less than or equal to
1595          *  (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1596          *  Block are assumed to have been received correctly.
1597          */
1598
1599         frags = sack->variable;
1600         gap = tsn - ctsn;
1601         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1602                 if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1603                     TSN_lte(gap, ntohs(frags[i].gab.end)))
1604                         goto pass;
1605         }
1606
1607         return 0;
1608 pass:
1609         return 1;
1610 }
1611
1612 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1613                                     int nskips, __u16 stream)
1614 {
1615         int i;
1616
1617         for (i = 0; i < nskips; i++) {
1618                 if (skiplist[i].stream == stream)
1619                         return i;
1620         }
1621         return i;
1622 }
1623
1624 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1625 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1626 {
1627         struct sctp_association *asoc = q->asoc;
1628         struct sctp_chunk *ftsn_chunk = NULL;
1629         struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1630         int nskips = 0;
1631         int skip_pos = 0;
1632         __u32 tsn;
1633         struct sctp_chunk *chunk;
1634         struct list_head *lchunk, *temp;
1635
1636         /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1637          * received SACK.
1638          *
1639          * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1640          * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1641          */
1642         if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1643                 asoc->adv_peer_ack_point = ctsn;
1644
1645         /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1646          * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1647          * the chunk next in the out-queue space is marked as "abandoned" as
1648          * shown in the following example:
1649          *
1650          * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1651          * and the Advanced.Peer.Ack.Point is updated to this value:
1652          *
1653          *   out-queue at the end of  ==>   out-queue after Adv.Ack.Point
1654          *   normal SACK processing           local advancement
1655          *                ...                           ...
1656          *   Adv.Ack.Pt-> 102 acked                     102 acked
1657          *                103 abandoned                 103 abandoned
1658          *                104 abandoned     Adv.Ack.P-> 104 abandoned
1659          *                105                           105
1660          *                106 acked                     106 acked
1661          *                ...                           ...
1662          *
1663          * In this example, the data sender successfully advanced the
1664          * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1665          */
1666         list_for_each_safe(lchunk, temp, &q->abandoned) {
1667                 chunk = list_entry(lchunk, struct sctp_chunk,
1668                                         transmitted_list);
1669                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1670
1671                 /* Remove any chunks in the abandoned queue that are acked by
1672                  * the ctsn.
1673                  */
1674                 if (TSN_lte(tsn, ctsn)) {
1675                         list_del_init(lchunk);
1676                         if (!chunk->tsn_gap_acked) {
1677                         chunk->transport->flight_size -=
1678                                                  sctp_data_size(chunk);
1679                         q->outstanding_bytes -= sctp_data_size(chunk);
1680                         }
1681                         sctp_chunk_free(chunk);
1682                 } else {
1683                         if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1684                                 asoc->adv_peer_ack_point = tsn;
1685                                 if (chunk->chunk_hdr->flags &
1686                                          SCTP_DATA_UNORDERED)
1687                                         continue;
1688                                 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1689                                                 nskips,
1690                                                 chunk->subh.data_hdr->stream);
1691                                 ftsn_skip_arr[skip_pos].stream =
1692                                         chunk->subh.data_hdr->stream;
1693                                 ftsn_skip_arr[skip_pos].ssn =
1694                                          chunk->subh.data_hdr->ssn;
1695                                 if (skip_pos == nskips)
1696                                         nskips++;
1697                                 if (nskips == 10)
1698                                         break;
1699                         } else
1700                                 break;
1701                 }
1702         }
1703
1704         /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1705          * is greater than the Cumulative TSN ACK carried in the received
1706          * SACK, the data sender MUST send the data receiver a FORWARD TSN
1707          * chunk containing the latest value of the
1708          * "Advanced.Peer.Ack.Point".
1709          *
1710          * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1711          * list each stream and sequence number in the forwarded TSN. This
1712          * information will enable the receiver to easily find any
1713          * stranded TSN's waiting on stream reorder queues. Each stream
1714          * SHOULD only be reported once; this means that if multiple
1715          * abandoned messages occur in the same stream then only the
1716          * highest abandoned stream sequence number is reported. If the
1717          * total size of the FORWARD TSN does NOT fit in a single MTU then
1718          * the sender of the FORWARD TSN SHOULD lower the
1719          * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1720          * single MTU.
1721          */
1722         if (asoc->adv_peer_ack_point > ctsn)
1723                 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1724                                               nskips, &ftsn_skip_arr[0]);
1725
1726         if (ftsn_chunk) {
1727                 __skb_queue_tail(&q->control, (struct sk_buff *)ftsn_chunk);
1728                 SCTP_INC_STATS(SctpOutCtrlChunks);
1729         }
1730 }