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         list_add(&ch->list, &q->out_chunk_list);
  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 = NULL;
  87
  88         if (!list_empty(&q->out_chunk_list)) {
  89                 struct list_head *entry = q->out_chunk_list.next;
  90
  91                 ch = list_entry(entry, struct sctp_chunk, list);
  92                 list_del_init(entry);
  93                 q->out_qlen -= ch->skb->len;
  94         }
  95         return ch;
  96 }
  97 /* Add data chunk to the end of the queue. */
  98 static inline void sctp_outq_tail_data(struct sctp_outq *q,
  99                                        struct sctp_chunk *ch)
 100 {
 101         list_add_tail(&ch->list, &q->out_chunk_list);
 102         q->out_qlen += ch->skb->len;
 103         return;
 104 }
 105
 106 /*
 107  * SFR-CACC algorithm:
 108  * D) If count_of_newacks is greater than or equal to 2
 109  * and t was not sent to the current primary then the
 110  * sender MUST NOT increment missing report count for t.
 111  */
 112 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary,
 113                                        struct sctp_transport *transport,
 114                                        int count_of_newacks)
 115 {
 116         if (count_of_newacks >=2 && transport != primary)
 117                 return 1;
 118         return 0;
 119 }
 120
 121 /*
 122  * SFR-CACC algorithm:
 123  * F) If count_of_newacks is less than 2, let d be the
 124  * destination to which t was sent. If cacc_saw_newack
 125  * is 0 for destination d, then the sender MUST NOT
 126  * increment missing report count for t.
 127  */
 128 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport,
 129                                        int count_of_newacks)
 130 {
 131         if (count_of_newacks < 2 && !transport->cacc.cacc_saw_newack)
 132                 return 1;
 133         return 0;
 134 }
 135
 136 /*
 137  * SFR-CACC algorithm:
 138  * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
 139  * execute steps C, D, F.
 140  *
 141  * C has been implemented in sctp_outq_sack
 142  */
 143 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
 144                                      struct sctp_transport *transport,
 145                                      int count_of_newacks)
 146 {
 147         if (!primary->cacc.cycling_changeover) {
 148                 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
 149                         return 1;
 150                 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
 151                         return 1;
 152                 return 0;
 153         }
 154         return 0;
 155 }
 156
 157 /*
 158  * SFR-CACC algorithm:
 159  * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
 160  * than next_tsn_at_change of the current primary, then
 161  * the sender MUST NOT increment missing report count
 162  * for t.
 163  */
 164 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
 165 {
 166         if (primary->cacc.cycling_changeover &&
 167             TSN_lt(tsn, primary->cacc.next_tsn_at_change))
 168                 return 1;
 169         return 0;
 170 }
 171
 172 /*
 173  * SFR-CACC algorithm:
 174  * 3) If the missing report count for TSN t is to be
 175  * incremented according to [RFC2960] and
 176  * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
 177  * then the sender MUST futher execute steps 3.1 and
 178  * 3.2 to determine if the missing report count for
 179  * TSN t SHOULD NOT be incremented.
 180  *
 181  * 3.3) If 3.1 and 3.2 do not dictate that the missing
 182  * report count for t should not be incremented, then
 183  * the sender SOULD increment missing report count for
 184  * t (according to [RFC2960] and [SCTP_STEWART_2002]).
 185  */
 186 static inline int sctp_cacc_skip(struct sctp_transport *primary,
 187                                  struct sctp_transport *transport,
 188                                  int count_of_newacks,
 189                                  __u32 tsn)
 190 {
 191         if (primary->cacc.changeover_active &&
 192             (sctp_cacc_skip_3_1(primary, transport, count_of_newacks)
 193              || sctp_cacc_skip_3_2(primary, tsn)))
 194                 return 1;
 195         return 0;
 196 }
 197
 198 /* Initialize an existing sctp_outq.  This does the boring stuff.
 199  * You still need to define handlers if you really want to DO
 200  * something with this structure...
 201  */
 202 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
 203 {
 204         q->asoc = asoc;
 205         INIT_LIST_HEAD(&q->out_chunk_list);
 206         INIT_LIST_HEAD(&q->control_chunk_list);
 207         INIT_LIST_HEAD(&q->retransmit);
 208         INIT_LIST_HEAD(&q->sacked);
 209         INIT_LIST_HEAD(&q->abandoned);
 210
 211         q->outstanding_bytes = 0;
 212         q->empty = 1;
 213         q->cork  = 0;
 214
 215         q->malloced = 0;
 216         q->out_qlen = 0;
 217 }
 218
 219 /* Free the outqueue structure and any related pending chunks.
 220  */
 221 void sctp_outq_teardown(struct sctp_outq *q)
 222 {
 223         struct sctp_transport *transport;
 224         struct list_head *lchunk, *pos, *temp;
 225         struct sctp_chunk *chunk, *tmp;
 226
 227         /* Throw away unacknowledged chunks. */
 228         list_for_each(pos, &q->asoc->peer.transport_addr_list) {
 229                 transport = list_entry(pos, struct sctp_transport, transports);
 230                 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
 231                         chunk = list_entry(lchunk, struct sctp_chunk,
 232                                            transmitted_list);
 233                         /* Mark as part of a failed message. */
 234                         sctp_chunk_fail(chunk, q->error);
 235                         sctp_chunk_free(chunk);
 236                 }
 237         }
 238
 239         /* Throw away chunks that have been gap ACKed.  */
 240         list_for_each_safe(lchunk, temp, &q->sacked) {
 241                 list_del_init(lchunk);
 242                 chunk = list_entry(lchunk, struct sctp_chunk,
 243                                    transmitted_list);
 244                 sctp_chunk_fail(chunk, q->error);
 245                 sctp_chunk_free(chunk);
 246         }
 247
 248         /* Throw away any chunks in the retransmit queue. */
 249         list_for_each_safe(lchunk, temp, &q->retransmit) {
 250                 list_del_init(lchunk);
 251                 chunk = list_entry(lchunk, struct sctp_chunk,
 252                                    transmitted_list);
 253                 sctp_chunk_fail(chunk, q->error);
 254                 sctp_chunk_free(chunk);
 255         }
 256
 257         /* Throw away any chunks that are in the abandoned queue. */
 258         list_for_each_safe(lchunk, temp, &q->abandoned) {
 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 leftover data chunks. */
 267         while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
 268
 269                 /* Mark as send failure. */
 270                 sctp_chunk_fail(chunk, q->error);
 271                 sctp_chunk_free(chunk);
 272         }
 273
 274         q->error = 0;
 275
 276         /* Throw away any leftover control chunks. */
 277         list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
 278                 list_del_init(&chunk->list);
 279                 sctp_chunk_free(chunk);
 280         }
 281 }
 282
 283 /* Free the outqueue structure and any related pending chunks.  */
 284 void sctp_outq_free(struct sctp_outq *q)
 285 {
 286         /* Throw away leftover chunks. */
 287         sctp_outq_teardown(q);
 288
 289         /* If we were kmalloc()'d, free the memory.  */
 290         if (q->malloced)
 291                 kfree(q);
 292 }
 293
 294 /* Put a new chunk in an sctp_outq.  */
 295 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk)
 296 {
 297         int error = 0;
 298
 299         SCTP_DEBUG_PRINTK("sctp_outq_tail(%p, %p[%s])\n",
 300                           q, chunk, chunk && chunk->chunk_hdr ?
 301                           sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
 302                           : "Illegal Chunk");
 303
 304         /* If it is data, queue it up, otherwise, send it
 305          * immediately.
 306          */
 307         if (SCTP_CID_DATA == chunk->chunk_hdr->type) {
 308                 /* Is it OK to queue data chunks?  */
 309                 /* From 9. Termination of Association
 310                  *
 311                  * When either endpoint performs a shutdown, the
 312                  * association on each peer will stop accepting new
 313                  * data from its user and only deliver data in queue
 314                  * at the time of sending or receiving the SHUTDOWN
 315                  * chunk.
 316                  */
 317                 switch (q->asoc->state) {
 318                 case SCTP_STATE_EMPTY:
 319                 case SCTP_STATE_CLOSED:
 320                 case SCTP_STATE_SHUTDOWN_PENDING:
 321                 case SCTP_STATE_SHUTDOWN_SENT:
 322                 case SCTP_STATE_SHUTDOWN_RECEIVED:
 323                 case SCTP_STATE_SHUTDOWN_ACK_SENT:
 324                         /* Cannot send after transport endpoint shutdown */
 325                         error = -ESHUTDOWN;
 326                         break;
 327
 328                 default:
 329                         SCTP_DEBUG_PRINTK("outqueueing (%p, %p[%s])\n",
 330                           q, chunk, chunk && chunk->chunk_hdr ?
 331                           sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
 332                           : "Illegal Chunk");
 333
 334                         sctp_outq_tail_data(q, chunk);
 335                         if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
 336                                 SCTP_INC_STATS(SCTP_MIB_OUTUNORDERCHUNKS);
 337                         else
 338                                 SCTP_INC_STATS(SCTP_MIB_OUTORDERCHUNKS);
 339                         q->empty = 0;
 340                         break;
 341                 }
 342         } else {
 343                 list_add_tail(&chunk->list, &q->control_chunk_list);
 344                 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
 345         }
 346
 347         if (error < 0)
 348                 return error;
 349
 350         if (!q->cork)
 351                 error = sctp_outq_flush(q, 0);
 352
 353         return error;
 354 }
 355
 356 /* Insert a chunk into the sorted list based on the TSNs.  The retransmit list
 357  * and the abandoned list are in ascending order.
 358  */
 359 static void sctp_insert_list(struct list_head *head, struct list_head *new)
 360 {
 361         struct list_head *pos;
 362         struct sctp_chunk *nchunk, *lchunk;
 363         __u32 ntsn, ltsn;
 364         int done = 0;
 365
 366         nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
 367         ntsn = ntohl(nchunk->subh.data_hdr->tsn);
 368
 369         list_for_each(pos, head) {
 370                 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
 371                 ltsn = ntohl(lchunk->subh.data_hdr->tsn);
 372                 if (TSN_lt(ntsn, ltsn)) {
 373                         list_add(new, pos->prev);
 374                         done = 1;
 375                         break;
 376                 }
 377         }
 378         if (!done)
 379                 list_add_tail(new, head);
 380 }
 381
 382 /* Mark all the eligible packets on a transport for retransmission.  */
 383 void sctp_retransmit_mark(struct sctp_outq *q,
 384                           struct sctp_transport *transport,
 385                           __u8 reason)
 386 {
 387         struct list_head *lchunk, *ltemp;
 388         struct sctp_chunk *chunk;
 389
 390         /* Walk through the specified transmitted queue.  */
 391         list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
 392                 chunk = list_entry(lchunk, struct sctp_chunk,
 393                                    transmitted_list);
 394
 395                 /* If the chunk is abandoned, move it to abandoned list. */
 396                 if (sctp_chunk_abandoned(chunk)) {
 397                         list_del_init(lchunk);
 398                         sctp_insert_list(&q->abandoned, lchunk);
 399
 400                         /* If this chunk has not been previousely acked,
 401                          * stop considering it 'outstanding'.  Our peer
 402                          * will most likely never see it since it will
 403                          * not be retransmitted
 404                          */
 405                         if (!chunk->tsn_gap_acked) {
 406                                 chunk->transport->flight_size -=
 407                                                 sctp_data_size(chunk);
 408                                 q->outstanding_bytes -= sctp_data_size(chunk);
 409                                 q->asoc->peer.rwnd += (sctp_data_size(chunk) +
 410                                                         sizeof(struct sk_buff));
 411                         }
 412                         continue;
 413                 }
 414
 415                 /* If we are doing  retransmission due to a timeout or pmtu
 416                  * discovery, only the  chunks that are not yet acked should
 417                  * be added to the retransmit queue.
 418                  */
 419                 if ((reason == SCTP_RTXR_FAST_RTX  &&
 420                             (chunk->fast_retransmit > 0)) ||
 421                     (reason != SCTP_RTXR_FAST_RTX  && !chunk->tsn_gap_acked)) {
 422                         /* If this chunk was sent less then 1 rto ago, do not
 423                          * retransmit this chunk, but give the peer time
 424                          * to acknowlege it.  Do this only when
 425                          * retransmitting due to T3 timeout.
 426                          */
 427                         if (reason == SCTP_RTXR_T3_RTX &&
 428                             (jiffies - chunk->sent_at) < transport->last_rto)
 429                                 continue;
 430
 431                         /* RFC 2960 6.2.1 Processing a Received SACK
 432                          *
 433                          * C) Any time a DATA chunk is marked for
 434                          * retransmission (via either T3-rtx timer expiration
 435                          * (Section 6.3.3) or via fast retransmit
 436                          * (Section 7.2.4)), add the data size of those
 437                          * chunks to the rwnd.
 438                          */
 439                         q->asoc->peer.rwnd += (sctp_data_size(chunk) +
 440                                                 sizeof(struct sk_buff));
 441                         q->outstanding_bytes -= sctp_data_size(chunk);
 442                         transport->flight_size -= sctp_data_size(chunk);
 443
 444                         /* sctpimpguide-05 Section 2.8.2
 445                          * M5) If a T3-rtx timer expires, the
 446                          * 'TSN.Missing.Report' of all affected TSNs is set
 447                          * to 0.
 448                          */
 449                         chunk->tsn_missing_report = 0;
 450
 451                         /* If a chunk that is being used for RTT measurement
 452                          * has to be retransmitted, we cannot use this chunk
 453                          * anymore for RTT measurements. Reset rto_pending so
 454                          * that a new RTT measurement is started when a new
 455                          * data chunk is sent.
 456                          */
 457                         if (chunk->rtt_in_progress) {
 458                                 chunk->rtt_in_progress = 0;
 459                                 transport->rto_pending = 0;
 460                         }
 461
 462                         /* Move the chunk to the retransmit queue. The chunks
 463                          * on the retransmit queue are always kept in order.
 464                          */
 465                         list_del_init(lchunk);
 466                         sctp_insert_list(&q->retransmit, lchunk);
 467                 }
 468         }
 469
 470         SCTP_DEBUG_PRINTK("%s: transport: %p, reason: %d, "
 471                           "cwnd: %d, ssthresh: %d, flight_size: %d, "
 472                           "pba: %d\n", __FUNCTION__,
 473                           transport, reason,
 474                           transport->cwnd, transport->ssthresh,
 475                           transport->flight_size,
 476                           transport->partial_bytes_acked);
 477
 478 }
 479
 480 /* Mark all the eligible packets on a transport for retransmission and force
 481  * one packet out.
 482  */
 483 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
 484                      sctp_retransmit_reason_t reason)
 485 {
 486         int error = 0;
 487
 488         switch(reason) {
 489         case SCTP_RTXR_T3_RTX:
 490                 SCTP_INC_STATS(SCTP_MIB_T3_RETRANSMITS);
 491                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
 492                 /* Update the retran path if the T3-rtx timer has expired for
 493                  * the current retran path.
 494                  */
 495                 if (transport == transport->asoc->peer.retran_path)
 496                         sctp_assoc_update_retran_path(transport->asoc);
 497                 break;
 498         case SCTP_RTXR_FAST_RTX:
 499                 SCTP_INC_STATS(SCTP_MIB_FAST_RETRANSMITS);
 500                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
 501                 break;
 502         case SCTP_RTXR_PMTUD:
 503                 SCTP_INC_STATS(SCTP_MIB_PMTUD_RETRANSMITS);
 504                 break;
 505         case SCTP_RTXR_T1_RTX:
 506                 SCTP_INC_STATS(SCTP_MIB_T1_RETRANSMITS);
 507                 break;
 508         default:
 509                 BUG();
 510         }
 511
 512         sctp_retransmit_mark(q, transport, reason);
 513
 514         /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
 515          * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
 516          * following the procedures outlined in C1 - C5.
 517          */
 518         sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
 519
 520         error = sctp_outq_flush(q, /* rtx_timeout */ 1);
 521
 522         if (error)
 523                 q->asoc->base.sk->sk_err = -error;
 524 }
 525
 526 /*
 527  * Transmit DATA chunks on the retransmit queue.  Upon return from
 528  * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
 529  * need to be transmitted by the caller.
 530  * We assume that pkt->transport has already been set.
 531  *
 532  * The return value is a normal kernel error return value.
 533  */
 534 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
 535                                int rtx_timeout, int *start_timer)
 536 {
 537         struct list_head *lqueue;
 538         struct list_head *lchunk, *lchunk1;
 539         struct sctp_transport *transport = pkt->transport;
 540         sctp_xmit_t status;
 541         struct sctp_chunk *chunk, *chunk1;
 542         struct sctp_association *asoc;
 543         int error = 0;
 544
 545         asoc = q->asoc;
 546         lqueue = &q->retransmit;
 547
 548         /* RFC 2960 6.3.3 Handle T3-rtx Expiration
 549          *
 550          * E3) Determine how many of the earliest (i.e., lowest TSN)
 551          * outstanding DATA chunks for the address for which the
 552          * T3-rtx has expired will fit into a single packet, subject
 553          * to the MTU constraint for the path corresponding to the
 554          * destination transport address to which the retransmission
 555          * is being sent (this may be different from the address for
 556          * which the timer expires [see Section 6.4]). Call this value
 557          * K. Bundle and retransmit those K DATA chunks in a single
 558          * packet to the destination endpoint.
 559          *
 560          * [Just to be painfully clear, if we are retransmitting
 561          * because a timeout just happened, we should send only ONE
 562          * packet of retransmitted data.]
 563          */
 564         lchunk = sctp_list_dequeue(lqueue);
 565
 566         while (lchunk) {
 567                 chunk = list_entry(lchunk, struct sctp_chunk,
 568                                    transmitted_list);
 569
 570                 /* Make sure that Gap Acked TSNs are not retransmitted.  A
 571                  * simple approach is just to move such TSNs out of the
 572                  * way and into a 'transmitted' queue and skip to the
 573                  * next chunk.
 574                  */
 575                 if (chunk->tsn_gap_acked) {
 576                         list_add_tail(lchunk, &transport->transmitted);
 577                         lchunk = sctp_list_dequeue(lqueue);
 578                         continue;
 579                 }
 580
 581                 /* Attempt to append this chunk to the packet. */
 582                 status = sctp_packet_append_chunk(pkt, chunk);
 583
 584                 switch (status) {
 585                 case SCTP_XMIT_PMTU_FULL:
 586                         /* Send this packet.  */
 587                         if ((error = sctp_packet_transmit(pkt)) == 0)
 588                                 *start_timer = 1;
 589
 590                         /* If we are retransmitting, we should only
 591                          * send a single packet.
 592                          */
 593                         if (rtx_timeout) {
 594                                 list_add(lchunk, lqueue);
 595                                 lchunk = NULL;
 596                         }
 597
 598                         /* Bundle lchunk in the next round.  */
 599                         break;
 600
 601                 case SCTP_XMIT_RWND_FULL:
 602                         /* Send this packet. */
 603                         if ((error = sctp_packet_transmit(pkt)) == 0)
 604                                 *start_timer = 1;
 605
 606                         /* Stop sending DATA as there is no more room
 607                          * at the receiver.
 608                          */
 609                         list_add(lchunk, lqueue);
 610                         lchunk = NULL;
 611                         break;
 612
 613                 case SCTP_XMIT_NAGLE_DELAY:
 614                         /* Send this packet. */
 615                         if ((error = sctp_packet_transmit(pkt)) == 0)
 616                                 *start_timer = 1;
 617
 618                         /* Stop sending DATA because of nagle delay. */
 619                         list_add(lchunk, lqueue);
 620                         lchunk = NULL;
 621                         break;
 622
 623                 default:
 624                         /* The append was successful, so add this chunk to
 625                          * the transmitted list.
 626                          */
 627                         list_add_tail(lchunk, &transport->transmitted);
 628
 629                         /* Mark the chunk as ineligible for fast retransmit
 630                          * after it is retransmitted.
 631                          */
 632                         if (chunk->fast_retransmit > 0)
 633                                 chunk->fast_retransmit = -1;
 634
 635                         *start_timer = 1;
 636                         q->empty = 0;
 637
 638                         /* Retrieve a new chunk to bundle. */
 639                         lchunk = sctp_list_dequeue(lqueue);
 640                         break;
 641                 }
 642
 643                 /* If we are here due to a retransmit timeout or a fast
 644                  * retransmit and if there are any chunks left in the retransmit
 645                  * queue that could not fit in the PMTU sized packet, they need
 646                  * to be marked as ineligible for a subsequent fast retransmit.
 647                  */
 648                 if (rtx_timeout && !lchunk) {
 649                         list_for_each(lchunk1, lqueue) {
 650                                 chunk1 = list_entry(lchunk1, struct sctp_chunk,
 651                                                     transmitted_list);
 652                                 if (chunk1->fast_retransmit > 0)
 653                                         chunk1->fast_retransmit = -1;
 654                         }
 655                 }
 656         }
 657
 658         return error;
 659 }
 660
 661 /* Cork the outqueue so queued chunks are really queued. */
 662 int sctp_outq_uncork(struct sctp_outq *q)
 663 {
 664         int error = 0;
 665         if (q->cork)
 666                 q->cork = 0;
 667         error = sctp_outq_flush(q, 0);
 668         return error;
 669 }
 670
 671 /*
 672  * Try to flush an outqueue.
 673  *
 674  * Description: Send everything in q which we legally can, subject to
 675  * congestion limitations.
 676  * * Note: This function can be called from multiple contexts so appropriate
 677  * locking concerns must be made.  Today we use the sock lock to protect
 678  * this function.
 679  */
 680 int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
 681 {
 682         struct sctp_packet *packet;
 683         struct sctp_packet singleton;
 684         struct sctp_association *asoc = q->asoc;
 685         __u16 sport = asoc->base.bind_addr.port;
 686         __u16 dport = asoc->peer.port;
 687         __u32 vtag = asoc->peer.i.init_tag;
 688         struct sctp_transport *transport = NULL;
 689         struct sctp_transport *new_transport;
 690         struct sctp_chunk *chunk, *tmp;
 691         sctp_xmit_t status;
 692         int error = 0;
 693         int start_timer = 0;
 694
 695         /* These transports have chunks to send. */
 696         struct list_head transport_list;
 697         struct list_head *ltransport;
 698
 699         INIT_LIST_HEAD(&transport_list);
 700         packet = NULL;
 701
 702         /*
 703          * 6.10 Bundling
 704          *   ...
 705          *   When bundling control chunks with DATA chunks, an
 706          *   endpoint MUST place control chunks first in the outbound
 707          *   SCTP packet.  The transmitter MUST transmit DATA chunks
 708          *   within a SCTP packet in increasing order of TSN.
 709          *   ...
 710          */
 711
 712         list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
 713                 list_del_init(&chunk->list);
 714
 715                 /* Pick the right transport to use. */
 716                 new_transport = chunk->transport;
 717
 718                 if (!new_transport) {
 719                         new_transport = asoc->peer.active_path;
 720                 } else if ((new_transport->state == SCTP_INACTIVE) ||
 721                            (new_transport->state == SCTP_UNCONFIRMED)) {
 722                         /* If the chunk is Heartbeat or Heartbeat Ack,
 723                          * send it to chunk->transport, even if it's
 724                          * inactive.
 725                          *
 726                          * 3.3.6 Heartbeat Acknowledgement:
 727                          * ...
 728                          * A HEARTBEAT ACK is always sent to the source IP
 729                          * address of the IP datagram containing the
 730                          * HEARTBEAT chunk to which this ack is responding.
 731                          * ...
 732                          */
 733                         if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
 734                             chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK)
 735                                 new_transport = asoc->peer.active_path;
 736                 }
 737
 738                 /* Are we switching transports?
 739                  * Take care of transport locks.
 740                  */
 741                 if (new_transport != transport) {
 742                         transport = new_transport;
 743                         if (list_empty(&transport->send_ready)) {
 744                                 list_add_tail(&transport->send_ready,
 745                                               &transport_list);
 746                         }
 747                         packet = &transport->packet;
 748                         sctp_packet_config(packet, vtag,
 749                                            asoc->peer.ecn_capable);
 750                 }
 751
 752                 switch (chunk->chunk_hdr->type) {
 753                 /*
 754                  * 6.10 Bundling
 755                  *   ...
 756                  *   An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
 757                  *   COMPLETE with any other chunks.  [Send them immediately.]
 758                  */
 759                 case SCTP_CID_INIT:
 760                 case SCTP_CID_INIT_ACK:
 761                 case SCTP_CID_SHUTDOWN_COMPLETE:
 762                         sctp_packet_init(&singleton, transport, sport, dport);
 763                         sctp_packet_config(&singleton, vtag, 0);
 764                         sctp_packet_append_chunk(&singleton, chunk);
 765                         error = sctp_packet_transmit(&singleton);
 766                         if (error < 0)
 767                                 return error;
 768                         break;
 769
 770                 case SCTP_CID_ABORT:
 771                 case SCTP_CID_SACK:
 772                 case SCTP_CID_HEARTBEAT:
 773                 case SCTP_CID_HEARTBEAT_ACK:
 774                 case SCTP_CID_SHUTDOWN:
 775                 case SCTP_CID_SHUTDOWN_ACK:
 776                 case SCTP_CID_ERROR:
 777                 case SCTP_CID_COOKIE_ECHO:
 778                 case SCTP_CID_COOKIE_ACK:
 779                 case SCTP_CID_ECN_ECNE:
 780                 case SCTP_CID_ECN_CWR:
 781                 case SCTP_CID_ASCONF:
 782                 case SCTP_CID_ASCONF_ACK:
 783                 case SCTP_CID_FWD_TSN:
 784                         sctp_packet_transmit_chunk(packet, chunk);
 785                         break;
 786
 787                 default:
 788                         /* We built a chunk with an illegal type! */
 789                         BUG();
 790                 }
 791         }
 792
 793         /* Is it OK to send data chunks?  */
 794         switch (asoc->state) {
 795         case SCTP_STATE_COOKIE_ECHOED:
 796                 /* Only allow bundling when this packet has a COOKIE-ECHO
 797                  * chunk.
 798                  */
 799                 if (!packet || !packet->has_cookie_echo)
 800                         break;
 801
 802                 /* fallthru */
 803         case SCTP_STATE_ESTABLISHED:
 804         case SCTP_STATE_SHUTDOWN_PENDING:
 805         case SCTP_STATE_SHUTDOWN_RECEIVED:
 806                 /*
 807                  * RFC 2960 6.1  Transmission of DATA Chunks
 808                  *
 809                  * C) When the time comes for the sender to transmit,
 810                  * before sending new DATA chunks, the sender MUST
 811                  * first transmit any outstanding DATA chunks which
 812                  * are marked for retransmission (limited by the
 813                  * current cwnd).
 814                  */
 815                 if (!list_empty(&q->retransmit)) {
 816                         if (transport == asoc->peer.retran_path)
 817                                 goto retran;
 818
 819                         /* Switch transports & prepare the packet.  */
 820
 821                         transport = asoc->peer.retran_path;
 822
 823                         if (list_empty(&transport->send_ready)) {
 824                                 list_add_tail(&transport->send_ready,
 825                                               &transport_list);
 826                         }
 827
 828                         packet = &transport->packet;
 829                         sctp_packet_config(packet, vtag,
 830                                            asoc->peer.ecn_capable);
 831                 retran:
 832                         error = sctp_outq_flush_rtx(q, packet,
 833                                                     rtx_timeout, &start_timer);
 834
 835                         if (start_timer)
 836                                 sctp_transport_reset_timers(transport);
 837
 838                         /* This can happen on COOKIE-ECHO resend.  Only
 839                          * one chunk can get bundled with a COOKIE-ECHO.
 840                          */
 841                         if (packet->has_cookie_echo)
 842                                 goto sctp_flush_out;
 843
 844                         /* Don't send new data if there is still data
 845                          * waiting to retransmit.
 846                          */
 847                         if (!list_empty(&q->retransmit))
 848                                 goto sctp_flush_out;
 849                 }
 850
 851                 /* Finally, transmit new packets.  */
 852                 start_timer = 0;
 853                 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
 854                         /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
 855                          * stream identifier.
 856                          */
 857                         if (chunk->sinfo.sinfo_stream >=
 858                             asoc->c.sinit_num_ostreams) {
 859
 860                                 /* Mark as failed send. */
 861                                 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
 862                                 sctp_chunk_free(chunk);
 863                                 continue;
 864                         }
 865
 866                         /* Has this chunk expired? */
 867                         if (sctp_chunk_abandoned(chunk)) {
 868                                 sctp_chunk_fail(chunk, 0);
 869                                 sctp_chunk_free(chunk);
 870                                 continue;
 871                         }
 872
 873                         /* If there is a specified transport, use it.
 874                          * Otherwise, we want to use the active path.
 875                          */
 876                         new_transport = chunk->transport;
 877                         if (!new_transport ||
 878                             ((new_transport->state == SCTP_INACTIVE) ||
 879                              (new_transport->state == SCTP_UNCONFIRMED)))
 880                                 new_transport = asoc->peer.active_path;
 881
 882                         /* Change packets if necessary.  */
 883                         if (new_transport != transport) {
 884                                 transport = new_transport;
 885
 886                                 /* Schedule to have this transport's
 887                                  * packet flushed.
 888                                  */
 889                                 if (list_empty(&transport->send_ready)) {
 890                                         list_add_tail(&transport->send_ready,
 891                                                       &transport_list);
 892                                 }
 893
 894                                 packet = &transport->packet;
 895                                 sctp_packet_config(packet, vtag,
 896                                                    asoc->peer.ecn_capable);
 897                         }
 898
 899                         SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
 900                                           q, chunk,
 901                                           chunk && chunk->chunk_hdr ?
 902                                           sctp_cname(SCTP_ST_CHUNK(
 903                                                   chunk->chunk_hdr->type))
 904                                           : "Illegal Chunk");
 905
 906                         SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
 907                                         "%p skb->users %d.\n",
 908                                         ntohl(chunk->subh.data_hdr->tsn),
 909                                         chunk->skb ?chunk->skb->head : NULL,
 910                                         chunk->skb ?
 911                                         atomic_read(&chunk->skb->users) : -1);
 912
 913                         /* Add the chunk to the packet.  */
 914                         status = sctp_packet_transmit_chunk(packet, chunk);
 915
 916                         switch (status) {
 917                         case SCTP_XMIT_PMTU_FULL:
 918                         case SCTP_XMIT_RWND_FULL:
 919                         case SCTP_XMIT_NAGLE_DELAY:
 920                                 /* We could not append this chunk, so put
 921                                  * the chunk back on the output queue.
 922                                  */
 923                                 SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
 924                                         "not transmit TSN: 0x%x, status: %d\n",
 925                                         ntohl(chunk->subh.data_hdr->tsn),
 926                                         status);
 927                                 sctp_outq_head_data(q, chunk);
 928                                 goto sctp_flush_out;
 929                                 break;
 930
 931                         case SCTP_XMIT_OK:
 932                                 break;
 933
 934                         default:
 935                                 BUG();
 936                         }
 937
 938                         /* BUG: We assume that the sctp_packet_transmit()
 939                          * call below will succeed all the time and add the
 940                          * chunk to the transmitted list and restart the
 941                          * timers.
 942                          * It is possible that the call can fail under OOM
 943                          * conditions.
 944                          *
 945                          * Is this really a problem?  Won't this behave
 946                          * like a lost TSN?
 947                          */
 948                         list_add_tail(&chunk->transmitted_list,
 949                                       &transport->transmitted);
 950
 951                         sctp_transport_reset_timers(transport);
 952
 953                         q->empty = 0;
 954
 955                         /* Only let one DATA chunk get bundled with a
 956                          * COOKIE-ECHO chunk.
 957                          */
 958                         if (packet->has_cookie_echo)
 959                                 goto sctp_flush_out;
 960                 }
 961                 break;
 962
 963         default:
 964                 /* Do nothing.  */
 965                 break;
 966         }
 967
 968 sctp_flush_out:
 969
 970         /* Before returning, examine all the transports touched in
 971          * this call.  Right now, we bluntly force clear all the
 972          * transports.  Things might change after we implement Nagle.
 973          * But such an examination is still required.
 974          *
 975          * --xguo
 976          */
 977         while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
 978                 struct sctp_transport *t = list_entry(ltransport,
 979                                                       struct sctp_transport,
 980                                                       send_ready);
 981                 packet = &t->packet;
 982                 if (!sctp_packet_empty(packet))
 983                         error = sctp_packet_transmit(packet);
 984         }
 985
 986         return error;
 987 }
 988
 989 /* Update unack_data based on the incoming SACK chunk */
 990 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
 991                                         struct sctp_sackhdr *sack)
 992 {
 993         sctp_sack_variable_t *frags;
 994         __u16 unack_data;
 995         int i;
 996
 997         unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
 998
 999         frags = sack->variable;
1000         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1001                 unack_data -= ((ntohs(frags[i].gab.end) -
1002                                 ntohs(frags[i].gab.start) + 1));
1003         }
1004
1005         assoc->unack_data = unack_data;
1006 }
1007
1008 /* Return the highest new tsn that is acknowledged by the given SACK chunk. */
1009 static __u32 sctp_highest_new_tsn(struct sctp_sackhdr *sack,
1010                                   struct sctp_association *asoc)
1011 {
1012         struct list_head *ltransport, *lchunk;
1013         struct sctp_transport *transport;
1014         struct sctp_chunk *chunk;
1015         __u32 highest_new_tsn, tsn;
1016         struct list_head *transport_list = &asoc->peer.transport_addr_list;
1017
1018         highest_new_tsn = ntohl(sack->cum_tsn_ack);
1019
1020         list_for_each(ltransport, transport_list) {
1021                 transport = list_entry(ltransport, struct sctp_transport,
1022                                        transports);
1023                 list_for_each(lchunk, &transport->transmitted) {
1024                         chunk = list_entry(lchunk, struct sctp_chunk,
1025                                            transmitted_list);
1026                         tsn = ntohl(chunk->subh.data_hdr->tsn);
1027
1028                         if (!chunk->tsn_gap_acked &&
1029                             TSN_lt(highest_new_tsn, tsn) &&
1030                             sctp_acked(sack, tsn))
1031                                 highest_new_tsn = tsn;
1032                 }
1033         }
1034
1035         return highest_new_tsn;
1036 }
1037
1038 /* This is where we REALLY process a SACK.
1039  *
1040  * Process the SACK against the outqueue.  Mostly, this just frees
1041  * things off the transmitted queue.
1042  */
1043 int sctp_outq_sack(struct sctp_outq *q, struct sctp_sackhdr *sack)
1044 {
1045         struct sctp_association *asoc = q->asoc;
1046         struct sctp_transport *transport;
1047         struct sctp_chunk *tchunk = NULL;
1048         struct list_head *lchunk, *transport_list, *pos, *temp;
1049         sctp_sack_variable_t *frags = sack->variable;
1050         __u32 sack_ctsn, ctsn, tsn;
1051         __u32 highest_tsn, highest_new_tsn;
1052         __u32 sack_a_rwnd;
1053         unsigned outstanding;
1054         struct sctp_transport *primary = asoc->peer.primary_path;
1055         int count_of_newacks = 0;
1056
1057         /* Grab the association's destination address list. */
1058         transport_list = &asoc->peer.transport_addr_list;
1059
1060         sack_ctsn = ntohl(sack->cum_tsn_ack);
1061
1062         /*
1063          * SFR-CACC algorithm:
1064          * On receipt of a SACK the sender SHOULD execute the
1065          * following statements.
1066          *
1067          * 1) If the cumulative ack in the SACK passes next tsn_at_change
1068          * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1069          * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1070          * all destinations.
1071          */
1072         if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1073                 primary->cacc.changeover_active = 0;
1074                 list_for_each(pos, transport_list) {
1075                         transport = list_entry(pos, struct sctp_transport,
1076                                         transports);
1077                         transport->cacc.cycling_changeover = 0;
1078                 }
1079         }
1080
1081         /*
1082          * SFR-CACC algorithm:
1083          * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1084          * is set the receiver of the SACK MUST take the following actions:
1085          *
1086          * A) Initialize the cacc_saw_newack to 0 for all destination
1087          * addresses.
1088          */
1089         if (sack->num_gap_ack_blocks &&
1090             primary->cacc.changeover_active) {
1091                 list_for_each(pos, transport_list) {
1092                         transport = list_entry(pos, struct sctp_transport,
1093                                         transports);
1094                         transport->cacc.cacc_saw_newack = 0;
1095                 }
1096         }
1097
1098         /* Get the highest TSN in the sack. */
1099         highest_tsn = sack_ctsn;
1100         if (sack->num_gap_ack_blocks)
1101                 highest_tsn +=
1102                     ntohs(frags[ntohs(sack->num_gap_ack_blocks) - 1].gab.end);
1103
1104         if (TSN_lt(asoc->highest_sacked, highest_tsn)) {
1105                 highest_new_tsn = highest_tsn;
1106                 asoc->highest_sacked = highest_tsn;
1107         } else {
1108                 highest_new_tsn = sctp_highest_new_tsn(sack, asoc);
1109         }
1110
1111         /* Run through the retransmit queue.  Credit bytes received
1112          * and free those chunks that we can.
1113          */
1114         sctp_check_transmitted(q, &q->retransmit, NULL, sack, highest_new_tsn);
1115         sctp_mark_missing(q, &q->retransmit, NULL, highest_new_tsn, 0);
1116
1117         /* Run through the transmitted queue.
1118          * Credit bytes received and free those chunks which we can.
1119          *
1120          * This is a MASSIVE candidate for optimization.
1121          */
1122         list_for_each(pos, transport_list) {
1123                 transport  = list_entry(pos, struct sctp_transport,
1124                                         transports);
1125                 sctp_check_transmitted(q, &transport->transmitted,
1126                                        transport, sack, highest_new_tsn);
1127                 /*
1128                  * SFR-CACC algorithm:
1129                  * C) Let count_of_newacks be the number of
1130                  * destinations for which cacc_saw_newack is set.
1131                  */
1132                 if (transport->cacc.cacc_saw_newack)
1133                         count_of_newacks ++;
1134         }
1135
1136         list_for_each(pos, transport_list) {
1137                 transport  = list_entry(pos, struct sctp_transport,
1138                                         transports);
1139                 sctp_mark_missing(q, &transport->transmitted, transport,
1140                                   highest_new_tsn, count_of_newacks);
1141         }
1142
1143         /* Move the Cumulative TSN Ack Point if appropriate.  */
1144         if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn))
1145                 asoc->ctsn_ack_point = sack_ctsn;
1146
1147         /* Update unack_data field in the assoc. */
1148         sctp_sack_update_unack_data(asoc, sack);
1149
1150         ctsn = asoc->ctsn_ack_point;
1151
1152         /* Throw away stuff rotting on the sack queue.  */
1153         list_for_each_safe(lchunk, temp, &q->sacked) {
1154                 tchunk = list_entry(lchunk, struct sctp_chunk,
1155                                     transmitted_list);
1156                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1157                 if (TSN_lte(tsn, ctsn))
1158                         sctp_chunk_free(tchunk);
1159         }
1160
1161         /* ii) Set rwnd equal to the newly received a_rwnd minus the
1162          *     number of bytes still outstanding after processing the
1163          *     Cumulative TSN Ack and the Gap Ack Blocks.
1164          */
1165
1166         sack_a_rwnd = ntohl(sack->a_rwnd);
1167         outstanding = q->outstanding_bytes;
1168
1169         if (outstanding < sack_a_rwnd)
1170                 sack_a_rwnd -= outstanding;
1171         else
1172                 sack_a_rwnd = 0;
1173
1174         asoc->peer.rwnd = sack_a_rwnd;
1175
1176         sctp_generate_fwdtsn(q, sack_ctsn);
1177
1178         SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n",
1179                           __FUNCTION__, sack_ctsn);
1180         SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, "
1181                           "%p is 0x%x. Adv peer ack point: 0x%x\n",
1182                           __FUNCTION__, asoc, ctsn, asoc->adv_peer_ack_point);
1183
1184         /* See if all chunks are acked.
1185          * Make sure the empty queue handler will get run later.
1186          */
1187         q->empty = (list_empty(&q->out_chunk_list) &&
1188                     list_empty(&q->control_chunk_list) &&
1189                     list_empty(&q->retransmit));
1190         if (!q->empty)
1191                 goto finish;
1192
1193         list_for_each(pos, transport_list) {
1194                 transport  = list_entry(pos, struct sctp_transport,
1195                                         transports);
1196                 q->empty = q->empty && list_empty(&transport->transmitted);
1197                 if (!q->empty)
1198                         goto finish;
1199         }
1200
1201         SCTP_DEBUG_PRINTK("sack queue is empty.\n");
1202 finish:
1203         return q->empty;
1204 }
1205
1206 /* Is the outqueue empty?  */
1207 int sctp_outq_is_empty(const struct sctp_outq *q)
1208 {
1209         return q->empty;
1210 }
1211
1212 /********************************************************************
1213  * 2nd Level Abstractions
1214  ********************************************************************/
1215
1216 /* Go through a transport's transmitted list or the association's retransmit
1217  * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1218  * The retransmit list will not have an associated transport.
1219  *
1220  * I added coherent debug information output.   --xguo
1221  *
1222  * Instead of printing 'sacked' or 'kept' for each TSN on the
1223  * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1224  * KEPT TSN6-TSN7, etc.
1225  */
1226 static void sctp_check_transmitted(struct sctp_outq *q,
1227                                    struct list_head *transmitted_queue,
1228                                    struct sctp_transport *transport,
1229                                    struct sctp_sackhdr *sack,
1230                                    __u32 highest_new_tsn_in_sack)
1231 {
1232         struct list_head *lchunk;
1233         struct sctp_chunk *tchunk;
1234         struct list_head tlist;
1235         __u32 tsn;
1236         __u32 sack_ctsn;
1237         __u32 rtt;
1238         __u8 restart_timer = 0;
1239         int bytes_acked = 0;
1240
1241         /* These state variables are for coherent debug output. --xguo */
1242
1243 #if SCTP_DEBUG
1244         __u32 dbg_ack_tsn = 0;  /* An ACKed TSN range starts here... */
1245         __u32 dbg_last_ack_tsn = 0;  /* ...and finishes here.        */
1246         __u32 dbg_kept_tsn = 0; /* An un-ACKed range starts here...  */
1247         __u32 dbg_last_kept_tsn = 0; /* ...and finishes here.        */
1248
1249         /* 0 : The last TSN was ACKed.
1250          * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1251          * -1: We need to initialize.
1252          */
1253         int dbg_prt_state = -1;
1254 #endif /* SCTP_DEBUG */
1255
1256         sack_ctsn = ntohl(sack->cum_tsn_ack);
1257
1258         INIT_LIST_HEAD(&tlist);
1259
1260         /* The while loop will skip empty transmitted queues. */
1261         while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1262                 tchunk = list_entry(lchunk, struct sctp_chunk,
1263                                     transmitted_list);
1264
1265                 if (sctp_chunk_abandoned(tchunk)) {
1266                         /* Move the chunk to abandoned list. */
1267                         sctp_insert_list(&q->abandoned, lchunk);
1268
1269                         /* If this chunk has not been acked, stop
1270                          * considering it as 'outstanding'.
1271                          */
1272                         if (!tchunk->tsn_gap_acked) {
1273                                 tchunk->transport->flight_size -=
1274                                                 sctp_data_size(tchunk);
1275                                 q->outstanding_bytes -= sctp_data_size(tchunk);
1276                         }
1277                         continue;
1278                 }
1279
1280                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1281                 if (sctp_acked(sack, tsn)) {
1282                         /* If this queue is the retransmit queue, the
1283                          * retransmit timer has already reclaimed
1284                          * the outstanding bytes for this chunk, so only
1285                          * count bytes associated with a transport.
1286                          */
1287                         if (transport) {
1288                                 /* If this chunk is being used for RTT
1289                                  * measurement, calculate the RTT and update
1290                                  * the RTO using this value.
1291                                  *
1292                                  * 6.3.1 C5) Karn's algorithm: RTT measurements
1293                                  * MUST NOT be made using packets that were
1294                                  * retransmitted (and thus for which it is
1295                                  * ambiguous whether the reply was for the
1296                                  * first instance of the packet or a later
1297                                  * instance).
1298                                  */
1299                                 if (!tchunk->tsn_gap_acked &&
1300                                     !tchunk->resent &&
1301                                     tchunk->rtt_in_progress) {
1302                                         tchunk->rtt_in_progress = 0;
1303                                         rtt = jiffies - tchunk->sent_at;
1304                                         sctp_transport_update_rto(transport,
1305                                                                   rtt);
1306                                 }
1307                         }
1308                         if (TSN_lte(tsn, sack_ctsn)) {
1309                                 /* RFC 2960  6.3.2 Retransmission Timer Rules
1310                                  *
1311                                  * R3) Whenever a SACK is received
1312                                  * that acknowledges the DATA chunk
1313                                  * with the earliest outstanding TSN
1314                                  * for that address, restart T3-rtx
1315                                  * timer for that address with its
1316                                  * current RTO.
1317                                  */
1318                                 restart_timer = 1;
1319
1320                                 if (!tchunk->tsn_gap_acked) {
1321                                         tchunk->tsn_gap_acked = 1;
1322                                         bytes_acked += sctp_data_size(tchunk);
1323                                         /*
1324                                          * SFR-CACC algorithm:
1325                                          * 2) If the SACK contains gap acks
1326                                          * and the flag CHANGEOVER_ACTIVE is
1327                                          * set the receiver of the SACK MUST
1328                                          * take the following action:
1329                                          *
1330                                          * B) For each TSN t being acked that
1331                                          * has not been acked in any SACK so
1332                                          * far, set cacc_saw_newack to 1 for
1333                                          * the destination that the TSN was
1334                                          * sent to.
1335                                          */
1336                                         if (transport &&
1337                                             sack->num_gap_ack_blocks &&
1338                                             q->asoc->peer.primary_path->cacc.
1339                                             changeover_active)
1340                                                 transport->cacc.cacc_saw_newack
1341                                                         = 1;
1342                                 }
1343
1344                                 list_add_tail(&tchunk->transmitted_list,
1345                                               &q->sacked);
1346                         } else {
1347                                 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1348                                  * M2) Each time a SACK arrives reporting
1349                                  * 'Stray DATA chunk(s)' record the highest TSN
1350                                  * reported as newly acknowledged, call this
1351                                  * value 'HighestTSNinSack'. A newly
1352                                  * acknowledged DATA chunk is one not
1353                                  * previously acknowledged in a SACK.
1354                                  *
1355                                  * When the SCTP sender of data receives a SACK
1356                                  * chunk that acknowledges, for the first time,
1357                                  * the receipt of a DATA chunk, all the still
1358                                  * unacknowledged DATA chunks whose TSN is
1359                                  * older than that newly acknowledged DATA
1360                                  * chunk, are qualified as 'Stray DATA chunks'.
1361                                  */
1362                                 if (!tchunk->tsn_gap_acked) {
1363                                         tchunk->tsn_gap_acked = 1;
1364                                         bytes_acked += sctp_data_size(tchunk);
1365                                 }
1366                                 list_add_tail(lchunk, &tlist);
1367                         }
1368
1369 #if SCTP_DEBUG
1370                         switch (dbg_prt_state) {
1371                         case 0: /* last TSN was ACKed */
1372                                 if (dbg_last_ack_tsn + 1 == tsn) {
1373                                         /* This TSN belongs to the
1374                                          * current ACK range.
1375                                          */
1376                                         break;
1377                                 }
1378
1379                                 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1380                                         /* Display the end of the
1381                                          * current range.
1382                                          */
1383                                         SCTP_DEBUG_PRINTK("-%08x",
1384                                                           dbg_last_ack_tsn);
1385                                 }
1386
1387                                 /* Start a new range.  */
1388                                 SCTP_DEBUG_PRINTK(",%08x", tsn);
1389                                 dbg_ack_tsn = tsn;
1390                                 break;
1391
1392                         case 1: /* The last TSN was NOT ACKed. */
1393                                 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1394                                         /* Display the end of current range. */
1395                                         SCTP_DEBUG_PRINTK("-%08x",
1396                                                           dbg_last_kept_tsn);
1397                                 }
1398
1399                                 SCTP_DEBUG_PRINTK("\n");
1400
1401                                 /* FALL THROUGH... */
1402                         default:
1403                                 /* This is the first-ever TSN we examined.  */
1404                                 /* Start a new range of ACK-ed TSNs.  */
1405                                 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn);
1406                                 dbg_prt_state = 0;
1407                                 dbg_ack_tsn = tsn;
1408                         }
1409
1410                         dbg_last_ack_tsn = tsn;
1411 #endif /* SCTP_DEBUG */
1412
1413                 } else {
1414                         if (tchunk->tsn_gap_acked) {
1415                                 SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1416                                                   "data TSN: 0x%x\n",
1417                                                   __FUNCTION__,
1418                                                   tsn);
1419                                 tchunk->tsn_gap_acked = 0;
1420
1421                                 bytes_acked -= sctp_data_size(tchunk);
1422
1423                                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1424                                  *
1425                                  * R4) Whenever a SACK is received missing a
1426                                  * TSN that was previously acknowledged via a
1427                                  * Gap Ack Block, start T3-rtx for the
1428                                  * destination address to which the DATA
1429                                  * chunk was originally
1430                                  * transmitted if it is not already running.
1431                                  */
1432                                 restart_timer = 1;
1433                         }
1434
1435                         list_add_tail(lchunk, &tlist);
1436
1437 #if SCTP_DEBUG
1438                         /* See the above comments on ACK-ed TSNs. */
1439                         switch (dbg_prt_state) {
1440                         case 1:
1441                                 if (dbg_last_kept_tsn + 1 == tsn)
1442                                         break;
1443
1444                                 if (dbg_last_kept_tsn != dbg_kept_tsn)
1445                                         SCTP_DEBUG_PRINTK("-%08x",
1446                                                           dbg_last_kept_tsn);
1447
1448                                 SCTP_DEBUG_PRINTK(",%08x", tsn);
1449                                 dbg_kept_tsn = tsn;
1450                                 break;
1451
1452                         case 0:
1453                                 if (dbg_last_ack_tsn != dbg_ack_tsn)
1454                                         SCTP_DEBUG_PRINTK("-%08x",
1455                                                           dbg_last_ack_tsn);
1456                                 SCTP_DEBUG_PRINTK("\n");
1457
1458                                 /* FALL THROUGH... */
1459                         default:
1460                                 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn);
1461                                 dbg_prt_state = 1;
1462                                 dbg_kept_tsn = tsn;
1463                         }
1464
1465                         dbg_last_kept_tsn = tsn;
1466 #endif /* SCTP_DEBUG */
1467                 }
1468         }
1469
1470 #if SCTP_DEBUG
1471         /* Finish off the last range, displaying its ending TSN.  */
1472         switch (dbg_prt_state) {
1473         case 0:
1474                 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1475                         SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_ack_tsn);
1476                 } else {
1477                         SCTP_DEBUG_PRINTK("\n");
1478                 }
1479         break;
1480
1481         case 1:
1482                 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1483                         SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_kept_tsn);
1484                 } else {
1485                         SCTP_DEBUG_PRINTK("\n");
1486                 }
1487         }
1488 #endif /* SCTP_DEBUG */
1489         if (transport) {
1490                 if (bytes_acked) {
1491                         /* 8.2. When an outstanding TSN is acknowledged,
1492                          * the endpoint shall clear the error counter of
1493                          * the destination transport address to which the
1494                          * DATA chunk was last sent.
1495                          * The association's overall error counter is
1496                          * also cleared.
1497                          */
1498                         transport->error_count = 0;
1499                         transport->asoc->overall_error_count = 0;
1500
1501                         /* Mark the destination transport address as
1502                          * active if it is not so marked.
1503                          */
1504                         if ((transport->state == SCTP_INACTIVE) ||
1505                             (transport->state == SCTP_UNCONFIRMED)) {
1506                                 sctp_assoc_control_transport(
1507                                         transport->asoc,
1508                                         transport,
1509                                         SCTP_TRANSPORT_UP,
1510                                         SCTP_RECEIVED_SACK);
1511                         }
1512
1513                         sctp_transport_raise_cwnd(transport, sack_ctsn,
1514                                                   bytes_acked);
1515
1516                         transport->flight_size -= bytes_acked;
1517                         q->outstanding_bytes -= bytes_acked;
1518                 } else {
1519                         /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1520                          * When a sender is doing zero window probing, it
1521                          * should not timeout the association if it continues
1522                          * to receive new packets from the receiver. The
1523                          * reason is that the receiver MAY keep its window
1524                          * closed for an indefinite time.
1525                          * A sender is doing zero window probing when the
1526                          * receiver's advertised window is zero, and there is
1527                          * only one data chunk in flight to the receiver.
1528                          */
1529                         if (!q->asoc->peer.rwnd &&
1530                             !list_empty(&tlist) &&
1531                             (sack_ctsn+2 == q->asoc->next_tsn)) {
1532                                 SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1533                                                   "window probe: %u\n",
1534                                                   __FUNCTION__, sack_ctsn);
1535                                 q->asoc->overall_error_count = 0;
1536                                 transport->error_count = 0;
1537                         }
1538                 }
1539
1540                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1541                  *
1542                  * R2) Whenever all outstanding data sent to an address have
1543                  * been acknowledged, turn off the T3-rtx timer of that
1544                  * address.
1545                  */
1546                 if (!transport->flight_size) {
1547                         if (timer_pending(&transport->T3_rtx_timer) &&
1548                             del_timer(&transport->T3_rtx_timer)) {
1549                                 sctp_transport_put(transport);
1550                         }
1551                 } else if (restart_timer) {
1552                         if (!mod_timer(&transport->T3_rtx_timer,
1553                                        jiffies + transport->rto))
1554                                 sctp_transport_hold(transport);
1555                 }
1556         }
1557
1558         list_splice(&tlist, transmitted_queue);
1559 }
1560
1561 /* Mark chunks as missing and consequently may get retransmitted. */
1562 static void sctp_mark_missing(struct sctp_outq *q,
1563                               struct list_head *transmitted_queue,
1564                               struct sctp_transport *transport,
1565                               __u32 highest_new_tsn_in_sack,
1566                               int count_of_newacks)
1567 {
1568         struct sctp_chunk *chunk;
1569         struct list_head *pos;
1570         __u32 tsn;
1571         char do_fast_retransmit = 0;
1572         struct sctp_transport *primary = q->asoc->peer.primary_path;
1573
1574         list_for_each(pos, transmitted_queue) {
1575
1576                 chunk = list_entry(pos, struct sctp_chunk, transmitted_list);
1577                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1578
1579                 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1580                  * 'Unacknowledged TSN's', if the TSN number of an
1581                  * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1582                  * value, increment the 'TSN.Missing.Report' count on that
1583                  * chunk if it has NOT been fast retransmitted or marked for
1584                  * fast retransmit already.
1585                  */
1586                 if (!chunk->fast_retransmit &&
1587                     !chunk->tsn_gap_acked &&
1588                     TSN_lt(tsn, highest_new_tsn_in_sack)) {
1589
1590                         /* SFR-CACC may require us to skip marking
1591                          * this chunk as missing.
1592                          */
1593                         if (!transport || !sctp_cacc_skip(primary, transport,
1594                                             count_of_newacks, tsn)) {
1595                                 chunk->tsn_missing_report++;
1596
1597                                 SCTP_DEBUG_PRINTK(
1598                                         "%s: TSN 0x%x missing counter: %d\n",
1599                                         __FUNCTION__, tsn,
1600                                         chunk->tsn_missing_report);
1601                         }
1602                 }
1603                 /*
1604                  * M4) If any DATA chunk is found to have a
1605                  * 'TSN.Missing.Report'
1606                  * value larger than or equal to 3, mark that chunk for
1607                  * retransmission and start the fast retransmit procedure.
1608                  */
1609
1610                 if (chunk->tsn_missing_report >= 3) {
1611                         chunk->fast_retransmit = 1;
1612                         do_fast_retransmit = 1;
1613                 }
1614         }
1615
1616         if (transport) {
1617                 if (do_fast_retransmit)
1618                         sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1619
1620                 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1621                                   "ssthresh: %d, flight_size: %d, pba: %d\n",
1622                                   __FUNCTION__, transport, transport->cwnd,
1623                                   transport->ssthresh, transport->flight_size,
1624                                   transport->partial_bytes_acked);
1625         }
1626 }
1627
1628 /* Is the given TSN acked by this packet?  */
1629 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1630 {
1631         int i;
1632         sctp_sack_variable_t *frags;
1633         __u16 gap;
1634         __u32 ctsn = ntohl(sack->cum_tsn_ack);
1635
1636         if (TSN_lte(tsn, ctsn))
1637                 goto pass;
1638
1639         /* 3.3.4 Selective Acknowledgement (SACK) (3):
1640          *
1641          * Gap Ack Blocks:
1642          *  These fields contain the Gap Ack Blocks. They are repeated
1643          *  for each Gap Ack Block up to the number of Gap Ack Blocks
1644          *  defined in the Number of Gap Ack Blocks field. All DATA
1645          *  chunks with TSNs greater than or equal to (Cumulative TSN
1646          *  Ack + Gap Ack Block Start) and less than or equal to
1647          *  (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1648          *  Block are assumed to have been received correctly.
1649          */
1650
1651         frags = sack->variable;
1652         gap = tsn - ctsn;
1653         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1654                 if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1655                     TSN_lte(gap, ntohs(frags[i].gab.end)))
1656                         goto pass;
1657         }
1658
1659         return 0;
1660 pass:
1661         return 1;
1662 }
1663
1664 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1665                                     int nskips, __be16 stream)
1666 {
1667         int i;
1668
1669         for (i = 0; i < nskips; i++) {
1670                 if (skiplist[i].stream == stream)
1671                         return i;
1672         }
1673         return i;
1674 }
1675
1676 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1677 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1678 {
1679         struct sctp_association *asoc = q->asoc;
1680         struct sctp_chunk *ftsn_chunk = NULL;
1681         struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1682         int nskips = 0;
1683         int skip_pos = 0;
1684         __u32 tsn;
1685         struct sctp_chunk *chunk;
1686         struct list_head *lchunk, *temp;
1687
1688         /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1689          * received SACK.
1690          *
1691          * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1692          * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1693          */
1694         if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1695                 asoc->adv_peer_ack_point = ctsn;
1696
1697         /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1698          * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1699          * the chunk next in the out-queue space is marked as "abandoned" as
1700          * shown in the following example:
1701          *
1702          * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1703          * and the Advanced.Peer.Ack.Point is updated to this value:
1704          *
1705          *   out-queue at the end of  ==>   out-queue after Adv.Ack.Point
1706          *   normal SACK processing           local advancement
1707          *                ...                           ...
1708          *   Adv.Ack.Pt-> 102 acked                     102 acked
1709          *                103 abandoned                 103 abandoned
1710          *                104 abandoned     Adv.Ack.P-> 104 abandoned
1711          *                105                           105
1712          *                106 acked                     106 acked
1713          *                ...                           ...
1714          *
1715          * In this example, the data sender successfully advanced the
1716          * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1717          */
1718         list_for_each_safe(lchunk, temp, &q->abandoned) {
1719                 chunk = list_entry(lchunk, struct sctp_chunk,
1720                                         transmitted_list);
1721                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1722
1723                 /* Remove any chunks in the abandoned queue that are acked by
1724                  * the ctsn.
1725                  */
1726                 if (TSN_lte(tsn, ctsn)) {
1727                         list_del_init(lchunk);
1728                         sctp_chunk_free(chunk);
1729                 } else {
1730                         if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1731                                 asoc->adv_peer_ack_point = tsn;
1732                                 if (chunk->chunk_hdr->flags &
1733                                          SCTP_DATA_UNORDERED)
1734                                         continue;
1735                                 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1736                                                 nskips,
1737                                                 chunk->subh.data_hdr->stream);
1738                                 ftsn_skip_arr[skip_pos].stream =
1739                                         chunk->subh.data_hdr->stream;
1740                                 ftsn_skip_arr[skip_pos].ssn =
1741                                          chunk->subh.data_hdr->ssn;
1742                                 if (skip_pos == nskips)
1743                                         nskips++;
1744                                 if (nskips == 10)
1745                                         break;
1746                         } else
1747                                 break;
1748                 }
1749         }
1750
1751         /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1752          * is greater than the Cumulative TSN ACK carried in the received
1753          * SACK, the data sender MUST send the data receiver a FORWARD TSN
1754          * chunk containing the latest value of the
1755          * "Advanced.Peer.Ack.Point".
1756          *
1757          * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1758          * list each stream and sequence number in the forwarded TSN. This
1759          * information will enable the receiver to easily find any
1760          * stranded TSN's waiting on stream reorder queues. Each stream
1761          * SHOULD only be reported once; this means that if multiple
1762          * abandoned messages occur in the same stream then only the
1763          * highest abandoned stream sequence number is reported. If the
1764          * total size of the FORWARD TSN does NOT fit in a single MTU then
1765          * the sender of the FORWARD TSN SHOULD lower the
1766          * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1767          * single MTU.
1768          */
1769         if (asoc->adv_peer_ack_point > ctsn)
1770                 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1771                                               nskips, &ftsn_skip_arr[0]);
1772
1773         if (ftsn_chunk) {
1774                 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1775                 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
1776         }
1777 }