net/sctp/outqueue.c

   1 /* SCTP kernel implementation
   2  * (C) Copyright IBM Corp. 2001, 2004
   3  * Copyright (c) 1999-2000 Cisco, Inc.
   4  * Copyright (c) 1999-2001 Motorola, Inc.
   5  * Copyright (c) 2001-2003 Intel Corp.
   6  *
   7  * This file is part of the SCTP kernel implementation
   8  *
   9  * These functions implement the sctp_outq class.   The outqueue handles
  10  * bundling and queueing of outgoing SCTP chunks.
  11  *
  12  * This SCTP 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  * This SCTP 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 <linux-sctp@vger.kernel.org>
  32  *
  33  * Written or modified by:
  34  *    La Monte H.P. Yarroll <piggy@acm.org>
  35  *    Karl Knutson          <karl@athena.chicago.il.us>
  36  *    Perry Melange         <pmelange@null.cc.uic.edu>
  37  *    Xingang Guo           <xingang.guo@intel.com>
  38  *    Hui Huang             <hui.huang@nokia.com>
  39  *    Sridhar Samudrala     <sri@us.ibm.com>
  40  *    Jon Grimm             <jgrimm@us.ibm.com>
  41  */
  42
  43 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  44
  45 #include <linux/types.h>
  46 #include <linux/list.h>   /* For struct list_head */
  47 #include <linux/socket.h>
  48 #include <linux/ip.h>
  49 #include <linux/slab.h>
  50 #include <net/sock.h>     /* For skb_set_owner_w */
  51
  52 #include <net/sctp/sctp.h>
  53 #include <net/sctp/sm.h>
  54
  55 /* Declare internal functions here.  */
  56 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn);
  57 static void sctp_check_transmitted(struct sctp_outq *q,
  58                                    struct list_head *transmitted_queue,
  59                                    struct sctp_transport *transport,
  60                                    union sctp_addr *saddr,
  61                                    struct sctp_sackhdr *sack,
  62                                    __u32 *highest_new_tsn);
  63
  64 static void sctp_mark_missing(struct sctp_outq *q,
  65                               struct list_head *transmitted_queue,
  66                               struct sctp_transport *transport,
  67                               __u32 highest_new_tsn,
  68                               int count_of_newacks);
  69
  70 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn);
  71
  72 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout);
  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 }
  81
  82 /* Take data from the front of the queue. */
  83 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
  84 {
  85         struct sctp_chunk *ch = NULL;
  86
  87         if (!list_empty(&q->out_chunk_list)) {
  88                 struct list_head *entry = q->out_chunk_list.next;
  89
  90                 ch = list_entry(entry, struct sctp_chunk, list);
  91                 list_del_init(entry);
  92                 q->out_qlen -= ch->skb->len;
  93         }
  94         return ch;
  95 }
  96 /* Add data chunk to the end of the queue. */
  97 static inline void sctp_outq_tail_data(struct sctp_outq *q,
  98                                        struct sctp_chunk *ch)
  99 {
 100         list_add_tail(&ch->list, &q->out_chunk_list);
 101         q->out_qlen += ch->skb->len;
 102 }
 103
 104 /*
 105  * SFR-CACC algorithm:
 106  * D) If count_of_newacks is greater than or equal to 2
 107  * and t was not sent to the current primary then the
 108  * sender MUST NOT increment missing report count for t.
 109  */
 110 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary,
 111                                        struct sctp_transport *transport,
 112                                        int count_of_newacks)
 113 {
 114         if (count_of_newacks >=2 && transport != primary)
 115                 return 1;
 116         return 0;
 117 }
 118
 119 /*
 120  * SFR-CACC algorithm:
 121  * F) If count_of_newacks is less than 2, let d be the
 122  * destination to which t was sent. If cacc_saw_newack
 123  * is 0 for destination d, then the sender MUST NOT
 124  * increment missing report count for t.
 125  */
 126 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport,
 127                                        int count_of_newacks)
 128 {
 129         if (count_of_newacks < 2 &&
 130                         (transport && !transport->cacc.cacc_saw_newack))
 131                 return 1;
 132         return 0;
 133 }
 134
 135 /*
 136  * SFR-CACC algorithm:
 137  * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
 138  * execute steps C, D, F.
 139  *
 140  * C has been implemented in sctp_outq_sack
 141  */
 142 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
 143                                      struct sctp_transport *transport,
 144                                      int count_of_newacks)
 145 {
 146         if (!primary->cacc.cycling_changeover) {
 147                 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
 148                         return 1;
 149                 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
 150                         return 1;
 151                 return 0;
 152         }
 153         return 0;
 154 }
 155
 156 /*
 157  * SFR-CACC algorithm:
 158  * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
 159  * than next_tsn_at_change of the current primary, then
 160  * the sender MUST NOT increment missing report count
 161  * for t.
 162  */
 163 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
 164 {
 165         if (primary->cacc.cycling_changeover &&
 166             TSN_lt(tsn, primary->cacc.next_tsn_at_change))
 167                 return 1;
 168         return 0;
 169 }
 170
 171 /*
 172  * SFR-CACC algorithm:
 173  * 3) If the missing report count for TSN t is to be
 174  * incremented according to [RFC2960] and
 175  * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
 176  * then the sender MUST further execute steps 3.1 and
 177  * 3.2 to determine if the missing report count for
 178  * TSN t SHOULD NOT be incremented.
 179  *
 180  * 3.3) If 3.1 and 3.2 do not dictate that the missing
 181  * report count for t should not be incremented, then
 182  * the sender SHOULD increment missing report count for
 183  * t (according to [RFC2960] and [SCTP_STEWART_2002]).
 184  */
 185 static inline int sctp_cacc_skip(struct sctp_transport *primary,
 186                                  struct sctp_transport *transport,
 187                                  int count_of_newacks,
 188                                  __u32 tsn)
 189 {
 190         if (primary->cacc.changeover_active &&
 191             (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) ||
 192              sctp_cacc_skip_3_2(primary, tsn)))
 193                 return 1;
 194         return 0;
 195 }
 196
 197 /* Initialize an existing sctp_outq.  This does the boring stuff.
 198  * You still need to define handlers if you really want to DO
 199  * something with this structure...
 200  */
 201 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
 202 {
 203         memset(q, 0, sizeof(struct sctp_outq));
 204
 205         q->asoc = asoc;
 206         INIT_LIST_HEAD(&q->out_chunk_list);
 207         INIT_LIST_HEAD(&q->control_chunk_list);
 208         INIT_LIST_HEAD(&q->retransmit);
 209         INIT_LIST_HEAD(&q->sacked);
 210         INIT_LIST_HEAD(&q->abandoned);
 211
 212         q->empty = 1;
 213 }
 214
 215 /* Free the outqueue structure and any related pending chunks.
 216  */
 217 static void __sctp_outq_teardown(struct sctp_outq *q)
 218 {
 219         struct sctp_transport *transport;
 220         struct list_head *lchunk, *temp;
 221         struct sctp_chunk *chunk, *tmp;
 222
 223         /* Throw away unacknowledged chunks. */
 224         list_for_each_entry(transport, &q->asoc->peer.transport_addr_list,
 225                         transports) {
 226                 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
 227                         chunk = list_entry(lchunk, struct sctp_chunk,
 228                                            transmitted_list);
 229                         /* Mark as part of a failed message. */
 230                         sctp_chunk_fail(chunk, q->error);
 231                         sctp_chunk_free(chunk);
 232                 }
 233         }
 234
 235         /* Throw away chunks that have been gap ACKed.  */
 236         list_for_each_safe(lchunk, temp, &q->sacked) {
 237                 list_del_init(lchunk);
 238                 chunk = list_entry(lchunk, struct sctp_chunk,
 239                                    transmitted_list);
 240                 sctp_chunk_fail(chunk, q->error);
 241                 sctp_chunk_free(chunk);
 242         }
 243
 244         /* Throw away any chunks in the retransmit queue. */
 245         list_for_each_safe(lchunk, temp, &q->retransmit) {
 246                 list_del_init(lchunk);
 247                 chunk = list_entry(lchunk, struct sctp_chunk,
 248                                    transmitted_list);
 249                 sctp_chunk_fail(chunk, q->error);
 250                 sctp_chunk_free(chunk);
 251         }
 252
 253         /* Throw away any chunks that are in the abandoned queue. */
 254         list_for_each_safe(lchunk, temp, &q->abandoned) {
 255                 list_del_init(lchunk);
 256                 chunk = list_entry(lchunk, struct sctp_chunk,
 257                                    transmitted_list);
 258                 sctp_chunk_fail(chunk, q->error);
 259                 sctp_chunk_free(chunk);
 260         }
 261
 262         /* Throw away any leftover data chunks. */
 263         while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
 264
 265                 /* Mark as send failure. */
 266                 sctp_chunk_fail(chunk, q->error);
 267                 sctp_chunk_free(chunk);
 268         }
 269
 270         /* Throw away any leftover control chunks. */
 271         list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
 272                 list_del_init(&chunk->list);
 273                 sctp_chunk_free(chunk);
 274         }
 275 }
 276
 277 void sctp_outq_teardown(struct sctp_outq *q)
 278 {
 279         __sctp_outq_teardown(q);
 280         sctp_outq_init(q->asoc, q);
 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
 290 /* Put a new chunk in an sctp_outq.  */
 291 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk)
 292 {
 293         struct net *net = sock_net(q->asoc->base.sk);
 294         int error = 0;
 295
 296         pr_debug("%s: outq:%p, chunk:%p[%s]\n", __func__, q, chunk,
 297                  chunk && chunk->chunk_hdr ?
 298                  sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
 299                  "illegal chunk");
 300
 301         /* If it is data, queue it up, otherwise, send it
 302          * immediately.
 303          */
 304         if (sctp_chunk_is_data(chunk)) {
 305                 /* Is it OK to queue data chunks?  */
 306                 /* From 9. Termination of Association
 307                  *
 308                  * When either endpoint performs a shutdown, the
 309                  * association on each peer will stop accepting new
 310                  * data from its user and only deliver data in queue
 311                  * at the time of sending or receiving the SHUTDOWN
 312                  * chunk.
 313                  */
 314                 switch (q->asoc->state) {
 315                 case SCTP_STATE_CLOSED:
 316                 case SCTP_STATE_SHUTDOWN_PENDING:
 317                 case SCTP_STATE_SHUTDOWN_SENT:
 318                 case SCTP_STATE_SHUTDOWN_RECEIVED:
 319                 case SCTP_STATE_SHUTDOWN_ACK_SENT:
 320                         /* Cannot send after transport endpoint shutdown */
 321                         error = -ESHUTDOWN;
 322                         break;
 323
 324                 default:
 325                         pr_debug("%s: outqueueing: outq:%p, chunk:%p[%s])\n",
 326                                  __func__, q, chunk, chunk && chunk->chunk_hdr ?
 327                                  sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
 328                                  "illegal chunk");
 329
 330                         sctp_outq_tail_data(q, chunk);
 331                         if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
 332                                 SCTP_INC_STATS(net, SCTP_MIB_OUTUNORDERCHUNKS);
 333                         else
 334                                 SCTP_INC_STATS(net, SCTP_MIB_OUTORDERCHUNKS);
 335                         q->empty = 0;
 336                         break;
 337                 }
 338         } else {
 339                 list_add_tail(&chunk->list, &q->control_chunk_list);
 340                 SCTP_INC_STATS(net, SCTP_MIB_OUTCTRLCHUNKS);
 341         }
 342
 343         if (error < 0)
 344                 return error;
 345
 346         if (!q->cork)
 347                 error = sctp_outq_flush(q, 0);
 348
 349         return error;
 350 }
 351
 352 /* Insert a chunk into the sorted list based on the TSNs.  The retransmit list
 353  * and the abandoned list are in ascending order.
 354  */
 355 static void sctp_insert_list(struct list_head *head, struct list_head *new)
 356 {
 357         struct list_head *pos;
 358         struct sctp_chunk *nchunk, *lchunk;
 359         __u32 ntsn, ltsn;
 360         int done = 0;
 361
 362         nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
 363         ntsn = ntohl(nchunk->subh.data_hdr->tsn);
 364
 365         list_for_each(pos, head) {
 366                 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
 367                 ltsn = ntohl(lchunk->subh.data_hdr->tsn);
 368                 if (TSN_lt(ntsn, ltsn)) {
 369                         list_add(new, pos->prev);
 370                         done = 1;
 371                         break;
 372                 }
 373         }
 374         if (!done)
 375                 list_add_tail(new, head);
 376 }
 377
 378 /* Mark all the eligible packets on a transport for retransmission.  */
 379 void sctp_retransmit_mark(struct sctp_outq *q,
 380                           struct sctp_transport *transport,
 381                           __u8 reason)
 382 {
 383         struct list_head *lchunk, *ltemp;
 384         struct sctp_chunk *chunk;
 385
 386         /* Walk through the specified transmitted queue.  */
 387         list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
 388                 chunk = list_entry(lchunk, struct sctp_chunk,
 389                                    transmitted_list);
 390
 391                 /* If the chunk is abandoned, move it to abandoned list. */
 392                 if (sctp_chunk_abandoned(chunk)) {
 393                         list_del_init(lchunk);
 394                         sctp_insert_list(&q->abandoned, lchunk);
 395
 396                         /* If this chunk has not been previousely acked,
 397                          * stop considering it 'outstanding'.  Our peer
 398                          * will most likely never see it since it will
 399                          * not be retransmitted
 400                          */
 401                         if (!chunk->tsn_gap_acked) {
 402                                 if (chunk->transport)
 403                                         chunk->transport->flight_size -=
 404                                                         sctp_data_size(chunk);
 405                                 q->outstanding_bytes -= sctp_data_size(chunk);
 406                                 q->asoc->peer.rwnd += sctp_data_size(chunk);
 407                         }
 408                         continue;
 409                 }
 410
 411                 /* If we are doing  retransmission due to a timeout or pmtu
 412                  * discovery, only the  chunks that are not yet acked should
 413                  * be added to the retransmit queue.
 414                  */
 415                 if ((reason == SCTP_RTXR_FAST_RTX  &&
 416                             (chunk->fast_retransmit == SCTP_NEED_FRTX)) ||
 417                     (reason != SCTP_RTXR_FAST_RTX  && !chunk->tsn_gap_acked)) {
 418                         /* RFC 2960 6.2.1 Processing a Received SACK
 419                          *
 420                          * C) Any time a DATA chunk is marked for
 421                          * retransmission (via either T3-rtx timer expiration
 422                          * (Section 6.3.3) or via fast retransmit
 423                          * (Section 7.2.4)), add the data size of those
 424                          * chunks to the rwnd.
 425                          */
 426                         q->asoc->peer.rwnd += sctp_data_size(chunk);
 427                         q->outstanding_bytes -= sctp_data_size(chunk);
 428                         if (chunk->transport)
 429                                 transport->flight_size -= sctp_data_size(chunk);
 430
 431                         /* sctpimpguide-05 Section 2.8.2
 432                          * M5) If a T3-rtx timer expires, the
 433                          * 'TSN.Missing.Report' of all affected TSNs is set
 434                          * to 0.
 435                          */
 436                         chunk->tsn_missing_report = 0;
 437
 438                         /* If a chunk that is being used for RTT measurement
 439                          * has to be retransmitted, we cannot use this chunk
 440                          * anymore for RTT measurements. Reset rto_pending so
 441                          * that a new RTT measurement is started when a new
 442                          * data chunk is sent.
 443                          */
 444                         if (chunk->rtt_in_progress) {
 445                                 chunk->rtt_in_progress = 0;
 446                                 transport->rto_pending = 0;
 447                         }
 448
 449                         chunk->resent = 1;
 450
 451                         /* Move the chunk to the retransmit queue. The chunks
 452                          * on the retransmit queue are always kept in order.
 453                          */
 454                         list_del_init(lchunk);
 455                         sctp_insert_list(&q->retransmit, lchunk);
 456                 }
 457         }
 458
 459         pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, "
 460                  "flight_size:%d, pba:%d\n", __func__, transport, reason,
 461                  transport->cwnd, transport->ssthresh, transport->flight_size,
 462                  transport->partial_bytes_acked);
 463 }
 464
 465 /* Mark all the eligible packets on a transport for retransmission and force
 466  * one packet out.
 467  */
 468 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
 469                      sctp_retransmit_reason_t reason)
 470 {
 471         struct net *net = sock_net(q->asoc->base.sk);
 472         int error = 0;
 473
 474         switch(reason) {
 475         case SCTP_RTXR_T3_RTX:
 476                 SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS);
 477                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
 478                 /* Update the retran path if the T3-rtx timer has expired for
 479                  * the current retran path.
 480                  */
 481                 if (transport == transport->asoc->peer.retran_path)
 482                         sctp_assoc_update_retran_path(transport->asoc);
 483                 transport->asoc->rtx_data_chunks +=
 484                         transport->asoc->unack_data;
 485                 break;
 486         case SCTP_RTXR_FAST_RTX:
 487                 SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS);
 488                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
 489                 q->fast_rtx = 1;
 490                 break;
 491         case SCTP_RTXR_PMTUD:
 492                 SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS);
 493                 break;
 494         case SCTP_RTXR_T1_RTX:
 495                 SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS);
 496                 transport->asoc->init_retries++;
 497                 break;
 498         default:
 499                 BUG();
 500         }
 501
 502         sctp_retransmit_mark(q, transport, reason);
 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         if (reason == SCTP_RTXR_T3_RTX)
 509                 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
 510
 511         /* Flush the queues only on timeout, since fast_rtx is only
 512          * triggered during sack processing and the queue
 513          * will be flushed at the end.
 514          */
 515         if (reason != SCTP_RTXR_FAST_RTX)
 516                 error = sctp_outq_flush(q, /* rtx_timeout */ 1);
 517
 518         if (error)
 519                 q->asoc->base.sk->sk_err = -error;
 520 }
 521
 522 /*
 523  * Transmit DATA chunks on the retransmit queue.  Upon return from
 524  * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
 525  * need to be transmitted by the caller.
 526  * We assume that pkt->transport has already been set.
 527  *
 528  * The return value is a normal kernel error return value.
 529  */
 530 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
 531                                int rtx_timeout, int *start_timer)
 532 {
 533         struct list_head *lqueue;
 534         struct sctp_transport *transport = pkt->transport;
 535         sctp_xmit_t status;
 536         struct sctp_chunk *chunk, *chunk1;
 537         int fast_rtx;
 538         int error = 0;
 539         int timer = 0;
 540         int done = 0;
 541
 542         lqueue = &q->retransmit;
 543         fast_rtx = q->fast_rtx;
 544
 545         /* This loop handles time-out retransmissions, fast retransmissions,
 546          * and retransmissions due to opening of whindow.
 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          * For fast retransmissions we also send only ONE packet.  However,
 565          * if we are just flushing the queue due to open window, we'll
 566          * try to send as much as possible.
 567          */
 568         list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
 569                 /* If the chunk is abandoned, move it to abandoned list. */
 570                 if (sctp_chunk_abandoned(chunk)) {
 571                         list_del_init(&chunk->transmitted_list);
 572                         sctp_insert_list(&q->abandoned,
 573                                          &chunk->transmitted_list);
 574                         continue;
 575                 }
 576
 577                 /* Make sure that Gap Acked TSNs are not retransmitted.  A
 578                  * simple approach is just to move such TSNs out of the
 579                  * way and into a 'transmitted' queue and skip to the
 580                  * next chunk.
 581                  */
 582                 if (chunk->tsn_gap_acked) {
 583                         list_move_tail(&chunk->transmitted_list,
 584                                        &transport->transmitted);
 585                         continue;
 586                 }
 587
 588                 /* If we are doing fast retransmit, ignore non-fast_rtransmit
 589                  * chunks
 590                  */
 591                 if (fast_rtx && !chunk->fast_retransmit)
 592                         continue;
 593
 594 redo:
 595                 /* Attempt to append this chunk to the packet. */
 596                 status = sctp_packet_append_chunk(pkt, chunk);
 597
 598                 switch (status) {
 599                 case SCTP_XMIT_PMTU_FULL:
 600                         if (!pkt->has_data && !pkt->has_cookie_echo) {
 601                                 /* If this packet did not contain DATA then
 602                                  * retransmission did not happen, so do it
 603                                  * again.  We'll ignore the error here since
 604                                  * control chunks are already freed so there
 605                                  * is nothing we can do.
 606                                  */
 607                                 sctp_packet_transmit(pkt);
 608                                 goto redo;
 609                         }
 610
 611                         /* Send this packet.  */
 612                         error = sctp_packet_transmit(pkt);
 613
 614                         /* If we are retransmitting, we should only
 615                          * send a single packet.
 616                          * Otherwise, try appending this chunk again.
 617                          */
 618                         if (rtx_timeout || fast_rtx)
 619                                 done = 1;
 620                         else
 621                                 goto redo;
 622
 623                         /* Bundle next chunk in the next round.  */
 624                         break;
 625
 626                 case SCTP_XMIT_RWND_FULL:
 627                         /* Send this packet. */
 628                         error = sctp_packet_transmit(pkt);
 629
 630                         /* Stop sending DATA as there is no more room
 631                          * at the receiver.
 632                          */
 633                         done = 1;
 634                         break;
 635
 636                 case SCTP_XMIT_NAGLE_DELAY:
 637                         /* Send this packet. */
 638                         error = sctp_packet_transmit(pkt);
 639
 640                         /* Stop sending DATA because of nagle delay. */
 641                         done = 1;
 642                         break;
 643
 644                 default:
 645                         /* The append was successful, so add this chunk to
 646                          * the transmitted list.
 647                          */
 648                         list_move_tail(&chunk->transmitted_list,
 649                                        &transport->transmitted);
 650
 651                         /* Mark the chunk as ineligible for fast retransmit
 652                          * after it is retransmitted.
 653                          */
 654                         if (chunk->fast_retransmit == SCTP_NEED_FRTX)
 655                                 chunk->fast_retransmit = SCTP_DONT_FRTX;
 656
 657                         q->empty = 0;
 658                         q->asoc->stats.rtxchunks++;
 659                         break;
 660                 }
 661
 662                 /* Set the timer if there were no errors */
 663                 if (!error && !timer)
 664                         timer = 1;
 665
 666                 if (done)
 667                         break;
 668         }
 669
 670         /* If we are here due to a retransmit timeout or a fast
 671          * retransmit and if there are any chunks left in the retransmit
 672          * queue that could not fit in the PMTU sized packet, they need
 673          * to be marked as ineligible for a subsequent fast retransmit.
 674          */
 675         if (rtx_timeout || fast_rtx) {
 676                 list_for_each_entry(chunk1, lqueue, transmitted_list) {
 677                         if (chunk1->fast_retransmit == SCTP_NEED_FRTX)
 678                                 chunk1->fast_retransmit = SCTP_DONT_FRTX;
 679                 }
 680         }
 681
 682         *start_timer = timer;
 683
 684         /* Clear fast retransmit hint */
 685         if (fast_rtx)
 686                 q->fast_rtx = 0;
 687
 688         return error;
 689 }
 690
 691 /* Cork the outqueue so queued chunks are really queued. */
 692 int sctp_outq_uncork(struct sctp_outq *q)
 693 {
 694         if (q->cork)
 695                 q->cork = 0;
 696
 697         return sctp_outq_flush(q, 0);
 698 }
 699
 700
 701 /*
 702  * Try to flush an outqueue.
 703  *
 704  * Description: Send everything in q which we legally can, subject to
 705  * congestion limitations.
 706  * * Note: This function can be called from multiple contexts so appropriate
 707  * locking concerns must be made.  Today we use the sock lock to protect
 708  * this function.
 709  */
 710 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
 711 {
 712         struct sctp_packet *packet;
 713         struct sctp_packet singleton;
 714         struct sctp_association *asoc = q->asoc;
 715         __u16 sport = asoc->base.bind_addr.port;
 716         __u16 dport = asoc->peer.port;
 717         __u32 vtag = asoc->peer.i.init_tag;
 718         struct sctp_transport *transport = NULL;
 719         struct sctp_transport *new_transport;
 720         struct sctp_chunk *chunk, *tmp;
 721         sctp_xmit_t status;
 722         int error = 0;
 723         int start_timer = 0;
 724         int one_packet = 0;
 725
 726         /* These transports have chunks to send. */
 727         struct list_head transport_list;
 728         struct list_head *ltransport;
 729
 730         INIT_LIST_HEAD(&transport_list);
 731         packet = NULL;
 732
 733         /*
 734          * 6.10 Bundling
 735          *   ...
 736          *   When bundling control chunks with DATA chunks, an
 737          *   endpoint MUST place control chunks first in the outbound
 738          *   SCTP packet.  The transmitter MUST transmit DATA chunks
 739          *   within a SCTP packet in increasing order of TSN.
 740          *   ...
 741          */
 742
 743         list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
 744                 /* RFC 5061, 5.3
 745                  * F1) This means that until such time as the ASCONF
 746                  * containing the add is acknowledged, the sender MUST
 747                  * NOT use the new IP address as a source for ANY SCTP
 748                  * packet except on carrying an ASCONF Chunk.
 749                  */
 750                 if (asoc->src_out_of_asoc_ok &&
 751                     chunk->chunk_hdr->type != SCTP_CID_ASCONF)
 752                         continue;
 753
 754                 list_del_init(&chunk->list);
 755
 756                 /* Pick the right transport to use. */
 757                 new_transport = chunk->transport;
 758
 759                 if (!new_transport) {
 760                         /*
 761                          * If we have a prior transport pointer, see if
 762                          * the destination address of the chunk
 763                          * matches the destination address of the
 764                          * current transport.  If not a match, then
 765                          * try to look up the transport with a given
 766                          * destination address.  We do this because
 767                          * after processing ASCONFs, we may have new
 768                          * transports created.
 769                          */
 770                         if (transport &&
 771                             sctp_cmp_addr_exact(&chunk->dest,
 772                                                 &transport->ipaddr))
 773                                         new_transport = transport;
 774                         else
 775                                 new_transport = sctp_assoc_lookup_paddr(asoc,
 776                                                                 &chunk->dest);
 777
 778                         /* if we still don't have a new transport, then
 779                          * use the current active path.
 780                          */
 781                         if (!new_transport)
 782                                 new_transport = asoc->peer.active_path;
 783                 } else if ((new_transport->state == SCTP_INACTIVE) ||
 784                            (new_transport->state == SCTP_UNCONFIRMED) ||
 785                            (new_transport->state == SCTP_PF)) {
 786                         /* If the chunk is Heartbeat or Heartbeat Ack,
 787                          * send it to chunk->transport, even if it's
 788                          * inactive.
 789                          *
 790                          * 3.3.6 Heartbeat Acknowledgement:
 791                          * ...
 792                          * A HEARTBEAT ACK is always sent to the source IP
 793                          * address of the IP datagram containing the
 794                          * HEARTBEAT chunk to which this ack is responding.
 795                          * ...
 796                          *
 797                          * ASCONF_ACKs also must be sent to the source.
 798                          */
 799                         if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
 800                             chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK &&
 801                             chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK)
 802                                 new_transport = asoc->peer.active_path;
 803                 }
 804
 805                 /* Are we switching transports?
 806                  * Take care of transport locks.
 807                  */
 808                 if (new_transport != transport) {
 809                         transport = new_transport;
 810                         if (list_empty(&transport->send_ready)) {
 811                                 list_add_tail(&transport->send_ready,
 812                                               &transport_list);
 813                         }
 814                         packet = &transport->packet;
 815                         sctp_packet_config(packet, vtag,
 816                                            asoc->peer.ecn_capable);
 817                 }
 818
 819                 switch (chunk->chunk_hdr->type) {
 820                 /*
 821                  * 6.10 Bundling
 822                  *   ...
 823                  *   An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
 824                  *   COMPLETE with any other chunks.  [Send them immediately.]
 825                  */
 826                 case SCTP_CID_INIT:
 827                 case SCTP_CID_INIT_ACK:
 828                 case SCTP_CID_SHUTDOWN_COMPLETE:
 829                         sctp_packet_init(&singleton, transport, sport, dport);
 830                         sctp_packet_config(&singleton, vtag, 0);
 831                         sctp_packet_append_chunk(&singleton, chunk);
 832                         error = sctp_packet_transmit(&singleton);
 833                         if (error < 0)
 834                                 return error;
 835                         break;
 836
 837                 case SCTP_CID_ABORT:
 838                         if (sctp_test_T_bit(chunk)) {
 839                                 packet->vtag = asoc->c.my_vtag;
 840                         }
 841                 /* The following chunks are "response" chunks, i.e.
 842                  * they are generated in response to something we
 843                  * received.  If we are sending these, then we can
 844                  * send only 1 packet containing these chunks.
 845                  */
 846                 case SCTP_CID_HEARTBEAT_ACK:
 847                 case SCTP_CID_SHUTDOWN_ACK:
 848                 case SCTP_CID_COOKIE_ACK:
 849                 case SCTP_CID_COOKIE_ECHO:
 850                 case SCTP_CID_ERROR:
 851                 case SCTP_CID_ECN_CWR:
 852                 case SCTP_CID_ASCONF_ACK:
 853                         one_packet = 1;
 854                         /* Fall through */
 855
 856                 case SCTP_CID_SACK:
 857                 case SCTP_CID_HEARTBEAT:
 858                 case SCTP_CID_SHUTDOWN:
 859                 case SCTP_CID_ECN_ECNE:
 860                 case SCTP_CID_ASCONF:
 861                 case SCTP_CID_FWD_TSN:
 862                         status = sctp_packet_transmit_chunk(packet, chunk,
 863                                                             one_packet);
 864                         if (status  != SCTP_XMIT_OK) {
 865                                 /* put the chunk back */
 866                                 list_add(&chunk->list, &q->control_chunk_list);
 867                         } else {
 868                                 asoc->stats.octrlchunks++;
 869                                 /* PR-SCTP C5) If a FORWARD TSN is sent, the
 870                                  * sender MUST assure that at least one T3-rtx
 871                                  * timer is running.
 872                                  */
 873                                 if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN)
 874                                         sctp_transport_reset_timers(transport);
 875                         }
 876                         break;
 877
 878                 default:
 879                         /* We built a chunk with an illegal type! */
 880                         BUG();
 881                 }
 882         }
 883
 884         if (q->asoc->src_out_of_asoc_ok)
 885                 goto sctp_flush_out;
 886
 887         /* Is it OK to send data chunks?  */
 888         switch (asoc->state) {
 889         case SCTP_STATE_COOKIE_ECHOED:
 890                 /* Only allow bundling when this packet has a COOKIE-ECHO
 891                  * chunk.
 892                  */
 893                 if (!packet || !packet->has_cookie_echo)
 894                         break;
 895
 896                 /* fallthru */
 897         case SCTP_STATE_ESTABLISHED:
 898         case SCTP_STATE_SHUTDOWN_PENDING:
 899         case SCTP_STATE_SHUTDOWN_RECEIVED:
 900                 /*
 901                  * RFC 2960 6.1  Transmission of DATA Chunks
 902                  *
 903                  * C) When the time comes for the sender to transmit,
 904                  * before sending new DATA chunks, the sender MUST
 905                  * first transmit any outstanding DATA chunks which
 906                  * are marked for retransmission (limited by the
 907                  * current cwnd).
 908                  */
 909                 if (!list_empty(&q->retransmit)) {
 910                         if (asoc->peer.retran_path->state == SCTP_UNCONFIRMED)
 911                                 goto sctp_flush_out;
 912                         if (transport == asoc->peer.retran_path)
 913                                 goto retran;
 914
 915                         /* Switch transports & prepare the packet.  */
 916
 917                         transport = asoc->peer.retran_path;
 918
 919                         if (list_empty(&transport->send_ready)) {
 920                                 list_add_tail(&transport->send_ready,
 921                                               &transport_list);
 922                         }
 923
 924                         packet = &transport->packet;
 925                         sctp_packet_config(packet, vtag,
 926                                            asoc->peer.ecn_capable);
 927                 retran:
 928                         error = sctp_outq_flush_rtx(q, packet,
 929                                                     rtx_timeout, &start_timer);
 930
 931                         if (start_timer)
 932                                 sctp_transport_reset_timers(transport);
 933
 934                         /* This can happen on COOKIE-ECHO resend.  Only
 935                          * one chunk can get bundled with a COOKIE-ECHO.
 936                          */
 937                         if (packet->has_cookie_echo)
 938                                 goto sctp_flush_out;
 939
 940                         /* Don't send new data if there is still data
 941                          * waiting to retransmit.
 942                          */
 943                         if (!list_empty(&q->retransmit))
 944                                 goto sctp_flush_out;
 945                 }
 946
 947                 /* Apply Max.Burst limitation to the current transport in
 948                  * case it will be used for new data.  We are going to
 949                  * rest it before we return, but we want to apply the limit
 950                  * to the currently queued data.
 951                  */
 952                 if (transport)
 953                         sctp_transport_burst_limited(transport);
 954
 955                 /* Finally, transmit new packets.  */
 956                 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
 957                         /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
 958                          * stream identifier.
 959                          */
 960                         if (chunk->sinfo.sinfo_stream >=
 961                             asoc->c.sinit_num_ostreams) {
 962
 963                                 /* Mark as failed send. */
 964                                 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
 965                                 sctp_chunk_free(chunk);
 966                                 continue;
 967                         }
 968
 969                         /* Has this chunk expired? */
 970                         if (sctp_chunk_abandoned(chunk)) {
 971                                 sctp_chunk_fail(chunk, 0);
 972                                 sctp_chunk_free(chunk);
 973                                 continue;
 974                         }
 975
 976                         /* If there is a specified transport, use it.
 977                          * Otherwise, we want to use the active path.
 978                          */
 979                         new_transport = chunk->transport;
 980                         if (!new_transport ||
 981                             ((new_transport->state == SCTP_INACTIVE) ||
 982                              (new_transport->state == SCTP_UNCONFIRMED) ||
 983                              (new_transport->state == SCTP_PF)))
 984                                 new_transport = asoc->peer.active_path;
 985                         if (new_transport->state == SCTP_UNCONFIRMED)
 986                                 continue;
 987
 988                         /* Change packets if necessary.  */
 989                         if (new_transport != transport) {
 990                                 transport = new_transport;
 991
 992                                 /* Schedule to have this transport's
 993                                  * packet flushed.
 994                                  */
 995                                 if (list_empty(&transport->send_ready)) {
 996                                         list_add_tail(&transport->send_ready,
 997                                                       &transport_list);
 998                                 }
 999
1000                                 packet = &transport->packet;
1001                                 sctp_packet_config(packet, vtag,
1002                                                    asoc->peer.ecn_capable);
1003                                 /* We've switched transports, so apply the
1004                                  * Burst limit to the new transport.
1005                                  */
1006                                 sctp_transport_burst_limited(transport);
1007                         }
1008
1009                         pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p "
1010                                  "skb->users:%d\n",
1011                                  __func__, q, chunk, chunk && chunk->chunk_hdr ?
1012                                  sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
1013                                  "illegal chunk", ntohl(chunk->subh.data_hdr->tsn),
1014                                  chunk->skb ? chunk->skb->head : NULL, chunk->skb ?
1015                                  atomic_read(&chunk->skb->users) : -1);
1016
1017                         /* Add the chunk to the packet.  */
1018                         status = sctp_packet_transmit_chunk(packet, chunk, 0);
1019
1020                         switch (status) {
1021                         case SCTP_XMIT_PMTU_FULL:
1022                         case SCTP_XMIT_RWND_FULL:
1023                         case SCTP_XMIT_NAGLE_DELAY:
1024                                 /* We could not append this chunk, so put
1025                                  * the chunk back on the output queue.
1026                                  */
1027                                 pr_debug("%s: could not transmit tsn:0x%x, status:%d\n",
1028                                          __func__, ntohl(chunk->subh.data_hdr->tsn),
1029                                          status);
1030
1031                                 sctp_outq_head_data(q, chunk);
1032                                 goto sctp_flush_out;
1033                                 break;
1034
1035                         case SCTP_XMIT_OK:
1036                                 /* The sender is in the SHUTDOWN-PENDING state,
1037                                  * The sender MAY set the I-bit in the DATA
1038                                  * chunk header.
1039                                  */
1040                                 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING)
1041                                         chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM;
1042                                 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
1043                                         asoc->stats.ouodchunks++;
1044                                 else
1045                                         asoc->stats.oodchunks++;
1046
1047                                 break;
1048
1049                         default:
1050                                 BUG();
1051                         }
1052
1053                         /* BUG: We assume that the sctp_packet_transmit()
1054                          * call below will succeed all the time and add the
1055                          * chunk to the transmitted list and restart the
1056                          * timers.
1057                          * It is possible that the call can fail under OOM
1058                          * conditions.
1059                          *
1060                          * Is this really a problem?  Won't this behave
1061                          * like a lost TSN?
1062                          */
1063                         list_add_tail(&chunk->transmitted_list,
1064                                       &transport->transmitted);
1065
1066                         sctp_transport_reset_timers(transport);
1067
1068                         q->empty = 0;
1069
1070                         /* Only let one DATA chunk get bundled with a
1071                          * COOKIE-ECHO chunk.
1072                          */
1073                         if (packet->has_cookie_echo)
1074                                 goto sctp_flush_out;
1075                 }
1076                 break;
1077
1078         default:
1079                 /* Do nothing.  */
1080                 break;
1081         }
1082
1083 sctp_flush_out:
1084
1085         /* Before returning, examine all the transports touched in
1086          * this call.  Right now, we bluntly force clear all the
1087          * transports.  Things might change after we implement Nagle.
1088          * But such an examination is still required.
1089          *
1090          * --xguo
1091          */
1092         while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
1093                 struct sctp_transport *t = list_entry(ltransport,
1094                                                       struct sctp_transport,
1095                                                       send_ready);
1096                 packet = &t->packet;
1097                 if (!sctp_packet_empty(packet))
1098                         error = sctp_packet_transmit(packet);
1099
1100                 /* Clear the burst limited state, if any */
1101                 sctp_transport_burst_reset(t);
1102         }
1103
1104         return error;
1105 }
1106
1107 /* Update unack_data based on the incoming SACK chunk */
1108 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1109                                         struct sctp_sackhdr *sack)
1110 {
1111         sctp_sack_variable_t *frags;
1112         __u16 unack_data;
1113         int i;
1114
1115         unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1116
1117         frags = sack->variable;
1118         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1119                 unack_data -= ((ntohs(frags[i].gab.end) -
1120                                 ntohs(frags[i].gab.start) + 1));
1121         }
1122
1123         assoc->unack_data = unack_data;
1124 }
1125
1126 /* This is where we REALLY process a SACK.
1127  *
1128  * Process the SACK against the outqueue.  Mostly, this just frees
1129  * things off the transmitted queue.
1130  */
1131 int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk)
1132 {
1133         struct sctp_association *asoc = q->asoc;
1134         struct sctp_sackhdr *sack = chunk->subh.sack_hdr;
1135         struct sctp_transport *transport;
1136         struct sctp_chunk *tchunk = NULL;
1137         struct list_head *lchunk, *transport_list, *temp;
1138         sctp_sack_variable_t *frags = sack->variable;
1139         __u32 sack_ctsn, ctsn, tsn;
1140         __u32 highest_tsn, highest_new_tsn;
1141         __u32 sack_a_rwnd;
1142         unsigned int outstanding;
1143         struct sctp_transport *primary = asoc->peer.primary_path;
1144         int count_of_newacks = 0;
1145         int gap_ack_blocks;
1146         u8 accum_moved = 0;
1147
1148         /* Grab the association's destination address list. */
1149         transport_list = &asoc->peer.transport_addr_list;
1150
1151         sack_ctsn = ntohl(sack->cum_tsn_ack);
1152         gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1153         asoc->stats.gapcnt += gap_ack_blocks;
1154         /*
1155          * SFR-CACC algorithm:
1156          * On receipt of a SACK the sender SHOULD execute the
1157          * following statements.
1158          *
1159          * 1) If the cumulative ack in the SACK passes next tsn_at_change
1160          * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1161          * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1162          * all destinations.
1163          * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1164          * is set the receiver of the SACK MUST take the following actions:
1165          *
1166          * A) Initialize the cacc_saw_newack to 0 for all destination
1167          * addresses.
1168          *
1169          * Only bother if changeover_active is set. Otherwise, this is
1170          * totally suboptimal to do on every SACK.
1171          */
1172         if (primary->cacc.changeover_active) {
1173                 u8 clear_cycling = 0;
1174
1175                 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1176                         primary->cacc.changeover_active = 0;
1177                         clear_cycling = 1;
1178                 }
1179
1180                 if (clear_cycling || gap_ack_blocks) {
1181                         list_for_each_entry(transport, transport_list,
1182                                         transports) {
1183                                 if (clear_cycling)
1184                                         transport->cacc.cycling_changeover = 0;
1185                                 if (gap_ack_blocks)
1186                                         transport->cacc.cacc_saw_newack = 0;
1187                         }
1188                 }
1189         }
1190
1191         /* Get the highest TSN in the sack. */
1192         highest_tsn = sack_ctsn;
1193         if (gap_ack_blocks)
1194                 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1195
1196         if (TSN_lt(asoc->highest_sacked, highest_tsn))
1197                 asoc->highest_sacked = highest_tsn;
1198
1199         highest_new_tsn = sack_ctsn;
1200
1201         /* Run through the retransmit queue.  Credit bytes received
1202          * and free those chunks that we can.
1203          */
1204         sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn);
1205
1206         /* Run through the transmitted queue.
1207          * Credit bytes received and free those chunks which we can.
1208          *
1209          * This is a MASSIVE candidate for optimization.
1210          */
1211         list_for_each_entry(transport, transport_list, transports) {
1212                 sctp_check_transmitted(q, &transport->transmitted,
1213                                        transport, &chunk->source, sack,
1214                                        &highest_new_tsn);
1215                 /*
1216                  * SFR-CACC algorithm:
1217                  * C) Let count_of_newacks be the number of
1218                  * destinations for which cacc_saw_newack is set.
1219                  */
1220                 if (transport->cacc.cacc_saw_newack)
1221                         count_of_newacks ++;
1222         }
1223
1224         /* Move the Cumulative TSN Ack Point if appropriate.  */
1225         if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) {
1226                 asoc->ctsn_ack_point = sack_ctsn;
1227                 accum_moved = 1;
1228         }
1229
1230         if (gap_ack_blocks) {
1231
1232                 if (asoc->fast_recovery && accum_moved)
1233                         highest_new_tsn = highest_tsn;
1234
1235                 list_for_each_entry(transport, transport_list, transports)
1236                         sctp_mark_missing(q, &transport->transmitted, transport,
1237                                           highest_new_tsn, count_of_newacks);
1238         }
1239
1240         /* Update unack_data field in the assoc. */
1241         sctp_sack_update_unack_data(asoc, sack);
1242
1243         ctsn = asoc->ctsn_ack_point;
1244
1245         /* Throw away stuff rotting on the sack queue.  */
1246         list_for_each_safe(lchunk, temp, &q->sacked) {
1247                 tchunk = list_entry(lchunk, struct sctp_chunk,
1248                                     transmitted_list);
1249                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1250                 if (TSN_lte(tsn, ctsn)) {
1251                         list_del_init(&tchunk->transmitted_list);
1252                         sctp_chunk_free(tchunk);
1253                 }
1254         }
1255
1256         /* ii) Set rwnd equal to the newly received a_rwnd minus the
1257          *     number of bytes still outstanding after processing the
1258          *     Cumulative TSN Ack and the Gap Ack Blocks.
1259          */
1260
1261         sack_a_rwnd = ntohl(sack->a_rwnd);
1262         outstanding = q->outstanding_bytes;
1263
1264         if (outstanding < sack_a_rwnd)
1265                 sack_a_rwnd -= outstanding;
1266         else
1267                 sack_a_rwnd = 0;
1268
1269         asoc->peer.rwnd = sack_a_rwnd;
1270
1271         sctp_generate_fwdtsn(q, sack_ctsn);
1272
1273         pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__, sack_ctsn);
1274         pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, "
1275                  "advertised peer ack point:0x%x\n", __func__, asoc, ctsn,
1276                  asoc->adv_peer_ack_point);
1277
1278         /* See if all chunks are acked.
1279          * Make sure the empty queue handler will get run later.
1280          */
1281         q->empty = (list_empty(&q->out_chunk_list) &&
1282                     list_empty(&q->retransmit));
1283         if (!q->empty)
1284                 goto finish;
1285
1286         list_for_each_entry(transport, transport_list, transports) {
1287                 q->empty = q->empty && list_empty(&transport->transmitted);
1288                 if (!q->empty)
1289                         goto finish;
1290         }
1291
1292         pr_debug("%s: sack queue is empty\n", __func__);
1293 finish:
1294         return q->empty;
1295 }
1296
1297 /* Is the outqueue empty?  */
1298 int sctp_outq_is_empty(const struct sctp_outq *q)
1299 {
1300         return q->empty;
1301 }
1302
1303 /********************************************************************
1304  * 2nd Level Abstractions
1305  ********************************************************************/
1306
1307 /* Go through a transport's transmitted list or the association's retransmit
1308  * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1309  * The retransmit list will not have an associated transport.
1310  *
1311  * I added coherent debug information output.   --xguo
1312  *
1313  * Instead of printing 'sacked' or 'kept' for each TSN on the
1314  * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1315  * KEPT TSN6-TSN7, etc.
1316  */
1317 static void sctp_check_transmitted(struct sctp_outq *q,
1318                                    struct list_head *transmitted_queue,
1319                                    struct sctp_transport *transport,
1320                                    union sctp_addr *saddr,
1321                                    struct sctp_sackhdr *sack,
1322                                    __u32 *highest_new_tsn_in_sack)
1323 {
1324         struct list_head *lchunk;
1325         struct sctp_chunk *tchunk;
1326         struct list_head tlist;
1327         __u32 tsn;
1328         __u32 sack_ctsn;
1329         __u32 rtt;
1330         __u8 restart_timer = 0;
1331         int bytes_acked = 0;
1332         int migrate_bytes = 0;
1333         bool forward_progress = false;
1334
1335         sack_ctsn = ntohl(sack->cum_tsn_ack);
1336
1337         INIT_LIST_HEAD(&tlist);
1338
1339         /* The while loop will skip empty transmitted queues. */
1340         while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1341                 tchunk = list_entry(lchunk, struct sctp_chunk,
1342                                     transmitted_list);
1343
1344                 if (sctp_chunk_abandoned(tchunk)) {
1345                         /* Move the chunk to abandoned list. */
1346                         sctp_insert_list(&q->abandoned, lchunk);
1347
1348                         /* If this chunk has not been acked, stop
1349                          * considering it as 'outstanding'.
1350                          */
1351                         if (!tchunk->tsn_gap_acked) {
1352                                 if (tchunk->transport)
1353                                         tchunk->transport->flight_size -=
1354                                                         sctp_data_size(tchunk);
1355                                 q->outstanding_bytes -= sctp_data_size(tchunk);
1356                         }
1357                         continue;
1358                 }
1359
1360                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1361                 if (sctp_acked(sack, tsn)) {
1362                         /* If this queue is the retransmit queue, the
1363                          * retransmit timer has already reclaimed
1364                          * the outstanding bytes for this chunk, so only
1365                          * count bytes associated with a transport.
1366                          */
1367                         if (transport) {
1368                                 /* If this chunk is being used for RTT
1369                                  * measurement, calculate the RTT and update
1370                                  * the RTO using this value.
1371                                  *
1372                                  * 6.3.1 C5) Karn's algorithm: RTT measurements
1373                                  * MUST NOT be made using packets that were
1374                                  * retransmitted (and thus for which it is
1375                                  * ambiguous whether the reply was for the
1376                                  * first instance of the packet or a later
1377                                  * instance).
1378                                  */
1379                                 if (!tchunk->tsn_gap_acked &&
1380                                     !tchunk->resent &&
1381                                     tchunk->rtt_in_progress) {
1382                                         tchunk->rtt_in_progress = 0;
1383                                         rtt = jiffies - tchunk->sent_at;
1384                                         sctp_transport_update_rto(transport,
1385                                                                   rtt);
1386                                 }
1387                         }
1388
1389                         /* If the chunk hasn't been marked as ACKED,
1390                          * mark it and account bytes_acked if the
1391                          * chunk had a valid transport (it will not
1392                          * have a transport if ASCONF had deleted it
1393                          * while DATA was outstanding).
1394                          */
1395                         if (!tchunk->tsn_gap_acked) {
1396                                 tchunk->tsn_gap_acked = 1;
1397                                 if (TSN_lt(*highest_new_tsn_in_sack, tsn))
1398                                         *highest_new_tsn_in_sack = tsn;
1399                                 bytes_acked += sctp_data_size(tchunk);
1400                                 if (!tchunk->transport)
1401                                         migrate_bytes += sctp_data_size(tchunk);
1402                                 forward_progress = true;
1403                         }
1404
1405                         if (TSN_lte(tsn, sack_ctsn)) {
1406                                 /* RFC 2960  6.3.2 Retransmission Timer Rules
1407                                  *
1408                                  * R3) Whenever a SACK is received
1409                                  * that acknowledges the DATA chunk
1410                                  * with the earliest outstanding TSN
1411                                  * for that address, restart T3-rtx
1412                                  * timer for that address with its
1413                                  * current RTO.
1414                                  */
1415                                 restart_timer = 1;
1416                                 forward_progress = true;
1417
1418                                 if (!tchunk->tsn_gap_acked) {
1419                                         /*
1420                                          * SFR-CACC algorithm:
1421                                          * 2) If the SACK contains gap acks
1422                                          * and the flag CHANGEOVER_ACTIVE is
1423                                          * set the receiver of the SACK MUST
1424                                          * take the following action:
1425                                          *
1426                                          * B) For each TSN t being acked that
1427                                          * has not been acked in any SACK so
1428                                          * far, set cacc_saw_newack to 1 for
1429                                          * the destination that the TSN was
1430                                          * sent to.
1431                                          */
1432                                         if (transport &&
1433                                             sack->num_gap_ack_blocks &&
1434                                             q->asoc->peer.primary_path->cacc.
1435                                             changeover_active)
1436                                                 transport->cacc.cacc_saw_newack
1437                                                         = 1;
1438                                 }
1439
1440                                 list_add_tail(&tchunk->transmitted_list,
1441                                               &q->sacked);
1442                         } else {
1443                                 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1444                                  * M2) Each time a SACK arrives reporting
1445                                  * 'Stray DATA chunk(s)' record the highest TSN
1446                                  * reported as newly acknowledged, call this
1447                                  * value 'HighestTSNinSack'. A newly
1448                                  * acknowledged DATA chunk is one not
1449                                  * previously acknowledged in a SACK.
1450                                  *
1451                                  * When the SCTP sender of data receives a SACK
1452                                  * chunk that acknowledges, for the first time,
1453                                  * the receipt of a DATA chunk, all the still
1454                                  * unacknowledged DATA chunks whose TSN is
1455                                  * older than that newly acknowledged DATA
1456                                  * chunk, are qualified as 'Stray DATA chunks'.
1457                                  */
1458                                 list_add_tail(lchunk, &tlist);
1459                         }
1460                 } else {
1461                         if (tchunk->tsn_gap_acked) {
1462                                 pr_debug("%s: receiver reneged on data TSN:0x%x\n",
1463                                          __func__, tsn);
1464
1465                                 tchunk->tsn_gap_acked = 0;
1466
1467                                 if (tchunk->transport)
1468                                         bytes_acked -= sctp_data_size(tchunk);
1469
1470                                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1471                                  *
1472                                  * R4) Whenever a SACK is received missing a
1473                                  * TSN that was previously acknowledged via a
1474                                  * Gap Ack Block, start T3-rtx for the
1475                                  * destination address to which the DATA
1476                                  * chunk was originally
1477                                  * transmitted if it is not already running.
1478                                  */
1479                                 restart_timer = 1;
1480                         }
1481
1482                         list_add_tail(lchunk, &tlist);
1483                 }
1484         }
1485
1486         if (transport) {
1487                 if (bytes_acked) {
1488                         struct sctp_association *asoc = transport->asoc;
1489
1490                         /* We may have counted DATA that was migrated
1491                          * to this transport due to DEL-IP operation.
1492                          * Subtract those bytes, since the were never
1493                          * send on this transport and shouldn't be
1494                          * credited to this transport.
1495                          */
1496                         bytes_acked -= migrate_bytes;
1497
1498                         /* 8.2. When an outstanding TSN is acknowledged,
1499                          * the endpoint shall clear the error counter of
1500                          * the destination transport address to which the
1501                          * DATA chunk was last sent.
1502                          * The association's overall error counter is
1503                          * also cleared.
1504                          */
1505                         transport->error_count = 0;
1506                         transport->asoc->overall_error_count = 0;
1507                         forward_progress = true;
1508
1509                         /*
1510                          * While in SHUTDOWN PENDING, we may have started
1511                          * the T5 shutdown guard timer after reaching the
1512                          * retransmission limit. Stop that timer as soon
1513                          * as the receiver acknowledged any data.
1514                          */
1515                         if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING &&
1516                             del_timer(&asoc->timers
1517                                 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]))
1518                                         sctp_association_put(asoc);
1519
1520                         /* Mark the destination transport address as
1521                          * active if it is not so marked.
1522                          */
1523                         if ((transport->state == SCTP_INACTIVE ||
1524                              transport->state == SCTP_UNCONFIRMED) &&
1525                             sctp_cmp_addr_exact(&transport->ipaddr, saddr)) {
1526                                 sctp_assoc_control_transport(
1527                                         transport->asoc,
1528                                         transport,
1529                                         SCTP_TRANSPORT_UP,
1530                                         SCTP_RECEIVED_SACK);
1531                         }
1532
1533                         sctp_transport_raise_cwnd(transport, sack_ctsn,
1534                                                   bytes_acked);
1535
1536                         transport->flight_size -= bytes_acked;
1537                         if (transport->flight_size == 0)
1538                                 transport->partial_bytes_acked = 0;
1539                         q->outstanding_bytes -= bytes_acked + migrate_bytes;
1540                 } else {
1541                         /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1542                          * When a sender is doing zero window probing, it
1543                          * should not timeout the association if it continues
1544                          * to receive new packets from the receiver. The
1545                          * reason is that the receiver MAY keep its window
1546                          * closed for an indefinite time.
1547                          * A sender is doing zero window probing when the
1548                          * receiver's advertised window is zero, and there is
1549                          * only one data chunk in flight to the receiver.
1550                          *
1551                          * Allow the association to timeout while in SHUTDOWN
1552                          * PENDING or SHUTDOWN RECEIVED in case the receiver
1553                          * stays in zero window mode forever.
1554                          */
1555                         if (!q->asoc->peer.rwnd &&
1556                             !list_empty(&tlist) &&
1557                             (sack_ctsn+2 == q->asoc->next_tsn) &&
1558                             q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) {
1559                                 pr_debug("%s: sack received for zero window "
1560                                          "probe:%u\n", __func__, sack_ctsn);
1561
1562                                 q->asoc->overall_error_count = 0;
1563                                 transport->error_count = 0;
1564                         }
1565                 }
1566
1567                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1568                  *
1569                  * R2) Whenever all outstanding data sent to an address have
1570                  * been acknowledged, turn off the T3-rtx timer of that
1571                  * address.
1572                  */
1573                 if (!transport->flight_size) {
1574                         if (del_timer(&transport->T3_rtx_timer))
1575                                 sctp_transport_put(transport);
1576                 } else if (restart_timer) {
1577                         if (!mod_timer(&transport->T3_rtx_timer,
1578                                        jiffies + transport->rto))
1579                                 sctp_transport_hold(transport);
1580                 }
1581
1582                 if (forward_progress) {
1583                         if (transport->dst)
1584                                 dst_confirm(transport->dst);
1585                 }
1586         }
1587
1588         list_splice(&tlist, transmitted_queue);
1589 }
1590
1591 /* Mark chunks as missing and consequently may get retransmitted. */
1592 static void sctp_mark_missing(struct sctp_outq *q,
1593                               struct list_head *transmitted_queue,
1594                               struct sctp_transport *transport,
1595                               __u32 highest_new_tsn_in_sack,
1596                               int count_of_newacks)
1597 {
1598         struct sctp_chunk *chunk;
1599         __u32 tsn;
1600         char do_fast_retransmit = 0;
1601         struct sctp_association *asoc = q->asoc;
1602         struct sctp_transport *primary = asoc->peer.primary_path;
1603
1604         list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1605
1606                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1607
1608                 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1609                  * 'Unacknowledged TSN's', if the TSN number of an
1610                  * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1611                  * value, increment the 'TSN.Missing.Report' count on that
1612                  * chunk if it has NOT been fast retransmitted or marked for
1613                  * fast retransmit already.
1614                  */
1615                 if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
1616                     !chunk->tsn_gap_acked &&
1617                     TSN_lt(tsn, highest_new_tsn_in_sack)) {
1618
1619                         /* SFR-CACC may require us to skip marking
1620                          * this chunk as missing.
1621                          */
1622                         if (!transport || !sctp_cacc_skip(primary,
1623                                                 chunk->transport,
1624                                                 count_of_newacks, tsn)) {
1625                                 chunk->tsn_missing_report++;
1626
1627                                 pr_debug("%s: tsn:0x%x missing counter:%d\n",
1628                                          __func__, tsn, chunk->tsn_missing_report);
1629                         }
1630                 }
1631                 /*
1632                  * M4) If any DATA chunk is found to have a
1633                  * 'TSN.Missing.Report'
1634                  * value larger than or equal to 3, mark that chunk for
1635                  * retransmission and start the fast retransmit procedure.
1636                  */
1637
1638                 if (chunk->tsn_missing_report >= 3) {
1639                         chunk->fast_retransmit = SCTP_NEED_FRTX;
1640                         do_fast_retransmit = 1;
1641                 }
1642         }
1643
1644         if (transport) {
1645                 if (do_fast_retransmit)
1646                         sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1647
1648                 pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, "
1649                          "flight_size:%d, pba:%d\n",  __func__, transport,
1650                          transport->cwnd, transport->ssthresh,
1651                          transport->flight_size, transport->partial_bytes_acked);
1652         }
1653 }
1654
1655 /* Is the given TSN acked by this packet?  */
1656 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1657 {
1658         int i;
1659         sctp_sack_variable_t *frags;
1660         __u16 gap;
1661         __u32 ctsn = ntohl(sack->cum_tsn_ack);
1662
1663         if (TSN_lte(tsn, ctsn))
1664                 goto pass;
1665
1666         /* 3.3.4 Selective Acknowledgement (SACK) (3):
1667          *
1668          * Gap Ack Blocks:
1669          *  These fields contain the Gap Ack Blocks. They are repeated
1670          *  for each Gap Ack Block up to the number of Gap Ack Blocks
1671          *  defined in the Number of Gap Ack Blocks field. All DATA
1672          *  chunks with TSNs greater than or equal to (Cumulative TSN
1673          *  Ack + Gap Ack Block Start) and less than or equal to
1674          *  (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1675          *  Block are assumed to have been received correctly.
1676          */
1677
1678         frags = sack->variable;
1679         gap = tsn - ctsn;
1680         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1681                 if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1682                     TSN_lte(gap, ntohs(frags[i].gab.end)))
1683                         goto pass;
1684         }
1685
1686         return 0;
1687 pass:
1688         return 1;
1689 }
1690
1691 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1692                                     int nskips, __be16 stream)
1693 {
1694         int i;
1695
1696         for (i = 0; i < nskips; i++) {
1697                 if (skiplist[i].stream == stream)
1698                         return i;
1699         }
1700         return i;
1701 }
1702
1703 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1704 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1705 {
1706         struct sctp_association *asoc = q->asoc;
1707         struct sctp_chunk *ftsn_chunk = NULL;
1708         struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1709         int nskips = 0;
1710         int skip_pos = 0;
1711         __u32 tsn;
1712         struct sctp_chunk *chunk;
1713         struct list_head *lchunk, *temp;
1714
1715         if (!asoc->peer.prsctp_capable)
1716                 return;
1717
1718         /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1719          * received SACK.
1720          *
1721          * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1722          * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1723          */
1724         if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1725                 asoc->adv_peer_ack_point = ctsn;
1726
1727         /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1728          * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1729          * the chunk next in the out-queue space is marked as "abandoned" as
1730          * shown in the following example:
1731          *
1732          * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1733          * and the Advanced.Peer.Ack.Point is updated to this value:
1734          *
1735          *   out-queue at the end of  ==>   out-queue after Adv.Ack.Point
1736          *   normal SACK processing           local advancement
1737          *                ...                           ...
1738          *   Adv.Ack.Pt-> 102 acked                     102 acked
1739          *                103 abandoned                 103 abandoned
1740          *                104 abandoned     Adv.Ack.P-> 104 abandoned
1741          *                105                           105
1742          *                106 acked                     106 acked
1743          *                ...                           ...
1744          *
1745          * In this example, the data sender successfully advanced the
1746          * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1747          */
1748         list_for_each_safe(lchunk, temp, &q->abandoned) {
1749                 chunk = list_entry(lchunk, struct sctp_chunk,
1750                                         transmitted_list);
1751                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1752
1753                 /* Remove any chunks in the abandoned queue that are acked by
1754                  * the ctsn.
1755                  */
1756                 if (TSN_lte(tsn, ctsn)) {
1757                         list_del_init(lchunk);
1758                         sctp_chunk_free(chunk);
1759                 } else {
1760                         if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1761                                 asoc->adv_peer_ack_point = tsn;
1762                                 if (chunk->chunk_hdr->flags &
1763                                          SCTP_DATA_UNORDERED)
1764                                         continue;
1765                                 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1766                                                 nskips,
1767                                                 chunk->subh.data_hdr->stream);
1768                                 ftsn_skip_arr[skip_pos].stream =
1769                                         chunk->subh.data_hdr->stream;
1770                                 ftsn_skip_arr[skip_pos].ssn =
1771                                          chunk->subh.data_hdr->ssn;
1772                                 if (skip_pos == nskips)
1773                                         nskips++;
1774                                 if (nskips == 10)
1775                                         break;
1776                         } else
1777                                 break;
1778                 }
1779         }
1780
1781         /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1782          * is greater than the Cumulative TSN ACK carried in the received
1783          * SACK, the data sender MUST send the data receiver a FORWARD TSN
1784          * chunk containing the latest value of the
1785          * "Advanced.Peer.Ack.Point".
1786          *
1787          * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1788          * list each stream and sequence number in the forwarded TSN. This
1789          * information will enable the receiver to easily find any
1790          * stranded TSN's waiting on stream reorder queues. Each stream
1791          * SHOULD only be reported once; this means that if multiple
1792          * abandoned messages occur in the same stream then only the
1793          * highest abandoned stream sequence number is reported. If the
1794          * total size of the FORWARD TSN does NOT fit in a single MTU then
1795          * the sender of the FORWARD TSN SHOULD lower the
1796          * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1797          * single MTU.
1798          */
1799         if (asoc->adv_peer_ack_point > ctsn)
1800                 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1801                                               nskips, &ftsn_skip_arr[0]);
1802
1803         if (ftsn_chunk) {
1804                 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1805                 SCTP_INC_STATS(sock_net(asoc->base.sk), SCTP_MIB_OUTCTRLCHUNKS);
1806         }
1807 }