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, see
  26  * <http://www.gnu.org/licenses/>.
  27  *
  28  * Please send any bug reports or fixes you make to the
  29  * email address(es):
  30  *    lksctp developers <linux-sctp@vger.kernel.org>
  31  *
  32  * Written or modified by:
  33  *    La Monte H.P. Yarroll <piggy@acm.org>
  34  *    Karl Knutson          <karl@athena.chicago.il.us>
  35  *    Perry Melange         <pmelange@null.cc.uic.edu>
  36  *    Xingang Guo           <xingang.guo@intel.com>
  37  *    Hui Huang             <hui.huang@nokia.com>
  38  *    Sridhar Samudrala     <sri@us.ibm.com>
  39  *    Jon Grimm             <jgrimm@us.ibm.com>
  40  */
  41
  42 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  43
  44 #include <linux/types.h>
  45 #include <linux/list.h>   /* For struct list_head */
  46 #include <linux/socket.h>
  47 #include <linux/ip.h>
  48 #include <linux/slab.h>
  49 #include <net/sock.h>     /* For skb_set_owner_w */
  50
  51 #include <net/sctp/sctp.h>
  52 #include <net/sctp/sm.h>
  53
  54 /* Declare internal functions here.  */
  55 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn);
  56 static void sctp_check_transmitted(struct sctp_outq *q,
  57                                    struct list_head *transmitted_queue,
  58                                    struct sctp_transport *transport,
  59                                    union sctp_addr *saddr,
  60                                    struct sctp_sackhdr *sack,
  61                                    __u32 *highest_new_tsn);
  62
  63 static void sctp_mark_missing(struct sctp_outq *q,
  64                               struct list_head *transmitted_queue,
  65                               struct sctp_transport *transport,
  66                               __u32 highest_new_tsn,
  67                               int count_of_newacks);
  68
  69 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn);
  70
  71 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp);
  72
  73 /* Add data to the front of the queue. */
  74 static inline void sctp_outq_head_data(struct sctp_outq *q,
  75                                         struct sctp_chunk *ch)
  76 {
  77         list_add(&ch->list, &q->out_chunk_list);
  78         q->out_qlen += ch->skb->len;
  79 }
  80
  81 /* Take data from the front of the queue. */
  82 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
  83 {
  84         struct sctp_chunk *ch = NULL;
  85
  86         if (!list_empty(&q->out_chunk_list)) {
  87                 struct list_head *entry = q->out_chunk_list.next;
  88
  89                 ch = list_entry(entry, struct sctp_chunk, list);
  90                 list_del_init(entry);
  91                 q->out_qlen -= ch->skb->len;
  92         }
  93         return ch;
  94 }
  95 /* Add data chunk to the end of the queue. */
  96 static inline void sctp_outq_tail_data(struct sctp_outq *q,
  97                                        struct sctp_chunk *ch)
  98 {
  99         list_add_tail(&ch->list, &q->out_chunk_list);
 100         q->out_qlen += ch->skb->len;
 101 }
 102
 103 /*
 104  * SFR-CACC algorithm:
 105  * D) If count_of_newacks is greater than or equal to 2
 106  * and t was not sent to the current primary then the
 107  * sender MUST NOT increment missing report count for t.
 108  */
 109 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary,
 110                                        struct sctp_transport *transport,
 111                                        int count_of_newacks)
 112 {
 113         if (count_of_newacks >= 2 && transport != primary)
 114                 return 1;
 115         return 0;
 116 }
 117
 118 /*
 119  * SFR-CACC algorithm:
 120  * F) If count_of_newacks is less than 2, let d be the
 121  * destination to which t was sent. If cacc_saw_newack
 122  * is 0 for destination d, then the sender MUST NOT
 123  * increment missing report count for t.
 124  */
 125 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport,
 126                                        int count_of_newacks)
 127 {
 128         if (count_of_newacks < 2 &&
 129                         (transport && !transport->cacc.cacc_saw_newack))
 130                 return 1;
 131         return 0;
 132 }
 133
 134 /*
 135  * SFR-CACC algorithm:
 136  * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
 137  * execute steps C, D, F.
 138  *
 139  * C has been implemented in sctp_outq_sack
 140  */
 141 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
 142                                      struct sctp_transport *transport,
 143                                      int count_of_newacks)
 144 {
 145         if (!primary->cacc.cycling_changeover) {
 146                 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
 147                         return 1;
 148                 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
 149                         return 1;
 150                 return 0;
 151         }
 152         return 0;
 153 }
 154
 155 /*
 156  * SFR-CACC algorithm:
 157  * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
 158  * than next_tsn_at_change of the current primary, then
 159  * the sender MUST NOT increment missing report count
 160  * for t.
 161  */
 162 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
 163 {
 164         if (primary->cacc.cycling_changeover &&
 165             TSN_lt(tsn, primary->cacc.next_tsn_at_change))
 166                 return 1;
 167         return 0;
 168 }
 169
 170 /*
 171  * SFR-CACC algorithm:
 172  * 3) If the missing report count for TSN t is to be
 173  * incremented according to [RFC2960] and
 174  * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
 175  * then the sender MUST further execute steps 3.1 and
 176  * 3.2 to determine if the missing report count for
 177  * TSN t SHOULD NOT be incremented.
 178  *
 179  * 3.3) If 3.1 and 3.2 do not dictate that the missing
 180  * report count for t should not be incremented, then
 181  * the sender SHOULD increment missing report count for
 182  * t (according to [RFC2960] and [SCTP_STEWART_2002]).
 183  */
 184 static inline int sctp_cacc_skip(struct sctp_transport *primary,
 185                                  struct sctp_transport *transport,
 186                                  int count_of_newacks,
 187                                  __u32 tsn)
 188 {
 189         if (primary->cacc.changeover_active &&
 190             (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) ||
 191              sctp_cacc_skip_3_2(primary, tsn)))
 192                 return 1;
 193         return 0;
 194 }
 195
 196 /* Initialize an existing sctp_outq.  This does the boring stuff.
 197  * You still need to define handlers if you really want to DO
 198  * something with this structure...
 199  */
 200 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
 201 {
 202         memset(q, 0, sizeof(struct sctp_outq));
 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
 212 /* Free the outqueue structure and any related pending chunks.
 213  */
 214 static void __sctp_outq_teardown(struct sctp_outq *q)
 215 {
 216         struct sctp_transport *transport;
 217         struct list_head *lchunk, *temp;
 218         struct sctp_chunk *chunk, *tmp;
 219
 220         /* Throw away unacknowledged chunks. */
 221         list_for_each_entry(transport, &q->asoc->peer.transport_addr_list,
 222                         transports) {
 223                 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
 224                         chunk = list_entry(lchunk, struct sctp_chunk,
 225                                            transmitted_list);
 226                         /* Mark as part of a failed message. */
 227                         sctp_chunk_fail(chunk, q->error);
 228                         sctp_chunk_free(chunk);
 229                 }
 230         }
 231
 232         /* Throw away chunks that have been gap ACKed.  */
 233         list_for_each_safe(lchunk, temp, &q->sacked) {
 234                 list_del_init(lchunk);
 235                 chunk = list_entry(lchunk, struct sctp_chunk,
 236                                    transmitted_list);
 237                 sctp_chunk_fail(chunk, q->error);
 238                 sctp_chunk_free(chunk);
 239         }
 240
 241         /* Throw away any chunks in the retransmit queue. */
 242         list_for_each_safe(lchunk, temp, &q->retransmit) {
 243                 list_del_init(lchunk);
 244                 chunk = list_entry(lchunk, struct sctp_chunk,
 245                                    transmitted_list);
 246                 sctp_chunk_fail(chunk, q->error);
 247                 sctp_chunk_free(chunk);
 248         }
 249
 250         /* Throw away any chunks that are in the abandoned queue. */
 251         list_for_each_safe(lchunk, temp, &q->abandoned) {
 252                 list_del_init(lchunk);
 253                 chunk = list_entry(lchunk, struct sctp_chunk,
 254                                    transmitted_list);
 255                 sctp_chunk_fail(chunk, q->error);
 256                 sctp_chunk_free(chunk);
 257         }
 258
 259         /* Throw away any leftover data chunks. */
 260         while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
 261
 262                 /* Mark as send failure. */
 263                 sctp_chunk_fail(chunk, q->error);
 264                 sctp_chunk_free(chunk);
 265         }
 266
 267         /* Throw away any leftover control chunks. */
 268         list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
 269                 list_del_init(&chunk->list);
 270                 sctp_chunk_free(chunk);
 271         }
 272 }
 273
 274 void sctp_outq_teardown(struct sctp_outq *q)
 275 {
 276         __sctp_outq_teardown(q);
 277         sctp_outq_init(q->asoc, q);
 278 }
 279
 280 /* Free the outqueue structure and any related pending chunks.  */
 281 void sctp_outq_free(struct sctp_outq *q)
 282 {
 283         /* Throw away leftover chunks. */
 284         __sctp_outq_teardown(q);
 285 }
 286
 287 /* Put a new chunk in an sctp_outq.  */
 288 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk, gfp_t gfp)
 289 {
 290         struct net *net = sock_net(q->asoc->base.sk);
 291         int error = 0;
 292
 293         pr_debug("%s: outq:%p, chunk:%p[%s]\n", __func__, q, chunk,
 294                  chunk && chunk->chunk_hdr ?
 295                  sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
 296                  "illegal chunk");
 297
 298         /* If it is data, queue it up, otherwise, send it
 299          * immediately.
 300          */
 301         if (sctp_chunk_is_data(chunk)) {
 302                 /* Is it OK to queue data chunks?  */
 303                 /* From 9. Termination of Association
 304                  *
 305                  * When either endpoint performs a shutdown, the
 306                  * association on each peer will stop accepting new
 307                  * data from its user and only deliver data in queue
 308                  * at the time of sending or receiving the SHUTDOWN
 309                  * chunk.
 310                  */
 311                 switch (q->asoc->state) {
 312                 case SCTP_STATE_CLOSED:
 313                 case SCTP_STATE_SHUTDOWN_PENDING:
 314                 case SCTP_STATE_SHUTDOWN_SENT:
 315                 case SCTP_STATE_SHUTDOWN_RECEIVED:
 316                 case SCTP_STATE_SHUTDOWN_ACK_SENT:
 317                         /* Cannot send after transport endpoint shutdown */
 318                         error = -ESHUTDOWN;
 319                         break;
 320
 321                 default:
 322                         pr_debug("%s: outqueueing: outq:%p, chunk:%p[%s])\n",
 323                                  __func__, q, chunk, chunk && chunk->chunk_hdr ?
 324                                  sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
 325                                  "illegal chunk");
 326
 327                         sctp_chunk_hold(chunk);
 328                         sctp_outq_tail_data(q, chunk);
 329                         if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
 330                                 SCTP_INC_STATS(net, SCTP_MIB_OUTUNORDERCHUNKS);
 331                         else
 332                                 SCTP_INC_STATS(net, SCTP_MIB_OUTORDERCHUNKS);
 333                         break;
 334                 }
 335         } else {
 336                 list_add_tail(&chunk->list, &q->control_chunk_list);
 337                 SCTP_INC_STATS(net, SCTP_MIB_OUTCTRLCHUNKS);
 338         }
 339
 340         if (error < 0)
 341                 return error;
 342
 343         if (!q->cork)
 344                 error = sctp_outq_flush(q, 0, gfp);
 345
 346         return error;
 347 }
 348
 349 /* Insert a chunk into the sorted list based on the TSNs.  The retransmit list
 350  * and the abandoned list are in ascending order.
 351  */
 352 static void sctp_insert_list(struct list_head *head, struct list_head *new)
 353 {
 354         struct list_head *pos;
 355         struct sctp_chunk *nchunk, *lchunk;
 356         __u32 ntsn, ltsn;
 357         int done = 0;
 358
 359         nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
 360         ntsn = ntohl(nchunk->subh.data_hdr->tsn);
 361
 362         list_for_each(pos, head) {
 363                 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
 364                 ltsn = ntohl(lchunk->subh.data_hdr->tsn);
 365                 if (TSN_lt(ntsn, ltsn)) {
 366                         list_add(new, pos->prev);
 367                         done = 1;
 368                         break;
 369                 }
 370         }
 371         if (!done)
 372                 list_add_tail(new, head);
 373 }
 374
 375 /* Mark all the eligible packets on a transport for retransmission.  */
 376 void sctp_retransmit_mark(struct sctp_outq *q,
 377                           struct sctp_transport *transport,
 378                           __u8 reason)
 379 {
 380         struct list_head *lchunk, *ltemp;
 381         struct sctp_chunk *chunk;
 382
 383         /* Walk through the specified transmitted queue.  */
 384         list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
 385                 chunk = list_entry(lchunk, struct sctp_chunk,
 386                                    transmitted_list);
 387
 388                 /* If the chunk is abandoned, move it to abandoned list. */
 389                 if (sctp_chunk_abandoned(chunk)) {
 390                         list_del_init(lchunk);
 391                         sctp_insert_list(&q->abandoned, lchunk);
 392
 393                         /* If this chunk has not been previousely acked,
 394                          * stop considering it 'outstanding'.  Our peer
 395                          * will most likely never see it since it will
 396                          * not be retransmitted
 397                          */
 398                         if (!chunk->tsn_gap_acked) {
 399                                 if (chunk->transport)
 400                                         chunk->transport->flight_size -=
 401                                                         sctp_data_size(chunk);
 402                                 q->outstanding_bytes -= sctp_data_size(chunk);
 403                                 q->asoc->peer.rwnd += sctp_data_size(chunk);
 404                         }
 405                         continue;
 406                 }
 407
 408                 /* If we are doing  retransmission due to a timeout or pmtu
 409                  * discovery, only the  chunks that are not yet acked should
 410                  * be added to the retransmit queue.
 411                  */
 412                 if ((reason == SCTP_RTXR_FAST_RTX  &&
 413                             (chunk->fast_retransmit == SCTP_NEED_FRTX)) ||
 414                     (reason != SCTP_RTXR_FAST_RTX  && !chunk->tsn_gap_acked)) {
 415                         /* RFC 2960 6.2.1 Processing a Received SACK
 416                          *
 417                          * C) Any time a DATA chunk is marked for
 418                          * retransmission (via either T3-rtx timer expiration
 419                          * (Section 6.3.3) or via fast retransmit
 420                          * (Section 7.2.4)), add the data size of those
 421                          * chunks to the rwnd.
 422                          */
 423                         q->asoc->peer.rwnd += sctp_data_size(chunk);
 424                         q->outstanding_bytes -= sctp_data_size(chunk);
 425                         if (chunk->transport)
 426                                 transport->flight_size -= sctp_data_size(chunk);
 427
 428                         /* sctpimpguide-05 Section 2.8.2
 429                          * M5) If a T3-rtx timer expires, the
 430                          * 'TSN.Missing.Report' of all affected TSNs is set
 431                          * to 0.
 432                          */
 433                         chunk->tsn_missing_report = 0;
 434
 435                         /* If a chunk that is being used for RTT measurement
 436                          * has to be retransmitted, we cannot use this chunk
 437                          * anymore for RTT measurements. Reset rto_pending so
 438                          * that a new RTT measurement is started when a new
 439                          * data chunk is sent.
 440                          */
 441                         if (chunk->rtt_in_progress) {
 442                                 chunk->rtt_in_progress = 0;
 443                                 transport->rto_pending = 0;
 444                         }
 445
 446                         chunk->resent = 1;
 447
 448                         /* Move the chunk to the retransmit queue. The chunks
 449                          * on the retransmit queue are always kept in order.
 450                          */
 451                         list_del_init(lchunk);
 452                         sctp_insert_list(&q->retransmit, lchunk);
 453                 }
 454         }
 455
 456         pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, "
 457                  "flight_size:%d, pba:%d\n", __func__, transport, reason,
 458                  transport->cwnd, transport->ssthresh, transport->flight_size,
 459                  transport->partial_bytes_acked);
 460 }
 461
 462 /* Mark all the eligible packets on a transport for retransmission and force
 463  * one packet out.
 464  */
 465 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
 466                      sctp_retransmit_reason_t reason)
 467 {
 468         struct net *net = sock_net(q->asoc->base.sk);
 469         int error = 0;
 470
 471         switch (reason) {
 472         case SCTP_RTXR_T3_RTX:
 473                 SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS);
 474                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
 475                 /* Update the retran path if the T3-rtx timer has expired for
 476                  * the current retran path.
 477                  */
 478                 if (transport == transport->asoc->peer.retran_path)
 479                         sctp_assoc_update_retran_path(transport->asoc);
 480                 transport->asoc->rtx_data_chunks +=
 481                         transport->asoc->unack_data;
 482                 break;
 483         case SCTP_RTXR_FAST_RTX:
 484                 SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS);
 485                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
 486                 q->fast_rtx = 1;
 487                 break;
 488         case SCTP_RTXR_PMTUD:
 489                 SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS);
 490                 break;
 491         case SCTP_RTXR_T1_RTX:
 492                 SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS);
 493                 transport->asoc->init_retries++;
 494                 break;
 495         default:
 496                 BUG();
 497         }
 498
 499         sctp_retransmit_mark(q, transport, reason);
 500
 501         /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
 502          * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
 503          * following the procedures outlined in C1 - C5.
 504          */
 505         if (reason == SCTP_RTXR_T3_RTX)
 506                 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
 507
 508         /* Flush the queues only on timeout, since fast_rtx is only
 509          * triggered during sack processing and the queue
 510          * will be flushed at the end.
 511          */
 512         if (reason != SCTP_RTXR_FAST_RTX)
 513                 error = sctp_outq_flush(q, /* rtx_timeout */ 1, GFP_ATOMIC);
 514
 515         if (error)
 516                 q->asoc->base.sk->sk_err = -error;
 517 }
 518
 519 /*
 520  * Transmit DATA chunks on the retransmit queue.  Upon return from
 521  * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
 522  * need to be transmitted by the caller.
 523  * We assume that pkt->transport has already been set.
 524  *
 525  * The return value is a normal kernel error return value.
 526  */
 527 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
 528                                int rtx_timeout, int *start_timer)
 529 {
 530         struct list_head *lqueue;
 531         struct sctp_transport *transport = pkt->transport;
 532         sctp_xmit_t status;
 533         struct sctp_chunk *chunk, *chunk1;
 534         int fast_rtx;
 535         int error = 0;
 536         int timer = 0;
 537         int done = 0;
 538
 539         lqueue = &q->retransmit;
 540         fast_rtx = q->fast_rtx;
 541
 542         /* This loop handles time-out retransmissions, fast retransmissions,
 543          * and retransmissions due to opening of whindow.
 544          *
 545          * RFC 2960 6.3.3 Handle T3-rtx Expiration
 546          *
 547          * E3) Determine how many of the earliest (i.e., lowest TSN)
 548          * outstanding DATA chunks for the address for which the
 549          * T3-rtx has expired will fit into a single packet, subject
 550          * to the MTU constraint for the path corresponding to the
 551          * destination transport address to which the retransmission
 552          * is being sent (this may be different from the address for
 553          * which the timer expires [see Section 6.4]). Call this value
 554          * K. Bundle and retransmit those K DATA chunks in a single
 555          * packet to the destination endpoint.
 556          *
 557          * [Just to be painfully clear, if we are retransmitting
 558          * because a timeout just happened, we should send only ONE
 559          * packet of retransmitted data.]
 560          *
 561          * For fast retransmissions we also send only ONE packet.  However,
 562          * if we are just flushing the queue due to open window, we'll
 563          * try to send as much as possible.
 564          */
 565         list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
 566                 /* If the chunk is abandoned, move it to abandoned list. */
 567                 if (sctp_chunk_abandoned(chunk)) {
 568                         list_del_init(&chunk->transmitted_list);
 569                         sctp_insert_list(&q->abandoned,
 570                                          &chunk->transmitted_list);
 571                         continue;
 572                 }
 573
 574                 /* Make sure that Gap Acked TSNs are not retransmitted.  A
 575                  * simple approach is just to move such TSNs out of the
 576                  * way and into a 'transmitted' queue and skip to the
 577                  * next chunk.
 578                  */
 579                 if (chunk->tsn_gap_acked) {
 580                         list_move_tail(&chunk->transmitted_list,
 581                                        &transport->transmitted);
 582                         continue;
 583                 }
 584
 585                 /* If we are doing fast retransmit, ignore non-fast_rtransmit
 586                  * chunks
 587                  */
 588                 if (fast_rtx && !chunk->fast_retransmit)
 589                         continue;
 590
 591 redo:
 592                 /* Attempt to append this chunk to the packet. */
 593                 status = sctp_packet_append_chunk(pkt, chunk);
 594
 595                 switch (status) {
 596                 case SCTP_XMIT_PMTU_FULL:
 597                         if (!pkt->has_data && !pkt->has_cookie_echo) {
 598                                 /* If this packet did not contain DATA then
 599                                  * retransmission did not happen, so do it
 600                                  * again.  We'll ignore the error here since
 601                                  * control chunks are already freed so there
 602                                  * is nothing we can do.
 603                                  */
 604                                 sctp_packet_transmit(pkt, GFP_ATOMIC);
 605                                 goto redo;
 606                         }
 607
 608                         /* Send this packet.  */
 609                         error = sctp_packet_transmit(pkt, GFP_ATOMIC);
 610
 611                         /* If we are retransmitting, we should only
 612                          * send a single packet.
 613                          * Otherwise, try appending this chunk again.
 614                          */
 615                         if (rtx_timeout || fast_rtx)
 616                                 done = 1;
 617                         else
 618                                 goto redo;
 619
 620                         /* Bundle next chunk in the next round.  */
 621                         break;
 622
 623                 case SCTP_XMIT_RWND_FULL:
 624                         /* Send this packet. */
 625                         error = sctp_packet_transmit(pkt, GFP_ATOMIC);
 626
 627                         /* Stop sending DATA as there is no more room
 628                          * at the receiver.
 629                          */
 630                         done = 1;
 631                         break;
 632
 633                 case SCTP_XMIT_DELAY:
 634                         /* Send this packet. */
 635                         error = sctp_packet_transmit(pkt, GFP_ATOMIC);
 636
 637                         /* Stop sending DATA because of nagle delay. */
 638                         done = 1;
 639                         break;
 640
 641                 default:
 642                         /* The append was successful, so add this chunk to
 643                          * the transmitted list.
 644                          */
 645                         list_move_tail(&chunk->transmitted_list,
 646                                        &transport->transmitted);
 647
 648                         /* Mark the chunk as ineligible for fast retransmit
 649                          * after it is retransmitted.
 650                          */
 651                         if (chunk->fast_retransmit == SCTP_NEED_FRTX)
 652                                 chunk->fast_retransmit = SCTP_DONT_FRTX;
 653
 654                         q->asoc->stats.rtxchunks++;
 655                         break;
 656                 }
 657
 658                 /* Set the timer if there were no errors */
 659                 if (!error && !timer)
 660                         timer = 1;
 661
 662                 if (done)
 663                         break;
 664         }
 665
 666         /* If we are here due to a retransmit timeout or a fast
 667          * retransmit and if there are any chunks left in the retransmit
 668          * queue that could not fit in the PMTU sized packet, they need
 669          * to be marked as ineligible for a subsequent fast retransmit.
 670          */
 671         if (rtx_timeout || fast_rtx) {
 672                 list_for_each_entry(chunk1, lqueue, transmitted_list) {
 673                         if (chunk1->fast_retransmit == SCTP_NEED_FRTX)
 674                                 chunk1->fast_retransmit = SCTP_DONT_FRTX;
 675                 }
 676         }
 677
 678         *start_timer = timer;
 679
 680         /* Clear fast retransmit hint */
 681         if (fast_rtx)
 682                 q->fast_rtx = 0;
 683
 684         return error;
 685 }
 686
 687 /* Cork the outqueue so queued chunks are really queued. */
 688 int sctp_outq_uncork(struct sctp_outq *q, gfp_t gfp)
 689 {
 690         if (q->cork)
 691                 q->cork = 0;
 692
 693         return sctp_outq_flush(q, 0, gfp);
 694 }
 695
 696
 697 /*
 698  * Try to flush an outqueue.
 699  *
 700  * Description: Send everything in q which we legally can, subject to
 701  * congestion limitations.
 702  * * Note: This function can be called from multiple contexts so appropriate
 703  * locking concerns must be made.  Today we use the sock lock to protect
 704  * this function.
 705  */
 706 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp)
 707 {
 708         struct sctp_packet *packet;
 709         struct sctp_packet singleton;
 710         struct sctp_association *asoc = q->asoc;
 711         __u16 sport = asoc->base.bind_addr.port;
 712         __u16 dport = asoc->peer.port;
 713         __u32 vtag = asoc->peer.i.init_tag;
 714         struct sctp_transport *transport = NULL;
 715         struct sctp_transport *new_transport;
 716         struct sctp_chunk *chunk, *tmp;
 717         sctp_xmit_t status;
 718         int error = 0;
 719         int start_timer = 0;
 720         int one_packet = 0;
 721
 722         /* These transports have chunks to send. */
 723         struct list_head transport_list;
 724         struct list_head *ltransport;
 725
 726         INIT_LIST_HEAD(&transport_list);
 727         packet = NULL;
 728
 729         /*
 730          * 6.10 Bundling
 731          *   ...
 732          *   When bundling control chunks with DATA chunks, an
 733          *   endpoint MUST place control chunks first in the outbound
 734          *   SCTP packet.  The transmitter MUST transmit DATA chunks
 735          *   within a SCTP packet in increasing order of TSN.
 736          *   ...
 737          */
 738
 739         list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
 740                 /* RFC 5061, 5.3
 741                  * F1) This means that until such time as the ASCONF
 742                  * containing the add is acknowledged, the sender MUST
 743                  * NOT use the new IP address as a source for ANY SCTP
 744                  * packet except on carrying an ASCONF Chunk.
 745                  */
 746                 if (asoc->src_out_of_asoc_ok &&
 747                     chunk->chunk_hdr->type != SCTP_CID_ASCONF)
 748                         continue;
 749
 750                 list_del_init(&chunk->list);
 751
 752                 /* Pick the right transport to use. */
 753                 new_transport = chunk->transport;
 754
 755                 if (!new_transport) {
 756                         /*
 757                          * If we have a prior transport pointer, see if
 758                          * the destination address of the chunk
 759                          * matches the destination address of the
 760                          * current transport.  If not a match, then
 761                          * try to look up the transport with a given
 762                          * destination address.  We do this because
 763                          * after processing ASCONFs, we may have new
 764                          * transports created.
 765                          */
 766                         if (transport &&
 767                             sctp_cmp_addr_exact(&chunk->dest,
 768                                                 &transport->ipaddr))
 769                                         new_transport = transport;
 770                         else
 771                                 new_transport = sctp_assoc_lookup_paddr(asoc,
 772                                                                 &chunk->dest);
 773
 774                         /* if we still don't have a new transport, then
 775                          * use the current active path.
 776                          */
 777                         if (!new_transport)
 778                                 new_transport = asoc->peer.active_path;
 779                 } else if ((new_transport->state == SCTP_INACTIVE) ||
 780                            (new_transport->state == SCTP_UNCONFIRMED) ||
 781                            (new_transport->state == SCTP_PF)) {
 782                         /* If the chunk is Heartbeat or Heartbeat Ack,
 783                          * send it to chunk->transport, even if it's
 784                          * inactive.
 785                          *
 786                          * 3.3.6 Heartbeat Acknowledgement:
 787                          * ...
 788                          * A HEARTBEAT ACK is always sent to the source IP
 789                          * address of the IP datagram containing the
 790                          * HEARTBEAT chunk to which this ack is responding.
 791                          * ...
 792                          *
 793                          * ASCONF_ACKs also must be sent to the source.
 794                          */
 795                         if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
 796                             chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK &&
 797                             chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK)
 798                                 new_transport = asoc->peer.active_path;
 799                 }
 800
 801                 /* Are we switching transports?
 802                  * Take care of transport locks.
 803                  */
 804                 if (new_transport != transport) {
 805                         transport = new_transport;
 806                         if (list_empty(&transport->send_ready)) {
 807                                 list_add_tail(&transport->send_ready,
 808                                               &transport_list);
 809                         }
 810                         packet = &transport->packet;
 811                         sctp_packet_config(packet, vtag,
 812                                            asoc->peer.ecn_capable);
 813                 }
 814
 815                 switch (chunk->chunk_hdr->type) {
 816                 /*
 817                  * 6.10 Bundling
 818                  *   ...
 819                  *   An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
 820                  *   COMPLETE with any other chunks.  [Send them immediately.]
 821                  */
 822                 case SCTP_CID_INIT:
 823                 case SCTP_CID_INIT_ACK:
 824                 case SCTP_CID_SHUTDOWN_COMPLETE:
 825                         sctp_packet_init(&singleton, transport, sport, dport);
 826                         sctp_packet_config(&singleton, vtag, 0);
 827                         sctp_packet_append_chunk(&singleton, chunk);
 828                         error = sctp_packet_transmit(&singleton, gfp);
 829                         if (error < 0)
 830                                 return error;
 831                         break;
 832
 833                 case SCTP_CID_ABORT:
 834                         if (sctp_test_T_bit(chunk)) {
 835                                 packet->vtag = asoc->c.my_vtag;
 836                         }
 837                 /* The following chunks are "response" chunks, i.e.
 838                  * they are generated in response to something we
 839                  * received.  If we are sending these, then we can
 840                  * send only 1 packet containing these chunks.
 841                  */
 842                 case SCTP_CID_HEARTBEAT_ACK:
 843                 case SCTP_CID_SHUTDOWN_ACK:
 844                 case SCTP_CID_COOKIE_ACK:
 845                 case SCTP_CID_COOKIE_ECHO:
 846                 case SCTP_CID_ERROR:
 847                 case SCTP_CID_ECN_CWR:
 848                 case SCTP_CID_ASCONF_ACK:
 849                         one_packet = 1;
 850                         /* Fall through */
 851
 852                 case SCTP_CID_SACK:
 853                 case SCTP_CID_HEARTBEAT:
 854                 case SCTP_CID_SHUTDOWN:
 855                 case SCTP_CID_ECN_ECNE:
 856                 case SCTP_CID_ASCONF:
 857                 case SCTP_CID_FWD_TSN:
 858                         status = sctp_packet_transmit_chunk(packet, chunk,
 859                                                             one_packet, gfp);
 860                         if (status  != SCTP_XMIT_OK) {
 861                                 /* put the chunk back */
 862                                 list_add(&chunk->list, &q->control_chunk_list);
 863                         } else {
 864                                 asoc->stats.octrlchunks++;
 865                                 /* PR-SCTP C5) If a FORWARD TSN is sent, the
 866                                  * sender MUST assure that at least one T3-rtx
 867                                  * timer is running.
 868                                  */
 869                                 if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN) {
 870                                         sctp_transport_reset_t3_rtx(transport);
 871                                         transport->last_time_sent = jiffies;
 872                                 }
 873                         }
 874                         break;
 875
 876                 default:
 877                         /* We built a chunk with an illegal type! */
 878                         BUG();
 879                 }
 880         }
 881
 882         if (q->asoc->src_out_of_asoc_ok)
 883                 goto sctp_flush_out;
 884
 885         /* Is it OK to send data chunks?  */
 886         switch (asoc->state) {
 887         case SCTP_STATE_COOKIE_ECHOED:
 888                 /* Only allow bundling when this packet has a COOKIE-ECHO
 889                  * chunk.
 890                  */
 891                 if (!packet || !packet->has_cookie_echo)
 892                         break;
 893
 894                 /* fallthru */
 895         case SCTP_STATE_ESTABLISHED:
 896         case SCTP_STATE_SHUTDOWN_PENDING:
 897         case SCTP_STATE_SHUTDOWN_RECEIVED:
 898                 /*
 899                  * RFC 2960 6.1  Transmission of DATA Chunks
 900                  *
 901                  * C) When the time comes for the sender to transmit,
 902                  * before sending new DATA chunks, the sender MUST
 903                  * first transmit any outstanding DATA chunks which
 904                  * are marked for retransmission (limited by the
 905                  * current cwnd).
 906                  */
 907                 if (!list_empty(&q->retransmit)) {
 908                         if (asoc->peer.retran_path->state == SCTP_UNCONFIRMED)
 909                                 goto sctp_flush_out;
 910                         if (transport == asoc->peer.retran_path)
 911                                 goto retran;
 912
 913                         /* Switch transports & prepare the packet.  */
 914
 915                         transport = asoc->peer.retran_path;
 916
 917                         if (list_empty(&transport->send_ready)) {
 918                                 list_add_tail(&transport->send_ready,
 919                                               &transport_list);
 920                         }
 921
 922                         packet = &transport->packet;
 923                         sctp_packet_config(packet, vtag,
 924                                            asoc->peer.ecn_capable);
 925                 retran:
 926                         error = sctp_outq_flush_rtx(q, packet,
 927                                                     rtx_timeout, &start_timer);
 928
 929                         if (start_timer) {
 930                                 sctp_transport_reset_t3_rtx(transport);
 931                                 transport->last_time_sent = jiffies;
 932                         }
 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                                 WARN_ONCE(1, "Atempt to send packet on unconfirmed path.");
 987                                 sctp_chunk_fail(chunk, 0);
 988                                 sctp_chunk_free(chunk);
 989                                 continue;
 990                         }
 991
 992                         /* Change packets if necessary.  */
 993                         if (new_transport != transport) {
 994                                 transport = new_transport;
 995
 996                                 /* Schedule to have this transport's
 997                                  * packet flushed.
 998                                  */
 999                                 if (list_empty(&transport->send_ready)) {
1000                                         list_add_tail(&transport->send_ready,
1001                                                       &transport_list);
1002                                 }
1003
1004                                 packet = &transport->packet;
1005                                 sctp_packet_config(packet, vtag,
1006                                                    asoc->peer.ecn_capable);
1007                                 /* We've switched transports, so apply the
1008                                  * Burst limit to the new transport.
1009                                  */
1010                                 sctp_transport_burst_limited(transport);
1011                         }
1012
1013                         pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p "
1014                                  "skb->users:%d\n",
1015                                  __func__, q, chunk, chunk && chunk->chunk_hdr ?
1016                                  sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
1017                                  "illegal chunk", ntohl(chunk->subh.data_hdr->tsn),
1018                                  chunk->skb ? chunk->skb->head : NULL, chunk->skb ?
1019                                  atomic_read(&chunk->skb->users) : -1);
1020
1021                         /* Add the chunk to the packet.  */
1022                         status = sctp_packet_transmit_chunk(packet, chunk, 0, gfp);
1023
1024                         switch (status) {
1025                         case SCTP_XMIT_PMTU_FULL:
1026                         case SCTP_XMIT_RWND_FULL:
1027                         case SCTP_XMIT_DELAY:
1028                                 /* We could not append this chunk, so put
1029                                  * the chunk back on the output queue.
1030                                  */
1031                                 pr_debug("%s: could not transmit tsn:0x%x, status:%d\n",
1032                                          __func__, ntohl(chunk->subh.data_hdr->tsn),
1033                                          status);
1034
1035                                 sctp_outq_head_data(q, chunk);
1036                                 goto sctp_flush_out;
1037
1038                         case SCTP_XMIT_OK:
1039                                 /* The sender is in the SHUTDOWN-PENDING state,
1040                                  * The sender MAY set the I-bit in the DATA
1041                                  * chunk header.
1042                                  */
1043                                 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING)
1044                                         chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM;
1045                                 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
1046                                         asoc->stats.ouodchunks++;
1047                                 else
1048                                         asoc->stats.oodchunks++;
1049
1050                                 break;
1051
1052                         default:
1053                                 BUG();
1054                         }
1055
1056                         /* BUG: We assume that the sctp_packet_transmit()
1057                          * call below will succeed all the time and add the
1058                          * chunk to the transmitted list and restart the
1059                          * timers.
1060                          * It is possible that the call can fail under OOM
1061                          * conditions.
1062                          *
1063                          * Is this really a problem?  Won't this behave
1064                          * like a lost TSN?
1065                          */
1066                         list_add_tail(&chunk->transmitted_list,
1067                                       &transport->transmitted);
1068
1069                         sctp_transport_reset_t3_rtx(transport);
1070                         transport->last_time_sent = jiffies;
1071
1072                         /* Only let one DATA chunk get bundled with a
1073                          * COOKIE-ECHO chunk.
1074                          */
1075                         if (packet->has_cookie_echo)
1076                                 goto sctp_flush_out;
1077                 }
1078                 break;
1079
1080         default:
1081                 /* Do nothing.  */
1082                 break;
1083         }
1084
1085 sctp_flush_out:
1086
1087         /* Before returning, examine all the transports touched in
1088          * this call.  Right now, we bluntly force clear all the
1089          * transports.  Things might change after we implement Nagle.
1090          * But such an examination is still required.
1091          *
1092          * --xguo
1093          */
1094         while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL) {
1095                 struct sctp_transport *t = list_entry(ltransport,
1096                                                       struct sctp_transport,
1097                                                       send_ready);
1098                 packet = &t->packet;
1099                 if (!sctp_packet_empty(packet))
1100                         error = sctp_packet_transmit(packet, gfp);
1101
1102                 /* Clear the burst limited state, if any */
1103                 sctp_transport_burst_reset(t);
1104         }
1105
1106         return error;
1107 }
1108
1109 /* Update unack_data based on the incoming SACK chunk */
1110 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1111                                         struct sctp_sackhdr *sack)
1112 {
1113         sctp_sack_variable_t *frags;
1114         __u16 unack_data;
1115         int i;
1116
1117         unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1118
1119         frags = sack->variable;
1120         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1121                 unack_data -= ((ntohs(frags[i].gab.end) -
1122                                 ntohs(frags[i].gab.start) + 1));
1123         }
1124
1125         assoc->unack_data = unack_data;
1126 }
1127
1128 /* This is where we REALLY process a SACK.
1129  *
1130  * Process the SACK against the outqueue.  Mostly, this just frees
1131  * things off the transmitted queue.
1132  */
1133 int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk)
1134 {
1135         struct sctp_association *asoc = q->asoc;
1136         struct sctp_sackhdr *sack = chunk->subh.sack_hdr;
1137         struct sctp_transport *transport;
1138         struct sctp_chunk *tchunk = NULL;
1139         struct list_head *lchunk, *transport_list, *temp;
1140         sctp_sack_variable_t *frags = sack->variable;
1141         __u32 sack_ctsn, ctsn, tsn;
1142         __u32 highest_tsn, highest_new_tsn;
1143         __u32 sack_a_rwnd;
1144         unsigned int outstanding;
1145         struct sctp_transport *primary = asoc->peer.primary_path;
1146         int count_of_newacks = 0;
1147         int gap_ack_blocks;
1148         u8 accum_moved = 0;
1149
1150         /* Grab the association's destination address list. */
1151         transport_list = &asoc->peer.transport_addr_list;
1152
1153         sack_ctsn = ntohl(sack->cum_tsn_ack);
1154         gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1155         asoc->stats.gapcnt += gap_ack_blocks;
1156         /*
1157          * SFR-CACC algorithm:
1158          * On receipt of a SACK the sender SHOULD execute the
1159          * following statements.
1160          *
1161          * 1) If the cumulative ack in the SACK passes next tsn_at_change
1162          * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1163          * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1164          * all destinations.
1165          * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1166          * is set the receiver of the SACK MUST take the following actions:
1167          *
1168          * A) Initialize the cacc_saw_newack to 0 for all destination
1169          * addresses.
1170          *
1171          * Only bother if changeover_active is set. Otherwise, this is
1172          * totally suboptimal to do on every SACK.
1173          */
1174         if (primary->cacc.changeover_active) {
1175                 u8 clear_cycling = 0;
1176
1177                 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1178                         primary->cacc.changeover_active = 0;
1179                         clear_cycling = 1;
1180                 }
1181
1182                 if (clear_cycling || gap_ack_blocks) {
1183                         list_for_each_entry(transport, transport_list,
1184                                         transports) {
1185                                 if (clear_cycling)
1186                                         transport->cacc.cycling_changeover = 0;
1187                                 if (gap_ack_blocks)
1188                                         transport->cacc.cacc_saw_newack = 0;
1189                         }
1190                 }
1191         }
1192
1193         /* Get the highest TSN in the sack. */
1194         highest_tsn = sack_ctsn;
1195         if (gap_ack_blocks)
1196                 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1197
1198         if (TSN_lt(asoc->highest_sacked, highest_tsn))
1199                 asoc->highest_sacked = highest_tsn;
1200
1201         highest_new_tsn = sack_ctsn;
1202
1203         /* Run through the retransmit queue.  Credit bytes received
1204          * and free those chunks that we can.
1205          */
1206         sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn);
1207
1208         /* Run through the transmitted queue.
1209          * Credit bytes received and free those chunks which we can.
1210          *
1211          * This is a MASSIVE candidate for optimization.
1212          */
1213         list_for_each_entry(transport, transport_list, transports) {
1214                 sctp_check_transmitted(q, &transport->transmitted,
1215                                        transport, &chunk->source, sack,
1216                                        &highest_new_tsn);
1217                 /*
1218                  * SFR-CACC algorithm:
1219                  * C) Let count_of_newacks be the number of
1220                  * destinations for which cacc_saw_newack is set.
1221                  */
1222                 if (transport->cacc.cacc_saw_newack)
1223                         count_of_newacks++;
1224         }
1225
1226         /* Move the Cumulative TSN Ack Point if appropriate.  */
1227         if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) {
1228                 asoc->ctsn_ack_point = sack_ctsn;
1229                 accum_moved = 1;
1230         }
1231
1232         if (gap_ack_blocks) {
1233
1234                 if (asoc->fast_recovery && accum_moved)
1235                         highest_new_tsn = highest_tsn;
1236
1237                 list_for_each_entry(transport, transport_list, transports)
1238                         sctp_mark_missing(q, &transport->transmitted, transport,
1239                                           highest_new_tsn, count_of_newacks);
1240         }
1241
1242         /* Update unack_data field in the assoc. */
1243         sctp_sack_update_unack_data(asoc, sack);
1244
1245         ctsn = asoc->ctsn_ack_point;
1246
1247         /* Throw away stuff rotting on the sack queue.  */
1248         list_for_each_safe(lchunk, temp, &q->sacked) {
1249                 tchunk = list_entry(lchunk, struct sctp_chunk,
1250                                     transmitted_list);
1251                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1252                 if (TSN_lte(tsn, ctsn)) {
1253                         list_del_init(&tchunk->transmitted_list);
1254                         sctp_chunk_free(tchunk);
1255                 }
1256         }
1257
1258         /* ii) Set rwnd equal to the newly received a_rwnd minus the
1259          *     number of bytes still outstanding after processing the
1260          *     Cumulative TSN Ack and the Gap Ack Blocks.
1261          */
1262
1263         sack_a_rwnd = ntohl(sack->a_rwnd);
1264         asoc->peer.zero_window_announced = !sack_a_rwnd;
1265         outstanding = q->outstanding_bytes;
1266
1267         if (outstanding < sack_a_rwnd)
1268                 sack_a_rwnd -= outstanding;
1269         else
1270                 sack_a_rwnd = 0;
1271
1272         asoc->peer.rwnd = sack_a_rwnd;
1273
1274         sctp_generate_fwdtsn(q, sack_ctsn);
1275
1276         pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__, sack_ctsn);
1277         pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, "
1278                  "advertised peer ack point:0x%x\n", __func__, asoc, ctsn,
1279                  asoc->adv_peer_ack_point);
1280
1281         return sctp_outq_is_empty(q);
1282 }
1283
1284 /* Is the outqueue empty?
1285  * The queue is empty when we have not pending data, no in-flight data
1286  * and nothing pending retransmissions.
1287  */
1288 int sctp_outq_is_empty(const struct sctp_outq *q)
1289 {
1290         return q->out_qlen == 0 && q->outstanding_bytes == 0 &&
1291                list_empty(&q->retransmit);
1292 }
1293
1294 /********************************************************************
1295  * 2nd Level Abstractions
1296  ********************************************************************/
1297
1298 /* Go through a transport's transmitted list or the association's retransmit
1299  * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1300  * The retransmit list will not have an associated transport.
1301  *
1302  * I added coherent debug information output.   --xguo
1303  *
1304  * Instead of printing 'sacked' or 'kept' for each TSN on the
1305  * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1306  * KEPT TSN6-TSN7, etc.
1307  */
1308 static void sctp_check_transmitted(struct sctp_outq *q,
1309                                    struct list_head *transmitted_queue,
1310                                    struct sctp_transport *transport,
1311                                    union sctp_addr *saddr,
1312                                    struct sctp_sackhdr *sack,
1313                                    __u32 *highest_new_tsn_in_sack)
1314 {
1315         struct list_head *lchunk;
1316         struct sctp_chunk *tchunk;
1317         struct list_head tlist;
1318         __u32 tsn;
1319         __u32 sack_ctsn;
1320         __u32 rtt;
1321         __u8 restart_timer = 0;
1322         int bytes_acked = 0;
1323         int migrate_bytes = 0;
1324         bool forward_progress = false;
1325
1326         sack_ctsn = ntohl(sack->cum_tsn_ack);
1327
1328         INIT_LIST_HEAD(&tlist);
1329
1330         /* The while loop will skip empty transmitted queues. */
1331         while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1332                 tchunk = list_entry(lchunk, struct sctp_chunk,
1333                                     transmitted_list);
1334
1335                 if (sctp_chunk_abandoned(tchunk)) {
1336                         /* Move the chunk to abandoned list. */
1337                         sctp_insert_list(&q->abandoned, lchunk);
1338
1339                         /* If this chunk has not been acked, stop
1340                          * considering it as 'outstanding'.
1341                          */
1342                         if (!tchunk->tsn_gap_acked) {
1343                                 if (tchunk->transport)
1344                                         tchunk->transport->flight_size -=
1345                                                         sctp_data_size(tchunk);
1346                                 q->outstanding_bytes -= sctp_data_size(tchunk);
1347                         }
1348                         continue;
1349                 }
1350
1351                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1352                 if (sctp_acked(sack, tsn)) {
1353                         /* If this queue is the retransmit queue, the
1354                          * retransmit timer has already reclaimed
1355                          * the outstanding bytes for this chunk, so only
1356                          * count bytes associated with a transport.
1357                          */
1358                         if (transport) {
1359                                 /* If this chunk is being used for RTT
1360                                  * measurement, calculate the RTT and update
1361                                  * the RTO using this value.
1362                                  *
1363                                  * 6.3.1 C5) Karn's algorithm: RTT measurements
1364                                  * MUST NOT be made using packets that were
1365                                  * retransmitted (and thus for which it is
1366                                  * ambiguous whether the reply was for the
1367                                  * first instance of the packet or a later
1368                                  * instance).
1369                                  */
1370                                 if (!tchunk->tsn_gap_acked &&
1371                                     !tchunk->resent &&
1372                                     tchunk->rtt_in_progress) {
1373                                         tchunk->rtt_in_progress = 0;
1374                                         rtt = jiffies - tchunk->sent_at;
1375                                         sctp_transport_update_rto(transport,
1376                                                                   rtt);
1377                                 }
1378                         }
1379
1380                         /* If the chunk hasn't been marked as ACKED,
1381                          * mark it and account bytes_acked if the
1382                          * chunk had a valid transport (it will not
1383                          * have a transport if ASCONF had deleted it
1384                          * while DATA was outstanding).
1385                          */
1386                         if (!tchunk->tsn_gap_acked) {
1387                                 tchunk->tsn_gap_acked = 1;
1388                                 if (TSN_lt(*highest_new_tsn_in_sack, tsn))
1389                                         *highest_new_tsn_in_sack = tsn;
1390                                 bytes_acked += sctp_data_size(tchunk);
1391                                 if (!tchunk->transport)
1392                                         migrate_bytes += sctp_data_size(tchunk);
1393                                 forward_progress = true;
1394                         }
1395
1396                         if (TSN_lte(tsn, sack_ctsn)) {
1397                                 /* RFC 2960  6.3.2 Retransmission Timer Rules
1398                                  *
1399                                  * R3) Whenever a SACK is received
1400                                  * that acknowledges the DATA chunk
1401                                  * with the earliest outstanding TSN
1402                                  * for that address, restart T3-rtx
1403                                  * timer for that address with its
1404                                  * current RTO.
1405                                  */
1406                                 restart_timer = 1;
1407                                 forward_progress = true;
1408
1409                                 if (!tchunk->tsn_gap_acked) {
1410                                         /*
1411                                          * SFR-CACC algorithm:
1412                                          * 2) If the SACK contains gap acks
1413                                          * and the flag CHANGEOVER_ACTIVE is
1414                                          * set the receiver of the SACK MUST
1415                                          * take the following action:
1416                                          *
1417                                          * B) For each TSN t being acked that
1418                                          * has not been acked in any SACK so
1419                                          * far, set cacc_saw_newack to 1 for
1420                                          * the destination that the TSN was
1421                                          * sent to.
1422                                          */
1423                                         if (transport &&
1424                                             sack->num_gap_ack_blocks &&
1425                                             q->asoc->peer.primary_path->cacc.
1426                                             changeover_active)
1427                                                 transport->cacc.cacc_saw_newack
1428                                                         = 1;
1429                                 }
1430
1431                                 list_add_tail(&tchunk->transmitted_list,
1432                                               &q->sacked);
1433                         } else {
1434                                 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1435                                  * M2) Each time a SACK arrives reporting
1436                                  * 'Stray DATA chunk(s)' record the highest TSN
1437                                  * reported as newly acknowledged, call this
1438                                  * value 'HighestTSNinSack'. A newly
1439                                  * acknowledged DATA chunk is one not
1440                                  * previously acknowledged in a SACK.
1441                                  *
1442                                  * When the SCTP sender of data receives a SACK
1443                                  * chunk that acknowledges, for the first time,
1444                                  * the receipt of a DATA chunk, all the still
1445                                  * unacknowledged DATA chunks whose TSN is
1446                                  * older than that newly acknowledged DATA
1447                                  * chunk, are qualified as 'Stray DATA chunks'.
1448                                  */
1449                                 list_add_tail(lchunk, &tlist);
1450                         }
1451                 } else {
1452                         if (tchunk->tsn_gap_acked) {
1453                                 pr_debug("%s: receiver reneged on data TSN:0x%x\n",
1454                                          __func__, tsn);
1455
1456                                 tchunk->tsn_gap_acked = 0;
1457
1458                                 if (tchunk->transport)
1459                                         bytes_acked -= sctp_data_size(tchunk);
1460
1461                                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1462                                  *
1463                                  * R4) Whenever a SACK is received missing a
1464                                  * TSN that was previously acknowledged via a
1465                                  * Gap Ack Block, start T3-rtx for the
1466                                  * destination address to which the DATA
1467                                  * chunk was originally
1468                                  * transmitted if it is not already running.
1469                                  */
1470                                 restart_timer = 1;
1471                         }
1472
1473                         list_add_tail(lchunk, &tlist);
1474                 }
1475         }
1476
1477         if (transport) {
1478                 if (bytes_acked) {
1479                         struct sctp_association *asoc = transport->asoc;
1480
1481                         /* We may have counted DATA that was migrated
1482                          * to this transport due to DEL-IP operation.
1483                          * Subtract those bytes, since the were never
1484                          * send on this transport and shouldn't be
1485                          * credited to this transport.
1486                          */
1487                         bytes_acked -= migrate_bytes;
1488
1489                         /* 8.2. When an outstanding TSN is acknowledged,
1490                          * the endpoint shall clear the error counter of
1491                          * the destination transport address to which the
1492                          * DATA chunk was last sent.
1493                          * The association's overall error counter is
1494                          * also cleared.
1495                          */
1496                         transport->error_count = 0;
1497                         transport->asoc->overall_error_count = 0;
1498                         forward_progress = true;
1499
1500                         /*
1501                          * While in SHUTDOWN PENDING, we may have started
1502                          * the T5 shutdown guard timer after reaching the
1503                          * retransmission limit. Stop that timer as soon
1504                          * as the receiver acknowledged any data.
1505                          */
1506                         if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING &&
1507                             del_timer(&asoc->timers
1508                                 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]))
1509                                         sctp_association_put(asoc);
1510
1511                         /* Mark the destination transport address as
1512                          * active if it is not so marked.
1513                          */
1514                         if ((transport->state == SCTP_INACTIVE ||
1515                              transport->state == SCTP_UNCONFIRMED) &&
1516                             sctp_cmp_addr_exact(&transport->ipaddr, saddr)) {
1517                                 sctp_assoc_control_transport(
1518                                         transport->asoc,
1519                                         transport,
1520                                         SCTP_TRANSPORT_UP,
1521                                         SCTP_RECEIVED_SACK);
1522                         }
1523
1524                         sctp_transport_raise_cwnd(transport, sack_ctsn,
1525                                                   bytes_acked);
1526
1527                         transport->flight_size -= bytes_acked;
1528                         if (transport->flight_size == 0)
1529                                 transport->partial_bytes_acked = 0;
1530                         q->outstanding_bytes -= bytes_acked + migrate_bytes;
1531                 } else {
1532                         /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1533                          * When a sender is doing zero window probing, it
1534                          * should not timeout the association if it continues
1535                          * to receive new packets from the receiver. The
1536                          * reason is that the receiver MAY keep its window
1537                          * closed for an indefinite time.
1538                          * A sender is doing zero window probing when the
1539                          * receiver's advertised window is zero, and there is
1540                          * only one data chunk in flight to the receiver.
1541                          *
1542                          * Allow the association to timeout while in SHUTDOWN
1543                          * PENDING or SHUTDOWN RECEIVED in case the receiver
1544                          * stays in zero window mode forever.
1545                          */
1546                         if (!q->asoc->peer.rwnd &&
1547                             !list_empty(&tlist) &&
1548                             (sack_ctsn+2 == q->asoc->next_tsn) &&
1549                             q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) {
1550                                 pr_debug("%s: sack received for zero window "
1551                                          "probe:%u\n", __func__, sack_ctsn);
1552
1553                                 q->asoc->overall_error_count = 0;
1554                                 transport->error_count = 0;
1555                         }
1556                 }
1557
1558                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1559                  *
1560                  * R2) Whenever all outstanding data sent to an address have
1561                  * been acknowledged, turn off the T3-rtx timer of that
1562                  * address.
1563                  */
1564                 if (!transport->flight_size) {
1565                         if (del_timer(&transport->T3_rtx_timer))
1566                                 sctp_transport_put(transport);
1567                 } else if (restart_timer) {
1568                         if (!mod_timer(&transport->T3_rtx_timer,
1569                                        jiffies + transport->rto))
1570                                 sctp_transport_hold(transport);
1571                 }
1572
1573                 if (forward_progress) {
1574                         if (transport->dst)
1575                                 dst_confirm(transport->dst);
1576                 }
1577         }
1578
1579         list_splice(&tlist, transmitted_queue);
1580 }
1581
1582 /* Mark chunks as missing and consequently may get retransmitted. */
1583 static void sctp_mark_missing(struct sctp_outq *q,
1584                               struct list_head *transmitted_queue,
1585                               struct sctp_transport *transport,
1586                               __u32 highest_new_tsn_in_sack,
1587                               int count_of_newacks)
1588 {
1589         struct sctp_chunk *chunk;
1590         __u32 tsn;
1591         char do_fast_retransmit = 0;
1592         struct sctp_association *asoc = q->asoc;
1593         struct sctp_transport *primary = asoc->peer.primary_path;
1594
1595         list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1596
1597                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1598
1599                 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1600                  * 'Unacknowledged TSN's', if the TSN number of an
1601                  * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1602                  * value, increment the 'TSN.Missing.Report' count on that
1603                  * chunk if it has NOT been fast retransmitted or marked for
1604                  * fast retransmit already.
1605                  */
1606                 if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
1607                     !chunk->tsn_gap_acked &&
1608                     TSN_lt(tsn, highest_new_tsn_in_sack)) {
1609
1610                         /* SFR-CACC may require us to skip marking
1611                          * this chunk as missing.
1612                          */
1613                         if (!transport || !sctp_cacc_skip(primary,
1614                                                 chunk->transport,
1615                                                 count_of_newacks, tsn)) {
1616                                 chunk->tsn_missing_report++;
1617
1618                                 pr_debug("%s: tsn:0x%x missing counter:%d\n",
1619                                          __func__, tsn, chunk->tsn_missing_report);
1620                         }
1621                 }
1622                 /*
1623                  * M4) If any DATA chunk is found to have a
1624                  * 'TSN.Missing.Report'
1625                  * value larger than or equal to 3, mark that chunk for
1626                  * retransmission and start the fast retransmit procedure.
1627                  */
1628
1629                 if (chunk->tsn_missing_report >= 3) {
1630                         chunk->fast_retransmit = SCTP_NEED_FRTX;
1631                         do_fast_retransmit = 1;
1632                 }
1633         }
1634
1635         if (transport) {
1636                 if (do_fast_retransmit)
1637                         sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1638
1639                 pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, "
1640                          "flight_size:%d, pba:%d\n",  __func__, transport,
1641                          transport->cwnd, transport->ssthresh,
1642                          transport->flight_size, transport->partial_bytes_acked);
1643         }
1644 }
1645
1646 /* Is the given TSN acked by this packet?  */
1647 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1648 {
1649         int i;
1650         sctp_sack_variable_t *frags;
1651         __u16 gap;
1652         __u32 ctsn = ntohl(sack->cum_tsn_ack);
1653
1654         if (TSN_lte(tsn, ctsn))
1655                 goto pass;
1656
1657         /* 3.3.4 Selective Acknowledgement (SACK) (3):
1658          *
1659          * Gap Ack Blocks:
1660          *  These fields contain the Gap Ack Blocks. They are repeated
1661          *  for each Gap Ack Block up to the number of Gap Ack Blocks
1662          *  defined in the Number of Gap Ack Blocks field. All DATA
1663          *  chunks with TSNs greater than or equal to (Cumulative TSN
1664          *  Ack + Gap Ack Block Start) and less than or equal to
1665          *  (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1666          *  Block are assumed to have been received correctly.
1667          */
1668
1669         frags = sack->variable;
1670         gap = tsn - ctsn;
1671         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1672                 if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1673                     TSN_lte(gap, ntohs(frags[i].gab.end)))
1674                         goto pass;
1675         }
1676
1677         return 0;
1678 pass:
1679         return 1;
1680 }
1681
1682 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1683                                     int nskips, __be16 stream)
1684 {
1685         int i;
1686
1687         for (i = 0; i < nskips; i++) {
1688                 if (skiplist[i].stream == stream)
1689                         return i;
1690         }
1691         return i;
1692 }
1693
1694 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1695 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1696 {
1697         struct sctp_association *asoc = q->asoc;
1698         struct sctp_chunk *ftsn_chunk = NULL;
1699         struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1700         int nskips = 0;
1701         int skip_pos = 0;
1702         __u32 tsn;
1703         struct sctp_chunk *chunk;
1704         struct list_head *lchunk, *temp;
1705
1706         if (!asoc->peer.prsctp_capable)
1707                 return;
1708
1709         /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1710          * received SACK.
1711          *
1712          * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1713          * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1714          */
1715         if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1716                 asoc->adv_peer_ack_point = ctsn;
1717
1718         /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1719          * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1720          * the chunk next in the out-queue space is marked as "abandoned" as
1721          * shown in the following example:
1722          *
1723          * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1724          * and the Advanced.Peer.Ack.Point is updated to this value:
1725          *
1726          *   out-queue at the end of  ==>   out-queue after Adv.Ack.Point
1727          *   normal SACK processing           local advancement
1728          *                ...                           ...
1729          *   Adv.Ack.Pt-> 102 acked                     102 acked
1730          *                103 abandoned                 103 abandoned
1731          *                104 abandoned     Adv.Ack.P-> 104 abandoned
1732          *                105                           105
1733          *                106 acked                     106 acked
1734          *                ...                           ...
1735          *
1736          * In this example, the data sender successfully advanced the
1737          * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1738          */
1739         list_for_each_safe(lchunk, temp, &q->abandoned) {
1740                 chunk = list_entry(lchunk, struct sctp_chunk,
1741                                         transmitted_list);
1742                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1743
1744                 /* Remove any chunks in the abandoned queue that are acked by
1745                  * the ctsn.
1746                  */
1747                 if (TSN_lte(tsn, ctsn)) {
1748                         list_del_init(lchunk);
1749                         sctp_chunk_free(chunk);
1750                 } else {
1751                         if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1752                                 asoc->adv_peer_ack_point = tsn;
1753                                 if (chunk->chunk_hdr->flags &
1754                                          SCTP_DATA_UNORDERED)
1755                                         continue;
1756                                 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1757                                                 nskips,
1758                                                 chunk->subh.data_hdr->stream);
1759                                 ftsn_skip_arr[skip_pos].stream =
1760                                         chunk->subh.data_hdr->stream;
1761                                 ftsn_skip_arr[skip_pos].ssn =
1762                                          chunk->subh.data_hdr->ssn;
1763                                 if (skip_pos == nskips)
1764                                         nskips++;
1765                                 if (nskips == 10)
1766                                         break;
1767                         } else
1768                                 break;
1769                 }
1770         }
1771
1772         /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1773          * is greater than the Cumulative TSN ACK carried in the received
1774          * SACK, the data sender MUST send the data receiver a FORWARD TSN
1775          * chunk containing the latest value of the
1776          * "Advanced.Peer.Ack.Point".
1777          *
1778          * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1779          * list each stream and sequence number in the forwarded TSN. This
1780          * information will enable the receiver to easily find any
1781          * stranded TSN's waiting on stream reorder queues. Each stream
1782          * SHOULD only be reported once; this means that if multiple
1783          * abandoned messages occur in the same stream then only the
1784          * highest abandoned stream sequence number is reported. If the
1785          * total size of the FORWARD TSN does NOT fit in a single MTU then
1786          * the sender of the FORWARD TSN SHOULD lower the
1787          * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1788          * single MTU.
1789          */
1790         if (asoc->adv_peer_ack_point > ctsn)
1791                 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1792                                               nskips, &ftsn_skip_arr[0]);
1793
1794         if (ftsn_chunk) {
1795                 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1796                 SCTP_INC_STATS(sock_net(asoc->base.sk), SCTP_MIB_OUTCTRLCHUNKS);
1797         }
1798 }