41abfd17627e1a3b0a75029acf1f78031f1c8447
[linux-2.6.git] / net / sctp / outqueue.c
1 /* SCTP kernel reference Implementation
2  * (C) Copyright IBM Corp. 2001, 2004
3  * Copyright (c) 1999-2000 Cisco, Inc.
4  * Copyright (c) 1999-2001 Motorola, Inc.
5  * Copyright (c) 2001-2003 Intel Corp.
6  *
7  * This file is part of the SCTP kernel reference Implementation
8  *
9  * These functions implement the sctp_outq class.   The outqueue handles
10  * bundling and queueing of outgoing SCTP chunks.
11  *
12  * The SCTP reference implementation is free software;
13  * you can redistribute it and/or modify it under the terms of
14  * the GNU General Public License as published by
15  * the Free Software Foundation; either version 2, or (at your option)
16  * any later version.
17  *
18  * The SCTP reference implementation is distributed in the hope that it
19  * will be useful, but WITHOUT ANY WARRANTY; without even the implied
20  *                 ************************
21  * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
22  * See the GNU General Public License for more details.
23  *
24  * You should have received a copy of the GNU General Public License
25  * along with GNU CC; see the file COPYING.  If not, write to
26  * the Free Software Foundation, 59 Temple Place - Suite 330,
27  * Boston, MA 02111-1307, USA.
28  *
29  * Please send any bug reports or fixes you make to the
30  * email address(es):
31  *    lksctp developers <lksctp-developers@lists.sourceforge.net>
32  *
33  * Or submit a bug report through the following website:
34  *    http://www.sf.net/projects/lksctp
35  *
36  * Written or modified by:
37  *    La Monte H.P. Yarroll <piggy@acm.org>
38  *    Karl Knutson          <karl@athena.chicago.il.us>
39  *    Perry Melange         <pmelange@null.cc.uic.edu>
40  *    Xingang Guo           <xingang.guo@intel.com>
41  *    Hui Huang             <hui.huang@nokia.com>
42  *    Sridhar Samudrala     <sri@us.ibm.com>
43  *    Jon Grimm             <jgrimm@us.ibm.com>
44  *
45  * Any bugs reported given to us we will try to fix... any fixes shared will
46  * be incorporated into the next SCTP release.
47  */
48
49 #include <linux/types.h>
50 #include <linux/list.h>   /* For struct list_head */
51 #include <linux/socket.h>
52 #include <linux/ip.h>
53 #include <net/sock.h>     /* For skb_set_owner_w */
54
55 #include <net/sctp/sctp.h>
56 #include <net/sctp/sm.h>
57
58 /* Declare internal functions here.  */
59 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn);
60 static void sctp_check_transmitted(struct sctp_outq *q,
61                                    struct list_head *transmitted_queue,
62                                    struct sctp_transport *transport,
63                                    struct sctp_sackhdr *sack,
64                                    __u32 highest_new_tsn);
65
66 static void sctp_mark_missing(struct sctp_outq *q,
67                               struct list_head *transmitted_queue,
68                               struct sctp_transport *transport,
69                               __u32 highest_new_tsn,
70                               int count_of_newacks);
71
72 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn);
73
74 /* Add data to the front of the queue. */
75 static inline void sctp_outq_head_data(struct sctp_outq *q,
76                                         struct sctp_chunk *ch)
77 {
78         list_add(&ch->list, &q->out_chunk_list);
79         q->out_qlen += ch->skb->len;
80         return;
81 }
82
83 /* Take data from the front of the queue. */
84 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
85 {
86         struct sctp_chunk *ch = NULL;
87
88         if (!list_empty(&q->out_chunk_list)) {
89                 struct list_head *entry = q->out_chunk_list.next;
90
91                 ch = list_entry(entry, struct sctp_chunk, list);
92                 list_del_init(entry);
93                 q->out_qlen -= ch->skb->len;
94         }
95         return ch;
96 }
97 /* Add data chunk to the end of the queue. */
98 static inline void sctp_outq_tail_data(struct sctp_outq *q,
99                                        struct sctp_chunk *ch)
100 {
101         list_add_tail(&ch->list, &q->out_chunk_list);
102         q->out_qlen += ch->skb->len;
103         return;
104 }
105
106 /*
107  * SFR-CACC algorithm:
108  * D) If count_of_newacks is greater than or equal to 2
109  * and t was not sent to the current primary then the
110  * sender MUST NOT increment missing report count for t.
111  */
112 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary,
113                                        struct sctp_transport *transport,
114                                        int count_of_newacks)
115 {
116         if (count_of_newacks >=2 && transport != primary)
117                 return 1;
118         return 0;
119 }
120
121 /*
122  * SFR-CACC algorithm:
123  * F) If count_of_newacks is less than 2, let d be the
124  * destination to which t was sent. If cacc_saw_newack
125  * is 0 for destination d, then the sender MUST NOT
126  * increment missing report count for t.
127  */
128 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport,
129                                        int count_of_newacks)
130 {
131         if (count_of_newacks < 2 && !transport->cacc.cacc_saw_newack)
132                 return 1;
133         return 0;
134 }
135
136 /*
137  * SFR-CACC algorithm:
138  * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
139  * execute steps C, D, F.
140  *
141  * C has been implemented in sctp_outq_sack
142  */
143 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
144                                      struct sctp_transport *transport,
145                                      int count_of_newacks)
146 {
147         if (!primary->cacc.cycling_changeover) {
148                 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
149                         return 1;
150                 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
151                         return 1;
152                 return 0;
153         }
154         return 0;
155 }
156
157 /*
158  * SFR-CACC algorithm:
159  * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
160  * than next_tsn_at_change of the current primary, then
161  * the sender MUST NOT increment missing report count
162  * for t.
163  */
164 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
165 {
166         if (primary->cacc.cycling_changeover &&
167             TSN_lt(tsn, primary->cacc.next_tsn_at_change))
168                 return 1;
169         return 0;
170 }
171
172 /*
173  * SFR-CACC algorithm:
174  * 3) If the missing report count for TSN t is to be
175  * incremented according to [RFC2960] and
176  * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
177  * then the sender MUST futher execute steps 3.1 and
178  * 3.2 to determine if the missing report count for
179  * TSN t SHOULD NOT be incremented.
180  *
181  * 3.3) If 3.1 and 3.2 do not dictate that the missing
182  * report count for t should not be incremented, then
183  * the sender SOULD increment missing report count for
184  * t (according to [RFC2960] and [SCTP_STEWART_2002]).
185  */
186 static inline int sctp_cacc_skip(struct sctp_transport *primary,
187                                  struct sctp_transport *transport,
188                                  int count_of_newacks,
189                                  __u32 tsn)
190 {
191         if (primary->cacc.changeover_active &&
192             (sctp_cacc_skip_3_1(primary, transport, count_of_newacks)
193              || sctp_cacc_skip_3_2(primary, tsn)))
194                 return 1;
195         return 0;
196 }
197
198 /* Initialize an existing sctp_outq.  This does the boring stuff.
199  * You still need to define handlers if you really want to DO
200  * something with this structure...
201  */
202 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
203 {
204         q->asoc = asoc;
205         INIT_LIST_HEAD(&q->out_chunk_list);
206         INIT_LIST_HEAD(&q->control_chunk_list);
207         INIT_LIST_HEAD(&q->retransmit);
208         INIT_LIST_HEAD(&q->sacked);
209         INIT_LIST_HEAD(&q->abandoned);
210
211         q->outstanding_bytes = 0;
212         q->empty = 1;
213         q->cork  = 0;
214
215         q->malloced = 0;
216         q->out_qlen = 0;
217 }
218
219 /* Free the outqueue structure and any related pending chunks.
220  */
221 void sctp_outq_teardown(struct sctp_outq *q)
222 {
223         struct sctp_transport *transport;
224         struct list_head *lchunk, *pos, *temp;
225         struct sctp_chunk *chunk, *tmp;
226
227         /* Throw away unacknowledged chunks. */
228         list_for_each(pos, &q->asoc->peer.transport_addr_list) {
229                 transport = list_entry(pos, struct sctp_transport, transports);
230                 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
231                         chunk = list_entry(lchunk, struct sctp_chunk,
232                                            transmitted_list);
233                         /* Mark as part of a failed message. */
234                         sctp_chunk_fail(chunk, q->error);
235                         sctp_chunk_free(chunk);
236                 }
237         }
238
239         /* Throw away chunks that have been gap ACKed.  */
240         list_for_each_safe(lchunk, temp, &q->sacked) {
241                 list_del_init(lchunk);
242                 chunk = list_entry(lchunk, struct sctp_chunk,
243                                    transmitted_list);
244                 sctp_chunk_fail(chunk, q->error);
245                 sctp_chunk_free(chunk);
246         }
247
248         /* Throw away any chunks in the retransmit queue. */
249         list_for_each_safe(lchunk, temp, &q->retransmit) {
250                 list_del_init(lchunk);
251                 chunk = list_entry(lchunk, struct sctp_chunk,
252                                    transmitted_list);
253                 sctp_chunk_fail(chunk, q->error);
254                 sctp_chunk_free(chunk);
255         }
256
257         /* Throw away any chunks that are in the abandoned queue. */
258         list_for_each_safe(lchunk, temp, &q->abandoned) {
259                 list_del_init(lchunk);
260                 chunk = list_entry(lchunk, struct sctp_chunk,
261                                    transmitted_list);
262                 sctp_chunk_fail(chunk, q->error);
263                 sctp_chunk_free(chunk);
264         }
265
266         /* Throw away any leftover data chunks. */
267         while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
268
269                 /* Mark as send failure. */
270                 sctp_chunk_fail(chunk, q->error);
271                 sctp_chunk_free(chunk);
272         }
273
274         q->error = 0;
275
276         /* Throw away any leftover control chunks. */
277         list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
278                 list_del_init(&chunk->list);
279                 sctp_chunk_free(chunk);
280         }
281 }
282
283 /* Free the outqueue structure and any related pending chunks.  */
284 void sctp_outq_free(struct sctp_outq *q)
285 {
286         /* Throw away leftover chunks. */
287         sctp_outq_teardown(q);
288
289         /* If we were kmalloc()'d, free the memory.  */
290         if (q->malloced)
291                 kfree(q);
292 }
293
294 /* Put a new chunk in an sctp_outq.  */
295 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk)
296 {
297         int error = 0;
298
299         SCTP_DEBUG_PRINTK("sctp_outq_tail(%p, %p[%s])\n",
300                           q, chunk, chunk && chunk->chunk_hdr ?
301                           sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
302                           : "Illegal Chunk");
303
304         /* If it is data, queue it up, otherwise, send it
305          * immediately.
306          */
307         if (SCTP_CID_DATA == chunk->chunk_hdr->type) {
308                 /* Is it OK to queue data chunks?  */
309                 /* From 9. Termination of Association
310                  *
311                  * When either endpoint performs a shutdown, the
312                  * association on each peer will stop accepting new
313                  * data from its user and only deliver data in queue
314                  * at the time of sending or receiving the SHUTDOWN
315                  * chunk.
316                  */
317                 switch (q->asoc->state) {
318                 case SCTP_STATE_EMPTY:
319                 case SCTP_STATE_CLOSED:
320                 case SCTP_STATE_SHUTDOWN_PENDING:
321                 case SCTP_STATE_SHUTDOWN_SENT:
322                 case SCTP_STATE_SHUTDOWN_RECEIVED:
323                 case SCTP_STATE_SHUTDOWN_ACK_SENT:
324                         /* Cannot send after transport endpoint shutdown */
325                         error = -ESHUTDOWN;
326                         break;
327
328                 default:
329                         SCTP_DEBUG_PRINTK("outqueueing (%p, %p[%s])\n",
330                           q, chunk, chunk && chunk->chunk_hdr ?
331                           sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
332                           : "Illegal Chunk");
333
334                         sctp_outq_tail_data(q, chunk);
335                         if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
336                                 SCTP_INC_STATS(SCTP_MIB_OUTUNORDERCHUNKS);
337                         else
338                                 SCTP_INC_STATS(SCTP_MIB_OUTORDERCHUNKS);
339                         q->empty = 0;
340                         break;
341                 };
342         } else {
343                 list_add_tail(&chunk->list, &q->control_chunk_list);
344                 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
345         }
346
347         if (error < 0)
348                 return error;
349
350         if (!q->cork)
351                 error = sctp_outq_flush(q, 0);
352
353         return error;
354 }
355
356 /* Insert a chunk into the sorted list based on the TSNs.  The retransmit list
357  * and the abandoned list are in ascending order.
358  */
359 static void sctp_insert_list(struct list_head *head, struct list_head *new)
360 {
361         struct list_head *pos;
362         struct sctp_chunk *nchunk, *lchunk;
363         __u32 ntsn, ltsn;
364         int done = 0;
365
366         nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
367         ntsn = ntohl(nchunk->subh.data_hdr->tsn);
368
369         list_for_each(pos, head) {
370                 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
371                 ltsn = ntohl(lchunk->subh.data_hdr->tsn);
372                 if (TSN_lt(ntsn, ltsn)) {
373                         list_add(new, pos->prev);
374                         done = 1;
375                         break;
376                 }
377         }
378         if (!done)
379                 list_add_tail(new, head);
380 }
381
382 /* Mark all the eligible packets on a transport for retransmission.  */
383 void sctp_retransmit_mark(struct sctp_outq *q,
384                           struct sctp_transport *transport,
385                           __u8 fast_retransmit)
386 {
387         struct list_head *lchunk, *ltemp;
388         struct sctp_chunk *chunk;
389
390         /* Walk through the specified transmitted queue.  */
391         list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
392                 chunk = list_entry(lchunk, struct sctp_chunk,
393                                    transmitted_list);
394
395                 /* If the chunk is abandoned, move it to abandoned list. */
396                 if (sctp_chunk_abandoned(chunk)) {
397                         list_del_init(lchunk);
398                         sctp_insert_list(&q->abandoned, lchunk);
399
400                         /* If this chunk has not been previousely acked,
401                          * stop considering it 'outstanding'.  Our peer
402                          * will most likely never see it since it will
403                          * not be retransmitted
404                          */
405                         if (!chunk->tsn_gap_acked) {
406                                 chunk->transport->flight_size -=
407                                                 sctp_data_size(chunk);
408                                 q->outstanding_bytes -= sctp_data_size(chunk);
409                                 q->asoc->peer.rwnd += (sctp_data_size(chunk) +
410                                                         sizeof(struct sk_buff));
411                         }
412                         continue;
413                 }
414
415                 /* If we are doing retransmission due to a fast retransmit,
416                  * only the chunk's that are marked for fast retransmit
417                  * should be added to the retransmit queue.  If we are doing
418                  * retransmission due to a timeout or pmtu discovery, only the
419                  * chunks that are not yet acked should be added to the
420                  * retransmit queue.
421                  */
422                 if ((fast_retransmit && (chunk->fast_retransmit > 0)) ||
423                    (!fast_retransmit && !chunk->tsn_gap_acked)) {
424                         /* RFC 2960 6.2.1 Processing a Received SACK
425                          *
426                          * C) Any time a DATA chunk is marked for
427                          * retransmission (via either T3-rtx timer expiration
428                          * (Section 6.3.3) or via fast retransmit
429                          * (Section 7.2.4)), add the data size of those
430                          * chunks to the rwnd.
431                          */
432                         q->asoc->peer.rwnd += (sctp_data_size(chunk) +
433                                                 sizeof(struct sk_buff));
434                         q->outstanding_bytes -= sctp_data_size(chunk);
435                         transport->flight_size -= sctp_data_size(chunk);
436
437                         /* sctpimpguide-05 Section 2.8.2
438                          * M5) If a T3-rtx timer expires, the
439                          * 'TSN.Missing.Report' of all affected TSNs is set
440                          * to 0.
441                          */
442                         chunk->tsn_missing_report = 0;
443
444                         /* If a chunk that is being used for RTT measurement
445                          * has to be retransmitted, we cannot use this chunk
446                          * anymore for RTT measurements. Reset rto_pending so
447                          * that a new RTT measurement is started when a new
448                          * data chunk is sent.
449                          */
450                         if (chunk->rtt_in_progress) {
451                                 chunk->rtt_in_progress = 0;
452                                 transport->rto_pending = 0;
453                         }
454
455                         /* Move the chunk to the retransmit queue. The chunks
456                          * on the retransmit queue are always kept in order.
457                          */
458                         list_del_init(lchunk);
459                         sctp_insert_list(&q->retransmit, lchunk);
460                 }
461         }
462
463         SCTP_DEBUG_PRINTK("%s: transport: %p, fast_retransmit: %d, "
464                           "cwnd: %d, ssthresh: %d, flight_size: %d, "
465                           "pba: %d\n", __FUNCTION__,
466                           transport, fast_retransmit,
467                           transport->cwnd, transport->ssthresh,
468                           transport->flight_size,
469                           transport->partial_bytes_acked);
470
471 }
472
473 /* Mark all the eligible packets on a transport for retransmission and force
474  * one packet out.
475  */
476 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
477                      sctp_retransmit_reason_t reason)
478 {
479         int error = 0;
480         __u8 fast_retransmit = 0;
481
482         switch(reason) {
483         case SCTP_RTXR_T3_RTX:
484                 SCTP_INC_STATS(SCTP_MIB_T3_RETRANSMITS);
485                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
486                 /* Update the retran path if the T3-rtx timer has expired for
487                  * the current retran path.
488                  */
489                 if (transport == transport->asoc->peer.retran_path)
490                         sctp_assoc_update_retran_path(transport->asoc);
491                 break;
492         case SCTP_RTXR_FAST_RTX:
493                 SCTP_INC_STATS(SCTP_MIB_FAST_RETRANSMITS);
494                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
495                 fast_retransmit = 1;
496                 break;
497         case SCTP_RTXR_PMTUD:
498                 SCTP_INC_STATS(SCTP_MIB_PMTUD_RETRANSMITS);
499                 break;
500         default:
501                 BUG();
502         }
503
504         sctp_retransmit_mark(q, transport, fast_retransmit);
505
506         /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
507          * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
508          * following the procedures outlined in C1 - C5.
509          */
510         sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
511
512         error = sctp_outq_flush(q, /* rtx_timeout */ 1);
513
514         if (error)
515                 q->asoc->base.sk->sk_err = -error;
516 }
517
518 /*
519  * Transmit DATA chunks on the retransmit queue.  Upon return from
520  * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
521  * need to be transmitted by the caller.
522  * We assume that pkt->transport has already been set.
523  *
524  * The return value is a normal kernel error return value.
525  */
526 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
527                                int rtx_timeout, int *start_timer)
528 {
529         struct list_head *lqueue;
530         struct list_head *lchunk, *lchunk1;
531         struct sctp_transport *transport = pkt->transport;
532         sctp_xmit_t status;
533         struct sctp_chunk *chunk, *chunk1;
534         struct sctp_association *asoc;
535         int error = 0;
536
537         asoc = q->asoc;
538         lqueue = &q->retransmit;
539
540         /* RFC 2960 6.3.3 Handle T3-rtx Expiration
541          *
542          * E3) Determine how many of the earliest (i.e., lowest TSN)
543          * outstanding DATA chunks for the address for which the
544          * T3-rtx has expired will fit into a single packet, subject
545          * to the MTU constraint for the path corresponding to the
546          * destination transport address to which the retransmission
547          * is being sent (this may be different from the address for
548          * which the timer expires [see Section 6.4]). Call this value
549          * K. Bundle and retransmit those K DATA chunks in a single
550          * packet to the destination endpoint.
551          *
552          * [Just to be painfully clear, if we are retransmitting
553          * because a timeout just happened, we should send only ONE
554          * packet of retransmitted data.]
555          */
556         lchunk = sctp_list_dequeue(lqueue);
557
558         while (lchunk) {
559                 chunk = list_entry(lchunk, struct sctp_chunk,
560                                    transmitted_list);
561
562                 /* Make sure that Gap Acked TSNs are not retransmitted.  A
563                  * simple approach is just to move such TSNs out of the
564                  * way and into a 'transmitted' queue and skip to the
565                  * next chunk.
566                  */
567                 if (chunk->tsn_gap_acked) {
568                         list_add_tail(lchunk, &transport->transmitted);
569                         lchunk = sctp_list_dequeue(lqueue);
570                         continue;
571                 }
572
573                 /* Attempt to append this chunk to the packet. */
574                 status = sctp_packet_append_chunk(pkt, chunk);
575
576                 switch (status) {
577                 case SCTP_XMIT_PMTU_FULL:
578                         /* Send this packet.  */
579                         if ((error = sctp_packet_transmit(pkt)) == 0)
580                                 *start_timer = 1;
581
582                         /* If we are retransmitting, we should only
583                          * send a single packet.
584                          */
585                         if (rtx_timeout) {
586                                 list_add(lchunk, lqueue);
587                                 lchunk = NULL;
588                         }
589
590                         /* Bundle lchunk in the next round.  */
591                         break;
592
593                 case SCTP_XMIT_RWND_FULL:
594                         /* Send this packet. */
595                         if ((error = sctp_packet_transmit(pkt)) == 0)
596                                 *start_timer = 1;
597
598                         /* Stop sending DATA as there is no more room
599                          * at the receiver.
600                          */
601                         list_add(lchunk, lqueue);
602                         lchunk = NULL;
603                         break;
604
605                 case SCTP_XMIT_NAGLE_DELAY:
606                         /* Send this packet. */
607                         if ((error = sctp_packet_transmit(pkt)) == 0)
608                                 *start_timer = 1;
609
610                         /* Stop sending DATA because of nagle delay. */
611                         list_add(lchunk, lqueue);
612                         lchunk = NULL;
613                         break;
614
615                 default:
616                         /* The append was successful, so add this chunk to
617                          * the transmitted list.
618                          */
619                         list_add_tail(lchunk, &transport->transmitted);
620
621                         /* Mark the chunk as ineligible for fast retransmit
622                          * after it is retransmitted.
623                          */
624                         if (chunk->fast_retransmit > 0)
625                                 chunk->fast_retransmit = -1;
626
627                         *start_timer = 1;
628                         q->empty = 0;
629
630                         /* Retrieve a new chunk to bundle. */
631                         lchunk = sctp_list_dequeue(lqueue);
632                         break;
633                 };
634
635                 /* If we are here due to a retransmit timeout or a fast
636                  * retransmit and if there are any chunks left in the retransmit
637                  * queue that could not fit in the PMTU sized packet, they need                  * to be marked as ineligible for a subsequent fast retransmit.
638                  */
639                 if (rtx_timeout && !lchunk) {
640                         list_for_each(lchunk1, lqueue) {
641                                 chunk1 = list_entry(lchunk1, struct sctp_chunk,
642                                                     transmitted_list);
643                                 if (chunk1->fast_retransmit > 0)
644                                         chunk1->fast_retransmit = -1;
645                         }
646                 }
647         }
648
649         return error;
650 }
651
652 /* Cork the outqueue so queued chunks are really queued. */
653 int sctp_outq_uncork(struct sctp_outq *q)
654 {
655         int error = 0;
656         if (q->cork) {
657                 q->cork = 0;
658                 error = sctp_outq_flush(q, 0);
659         }
660         return error;
661 }
662
663 /*
664  * Try to flush an outqueue.
665  *
666  * Description: Send everything in q which we legally can, subject to
667  * congestion limitations.
668  * * Note: This function can be called from multiple contexts so appropriate
669  * locking concerns must be made.  Today we use the sock lock to protect
670  * this function.
671  */
672 int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
673 {
674         struct sctp_packet *packet;
675         struct sctp_packet singleton;
676         struct sctp_association *asoc = q->asoc;
677         __u16 sport = asoc->base.bind_addr.port;
678         __u16 dport = asoc->peer.port;
679         __u32 vtag = asoc->peer.i.init_tag;
680         struct sctp_transport *transport = NULL;
681         struct sctp_transport *new_transport;
682         struct sctp_chunk *chunk, *tmp;
683         sctp_xmit_t status;
684         int error = 0;
685         int start_timer = 0;
686
687         /* These transports have chunks to send. */
688         struct list_head transport_list;
689         struct list_head *ltransport;
690
691         INIT_LIST_HEAD(&transport_list);
692         packet = NULL;
693
694         /*
695          * 6.10 Bundling
696          *   ...
697          *   When bundling control chunks with DATA chunks, an
698          *   endpoint MUST place control chunks first in the outbound
699          *   SCTP packet.  The transmitter MUST transmit DATA chunks
700          *   within a SCTP packet in increasing order of TSN.
701          *   ...
702          */
703
704         list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
705                 list_del_init(&chunk->list);
706
707                 /* Pick the right transport to use. */
708                 new_transport = chunk->transport;
709
710                 if (!new_transport) {
711                         new_transport = asoc->peer.active_path;
712                 } else if ((new_transport->state == SCTP_INACTIVE) ||
713                            (new_transport->state == SCTP_UNCONFIRMED)) {
714                         /* If the chunk is Heartbeat or Heartbeat Ack,
715                          * send it to chunk->transport, even if it's
716                          * inactive.
717                          *
718                          * 3.3.6 Heartbeat Acknowledgement:
719                          * ...
720                          * A HEARTBEAT ACK is always sent to the source IP
721                          * address of the IP datagram containing the
722                          * HEARTBEAT chunk to which this ack is responding.
723                          * ...
724                          */
725                         if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
726                             chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK)
727                                 new_transport = asoc->peer.active_path;
728                 }
729
730                 /* Are we switching transports?
731                  * Take care of transport locks.
732                  */
733                 if (new_transport != transport) {
734                         transport = new_transport;
735                         if (list_empty(&transport->send_ready)) {
736                                 list_add_tail(&transport->send_ready,
737                                               &transport_list);
738                         }
739                         packet = &transport->packet;
740                         sctp_packet_config(packet, vtag,
741                                            asoc->peer.ecn_capable);
742                 }
743
744                 switch (chunk->chunk_hdr->type) {
745                 /*
746                  * 6.10 Bundling
747                  *   ...
748                  *   An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
749                  *   COMPLETE with any other chunks.  [Send them immediately.]
750                  */
751                 case SCTP_CID_INIT:
752                 case SCTP_CID_INIT_ACK:
753                 case SCTP_CID_SHUTDOWN_COMPLETE:
754                         sctp_packet_init(&singleton, transport, sport, dport);
755                         sctp_packet_config(&singleton, vtag, 0);
756                         sctp_packet_append_chunk(&singleton, chunk);
757                         error = sctp_packet_transmit(&singleton);
758                         if (error < 0)
759                                 return error;
760                         break;
761
762                 case SCTP_CID_ABORT:
763                 case SCTP_CID_SACK:
764                 case SCTP_CID_HEARTBEAT:
765                 case SCTP_CID_HEARTBEAT_ACK:
766                 case SCTP_CID_SHUTDOWN:
767                 case SCTP_CID_SHUTDOWN_ACK:
768                 case SCTP_CID_ERROR:
769                 case SCTP_CID_COOKIE_ECHO:
770                 case SCTP_CID_COOKIE_ACK:
771                 case SCTP_CID_ECN_ECNE:
772                 case SCTP_CID_ECN_CWR:
773                 case SCTP_CID_ASCONF:
774                 case SCTP_CID_ASCONF_ACK:
775                 case SCTP_CID_FWD_TSN:
776                         sctp_packet_transmit_chunk(packet, chunk);
777                         break;
778
779                 default:
780                         /* We built a chunk with an illegal type! */
781                         BUG();
782                 };
783         }
784
785         /* Is it OK to send data chunks?  */
786         switch (asoc->state) {
787         case SCTP_STATE_COOKIE_ECHOED:
788                 /* Only allow bundling when this packet has a COOKIE-ECHO
789                  * chunk.
790                  */
791                 if (!packet || !packet->has_cookie_echo)
792                         break;
793
794                 /* fallthru */
795         case SCTP_STATE_ESTABLISHED:
796         case SCTP_STATE_SHUTDOWN_PENDING:
797         case SCTP_STATE_SHUTDOWN_RECEIVED:
798                 /*
799                  * RFC 2960 6.1  Transmission of DATA Chunks
800                  *
801                  * C) When the time comes for the sender to transmit,
802                  * before sending new DATA chunks, the sender MUST
803                  * first transmit any outstanding DATA chunks which
804                  * are marked for retransmission (limited by the
805                  * current cwnd).
806                  */
807                 if (!list_empty(&q->retransmit)) {
808                         if (transport == asoc->peer.retran_path)
809                                 goto retran;
810
811                         /* Switch transports & prepare the packet.  */
812
813                         transport = asoc->peer.retran_path;
814
815                         if (list_empty(&transport->send_ready)) {
816                                 list_add_tail(&transport->send_ready,
817                                               &transport_list);
818                         }
819
820                         packet = &transport->packet;
821                         sctp_packet_config(packet, vtag,
822                                            asoc->peer.ecn_capable);
823                 retran:
824                         error = sctp_outq_flush_rtx(q, packet,
825                                                     rtx_timeout, &start_timer);
826
827                         if (start_timer)
828                                 sctp_transport_reset_timers(transport);
829
830                         /* This can happen on COOKIE-ECHO resend.  Only
831                          * one chunk can get bundled with a COOKIE-ECHO.
832                          */
833                         if (packet->has_cookie_echo)
834                                 goto sctp_flush_out;
835
836                         /* Don't send new data if there is still data
837                          * waiting to retransmit.
838                          */
839                         if (!list_empty(&q->retransmit))
840                                 goto sctp_flush_out;
841                 }
842
843                 /* Finally, transmit new packets.  */
844                 start_timer = 0;
845                 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
846                         /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
847                          * stream identifier.
848                          */
849                         if (chunk->sinfo.sinfo_stream >=
850                             asoc->c.sinit_num_ostreams) {
851
852                                 /* Mark as failed send. */
853                                 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
854                                 sctp_chunk_free(chunk);
855                                 continue;
856                         }
857
858                         /* Has this chunk expired? */
859                         if (sctp_chunk_abandoned(chunk)) {
860                                 sctp_chunk_fail(chunk, 0);
861                                 sctp_chunk_free(chunk);
862                                 continue;
863                         }
864
865                         /* If there is a specified transport, use it.
866                          * Otherwise, we want to use the active path.
867                          */
868                         new_transport = chunk->transport;
869                         if (!new_transport ||
870                             ((new_transport->state == SCTP_INACTIVE) ||
871                              (new_transport->state == SCTP_UNCONFIRMED)))
872                                 new_transport = asoc->peer.active_path;
873
874                         /* Change packets if necessary.  */
875                         if (new_transport != transport) {
876                                 transport = new_transport;
877
878                                 /* Schedule to have this transport's
879                                  * packet flushed.
880                                  */
881                                 if (list_empty(&transport->send_ready)) {
882                                         list_add_tail(&transport->send_ready,
883                                                       &transport_list);
884                                 }
885
886                                 packet = &transport->packet;
887                                 sctp_packet_config(packet, vtag,
888                                                    asoc->peer.ecn_capable);
889                         }
890
891                         SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
892                                           q, chunk,
893                                           chunk && chunk->chunk_hdr ?
894                                           sctp_cname(SCTP_ST_CHUNK(
895                                                   chunk->chunk_hdr->type))
896                                           : "Illegal Chunk");
897
898                         SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
899                                         "%p skb->users %d.\n",
900                                         ntohl(chunk->subh.data_hdr->tsn),
901                                         chunk->skb ?chunk->skb->head : NULL,
902                                         chunk->skb ?
903                                         atomic_read(&chunk->skb->users) : -1);
904
905                         /* Add the chunk to the packet.  */
906                         status = sctp_packet_transmit_chunk(packet, chunk);
907
908                         switch (status) {
909                         case SCTP_XMIT_PMTU_FULL:
910                         case SCTP_XMIT_RWND_FULL:
911                         case SCTP_XMIT_NAGLE_DELAY:
912                                 /* We could not append this chunk, so put
913                                  * the chunk back on the output queue.
914                                  */
915                                 SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
916                                         "not transmit TSN: 0x%x, status: %d\n",
917                                         ntohl(chunk->subh.data_hdr->tsn),
918                                         status);
919                                 sctp_outq_head_data(q, chunk);
920                                 goto sctp_flush_out;
921                                 break;
922
923                         case SCTP_XMIT_OK:
924                                 break;
925
926                         default:
927                                 BUG();
928                         }
929
930                         /* BUG: We assume that the sctp_packet_transmit()
931                          * call below will succeed all the time and add the
932                          * chunk to the transmitted list and restart the
933                          * timers.
934                          * It is possible that the call can fail under OOM
935                          * conditions.
936                          *
937                          * Is this really a problem?  Won't this behave
938                          * like a lost TSN?
939                          */
940                         list_add_tail(&chunk->transmitted_list,
941                                       &transport->transmitted);
942
943                         sctp_transport_reset_timers(transport);
944
945                         q->empty = 0;
946
947                         /* Only let one DATA chunk get bundled with a
948                          * COOKIE-ECHO chunk.
949                          */
950                         if (packet->has_cookie_echo)
951                                 goto sctp_flush_out;
952                 }
953                 break;
954
955         default:
956                 /* Do nothing.  */
957                 break;
958         }
959
960 sctp_flush_out:
961
962         /* Before returning, examine all the transports touched in
963          * this call.  Right now, we bluntly force clear all the
964          * transports.  Things might change after we implement Nagle.
965          * But such an examination is still required.
966          *
967          * --xguo
968          */
969         while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
970                 struct sctp_transport *t = list_entry(ltransport,
971                                                       struct sctp_transport,
972                                                       send_ready);
973                 packet = &t->packet;
974                 if (!sctp_packet_empty(packet))
975                         error = sctp_packet_transmit(packet);
976         }
977
978         return error;
979 }
980
981 /* Update unack_data based on the incoming SACK chunk */
982 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
983                                         struct sctp_sackhdr *sack)
984 {
985         sctp_sack_variable_t *frags;
986         __u16 unack_data;
987         int i;
988
989         unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
990
991         frags = sack->variable;
992         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
993                 unack_data -= ((ntohs(frags[i].gab.end) -
994                                 ntohs(frags[i].gab.start) + 1));
995         }
996
997         assoc->unack_data = unack_data;
998 }
999
1000 /* Return the highest new tsn that is acknowledged by the given SACK chunk. */
1001 static __u32 sctp_highest_new_tsn(struct sctp_sackhdr *sack,
1002                                   struct sctp_association *asoc)
1003 {
1004         struct list_head *ltransport, *lchunk;
1005         struct sctp_transport *transport;
1006         struct sctp_chunk *chunk;
1007         __u32 highest_new_tsn, tsn;
1008         struct list_head *transport_list = &asoc->peer.transport_addr_list;
1009
1010         highest_new_tsn = ntohl(sack->cum_tsn_ack);
1011
1012         list_for_each(ltransport, transport_list) {
1013                 transport = list_entry(ltransport, struct sctp_transport,
1014                                        transports);
1015                 list_for_each(lchunk, &transport->transmitted) {
1016                         chunk = list_entry(lchunk, struct sctp_chunk,
1017                                            transmitted_list);
1018                         tsn = ntohl(chunk->subh.data_hdr->tsn);
1019
1020                         if (!chunk->tsn_gap_acked &&
1021                             TSN_lt(highest_new_tsn, tsn) &&
1022                             sctp_acked(sack, tsn))
1023                                 highest_new_tsn = tsn;
1024                 }
1025         }
1026
1027         return highest_new_tsn;
1028 }
1029
1030 /* This is where we REALLY process a SACK.
1031  *
1032  * Process the SACK against the outqueue.  Mostly, this just frees
1033  * things off the transmitted queue.
1034  */
1035 int sctp_outq_sack(struct sctp_outq *q, struct sctp_sackhdr *sack)
1036 {
1037         struct sctp_association *asoc = q->asoc;
1038         struct sctp_transport *transport;
1039         struct sctp_chunk *tchunk = NULL;
1040         struct list_head *lchunk, *transport_list, *pos, *temp;
1041         sctp_sack_variable_t *frags = sack->variable;
1042         __u32 sack_ctsn, ctsn, tsn;
1043         __u32 highest_tsn, highest_new_tsn;
1044         __u32 sack_a_rwnd;
1045         unsigned outstanding;
1046         struct sctp_transport *primary = asoc->peer.primary_path;
1047         int count_of_newacks = 0;
1048
1049         /* Grab the association's destination address list. */
1050         transport_list = &asoc->peer.transport_addr_list;
1051
1052         sack_ctsn = ntohl(sack->cum_tsn_ack);
1053
1054         /*
1055          * SFR-CACC algorithm:
1056          * On receipt of a SACK the sender SHOULD execute the
1057          * following statements.
1058          *
1059          * 1) If the cumulative ack in the SACK passes next tsn_at_change
1060          * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1061          * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1062          * all destinations.
1063          */
1064         if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1065                 primary->cacc.changeover_active = 0;
1066                 list_for_each(pos, transport_list) {
1067                         transport = list_entry(pos, struct sctp_transport,
1068                                         transports);
1069                         transport->cacc.cycling_changeover = 0;
1070                 }
1071         }
1072
1073         /*
1074          * SFR-CACC algorithm:
1075          * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1076          * is set the receiver of the SACK MUST take the following actions:
1077          *
1078          * A) Initialize the cacc_saw_newack to 0 for all destination
1079          * addresses.
1080          */
1081         if (sack->num_gap_ack_blocks &&
1082             primary->cacc.changeover_active) {
1083                 list_for_each(pos, transport_list) {
1084                         transport = list_entry(pos, struct sctp_transport,
1085                                         transports);
1086                         transport->cacc.cacc_saw_newack = 0;
1087                 }
1088         }
1089
1090         /* Get the highest TSN in the sack. */
1091         highest_tsn = sack_ctsn;
1092         if (sack->num_gap_ack_blocks)
1093                 highest_tsn +=
1094                     ntohs(frags[ntohs(sack->num_gap_ack_blocks) - 1].gab.end);
1095
1096         if (TSN_lt(asoc->highest_sacked, highest_tsn)) {
1097                 highest_new_tsn = highest_tsn;
1098                 asoc->highest_sacked = highest_tsn;
1099         } else {
1100                 highest_new_tsn = sctp_highest_new_tsn(sack, asoc);
1101         }
1102
1103         /* Run through the retransmit queue.  Credit bytes received
1104          * and free those chunks that we can.
1105          */
1106         sctp_check_transmitted(q, &q->retransmit, NULL, sack, highest_new_tsn);
1107         sctp_mark_missing(q, &q->retransmit, NULL, highest_new_tsn, 0);
1108
1109         /* Run through the transmitted queue.
1110          * Credit bytes received and free those chunks which we can.
1111          *
1112          * This is a MASSIVE candidate for optimization.
1113          */
1114         list_for_each(pos, transport_list) {
1115                 transport  = list_entry(pos, struct sctp_transport,
1116                                         transports);
1117                 sctp_check_transmitted(q, &transport->transmitted,
1118                                        transport, sack, highest_new_tsn);
1119                 /*
1120                  * SFR-CACC algorithm:
1121                  * C) Let count_of_newacks be the number of
1122                  * destinations for which cacc_saw_newack is set.
1123                  */
1124                 if (transport->cacc.cacc_saw_newack)
1125                         count_of_newacks ++;
1126         }
1127
1128         list_for_each(pos, transport_list) {
1129                 transport  = list_entry(pos, struct sctp_transport,
1130                                         transports);
1131                 sctp_mark_missing(q, &transport->transmitted, transport,
1132                                   highest_new_tsn, count_of_newacks);
1133         }
1134
1135         /* Move the Cumulative TSN Ack Point if appropriate.  */
1136         if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn))
1137                 asoc->ctsn_ack_point = sack_ctsn;
1138
1139         /* Update unack_data field in the assoc. */
1140         sctp_sack_update_unack_data(asoc, sack);
1141
1142         ctsn = asoc->ctsn_ack_point;
1143
1144         /* Throw away stuff rotting on the sack queue.  */
1145         list_for_each_safe(lchunk, temp, &q->sacked) {
1146                 tchunk = list_entry(lchunk, struct sctp_chunk,
1147                                     transmitted_list);
1148                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1149                 if (TSN_lte(tsn, ctsn))
1150                         sctp_chunk_free(tchunk);
1151         }
1152
1153         /* ii) Set rwnd equal to the newly received a_rwnd minus the
1154          *     number of bytes still outstanding after processing the
1155          *     Cumulative TSN Ack and the Gap Ack Blocks.
1156          */
1157
1158         sack_a_rwnd = ntohl(sack->a_rwnd);
1159         outstanding = q->outstanding_bytes;
1160
1161         if (outstanding < sack_a_rwnd)
1162                 sack_a_rwnd -= outstanding;
1163         else
1164                 sack_a_rwnd = 0;
1165
1166         asoc->peer.rwnd = sack_a_rwnd;
1167
1168         sctp_generate_fwdtsn(q, sack_ctsn);
1169
1170         SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n",
1171                           __FUNCTION__, sack_ctsn);
1172         SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, "
1173                           "%p is 0x%x. Adv peer ack point: 0x%x\n",
1174                           __FUNCTION__, asoc, ctsn, asoc->adv_peer_ack_point);
1175
1176         /* See if all chunks are acked.
1177          * Make sure the empty queue handler will get run later.
1178          */
1179         q->empty = (list_empty(&q->out_chunk_list) &&
1180                     list_empty(&q->control_chunk_list) &&
1181                     list_empty(&q->retransmit));
1182         if (!q->empty)
1183                 goto finish;
1184
1185         list_for_each(pos, transport_list) {
1186                 transport  = list_entry(pos, struct sctp_transport,
1187                                         transports);
1188                 q->empty = q->empty && list_empty(&transport->transmitted);
1189                 if (!q->empty)
1190                         goto finish;
1191         }
1192
1193         SCTP_DEBUG_PRINTK("sack queue is empty.\n");
1194 finish:
1195         return q->empty;
1196 }
1197
1198 /* Is the outqueue empty?  */
1199 int sctp_outq_is_empty(const struct sctp_outq *q)
1200 {
1201         return q->empty;
1202 }
1203
1204 /********************************************************************
1205  * 2nd Level Abstractions
1206  ********************************************************************/
1207
1208 /* Go through a transport's transmitted list or the association's retransmit
1209  * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1210  * The retransmit list will not have an associated transport.
1211  *
1212  * I added coherent debug information output.   --xguo
1213  *
1214  * Instead of printing 'sacked' or 'kept' for each TSN on the
1215  * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1216  * KEPT TSN6-TSN7, etc.
1217  */
1218 static void sctp_check_transmitted(struct sctp_outq *q,
1219                                    struct list_head *transmitted_queue,
1220                                    struct sctp_transport *transport,
1221                                    struct sctp_sackhdr *sack,
1222                                    __u32 highest_new_tsn_in_sack)
1223 {
1224         struct list_head *lchunk;
1225         struct sctp_chunk *tchunk;
1226         struct list_head tlist;
1227         __u32 tsn;
1228         __u32 sack_ctsn;
1229         __u32 rtt;
1230         __u8 restart_timer = 0;
1231         int bytes_acked = 0;
1232
1233         /* These state variables are for coherent debug output. --xguo */
1234
1235 #if SCTP_DEBUG
1236         __u32 dbg_ack_tsn = 0;  /* An ACKed TSN range starts here... */
1237         __u32 dbg_last_ack_tsn = 0;  /* ...and finishes here.        */
1238         __u32 dbg_kept_tsn = 0; /* An un-ACKed range starts here...  */
1239         __u32 dbg_last_kept_tsn = 0; /* ...and finishes here.        */
1240
1241         /* 0 : The last TSN was ACKed.
1242          * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1243          * -1: We need to initialize.
1244          */
1245         int dbg_prt_state = -1;
1246 #endif /* SCTP_DEBUG */
1247
1248         sack_ctsn = ntohl(sack->cum_tsn_ack);
1249
1250         INIT_LIST_HEAD(&tlist);
1251
1252         /* The while loop will skip empty transmitted queues. */
1253         while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1254                 tchunk = list_entry(lchunk, struct sctp_chunk,
1255                                     transmitted_list);
1256
1257                 if (sctp_chunk_abandoned(tchunk)) {
1258                         /* Move the chunk to abandoned list. */
1259                         sctp_insert_list(&q->abandoned, lchunk);
1260
1261                         /* If this chunk has not been acked, stop
1262                          * considering it as 'outstanding'.
1263                          */
1264                         if (!tchunk->tsn_gap_acked) {
1265                                 tchunk->transport->flight_size -=
1266                                                 sctp_data_size(tchunk);
1267                                 q->outstanding_bytes -= sctp_data_size(tchunk);
1268                         }
1269                         continue;
1270                 }
1271
1272                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1273                 if (sctp_acked(sack, tsn)) {
1274                         /* If this queue is the retransmit queue, the
1275                          * retransmit timer has already reclaimed
1276                          * the outstanding bytes for this chunk, so only
1277                          * count bytes associated with a transport.
1278                          */
1279                         if (transport) {
1280                                 /* If this chunk is being used for RTT
1281                                  * measurement, calculate the RTT and update
1282                                  * the RTO using this value.
1283                                  *
1284                                  * 6.3.1 C5) Karn's algorithm: RTT measurements
1285                                  * MUST NOT be made using packets that were
1286                                  * retransmitted (and thus for which it is
1287                                  * ambiguous whether the reply was for the
1288                                  * first instance of the packet or a later
1289                                  * instance).
1290                                  */
1291                                 if (!tchunk->tsn_gap_acked &&
1292                                     !tchunk->resent &&
1293                                     tchunk->rtt_in_progress) {
1294                                         tchunk->rtt_in_progress = 0;
1295                                         rtt = jiffies - tchunk->sent_at;
1296                                         sctp_transport_update_rto(transport,
1297                                                                   rtt);
1298                                 }
1299                         }
1300                         if (TSN_lte(tsn, sack_ctsn)) {
1301                                 /* RFC 2960  6.3.2 Retransmission Timer Rules
1302                                  *
1303                                  * R3) Whenever a SACK is received
1304                                  * that acknowledges the DATA chunk
1305                                  * with the earliest outstanding TSN
1306                                  * for that address, restart T3-rtx
1307                                  * timer for that address with its
1308                                  * current RTO.
1309                                  */
1310                                 restart_timer = 1;
1311
1312                                 if (!tchunk->tsn_gap_acked) {
1313                                         tchunk->tsn_gap_acked = 1;
1314                                         bytes_acked += sctp_data_size(tchunk);
1315                                         /*
1316                                          * SFR-CACC algorithm:
1317                                          * 2) If the SACK contains gap acks
1318                                          * and the flag CHANGEOVER_ACTIVE is
1319                                          * set the receiver of the SACK MUST
1320                                          * take the following action:
1321                                          *
1322                                          * B) For each TSN t being acked that
1323                                          * has not been acked in any SACK so
1324                                          * far, set cacc_saw_newack to 1 for
1325                                          * the destination that the TSN was
1326                                          * sent to.
1327                                          */
1328                                         if (transport &&
1329                                             sack->num_gap_ack_blocks &&
1330                                             q->asoc->peer.primary_path->cacc.
1331                                             changeover_active)
1332                                                 transport->cacc.cacc_saw_newack
1333                                                         = 1;
1334                                 }
1335
1336                                 list_add_tail(&tchunk->transmitted_list,
1337                                               &q->sacked);
1338                         } else {
1339                                 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1340                                  * M2) Each time a SACK arrives reporting
1341                                  * 'Stray DATA chunk(s)' record the highest TSN
1342                                  * reported as newly acknowledged, call this
1343                                  * value 'HighestTSNinSack'. A newly
1344                                  * acknowledged DATA chunk is one not
1345                                  * previously acknowledged in a SACK.
1346                                  *
1347                                  * When the SCTP sender of data receives a SACK
1348                                  * chunk that acknowledges, for the first time,
1349                                  * the receipt of a DATA chunk, all the still
1350                                  * unacknowledged DATA chunks whose TSN is
1351                                  * older than that newly acknowledged DATA
1352                                  * chunk, are qualified as 'Stray DATA chunks'.
1353                                  */
1354                                 if (!tchunk->tsn_gap_acked) {
1355                                         tchunk->tsn_gap_acked = 1;
1356                                         bytes_acked += sctp_data_size(tchunk);
1357                                 }
1358                                 list_add_tail(lchunk, &tlist);
1359                         }
1360
1361 #if SCTP_DEBUG
1362                         switch (dbg_prt_state) {
1363                         case 0: /* last TSN was ACKed */
1364                                 if (dbg_last_ack_tsn + 1 == tsn) {
1365                                         /* This TSN belongs to the
1366                                          * current ACK range.
1367                                          */
1368                                         break;
1369                                 }
1370
1371                                 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1372                                         /* Display the end of the
1373                                          * current range.
1374                                          */
1375                                         SCTP_DEBUG_PRINTK("-%08x",
1376                                                           dbg_last_ack_tsn);
1377                                 }
1378
1379                                 /* Start a new range.  */
1380                                 SCTP_DEBUG_PRINTK(",%08x", tsn);
1381                                 dbg_ack_tsn = tsn;
1382                                 break;
1383
1384                         case 1: /* The last TSN was NOT ACKed. */
1385                                 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1386                                         /* Display the end of current range. */
1387                                         SCTP_DEBUG_PRINTK("-%08x",
1388                                                           dbg_last_kept_tsn);
1389                                 }
1390
1391                                 SCTP_DEBUG_PRINTK("\n");
1392
1393                                 /* FALL THROUGH... */
1394                         default:
1395                                 /* This is the first-ever TSN we examined.  */
1396                                 /* Start a new range of ACK-ed TSNs.  */
1397                                 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn);
1398                                 dbg_prt_state = 0;
1399                                 dbg_ack_tsn = tsn;
1400                         };
1401
1402                         dbg_last_ack_tsn = tsn;
1403 #endif /* SCTP_DEBUG */
1404
1405                 } else {
1406                         if (tchunk->tsn_gap_acked) {
1407                                 SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1408                                                   "data TSN: 0x%x\n",
1409                                                   __FUNCTION__,
1410                                                   tsn);
1411                                 tchunk->tsn_gap_acked = 0;
1412
1413                                 bytes_acked -= sctp_data_size(tchunk);
1414
1415                                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1416                                  *
1417                                  * R4) Whenever a SACK is received missing a
1418                                  * TSN that was previously acknowledged via a
1419                                  * Gap Ack Block, start T3-rtx for the
1420                                  * destination address to which the DATA
1421                                  * chunk was originally
1422                                  * transmitted if it is not already running.
1423                                  */
1424                                 restart_timer = 1;
1425                         }
1426
1427                         list_add_tail(lchunk, &tlist);
1428
1429 #if SCTP_DEBUG
1430                         /* See the above comments on ACK-ed TSNs. */
1431                         switch (dbg_prt_state) {
1432                         case 1:
1433                                 if (dbg_last_kept_tsn + 1 == tsn)
1434                                         break;
1435
1436                                 if (dbg_last_kept_tsn != dbg_kept_tsn)
1437                                         SCTP_DEBUG_PRINTK("-%08x",
1438                                                           dbg_last_kept_tsn);
1439
1440                                 SCTP_DEBUG_PRINTK(",%08x", tsn);
1441                                 dbg_kept_tsn = tsn;
1442                                 break;
1443
1444                         case 0:
1445                                 if (dbg_last_ack_tsn != dbg_ack_tsn)
1446                                         SCTP_DEBUG_PRINTK("-%08x",
1447                                                           dbg_last_ack_tsn);
1448                                 SCTP_DEBUG_PRINTK("\n");
1449
1450                                 /* FALL THROUGH... */
1451                         default:
1452                                 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn);
1453                                 dbg_prt_state = 1;
1454                                 dbg_kept_tsn = tsn;
1455                         };
1456
1457                         dbg_last_kept_tsn = tsn;
1458 #endif /* SCTP_DEBUG */
1459                 }
1460         }
1461
1462 #if SCTP_DEBUG
1463         /* Finish off the last range, displaying its ending TSN.  */
1464         switch (dbg_prt_state) {
1465         case 0:
1466                 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1467                         SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_ack_tsn);
1468                 } else {
1469                         SCTP_DEBUG_PRINTK("\n");
1470                 }
1471         break;
1472
1473         case 1:
1474                 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1475                         SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_kept_tsn);
1476                 } else {
1477                         SCTP_DEBUG_PRINTK("\n");
1478                 }
1479         };
1480 #endif /* SCTP_DEBUG */
1481         if (transport) {
1482                 if (bytes_acked) {
1483                         /* 8.2. When an outstanding TSN is acknowledged,
1484                          * the endpoint shall clear the error counter of
1485                          * the destination transport address to which the
1486                          * DATA chunk was last sent.
1487                          * The association's overall error counter is
1488                          * also cleared.
1489                          */
1490                         transport->error_count = 0;
1491                         transport->asoc->overall_error_count = 0;
1492
1493                         /* Mark the destination transport address as
1494                          * active if it is not so marked.
1495                          */
1496                         if ((transport->state == SCTP_INACTIVE) ||
1497                             (transport->state == SCTP_UNCONFIRMED)) {
1498                                 sctp_assoc_control_transport(
1499                                         transport->asoc,
1500                                         transport,
1501                                         SCTP_TRANSPORT_UP,
1502                                         SCTP_RECEIVED_SACK);
1503                         }
1504
1505                         sctp_transport_raise_cwnd(transport, sack_ctsn,
1506                                                   bytes_acked);
1507
1508                         transport->flight_size -= bytes_acked;
1509                         q->outstanding_bytes -= bytes_acked;
1510                 } else {
1511                         /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1512                          * When a sender is doing zero window probing, it
1513                          * should not timeout the association if it continues
1514                          * to receive new packets from the receiver. The
1515                          * reason is that the receiver MAY keep its window
1516                          * closed for an indefinite time.
1517                          * A sender is doing zero window probing when the
1518                          * receiver's advertised window is zero, and there is
1519                          * only one data chunk in flight to the receiver.
1520                          */
1521                         if (!q->asoc->peer.rwnd &&
1522                             !list_empty(&tlist) &&
1523                             (sack_ctsn+2 == q->asoc->next_tsn)) {
1524                                 SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1525                                                   "window probe: %u\n",
1526                                                   __FUNCTION__, sack_ctsn);
1527                                 q->asoc->overall_error_count = 0;
1528                                 transport->error_count = 0;
1529                         }
1530                 }
1531
1532                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1533                  *
1534                  * R2) Whenever all outstanding data sent to an address have
1535                  * been acknowledged, turn off the T3-rtx timer of that
1536                  * address.
1537                  */
1538                 if (!transport->flight_size) {
1539                         if (timer_pending(&transport->T3_rtx_timer) &&
1540                             del_timer(&transport->T3_rtx_timer)) {
1541                                 sctp_transport_put(transport);
1542                         }
1543                 } else if (restart_timer) {
1544                         if (!mod_timer(&transport->T3_rtx_timer,
1545                                        jiffies + transport->rto))
1546                                 sctp_transport_hold(transport);
1547                 }
1548         }
1549
1550         list_splice(&tlist, transmitted_queue);
1551 }
1552
1553 /* Mark chunks as missing and consequently may get retransmitted. */
1554 static void sctp_mark_missing(struct sctp_outq *q,
1555                               struct list_head *transmitted_queue,
1556                               struct sctp_transport *transport,
1557                               __u32 highest_new_tsn_in_sack,
1558                               int count_of_newacks)
1559 {
1560         struct sctp_chunk *chunk;
1561         struct list_head *pos;
1562         __u32 tsn;
1563         char do_fast_retransmit = 0;
1564         struct sctp_transport *primary = q->asoc->peer.primary_path;
1565
1566         list_for_each(pos, transmitted_queue) {
1567
1568                 chunk = list_entry(pos, struct sctp_chunk, transmitted_list);
1569                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1570
1571                 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1572                  * 'Unacknowledged TSN's', if the TSN number of an
1573                  * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1574                  * value, increment the 'TSN.Missing.Report' count on that
1575                  * chunk if it has NOT been fast retransmitted or marked for
1576                  * fast retransmit already.
1577                  */
1578                 if (!chunk->fast_retransmit &&
1579                     !chunk->tsn_gap_acked &&
1580                     TSN_lt(tsn, highest_new_tsn_in_sack)) {
1581
1582                         /* SFR-CACC may require us to skip marking
1583                          * this chunk as missing.
1584                          */
1585                         if (!transport || !sctp_cacc_skip(primary, transport,
1586                                             count_of_newacks, tsn)) {
1587                                 chunk->tsn_missing_report++;
1588
1589                                 SCTP_DEBUG_PRINTK(
1590                                         "%s: TSN 0x%x missing counter: %d\n",
1591                                         __FUNCTION__, tsn,
1592                                         chunk->tsn_missing_report);
1593                         }
1594                 }
1595                 /*
1596                  * M4) If any DATA chunk is found to have a
1597                  * 'TSN.Missing.Report'
1598                  * value larger than or equal to 3, mark that chunk for
1599                  * retransmission and start the fast retransmit procedure.
1600                  */
1601
1602                 if (chunk->tsn_missing_report >= 3) {
1603                         chunk->fast_retransmit = 1;
1604                         do_fast_retransmit = 1;
1605                 }
1606         }
1607
1608         if (transport) {
1609                 if (do_fast_retransmit)
1610                         sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1611
1612                 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1613                                   "ssthresh: %d, flight_size: %d, pba: %d\n",
1614                                   __FUNCTION__, transport, transport->cwnd,
1615                                   transport->ssthresh, transport->flight_size,
1616                                   transport->partial_bytes_acked);
1617         }
1618 }
1619
1620 /* Is the given TSN acked by this packet?  */
1621 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1622 {
1623         int i;
1624         sctp_sack_variable_t *frags;
1625         __u16 gap;
1626         __u32 ctsn = ntohl(sack->cum_tsn_ack);
1627
1628         if (TSN_lte(tsn, ctsn))
1629                 goto pass;
1630
1631         /* 3.3.4 Selective Acknowledgement (SACK) (3):
1632          *
1633          * Gap Ack Blocks:
1634          *  These fields contain the Gap Ack Blocks. They are repeated
1635          *  for each Gap Ack Block up to the number of Gap Ack Blocks
1636          *  defined in the Number of Gap Ack Blocks field. All DATA
1637          *  chunks with TSNs greater than or equal to (Cumulative TSN
1638          *  Ack + Gap Ack Block Start) and less than or equal to
1639          *  (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1640          *  Block are assumed to have been received correctly.
1641          */
1642
1643         frags = sack->variable;
1644         gap = tsn - ctsn;
1645         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1646                 if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1647                     TSN_lte(gap, ntohs(frags[i].gab.end)))
1648                         goto pass;
1649         }
1650
1651         return 0;
1652 pass:
1653         return 1;
1654 }
1655
1656 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1657                                     int nskips, __be16 stream)
1658 {
1659         int i;
1660
1661         for (i = 0; i < nskips; i++) {
1662                 if (skiplist[i].stream == stream)
1663                         return i;
1664         }
1665         return i;
1666 }
1667
1668 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1669 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1670 {
1671         struct sctp_association *asoc = q->asoc;
1672         struct sctp_chunk *ftsn_chunk = NULL;
1673         struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1674         int nskips = 0;
1675         int skip_pos = 0;
1676         __u32 tsn;
1677         struct sctp_chunk *chunk;
1678         struct list_head *lchunk, *temp;
1679
1680         /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1681          * received SACK.
1682          *
1683          * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1684          * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1685          */
1686         if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1687                 asoc->adv_peer_ack_point = ctsn;
1688
1689         /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1690          * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1691          * the chunk next in the out-queue space is marked as "abandoned" as
1692          * shown in the following example:
1693          *
1694          * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1695          * and the Advanced.Peer.Ack.Point is updated to this value:
1696          *
1697          *   out-queue at the end of  ==>   out-queue after Adv.Ack.Point
1698          *   normal SACK processing           local advancement
1699          *                ...                           ...
1700          *   Adv.Ack.Pt-> 102 acked                     102 acked
1701          *                103 abandoned                 103 abandoned
1702          *                104 abandoned     Adv.Ack.P-> 104 abandoned
1703          *                105                           105
1704          *                106 acked                     106 acked
1705          *                ...                           ...
1706          *
1707          * In this example, the data sender successfully advanced the
1708          * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1709          */
1710         list_for_each_safe(lchunk, temp, &q->abandoned) {
1711                 chunk = list_entry(lchunk, struct sctp_chunk,
1712                                         transmitted_list);
1713                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1714
1715                 /* Remove any chunks in the abandoned queue that are acked by
1716                  * the ctsn.
1717                  */
1718                 if (TSN_lte(tsn, ctsn)) {
1719                         list_del_init(lchunk);
1720                         sctp_chunk_free(chunk);
1721                 } else {
1722                         if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1723                                 asoc->adv_peer_ack_point = tsn;
1724                                 if (chunk->chunk_hdr->flags &
1725                                          SCTP_DATA_UNORDERED)
1726                                         continue;
1727                                 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1728                                                 nskips,
1729                                                 chunk->subh.data_hdr->stream);
1730                                 ftsn_skip_arr[skip_pos].stream =
1731                                         chunk->subh.data_hdr->stream;
1732                                 ftsn_skip_arr[skip_pos].ssn =
1733                                          chunk->subh.data_hdr->ssn;
1734                                 if (skip_pos == nskips)
1735                                         nskips++;
1736                                 if (nskips == 10)
1737                                         break;
1738                         } else
1739                                 break;
1740                 }
1741         }
1742
1743         /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1744          * is greater than the Cumulative TSN ACK carried in the received
1745          * SACK, the data sender MUST send the data receiver a FORWARD TSN
1746          * chunk containing the latest value of the
1747          * "Advanced.Peer.Ack.Point".
1748          *
1749          * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1750          * list each stream and sequence number in the forwarded TSN. This
1751          * information will enable the receiver to easily find any
1752          * stranded TSN's waiting on stream reorder queues. Each stream
1753          * SHOULD only be reported once; this means that if multiple
1754          * abandoned messages occur in the same stream then only the
1755          * highest abandoned stream sequence number is reported. If the
1756          * total size of the FORWARD TSN does NOT fit in a single MTU then
1757          * the sender of the FORWARD TSN SHOULD lower the
1758          * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1759          * single MTU.
1760          */
1761         if (asoc->adv_peer_ack_point > ctsn)
1762                 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1763                                               nskips, &ftsn_skip_arr[0]);
1764
1765         if (ftsn_chunk) {
1766                 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1767                 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
1768         }
1769 }