Merge head 'drm-fixes' of master.kernel.org:/pub/scm/linux/kernel/git/airlied/drm-2.6
[linux-2.6] / 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                         continue;
400                 }
401
402                 /* If we are doing retransmission due to a fast retransmit,
403                  * only the chunk's that are marked for fast retransmit
404                  * should be added to the retransmit queue.  If we are doing
405                  * retransmission due to a timeout or pmtu discovery, only the
406                  * chunks that are not yet acked should be added to the
407                  * retransmit queue.
408                  */
409                 if ((fast_retransmit && chunk->fast_retransmit) ||
410                    (!fast_retransmit && !chunk->tsn_gap_acked)) {
411                         /* RFC 2960 6.2.1 Processing a Received SACK
412                          *
413                          * C) Any time a DATA chunk is marked for
414                          * retransmission (via either T3-rtx timer expiration
415                          * (Section 6.3.3) or via fast retransmit
416                          * (Section 7.2.4)), add the data size of those
417                          * chunks to the rwnd.
418                          */
419                         q->asoc->peer.rwnd += sctp_data_size(chunk);
420                         q->outstanding_bytes -= sctp_data_size(chunk);
421                         transport->flight_size -= sctp_data_size(chunk);
422
423                         /* sctpimpguide-05 Section 2.8.2
424                          * M5) If a T3-rtx timer expires, the
425                          * 'TSN.Missing.Report' of all affected TSNs is set
426                          * to 0.
427                          */
428                         chunk->tsn_missing_report = 0;
429
430                         /* If a chunk that is being used for RTT measurement
431                          * has to be retransmitted, we cannot use this chunk
432                          * anymore for RTT measurements. Reset rto_pending so
433                          * that a new RTT measurement is started when a new
434                          * data chunk is sent.
435                          */
436                         if (chunk->rtt_in_progress) {
437                                 chunk->rtt_in_progress = 0;
438                                 transport->rto_pending = 0;
439                         }
440
441                         /* Move the chunk to the retransmit queue. The chunks
442                          * on the retransmit queue are always kept in order.
443                          */
444                         list_del_init(lchunk);
445                         sctp_insert_list(&q->retransmit, lchunk);
446                 }
447         }
448
449         SCTP_DEBUG_PRINTK("%s: transport: %p, fast_retransmit: %d, "
450                           "cwnd: %d, ssthresh: %d, flight_size: %d, "
451                           "pba: %d\n", __FUNCTION__,
452                           transport, fast_retransmit,
453                           transport->cwnd, transport->ssthresh,
454                           transport->flight_size,
455                           transport->partial_bytes_acked);
456
457 }
458
459 /* Mark all the eligible packets on a transport for retransmission and force
460  * one packet out.
461  */
462 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
463                      sctp_retransmit_reason_t reason)
464 {
465         int error = 0;
466         __u8 fast_retransmit = 0;
467
468         switch(reason) {
469         case SCTP_RTXR_T3_RTX:
470                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
471                 /* Update the retran path if the T3-rtx timer has expired for
472                  * the current retran path.
473                  */
474                 if (transport == transport->asoc->peer.retran_path)
475                         sctp_assoc_update_retran_path(transport->asoc);
476                 break;
477         case SCTP_RTXR_FAST_RTX:
478                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
479                 fast_retransmit = 1;
480                 break;
481         case SCTP_RTXR_PMTUD:
482         default:
483                 break;
484         }
485
486         sctp_retransmit_mark(q, transport, fast_retransmit);
487
488         /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
489          * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
490          * following the procedures outlined in C1 - C5.
491          */
492         sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
493
494         error = sctp_outq_flush(q, /* rtx_timeout */ 1);
495
496         if (error)
497                 q->asoc->base.sk->sk_err = -error;
498 }
499
500 /*
501  * Transmit DATA chunks on the retransmit queue.  Upon return from
502  * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
503  * need to be transmitted by the caller.
504  * We assume that pkt->transport has already been set.
505  *
506  * The return value is a normal kernel error return value.
507  */
508 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
509                                int rtx_timeout, int *start_timer)
510 {
511         struct list_head *lqueue;
512         struct list_head *lchunk, *lchunk1;
513         struct sctp_transport *transport = pkt->transport;
514         sctp_xmit_t status;
515         struct sctp_chunk *chunk, *chunk1;
516         struct sctp_association *asoc;
517         int error = 0;
518
519         asoc = q->asoc;
520         lqueue = &q->retransmit;
521
522         /* RFC 2960 6.3.3 Handle T3-rtx Expiration
523          *
524          * E3) Determine how many of the earliest (i.e., lowest TSN)
525          * outstanding DATA chunks for the address for which the
526          * T3-rtx has expired will fit into a single packet, subject
527          * to the MTU constraint for the path corresponding to the
528          * destination transport address to which the retransmission
529          * is being sent (this may be different from the address for
530          * which the timer expires [see Section 6.4]). Call this value
531          * K. Bundle and retransmit those K DATA chunks in a single
532          * packet to the destination endpoint.
533          *
534          * [Just to be painfully clear, if we are retransmitting
535          * because a timeout just happened, we should send only ONE
536          * packet of retransmitted data.]
537          */
538         lchunk = sctp_list_dequeue(lqueue);
539
540         while (lchunk) {
541                 chunk = list_entry(lchunk, struct sctp_chunk,
542                                    transmitted_list);
543
544                 /* Make sure that Gap Acked TSNs are not retransmitted.  A
545                  * simple approach is just to move such TSNs out of the
546                  * way and into a 'transmitted' queue and skip to the
547                  * next chunk.
548                  */
549                 if (chunk->tsn_gap_acked) {
550                         list_add_tail(lchunk, &transport->transmitted);
551                         lchunk = sctp_list_dequeue(lqueue);
552                         continue;
553                 }
554
555                 /* Attempt to append this chunk to the packet. */
556                 status = sctp_packet_append_chunk(pkt, chunk);
557
558                 switch (status) {
559                 case SCTP_XMIT_PMTU_FULL:
560                         /* Send this packet.  */
561                         if ((error = sctp_packet_transmit(pkt)) == 0)
562                                 *start_timer = 1;
563
564                         /* If we are retransmitting, we should only
565                          * send a single packet.
566                          */
567                         if (rtx_timeout) {
568                                 list_add(lchunk, lqueue);
569                                 lchunk = NULL;
570                         }
571
572                         /* Bundle lchunk in the next round.  */
573                         break;
574
575                 case SCTP_XMIT_RWND_FULL:
576                         /* Send this packet. */
577                         if ((error = sctp_packet_transmit(pkt)) == 0)
578                                 *start_timer = 1;
579
580                         /* Stop sending DATA as there is no more room
581                          * at the receiver.
582                          */
583                         list_add(lchunk, lqueue);
584                         lchunk = NULL;
585                         break;
586
587                 case SCTP_XMIT_NAGLE_DELAY:
588                         /* Send this packet. */
589                         if ((error = sctp_packet_transmit(pkt)) == 0)
590                                 *start_timer = 1;
591
592                         /* Stop sending DATA because of nagle delay. */
593                         list_add(lchunk, lqueue);
594                         lchunk = NULL;
595                         break;
596
597                 default:
598                         /* The append was successful, so add this chunk to
599                          * the transmitted list.
600                          */
601                         list_add_tail(lchunk, &transport->transmitted);
602
603                         /* Mark the chunk as ineligible for fast retransmit 
604                          * after it is retransmitted.
605                          */
606                         chunk->fast_retransmit = 0;
607
608                         *start_timer = 1;
609                         q->empty = 0;
610
611                         /* Retrieve a new chunk to bundle. */
612                         lchunk = sctp_list_dequeue(lqueue);
613                         break;
614                 };
615
616                 /* If we are here due to a retransmit timeout or a fast
617                  * retransmit and if there are any chunks left in the retransmit
618                  * queue that could not fit in the PMTU sized packet, they need                  * to be marked as ineligible for a subsequent fast retransmit.
619                  */
620                 if (rtx_timeout && !lchunk) {
621                         list_for_each(lchunk1, lqueue) {
622                                 chunk1 = list_entry(lchunk1, struct sctp_chunk,
623                                                     transmitted_list);
624                                 chunk1->fast_retransmit = 0;
625                         }
626                 }
627         }
628
629         return error;
630 }
631
632 /* Cork the outqueue so queued chunks are really queued. */
633 int sctp_outq_uncork(struct sctp_outq *q)
634 {
635         int error = 0;
636         if (q->cork) {
637                 q->cork = 0;
638                 error = sctp_outq_flush(q, 0);
639         }
640         return error;
641 }
642
643 /*
644  * Try to flush an outqueue.
645  *
646  * Description: Send everything in q which we legally can, subject to
647  * congestion limitations.
648  * * Note: This function can be called from multiple contexts so appropriate
649  * locking concerns must be made.  Today we use the sock lock to protect
650  * this function.
651  */
652 int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
653 {
654         struct sctp_packet *packet;
655         struct sctp_packet singleton;
656         struct sctp_association *asoc = q->asoc;
657         __u16 sport = asoc->base.bind_addr.port;
658         __u16 dport = asoc->peer.port;
659         __u32 vtag = asoc->peer.i.init_tag;
660         struct sctp_transport *transport = NULL;
661         struct sctp_transport *new_transport;
662         struct sctp_chunk *chunk, *tmp;
663         sctp_xmit_t status;
664         int error = 0;
665         int start_timer = 0;
666
667         /* These transports have chunks to send. */
668         struct list_head transport_list;
669         struct list_head *ltransport;
670
671         INIT_LIST_HEAD(&transport_list);
672         packet = NULL;
673
674         /*
675          * 6.10 Bundling
676          *   ...
677          *   When bundling control chunks with DATA chunks, an
678          *   endpoint MUST place control chunks first in the outbound
679          *   SCTP packet.  The transmitter MUST transmit DATA chunks
680          *   within a SCTP packet in increasing order of TSN.
681          *   ...
682          */
683
684         list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
685                 list_del_init(&chunk->list);
686
687                 /* Pick the right transport to use. */
688                 new_transport = chunk->transport;
689
690                 if (!new_transport) {
691                         new_transport = asoc->peer.active_path;
692                 } else if (new_transport->state == SCTP_INACTIVE) {
693                         /* If the chunk is Heartbeat or Heartbeat Ack,
694                          * send it to chunk->transport, even if it's
695                          * inactive.
696                          *
697                          * 3.3.6 Heartbeat Acknowledgement:
698                          * ...  
699                          * A HEARTBEAT ACK is always sent to the source IP
700                          * address of the IP datagram containing the
701                          * HEARTBEAT chunk to which this ack is responding.
702                          * ...  
703                          */
704                         if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
705                             chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK)
706                                 new_transport = asoc->peer.active_path;
707                 }
708
709                 /* Are we switching transports?
710                  * Take care of transport locks.
711                  */
712                 if (new_transport != transport) {
713                         transport = new_transport;
714                         if (list_empty(&transport->send_ready)) {
715                                 list_add_tail(&transport->send_ready,
716                                               &transport_list);
717                         }
718                         packet = &transport->packet;
719                         sctp_packet_config(packet, vtag,
720                                            asoc->peer.ecn_capable);
721                 }
722
723                 switch (chunk->chunk_hdr->type) {
724                 /*
725                  * 6.10 Bundling
726                  *   ...
727                  *   An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
728                  *   COMPLETE with any other chunks.  [Send them immediately.]
729                  */
730                 case SCTP_CID_INIT:
731                 case SCTP_CID_INIT_ACK:
732                 case SCTP_CID_SHUTDOWN_COMPLETE:
733                         sctp_packet_init(&singleton, transport, sport, dport);
734                         sctp_packet_config(&singleton, vtag, 0);
735                         sctp_packet_append_chunk(&singleton, chunk);
736                         error = sctp_packet_transmit(&singleton);
737                         if (error < 0)
738                                 return error;
739                         break;
740
741                 case SCTP_CID_ABORT:
742                 case SCTP_CID_SACK:
743                 case SCTP_CID_HEARTBEAT:
744                 case SCTP_CID_HEARTBEAT_ACK:
745                 case SCTP_CID_SHUTDOWN:
746                 case SCTP_CID_SHUTDOWN_ACK:
747                 case SCTP_CID_ERROR:
748                 case SCTP_CID_COOKIE_ECHO:
749                 case SCTP_CID_COOKIE_ACK:
750                 case SCTP_CID_ECN_ECNE:
751                 case SCTP_CID_ECN_CWR:
752                 case SCTP_CID_ASCONF:
753                 case SCTP_CID_ASCONF_ACK:
754                 case SCTP_CID_FWD_TSN:
755                         sctp_packet_transmit_chunk(packet, chunk);
756                         break;
757
758                 default:
759                         /* We built a chunk with an illegal type! */
760                         BUG();
761                 };
762         }
763
764         /* Is it OK to send data chunks?  */
765         switch (asoc->state) {
766         case SCTP_STATE_COOKIE_ECHOED:
767                 /* Only allow bundling when this packet has a COOKIE-ECHO
768                  * chunk.
769                  */
770                 if (!packet || !packet->has_cookie_echo)
771                         break;
772
773                 /* fallthru */
774         case SCTP_STATE_ESTABLISHED:
775         case SCTP_STATE_SHUTDOWN_PENDING:
776         case SCTP_STATE_SHUTDOWN_RECEIVED:
777                 /*
778                  * RFC 2960 6.1  Transmission of DATA Chunks
779                  *
780                  * C) When the time comes for the sender to transmit,
781                  * before sending new DATA chunks, the sender MUST
782                  * first transmit any outstanding DATA chunks which
783                  * are marked for retransmission (limited by the
784                  * current cwnd).
785                  */
786                 if (!list_empty(&q->retransmit)) {
787                         if (transport == asoc->peer.retran_path)
788                                 goto retran;
789
790                         /* Switch transports & prepare the packet.  */
791
792                         transport = asoc->peer.retran_path;
793
794                         if (list_empty(&transport->send_ready)) {
795                                 list_add_tail(&transport->send_ready,
796                                               &transport_list);
797                         }
798
799                         packet = &transport->packet;
800                         sctp_packet_config(packet, vtag,
801                                            asoc->peer.ecn_capable);
802                 retran:
803                         error = sctp_outq_flush_rtx(q, packet,
804                                                     rtx_timeout, &start_timer);
805
806                         if (start_timer)
807                                 sctp_transport_reset_timers(transport);
808
809                         /* This can happen on COOKIE-ECHO resend.  Only
810                          * one chunk can get bundled with a COOKIE-ECHO.
811                          */
812                         if (packet->has_cookie_echo)
813                                 goto sctp_flush_out;
814
815                         /* Don't send new data if there is still data
816                          * waiting to retransmit.
817                          */
818                         if (!list_empty(&q->retransmit))
819                                 goto sctp_flush_out;
820                 }
821
822                 /* Finally, transmit new packets.  */
823                 start_timer = 0;
824                 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
825                         /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
826                          * stream identifier.
827                          */
828                         if (chunk->sinfo.sinfo_stream >=
829                             asoc->c.sinit_num_ostreams) {
830
831                                 /* Mark as failed send. */
832                                 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
833                                 sctp_chunk_free(chunk);
834                                 continue;
835                         }
836
837                         /* Has this chunk expired? */
838                         if (sctp_chunk_abandoned(chunk)) {
839                                 sctp_chunk_fail(chunk, 0);
840                                 sctp_chunk_free(chunk);
841                                 continue;
842                         }
843
844                         /* If there is a specified transport, use it.
845                          * Otherwise, we want to use the active path.
846                          */
847                         new_transport = chunk->transport;
848                         if (!new_transport ||
849                             new_transport->state == SCTP_INACTIVE)
850                                 new_transport = asoc->peer.active_path;
851
852                         /* Change packets if necessary.  */
853                         if (new_transport != transport) {
854                                 transport = new_transport;
855
856                                 /* Schedule to have this transport's
857                                  * packet flushed.
858                                  */
859                                 if (list_empty(&transport->send_ready)) {
860                                         list_add_tail(&transport->send_ready,
861                                                       &transport_list);
862                                 }
863
864                                 packet = &transport->packet;
865                                 sctp_packet_config(packet, vtag,
866                                                    asoc->peer.ecn_capable);
867                         }
868
869                         SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
870                                           q, chunk,
871                                           chunk && chunk->chunk_hdr ?
872                                           sctp_cname(SCTP_ST_CHUNK(
873                                                   chunk->chunk_hdr->type))
874                                           : "Illegal Chunk");
875
876                         SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
877                                         "%p skb->users %d.\n",
878                                         ntohl(chunk->subh.data_hdr->tsn),
879                                         chunk->skb ?chunk->skb->head : NULL,
880                                         chunk->skb ?
881                                         atomic_read(&chunk->skb->users) : -1);
882
883                         /* Add the chunk to the packet.  */
884                         status = sctp_packet_transmit_chunk(packet, chunk);
885
886                         switch (status) {
887                         case SCTP_XMIT_PMTU_FULL:
888                         case SCTP_XMIT_RWND_FULL:
889                         case SCTP_XMIT_NAGLE_DELAY:
890                                 /* We could not append this chunk, so put
891                                  * the chunk back on the output queue.
892                                  */
893                                 SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
894                                         "not transmit TSN: 0x%x, status: %d\n",
895                                         ntohl(chunk->subh.data_hdr->tsn),
896                                         status);
897                                 sctp_outq_head_data(q, chunk);
898                                 goto sctp_flush_out;
899                                 break;
900
901                         case SCTP_XMIT_OK:
902                                 break;
903
904                         default:
905                                 BUG();
906                         }
907
908                         /* BUG: We assume that the sctp_packet_transmit() 
909                          * call below will succeed all the time and add the
910                          * chunk to the transmitted list and restart the
911                          * timers.
912                          * It is possible that the call can fail under OOM
913                          * conditions.
914                          *
915                          * Is this really a problem?  Won't this behave
916                          * like a lost TSN?
917                          */
918                         list_add_tail(&chunk->transmitted_list,
919                                       &transport->transmitted);
920
921                         sctp_transport_reset_timers(transport);
922
923                         q->empty = 0;
924
925                         /* Only let one DATA chunk get bundled with a
926                          * COOKIE-ECHO chunk.
927                          */
928                         if (packet->has_cookie_echo)
929                                 goto sctp_flush_out;
930                 }
931                 break;
932
933         default:
934                 /* Do nothing.  */
935                 break;
936         }
937
938 sctp_flush_out:
939
940         /* Before returning, examine all the transports touched in
941          * this call.  Right now, we bluntly force clear all the
942          * transports.  Things might change after we implement Nagle.
943          * But such an examination is still required.
944          *
945          * --xguo
946          */
947         while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
948                 struct sctp_transport *t = list_entry(ltransport,
949                                                       struct sctp_transport,
950                                                       send_ready);
951                 packet = &t->packet;
952                 if (!sctp_packet_empty(packet))
953                         error = sctp_packet_transmit(packet);
954         }
955
956         return error;
957 }
958
959 /* Update unack_data based on the incoming SACK chunk */
960 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
961                                         struct sctp_sackhdr *sack)
962 {
963         sctp_sack_variable_t *frags;
964         __u16 unack_data;
965         int i;
966
967         unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
968
969         frags = sack->variable;
970         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
971                 unack_data -= ((ntohs(frags[i].gab.end) -
972                                 ntohs(frags[i].gab.start) + 1));
973         }
974
975         assoc->unack_data = unack_data;
976 }
977
978 /* Return the highest new tsn that is acknowledged by the given SACK chunk. */
979 static __u32 sctp_highest_new_tsn(struct sctp_sackhdr *sack,
980                                   struct sctp_association *asoc)
981 {
982         struct list_head *ltransport, *lchunk;
983         struct sctp_transport *transport;
984         struct sctp_chunk *chunk;
985         __u32 highest_new_tsn, tsn;
986         struct list_head *transport_list = &asoc->peer.transport_addr_list;
987
988         highest_new_tsn = ntohl(sack->cum_tsn_ack);
989
990         list_for_each(ltransport, transport_list) {
991                 transport = list_entry(ltransport, struct sctp_transport,
992                                        transports);
993                 list_for_each(lchunk, &transport->transmitted) {
994                         chunk = list_entry(lchunk, struct sctp_chunk,
995                                            transmitted_list);
996                         tsn = ntohl(chunk->subh.data_hdr->tsn);
997
998                         if (!chunk->tsn_gap_acked &&
999                             TSN_lt(highest_new_tsn, tsn) &&
1000                             sctp_acked(sack, tsn))
1001                                 highest_new_tsn = tsn;
1002                 }
1003         }
1004
1005         return highest_new_tsn;
1006 }
1007
1008 /* This is where we REALLY process a SACK.
1009  *
1010  * Process the SACK against the outqueue.  Mostly, this just frees
1011  * things off the transmitted queue.
1012  */
1013 int sctp_outq_sack(struct sctp_outq *q, struct sctp_sackhdr *sack)
1014 {
1015         struct sctp_association *asoc = q->asoc;
1016         struct sctp_transport *transport;
1017         struct sctp_chunk *tchunk = NULL;
1018         struct list_head *lchunk, *transport_list, *pos, *temp;
1019         sctp_sack_variable_t *frags = sack->variable;
1020         __u32 sack_ctsn, ctsn, tsn;
1021         __u32 highest_tsn, highest_new_tsn;
1022         __u32 sack_a_rwnd;
1023         unsigned outstanding;
1024         struct sctp_transport *primary = asoc->peer.primary_path;
1025         int count_of_newacks = 0;
1026
1027         /* Grab the association's destination address list. */
1028         transport_list = &asoc->peer.transport_addr_list;
1029
1030         sack_ctsn = ntohl(sack->cum_tsn_ack);
1031
1032         /*
1033          * SFR-CACC algorithm:
1034          * On receipt of a SACK the sender SHOULD execute the
1035          * following statements.
1036          *
1037          * 1) If the cumulative ack in the SACK passes next tsn_at_change
1038          * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1039          * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1040          * all destinations.
1041          */
1042         if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1043                 primary->cacc.changeover_active = 0;
1044                 list_for_each(pos, transport_list) {
1045                         transport = list_entry(pos, struct sctp_transport,
1046                                         transports);
1047                         transport->cacc.cycling_changeover = 0;
1048                 }
1049         }
1050
1051         /*
1052          * SFR-CACC algorithm:
1053          * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1054          * is set the receiver of the SACK MUST take the following actions:
1055          *
1056          * A) Initialize the cacc_saw_newack to 0 for all destination
1057          * addresses.
1058          */
1059         if (sack->num_gap_ack_blocks > 0 &&
1060             primary->cacc.changeover_active) {
1061                 list_for_each(pos, transport_list) {
1062                         transport = list_entry(pos, struct sctp_transport,
1063                                         transports);
1064                         transport->cacc.cacc_saw_newack = 0;
1065                 }
1066         }
1067
1068         /* Get the highest TSN in the sack. */
1069         highest_tsn = sack_ctsn;
1070         if (sack->num_gap_ack_blocks)
1071                 highest_tsn +=
1072                     ntohs(frags[ntohs(sack->num_gap_ack_blocks) - 1].gab.end);
1073
1074         if (TSN_lt(asoc->highest_sacked, highest_tsn)) {
1075                 highest_new_tsn = highest_tsn;
1076                 asoc->highest_sacked = highest_tsn;
1077         } else {
1078                 highest_new_tsn = sctp_highest_new_tsn(sack, asoc);
1079         }
1080
1081         /* Run through the retransmit queue.  Credit bytes received
1082          * and free those chunks that we can.
1083          */
1084         sctp_check_transmitted(q, &q->retransmit, NULL, sack, highest_new_tsn);
1085         sctp_mark_missing(q, &q->retransmit, NULL, highest_new_tsn, 0);
1086
1087         /* Run through the transmitted queue.
1088          * Credit bytes received and free those chunks which we can.
1089          *
1090          * This is a MASSIVE candidate for optimization.
1091          */
1092         list_for_each(pos, transport_list) {
1093                 transport  = list_entry(pos, struct sctp_transport,
1094                                         transports);
1095                 sctp_check_transmitted(q, &transport->transmitted,
1096                                        transport, sack, highest_new_tsn);
1097                 /*
1098                  * SFR-CACC algorithm:
1099                  * C) Let count_of_newacks be the number of
1100                  * destinations for which cacc_saw_newack is set.
1101                  */
1102                 if (transport->cacc.cacc_saw_newack)
1103                         count_of_newacks ++;
1104         }
1105
1106         list_for_each(pos, transport_list) {
1107                 transport  = list_entry(pos, struct sctp_transport,
1108                                         transports);
1109                 sctp_mark_missing(q, &transport->transmitted, transport,
1110                                   highest_new_tsn, count_of_newacks);
1111         }
1112
1113         /* Move the Cumulative TSN Ack Point if appropriate.  */
1114         if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn))
1115                 asoc->ctsn_ack_point = sack_ctsn;
1116
1117         /* Update unack_data field in the assoc. */
1118         sctp_sack_update_unack_data(asoc, sack);
1119
1120         ctsn = asoc->ctsn_ack_point;
1121
1122         /* Throw away stuff rotting on the sack queue.  */
1123         list_for_each_safe(lchunk, temp, &q->sacked) {
1124                 tchunk = list_entry(lchunk, struct sctp_chunk,
1125                                     transmitted_list);
1126                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1127                 if (TSN_lte(tsn, ctsn))
1128                         sctp_chunk_free(tchunk);
1129         }
1130
1131         /* ii) Set rwnd equal to the newly received a_rwnd minus the
1132          *     number of bytes still outstanding after processing the
1133          *     Cumulative TSN Ack and the Gap Ack Blocks.
1134          */
1135
1136         sack_a_rwnd = ntohl(sack->a_rwnd);
1137         outstanding = q->outstanding_bytes;
1138
1139         if (outstanding < sack_a_rwnd)
1140                 sack_a_rwnd -= outstanding;
1141         else
1142                 sack_a_rwnd = 0;
1143
1144         asoc->peer.rwnd = sack_a_rwnd;
1145
1146         sctp_generate_fwdtsn(q, sack_ctsn);
1147
1148         SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n",
1149                           __FUNCTION__, sack_ctsn);
1150         SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, "
1151                           "%p is 0x%x. Adv peer ack point: 0x%x\n",
1152                           __FUNCTION__, asoc, ctsn, asoc->adv_peer_ack_point);
1153
1154         /* See if all chunks are acked.
1155          * Make sure the empty queue handler will get run later.
1156          */
1157         q->empty = (list_empty(&q->out_chunk_list) &&
1158                     list_empty(&q->control_chunk_list) &&
1159                     list_empty(&q->retransmit));
1160         if (!q->empty)
1161                 goto finish;
1162
1163         list_for_each(pos, transport_list) {
1164                 transport  = list_entry(pos, struct sctp_transport,
1165                                         transports);
1166                 q->empty = q->empty && list_empty(&transport->transmitted);
1167                 if (!q->empty)
1168                         goto finish;
1169         }
1170
1171         SCTP_DEBUG_PRINTK("sack queue is empty.\n");
1172 finish:
1173         return q->empty;
1174 }
1175
1176 /* Is the outqueue empty?  */
1177 int sctp_outq_is_empty(const struct sctp_outq *q)
1178 {
1179         return q->empty;
1180 }
1181
1182 /********************************************************************
1183  * 2nd Level Abstractions
1184  ********************************************************************/
1185
1186 /* Go through a transport's transmitted list or the association's retransmit
1187  * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1188  * The retransmit list will not have an associated transport.
1189  *
1190  * I added coherent debug information output.   --xguo
1191  *
1192  * Instead of printing 'sacked' or 'kept' for each TSN on the
1193  * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1194  * KEPT TSN6-TSN7, etc.
1195  */
1196 static void sctp_check_transmitted(struct sctp_outq *q,
1197                                    struct list_head *transmitted_queue,
1198                                    struct sctp_transport *transport,
1199                                    struct sctp_sackhdr *sack,
1200                                    __u32 highest_new_tsn_in_sack)
1201 {
1202         struct list_head *lchunk;
1203         struct sctp_chunk *tchunk;
1204         struct list_head tlist;
1205         __u32 tsn;
1206         __u32 sack_ctsn;
1207         __u32 rtt;
1208         __u8 restart_timer = 0;
1209         int bytes_acked = 0;
1210
1211         /* These state variables are for coherent debug output. --xguo */
1212
1213 #if SCTP_DEBUG
1214         __u32 dbg_ack_tsn = 0;  /* An ACKed TSN range starts here... */
1215         __u32 dbg_last_ack_tsn = 0;  /* ...and finishes here.        */
1216         __u32 dbg_kept_tsn = 0; /* An un-ACKed range starts here...  */
1217         __u32 dbg_last_kept_tsn = 0; /* ...and finishes here.        */
1218
1219         /* 0 : The last TSN was ACKed.
1220          * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1221          * -1: We need to initialize.
1222          */
1223         int dbg_prt_state = -1;
1224 #endif /* SCTP_DEBUG */
1225
1226         sack_ctsn = ntohl(sack->cum_tsn_ack);
1227
1228         INIT_LIST_HEAD(&tlist);
1229
1230         /* The while loop will skip empty transmitted queues. */
1231         while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1232                 tchunk = list_entry(lchunk, struct sctp_chunk,
1233                                     transmitted_list);
1234
1235                 if (sctp_chunk_abandoned(tchunk)) {
1236                         /* Move the chunk to abandoned list. */
1237                         sctp_insert_list(&q->abandoned, lchunk);
1238                         continue;
1239                 }
1240
1241                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1242                 if (sctp_acked(sack, tsn)) {
1243                         /* If this queue is the retransmit queue, the
1244                          * retransmit timer has already reclaimed
1245                          * the outstanding bytes for this chunk, so only
1246                          * count bytes associated with a transport.
1247                          */
1248                         if (transport) {
1249                                 /* If this chunk is being used for RTT
1250                                  * measurement, calculate the RTT and update
1251                                  * the RTO using this value.
1252                                  *
1253                                  * 6.3.1 C5) Karn's algorithm: RTT measurements
1254                                  * MUST NOT be made using packets that were
1255                                  * retransmitted (and thus for which it is
1256                                  * ambiguous whether the reply was for the
1257                                  * first instance of the packet or a later
1258                                  * instance).
1259                                  */
1260                                 if (!tchunk->tsn_gap_acked &&
1261                                     !tchunk->resent &&
1262                                     tchunk->rtt_in_progress) {
1263                                         rtt = jiffies - tchunk->sent_at;
1264                                         sctp_transport_update_rto(transport,
1265                                                                   rtt);
1266                                 }
1267                         }
1268                         if (TSN_lte(tsn, sack_ctsn)) {
1269                                 /* RFC 2960  6.3.2 Retransmission Timer Rules
1270                                  *
1271                                  * R3) Whenever a SACK is received
1272                                  * that acknowledges the DATA chunk
1273                                  * with the earliest outstanding TSN
1274                                  * for that address, restart T3-rtx
1275                                  * timer for that address with its
1276                                  * current RTO.
1277                                  */
1278                                 restart_timer = 1;
1279
1280                                 if (!tchunk->tsn_gap_acked) {
1281                                         tchunk->tsn_gap_acked = 1;
1282                                         bytes_acked += sctp_data_size(tchunk);
1283                                         /*
1284                                          * SFR-CACC algorithm:
1285                                          * 2) If the SACK contains gap acks
1286                                          * and the flag CHANGEOVER_ACTIVE is
1287                                          * set the receiver of the SACK MUST
1288                                          * take the following action:
1289                                          *
1290                                          * B) For each TSN t being acked that
1291                                          * has not been acked in any SACK so
1292                                          * far, set cacc_saw_newack to 1 for
1293                                          * the destination that the TSN was
1294                                          * sent to.
1295                                          */
1296                                         if (transport &&
1297                                             sack->num_gap_ack_blocks &&
1298                                             q->asoc->peer.primary_path->cacc.
1299                                             changeover_active)
1300                                                 transport->cacc.cacc_saw_newack
1301                                                         = 1;
1302                                 }
1303
1304                                 list_add_tail(&tchunk->transmitted_list,
1305                                               &q->sacked);
1306                         } else {
1307                                 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1308                                  * M2) Each time a SACK arrives reporting
1309                                  * 'Stray DATA chunk(s)' record the highest TSN
1310                                  * reported as newly acknowledged, call this
1311                                  * value 'HighestTSNinSack'. A newly
1312                                  * acknowledged DATA chunk is one not
1313                                  * previously acknowledged in a SACK.
1314                                  *
1315                                  * When the SCTP sender of data receives a SACK
1316                                  * chunk that acknowledges, for the first time,
1317                                  * the receipt of a DATA chunk, all the still
1318                                  * unacknowledged DATA chunks whose TSN is
1319                                  * older than that newly acknowledged DATA
1320                                  * chunk, are qualified as 'Stray DATA chunks'.
1321                                  */
1322                                 if (!tchunk->tsn_gap_acked) {
1323                                         tchunk->tsn_gap_acked = 1;
1324                                         bytes_acked += sctp_data_size(tchunk);
1325                                 }
1326                                 list_add_tail(lchunk, &tlist);
1327                         }
1328
1329 #if SCTP_DEBUG
1330                         switch (dbg_prt_state) {
1331                         case 0: /* last TSN was ACKed */
1332                                 if (dbg_last_ack_tsn + 1 == tsn) {
1333                                         /* This TSN belongs to the
1334                                          * current ACK range.
1335                                          */
1336                                         break;
1337                                 }
1338
1339                                 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1340                                         /* Display the end of the
1341                                          * current range.
1342                                          */
1343                                         SCTP_DEBUG_PRINTK("-%08x",
1344                                                           dbg_last_ack_tsn);
1345                                 }
1346
1347                                 /* Start a new range.  */
1348                                 SCTP_DEBUG_PRINTK(",%08x", tsn);
1349                                 dbg_ack_tsn = tsn;
1350                                 break;
1351
1352                         case 1: /* The last TSN was NOT ACKed. */
1353                                 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1354                                         /* Display the end of current range. */
1355                                         SCTP_DEBUG_PRINTK("-%08x",
1356                                                           dbg_last_kept_tsn);
1357                                 }
1358
1359                                 SCTP_DEBUG_PRINTK("\n");
1360
1361                                 /* FALL THROUGH... */
1362                         default:
1363                                 /* This is the first-ever TSN we examined.  */
1364                                 /* Start a new range of ACK-ed TSNs.  */
1365                                 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn);
1366                                 dbg_prt_state = 0;
1367                                 dbg_ack_tsn = tsn;
1368                         };
1369
1370                         dbg_last_ack_tsn = tsn;
1371 #endif /* SCTP_DEBUG */
1372
1373                 } else {
1374                         if (tchunk->tsn_gap_acked) {
1375                                 SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1376                                                   "data TSN: 0x%x\n",
1377                                                   __FUNCTION__,
1378                                                   tsn);
1379                                 tchunk->tsn_gap_acked = 0;
1380
1381                                 bytes_acked -= sctp_data_size(tchunk);
1382
1383                                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1384                                  *
1385                                  * R4) Whenever a SACK is received missing a
1386                                  * TSN that was previously acknowledged via a
1387                                  * Gap Ack Block, start T3-rtx for the
1388                                  * destination address to which the DATA
1389                                  * chunk was originally
1390                                  * transmitted if it is not already running.
1391                                  */
1392                                 restart_timer = 1;
1393                         }
1394
1395                         list_add_tail(lchunk, &tlist);
1396
1397 #if SCTP_DEBUG
1398                         /* See the above comments on ACK-ed TSNs. */
1399                         switch (dbg_prt_state) {
1400                         case 1:
1401                                 if (dbg_last_kept_tsn + 1 == tsn)
1402                                         break;
1403
1404                                 if (dbg_last_kept_tsn != dbg_kept_tsn)
1405                                         SCTP_DEBUG_PRINTK("-%08x",
1406                                                           dbg_last_kept_tsn);
1407
1408                                 SCTP_DEBUG_PRINTK(",%08x", tsn);
1409                                 dbg_kept_tsn = tsn;
1410                                 break;
1411
1412                         case 0:
1413                                 if (dbg_last_ack_tsn != dbg_ack_tsn)
1414                                         SCTP_DEBUG_PRINTK("-%08x",
1415                                                           dbg_last_ack_tsn);
1416                                 SCTP_DEBUG_PRINTK("\n");
1417
1418                                 /* FALL THROUGH... */
1419                         default:
1420                                 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn);
1421                                 dbg_prt_state = 1;
1422                                 dbg_kept_tsn = tsn;
1423                         };
1424
1425                         dbg_last_kept_tsn = tsn;
1426 #endif /* SCTP_DEBUG */
1427                 }
1428         }
1429
1430 #if SCTP_DEBUG
1431         /* Finish off the last range, displaying its ending TSN.  */
1432         switch (dbg_prt_state) {
1433         case 0:
1434                 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1435                         SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_ack_tsn);
1436                 } else {
1437                         SCTP_DEBUG_PRINTK("\n");
1438                 }
1439         break;
1440
1441         case 1:
1442                 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1443                         SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_kept_tsn);
1444                 } else {
1445                         SCTP_DEBUG_PRINTK("\n");
1446                 }
1447         };
1448 #endif /* SCTP_DEBUG */
1449         if (transport) {
1450                 if (bytes_acked) {
1451                         /* 8.2. When an outstanding TSN is acknowledged,
1452                          * the endpoint shall clear the error counter of
1453                          * the destination transport address to which the
1454                          * DATA chunk was last sent.
1455                          * The association's overall error counter is
1456                          * also cleared.
1457                          */
1458                         transport->error_count = 0;
1459                         transport->asoc->overall_error_count = 0;
1460
1461                         /* Mark the destination transport address as
1462                          * active if it is not so marked.
1463                          */
1464                         if (transport->state == SCTP_INACTIVE) {
1465                                 sctp_assoc_control_transport(
1466                                         transport->asoc,
1467                                         transport,
1468                                         SCTP_TRANSPORT_UP,
1469                                         SCTP_RECEIVED_SACK);
1470                         }
1471
1472                         sctp_transport_raise_cwnd(transport, sack_ctsn,
1473                                                   bytes_acked);
1474
1475                         transport->flight_size -= bytes_acked;
1476                         q->outstanding_bytes -= bytes_acked;
1477                 } else {
1478                         /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1479                          * When a sender is doing zero window probing, it
1480                          * should not timeout the association if it continues
1481                          * to receive new packets from the receiver. The
1482                          * reason is that the receiver MAY keep its window
1483                          * closed for an indefinite time.
1484                          * A sender is doing zero window probing when the
1485                          * receiver's advertised window is zero, and there is
1486                          * only one data chunk in flight to the receiver.
1487                          */
1488                         if (!q->asoc->peer.rwnd &&
1489                             !list_empty(&tlist) &&
1490                             (sack_ctsn+2 == q->asoc->next_tsn)) {
1491                                 SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1492                                                   "window probe: %u\n",
1493                                                   __FUNCTION__, sack_ctsn);
1494                                 q->asoc->overall_error_count = 0;
1495                                 transport->error_count = 0;
1496                         }
1497                 }
1498
1499                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1500                  *
1501                  * R2) Whenever all outstanding data sent to an address have
1502                  * been acknowledged, turn off the T3-rtx timer of that
1503                  * address.
1504                  */
1505                 if (!transport->flight_size) {
1506                         if (timer_pending(&transport->T3_rtx_timer) &&
1507                             del_timer(&transport->T3_rtx_timer)) {
1508                                 sctp_transport_put(transport);
1509                         }
1510                 } else if (restart_timer) {
1511                         if (!mod_timer(&transport->T3_rtx_timer,
1512                                        jiffies + transport->rto))
1513                                 sctp_transport_hold(transport);
1514                 }
1515         }
1516
1517         list_splice(&tlist, transmitted_queue);
1518 }
1519
1520 /* Mark chunks as missing and consequently may get retransmitted. */
1521 static void sctp_mark_missing(struct sctp_outq *q,
1522                               struct list_head *transmitted_queue,
1523                               struct sctp_transport *transport,
1524                               __u32 highest_new_tsn_in_sack,
1525                               int count_of_newacks)
1526 {
1527         struct sctp_chunk *chunk;
1528         struct list_head *pos;
1529         __u32 tsn;
1530         char do_fast_retransmit = 0;
1531         struct sctp_transport *primary = q->asoc->peer.primary_path;
1532
1533         list_for_each(pos, transmitted_queue) {
1534
1535                 chunk = list_entry(pos, struct sctp_chunk, transmitted_list);
1536                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1537
1538                 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1539                  * 'Unacknowledged TSN's', if the TSN number of an
1540                  * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1541                  * value, increment the 'TSN.Missing.Report' count on that
1542                  * chunk if it has NOT been fast retransmitted or marked for
1543                  * fast retransmit already.
1544                  */
1545                 if (!chunk->fast_retransmit &&
1546                     !chunk->tsn_gap_acked &&
1547                     TSN_lt(tsn, highest_new_tsn_in_sack)) {
1548
1549                         /* SFR-CACC may require us to skip marking
1550                          * this chunk as missing.
1551                          */
1552                         if (!transport || !sctp_cacc_skip(primary, transport,
1553                                             count_of_newacks, tsn)) {
1554                                 chunk->tsn_missing_report++;
1555
1556                                 SCTP_DEBUG_PRINTK(
1557                                         "%s: TSN 0x%x missing counter: %d\n",
1558                                         __FUNCTION__, tsn,
1559                                         chunk->tsn_missing_report);
1560                         }
1561                 }
1562                 /*
1563                  * M4) If any DATA chunk is found to have a
1564                  * 'TSN.Missing.Report'
1565                  * value larger than or equal to 4, mark that chunk for
1566                  * retransmission and start the fast retransmit procedure.
1567                  */
1568
1569                 if (chunk->tsn_missing_report >= 4) {
1570                         chunk->fast_retransmit = 1;
1571                         do_fast_retransmit = 1;
1572                 }
1573         }
1574
1575         if (transport) {
1576                 if (do_fast_retransmit)
1577                         sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1578
1579                 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1580                                   "ssthresh: %d, flight_size: %d, pba: %d\n",
1581                                   __FUNCTION__, transport, transport->cwnd,
1582                                   transport->ssthresh, transport->flight_size,
1583                                   transport->partial_bytes_acked);
1584         }
1585 }
1586
1587 /* Is the given TSN acked by this packet?  */
1588 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1589 {
1590         int i;
1591         sctp_sack_variable_t *frags;
1592         __u16 gap;
1593         __u32 ctsn = ntohl(sack->cum_tsn_ack);
1594
1595         if (TSN_lte(tsn, ctsn))
1596                 goto pass;
1597
1598         /* 3.3.4 Selective Acknowledgement (SACK) (3):
1599          *
1600          * Gap Ack Blocks:
1601          *  These fields contain the Gap Ack Blocks. They are repeated
1602          *  for each Gap Ack Block up to the number of Gap Ack Blocks
1603          *  defined in the Number of Gap Ack Blocks field. All DATA
1604          *  chunks with TSNs greater than or equal to (Cumulative TSN
1605          *  Ack + Gap Ack Block Start) and less than or equal to
1606          *  (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1607          *  Block are assumed to have been received correctly.
1608          */
1609
1610         frags = sack->variable;
1611         gap = tsn - ctsn;
1612         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1613                 if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1614                     TSN_lte(gap, ntohs(frags[i].gab.end)))
1615                         goto pass;
1616         }
1617
1618         return 0;
1619 pass:
1620         return 1;
1621 }
1622
1623 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1624                                     int nskips, __u16 stream)
1625 {
1626         int i;
1627
1628         for (i = 0; i < nskips; i++) {
1629                 if (skiplist[i].stream == stream)
1630                         return i;
1631         }
1632         return i;
1633 }
1634
1635 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1636 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1637 {
1638         struct sctp_association *asoc = q->asoc;
1639         struct sctp_chunk *ftsn_chunk = NULL;
1640         struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1641         int nskips = 0;
1642         int skip_pos = 0;
1643         __u32 tsn;
1644         struct sctp_chunk *chunk;
1645         struct list_head *lchunk, *temp;
1646
1647         /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1648          * received SACK.
1649          * 
1650          * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1651          * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1652          */
1653         if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1654                 asoc->adv_peer_ack_point = ctsn;
1655
1656         /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1657          * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1658          * the chunk next in the out-queue space is marked as "abandoned" as
1659          * shown in the following example:
1660          *
1661          * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1662          * and the Advanced.Peer.Ack.Point is updated to this value:
1663          * 
1664          *   out-queue at the end of  ==>   out-queue after Adv.Ack.Point
1665          *   normal SACK processing           local advancement
1666          *                ...                           ...
1667          *   Adv.Ack.Pt-> 102 acked                     102 acked
1668          *                103 abandoned                 103 abandoned
1669          *                104 abandoned     Adv.Ack.P-> 104 abandoned
1670          *                105                           105
1671          *                106 acked                     106 acked
1672          *                ...                           ...
1673          *
1674          * In this example, the data sender successfully advanced the
1675          * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1676          */
1677         list_for_each_safe(lchunk, temp, &q->abandoned) {
1678                 chunk = list_entry(lchunk, struct sctp_chunk,
1679                                         transmitted_list);
1680                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1681
1682                 /* Remove any chunks in the abandoned queue that are acked by
1683                  * the ctsn.
1684                  */ 
1685                 if (TSN_lte(tsn, ctsn)) {
1686                         list_del_init(lchunk);
1687                         if (!chunk->tsn_gap_acked) {
1688                                 chunk->transport->flight_size -=
1689                                         sctp_data_size(chunk);
1690                                 q->outstanding_bytes -= sctp_data_size(chunk);
1691                         }
1692                         sctp_chunk_free(chunk);
1693                 } else {
1694                         if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1695                                 asoc->adv_peer_ack_point = tsn;
1696                                 if (chunk->chunk_hdr->flags &
1697                                          SCTP_DATA_UNORDERED)
1698                                         continue;
1699                                 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1700                                                 nskips,
1701                                                 chunk->subh.data_hdr->stream);
1702                                 ftsn_skip_arr[skip_pos].stream =
1703                                         chunk->subh.data_hdr->stream;
1704                                 ftsn_skip_arr[skip_pos].ssn =
1705                                          chunk->subh.data_hdr->ssn;
1706                                 if (skip_pos == nskips)
1707                                         nskips++;
1708                                 if (nskips == 10)
1709                                         break;
1710                         } else
1711                                 break;
1712                 }
1713         }
1714
1715         /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1716          * is greater than the Cumulative TSN ACK carried in the received
1717          * SACK, the data sender MUST send the data receiver a FORWARD TSN
1718          * chunk containing the latest value of the
1719          * "Advanced.Peer.Ack.Point".
1720          *
1721          * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1722          * list each stream and sequence number in the forwarded TSN. This
1723          * information will enable the receiver to easily find any
1724          * stranded TSN's waiting on stream reorder queues. Each stream
1725          * SHOULD only be reported once; this means that if multiple
1726          * abandoned messages occur in the same stream then only the
1727          * highest abandoned stream sequence number is reported. If the
1728          * total size of the FORWARD TSN does NOT fit in a single MTU then
1729          * the sender of the FORWARD TSN SHOULD lower the
1730          * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1731          * single MTU.
1732          */
1733         if (asoc->adv_peer_ack_point > ctsn)
1734                 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1735                                               nskips, &ftsn_skip_arr[0]); 
1736
1737         if (ftsn_chunk) {
1738                 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1739                 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
1740         }
1741 }