sctp: Optimize SFR-CACC transport list walking during SACK processing
[linux-3.10.git] / net / sctp / outqueue.c
1 /* SCTP kernel implementation
2  * (C) Copyright IBM Corp. 2001, 2004
3  * Copyright (c) 1999-2000 Cisco, Inc.
4  * Copyright (c) 1999-2001 Motorola, Inc.
5  * Copyright (c) 2001-2003 Intel Corp.
6  *
7  * This file is part of the SCTP kernel implementation
8  *
9  * These functions implement the sctp_outq class.   The outqueue handles
10  * bundling and queueing of outgoing SCTP chunks.
11  *
12  * This SCTP implementation is free software;
13  * you can redistribute it and/or modify it under the terms of
14  * the GNU General Public License as published by
15  * the Free Software Foundation; either version 2, or (at your option)
16  * any later version.
17  *
18  * This SCTP implementation is distributed in the hope that it
19  * will be useful, but WITHOUT ANY WARRANTY; without even the implied
20  *                 ************************
21  * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
22  * See the GNU General Public License for more details.
23  *
24  * You should have received a copy of the GNU General Public License
25  * along with GNU CC; see the file COPYING.  If not, write to
26  * the Free Software Foundation, 59 Temple Place - Suite 330,
27  * Boston, MA 02111-1307, USA.
28  *
29  * Please send any bug reports or fixes you make to the
30  * email address(es):
31  *    lksctp developers <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 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout);
75
76 /* Add data to the front of the queue. */
77 static inline void sctp_outq_head_data(struct sctp_outq *q,
78                                         struct sctp_chunk *ch)
79 {
80         list_add(&ch->list, &q->out_chunk_list);
81         q->out_qlen += ch->skb->len;
82         return;
83 }
84
85 /* Take data from the front of the queue. */
86 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
87 {
88         struct sctp_chunk *ch = NULL;
89
90         if (!list_empty(&q->out_chunk_list)) {
91                 struct list_head *entry = q->out_chunk_list.next;
92
93                 ch = list_entry(entry, struct sctp_chunk, list);
94                 list_del_init(entry);
95                 q->out_qlen -= ch->skb->len;
96         }
97         return ch;
98 }
99 /* Add data chunk to the end of the queue. */
100 static inline void sctp_outq_tail_data(struct sctp_outq *q,
101                                        struct sctp_chunk *ch)
102 {
103         list_add_tail(&ch->list, &q->out_chunk_list);
104         q->out_qlen += ch->skb->len;
105         return;
106 }
107
108 /*
109  * SFR-CACC algorithm:
110  * D) If count_of_newacks is greater than or equal to 2
111  * and t was not sent to the current primary then the
112  * sender MUST NOT increment missing report count for t.
113  */
114 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary,
115                                        struct sctp_transport *transport,
116                                        int count_of_newacks)
117 {
118         if (count_of_newacks >=2 && transport != primary)
119                 return 1;
120         return 0;
121 }
122
123 /*
124  * SFR-CACC algorithm:
125  * F) If count_of_newacks is less than 2, let d be the
126  * destination to which t was sent. If cacc_saw_newack
127  * is 0 for destination d, then the sender MUST NOT
128  * increment missing report count for t.
129  */
130 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport,
131                                        int count_of_newacks)
132 {
133         if (count_of_newacks < 2 && !transport->cacc.cacc_saw_newack)
134                 return 1;
135         return 0;
136 }
137
138 /*
139  * SFR-CACC algorithm:
140  * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
141  * execute steps C, D, F.
142  *
143  * C has been implemented in sctp_outq_sack
144  */
145 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
146                                      struct sctp_transport *transport,
147                                      int count_of_newacks)
148 {
149         if (!primary->cacc.cycling_changeover) {
150                 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
151                         return 1;
152                 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
153                         return 1;
154                 return 0;
155         }
156         return 0;
157 }
158
159 /*
160  * SFR-CACC algorithm:
161  * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
162  * than next_tsn_at_change of the current primary, then
163  * the sender MUST NOT increment missing report count
164  * for t.
165  */
166 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
167 {
168         if (primary->cacc.cycling_changeover &&
169             TSN_lt(tsn, primary->cacc.next_tsn_at_change))
170                 return 1;
171         return 0;
172 }
173
174 /*
175  * SFR-CACC algorithm:
176  * 3) If the missing report count for TSN t is to be
177  * incremented according to [RFC2960] and
178  * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
179  * then the sender MUST futher execute steps 3.1 and
180  * 3.2 to determine if the missing report count for
181  * TSN t SHOULD NOT be incremented.
182  *
183  * 3.3) If 3.1 and 3.2 do not dictate that the missing
184  * report count for t should not be incremented, then
185  * the sender SOULD increment missing report count for
186  * t (according to [RFC2960] and [SCTP_STEWART_2002]).
187  */
188 static inline int sctp_cacc_skip(struct sctp_transport *primary,
189                                  struct sctp_transport *transport,
190                                  int count_of_newacks,
191                                  __u32 tsn)
192 {
193         if (primary->cacc.changeover_active &&
194             (sctp_cacc_skip_3_1(primary, transport, count_of_newacks)
195              || sctp_cacc_skip_3_2(primary, tsn)))
196                 return 1;
197         return 0;
198 }
199
200 /* Initialize an existing sctp_outq.  This does the boring stuff.
201  * You still need to define handlers if you really want to DO
202  * something with this structure...
203  */
204 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
205 {
206         q->asoc = asoc;
207         INIT_LIST_HEAD(&q->out_chunk_list);
208         INIT_LIST_HEAD(&q->control_chunk_list);
209         INIT_LIST_HEAD(&q->retransmit);
210         INIT_LIST_HEAD(&q->sacked);
211         INIT_LIST_HEAD(&q->abandoned);
212
213         q->fast_rtx = 0;
214         q->outstanding_bytes = 0;
215         q->empty = 1;
216         q->cork  = 0;
217
218         q->malloced = 0;
219         q->out_qlen = 0;
220 }
221
222 /* Free the outqueue structure and any related pending chunks.
223  */
224 void sctp_outq_teardown(struct sctp_outq *q)
225 {
226         struct sctp_transport *transport;
227         struct list_head *lchunk, *temp;
228         struct sctp_chunk *chunk, *tmp;
229
230         /* Throw away unacknowledged chunks. */
231         list_for_each_entry(transport, &q->asoc->peer.transport_addr_list,
232                         transports) {
233                 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
234                         chunk = list_entry(lchunk, struct sctp_chunk,
235                                            transmitted_list);
236                         /* Mark as part of a failed message. */
237                         sctp_chunk_fail(chunk, q->error);
238                         sctp_chunk_free(chunk);
239                 }
240         }
241
242         /* Throw away chunks that have been gap ACKed.  */
243         list_for_each_safe(lchunk, temp, &q->sacked) {
244                 list_del_init(lchunk);
245                 chunk = list_entry(lchunk, struct sctp_chunk,
246                                    transmitted_list);
247                 sctp_chunk_fail(chunk, q->error);
248                 sctp_chunk_free(chunk);
249         }
250
251         /* Throw away any chunks in the retransmit queue. */
252         list_for_each_safe(lchunk, temp, &q->retransmit) {
253                 list_del_init(lchunk);
254                 chunk = list_entry(lchunk, struct sctp_chunk,
255                                    transmitted_list);
256                 sctp_chunk_fail(chunk, q->error);
257                 sctp_chunk_free(chunk);
258         }
259
260         /* Throw away any chunks that are in the abandoned queue. */
261         list_for_each_safe(lchunk, temp, &q->abandoned) {
262                 list_del_init(lchunk);
263                 chunk = list_entry(lchunk, struct sctp_chunk,
264                                    transmitted_list);
265                 sctp_chunk_fail(chunk, q->error);
266                 sctp_chunk_free(chunk);
267         }
268
269         /* Throw away any leftover data chunks. */
270         while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
271
272                 /* Mark as send failure. */
273                 sctp_chunk_fail(chunk, q->error);
274                 sctp_chunk_free(chunk);
275         }
276
277         q->error = 0;
278
279         /* Throw away any leftover control chunks. */
280         list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
281                 list_del_init(&chunk->list);
282                 sctp_chunk_free(chunk);
283         }
284 }
285
286 /* Free the outqueue structure and any related pending chunks.  */
287 void sctp_outq_free(struct sctp_outq *q)
288 {
289         /* Throw away leftover chunks. */
290         sctp_outq_teardown(q);
291
292         /* If we were kmalloc()'d, free the memory.  */
293         if (q->malloced)
294                 kfree(q);
295 }
296
297 /* Put a new chunk in an sctp_outq.  */
298 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk)
299 {
300         int error = 0;
301
302         SCTP_DEBUG_PRINTK("sctp_outq_tail(%p, %p[%s])\n",
303                           q, chunk, chunk && chunk->chunk_hdr ?
304                           sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
305                           : "Illegal Chunk");
306
307         /* If it is data, queue it up, otherwise, send it
308          * immediately.
309          */
310         if (SCTP_CID_DATA == chunk->chunk_hdr->type) {
311                 /* Is it OK to queue data chunks?  */
312                 /* From 9. Termination of Association
313                  *
314                  * When either endpoint performs a shutdown, the
315                  * association on each peer will stop accepting new
316                  * data from its user and only deliver data in queue
317                  * at the time of sending or receiving the SHUTDOWN
318                  * chunk.
319                  */
320                 switch (q->asoc->state) {
321                 case SCTP_STATE_EMPTY:
322                 case SCTP_STATE_CLOSED:
323                 case SCTP_STATE_SHUTDOWN_PENDING:
324                 case SCTP_STATE_SHUTDOWN_SENT:
325                 case SCTP_STATE_SHUTDOWN_RECEIVED:
326                 case SCTP_STATE_SHUTDOWN_ACK_SENT:
327                         /* Cannot send after transport endpoint shutdown */
328                         error = -ESHUTDOWN;
329                         break;
330
331                 default:
332                         SCTP_DEBUG_PRINTK("outqueueing (%p, %p[%s])\n",
333                           q, chunk, chunk && chunk->chunk_hdr ?
334                           sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
335                           : "Illegal Chunk");
336
337                         sctp_outq_tail_data(q, chunk);
338                         if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
339                                 SCTP_INC_STATS(SCTP_MIB_OUTUNORDERCHUNKS);
340                         else
341                                 SCTP_INC_STATS(SCTP_MIB_OUTORDERCHUNKS);
342                         q->empty = 0;
343                         break;
344                 }
345         } else {
346                 list_add_tail(&chunk->list, &q->control_chunk_list);
347                 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
348         }
349
350         if (error < 0)
351                 return error;
352
353         if (!q->cork)
354                 error = sctp_outq_flush(q, 0);
355
356         return error;
357 }
358
359 /* Insert a chunk into the sorted list based on the TSNs.  The retransmit list
360  * and the abandoned list are in ascending order.
361  */
362 static void sctp_insert_list(struct list_head *head, struct list_head *new)
363 {
364         struct list_head *pos;
365         struct sctp_chunk *nchunk, *lchunk;
366         __u32 ntsn, ltsn;
367         int done = 0;
368
369         nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
370         ntsn = ntohl(nchunk->subh.data_hdr->tsn);
371
372         list_for_each(pos, head) {
373                 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
374                 ltsn = ntohl(lchunk->subh.data_hdr->tsn);
375                 if (TSN_lt(ntsn, ltsn)) {
376                         list_add(new, pos->prev);
377                         done = 1;
378                         break;
379                 }
380         }
381         if (!done)
382                 list_add_tail(new, head);
383 }
384
385 /* Mark all the eligible packets on a transport for retransmission.  */
386 void sctp_retransmit_mark(struct sctp_outq *q,
387                           struct sctp_transport *transport,
388                           __u8 reason)
389 {
390         struct list_head *lchunk, *ltemp;
391         struct sctp_chunk *chunk;
392
393         /* Walk through the specified transmitted queue.  */
394         list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
395                 chunk = list_entry(lchunk, struct sctp_chunk,
396                                    transmitted_list);
397
398                 /* If the chunk is abandoned, move it to abandoned list. */
399                 if (sctp_chunk_abandoned(chunk)) {
400                         list_del_init(lchunk);
401                         sctp_insert_list(&q->abandoned, lchunk);
402
403                         /* If this chunk has not been previousely acked,
404                          * stop considering it 'outstanding'.  Our peer
405                          * will most likely never see it since it will
406                          * not be retransmitted
407                          */
408                         if (!chunk->tsn_gap_acked) {
409                                 chunk->transport->flight_size -=
410                                                 sctp_data_size(chunk);
411                                 q->outstanding_bytes -= sctp_data_size(chunk);
412                                 q->asoc->peer.rwnd += (sctp_data_size(chunk) +
413                                                         sizeof(struct sk_buff));
414                         }
415                         continue;
416                 }
417
418                 /* If we are doing  retransmission due to a timeout or pmtu
419                  * discovery, only the  chunks that are not yet acked should
420                  * be added to the retransmit queue.
421                  */
422                 if ((reason == SCTP_RTXR_FAST_RTX  &&
423                             (chunk->fast_retransmit > 0)) ||
424                     (reason != SCTP_RTXR_FAST_RTX  && !chunk->tsn_gap_acked)) {
425                         /* If this chunk was sent less then 1 rto ago, do not
426                          * retransmit this chunk, but give the peer time
427                          * to acknowlege it.  Do this only when
428                          * retransmitting due to T3 timeout.
429                          */
430                         if (reason == SCTP_RTXR_T3_RTX &&
431                             (jiffies - chunk->sent_at) < transport->last_rto)
432                                 continue;
433
434                         /* RFC 2960 6.2.1 Processing a Received SACK
435                          *
436                          * C) Any time a DATA chunk is marked for
437                          * retransmission (via either T3-rtx timer expiration
438                          * (Section 6.3.3) or via fast retransmit
439                          * (Section 7.2.4)), add the data size of those
440                          * chunks to the rwnd.
441                          */
442                         q->asoc->peer.rwnd += (sctp_data_size(chunk) +
443                                                 sizeof(struct sk_buff));
444                         q->outstanding_bytes -= sctp_data_size(chunk);
445                         transport->flight_size -= sctp_data_size(chunk);
446
447                         /* sctpimpguide-05 Section 2.8.2
448                          * M5) If a T3-rtx timer expires, the
449                          * 'TSN.Missing.Report' of all affected TSNs is set
450                          * to 0.
451                          */
452                         chunk->tsn_missing_report = 0;
453
454                         /* If a chunk that is being used for RTT measurement
455                          * has to be retransmitted, we cannot use this chunk
456                          * anymore for RTT measurements. Reset rto_pending so
457                          * that a new RTT measurement is started when a new
458                          * data chunk is sent.
459                          */
460                         if (chunk->rtt_in_progress) {
461                                 chunk->rtt_in_progress = 0;
462                                 transport->rto_pending = 0;
463                         }
464
465                         /* Move the chunk to the retransmit queue. The chunks
466                          * on the retransmit queue are always kept in order.
467                          */
468                         list_del_init(lchunk);
469                         sctp_insert_list(&q->retransmit, lchunk);
470                 }
471         }
472
473         SCTP_DEBUG_PRINTK("%s: transport: %p, reason: %d, "
474                           "cwnd: %d, ssthresh: %d, flight_size: %d, "
475                           "pba: %d\n", __func__,
476                           transport, reason,
477                           transport->cwnd, transport->ssthresh,
478                           transport->flight_size,
479                           transport->partial_bytes_acked);
480
481 }
482
483 /* Mark all the eligible packets on a transport for retransmission and force
484  * one packet out.
485  */
486 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
487                      sctp_retransmit_reason_t reason)
488 {
489         int error = 0;
490
491         switch(reason) {
492         case SCTP_RTXR_T3_RTX:
493                 SCTP_INC_STATS(SCTP_MIB_T3_RETRANSMITS);
494                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
495                 /* Update the retran path if the T3-rtx timer has expired for
496                  * the current retran path.
497                  */
498                 if (transport == transport->asoc->peer.retran_path)
499                         sctp_assoc_update_retran_path(transport->asoc);
500                 transport->asoc->rtx_data_chunks +=
501                         transport->asoc->unack_data;
502                 break;
503         case SCTP_RTXR_FAST_RTX:
504                 SCTP_INC_STATS(SCTP_MIB_FAST_RETRANSMITS);
505                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
506                 q->fast_rtx = 1;
507                 break;
508         case SCTP_RTXR_PMTUD:
509                 SCTP_INC_STATS(SCTP_MIB_PMTUD_RETRANSMITS);
510                 break;
511         case SCTP_RTXR_T1_RTX:
512                 SCTP_INC_STATS(SCTP_MIB_T1_RETRANSMITS);
513                 transport->asoc->init_retries++;
514                 break;
515         default:
516                 BUG();
517         }
518
519         sctp_retransmit_mark(q, transport, reason);
520
521         /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
522          * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
523          * following the procedures outlined in C1 - C5.
524          */
525         if (reason == SCTP_RTXR_T3_RTX)
526                 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
527
528         /* Flush the queues only on timeout, since fast_rtx is only
529          * triggered during sack processing and the queue
530          * will be flushed at the end.
531          */
532         if (reason != SCTP_RTXR_FAST_RTX)
533                 error = sctp_outq_flush(q, /* rtx_timeout */ 1);
534
535         if (error)
536                 q->asoc->base.sk->sk_err = -error;
537 }
538
539 /*
540  * Transmit DATA chunks on the retransmit queue.  Upon return from
541  * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
542  * need to be transmitted by the caller.
543  * We assume that pkt->transport has already been set.
544  *
545  * The return value is a normal kernel error return value.
546  */
547 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
548                                int rtx_timeout, int *start_timer)
549 {
550         struct list_head *lqueue;
551         struct sctp_transport *transport = pkt->transport;
552         sctp_xmit_t status;
553         struct sctp_chunk *chunk, *chunk1;
554         struct sctp_association *asoc;
555         int fast_rtx;
556         int error = 0;
557         int timer = 0;
558         int done = 0;
559
560         asoc = q->asoc;
561         lqueue = &q->retransmit;
562         fast_rtx = q->fast_rtx;
563
564         /* This loop handles time-out retransmissions, fast retransmissions,
565          * and retransmissions due to opening of whindow.
566          *
567          * RFC 2960 6.3.3 Handle T3-rtx Expiration
568          *
569          * E3) Determine how many of the earliest (i.e., lowest TSN)
570          * outstanding DATA chunks for the address for which the
571          * T3-rtx has expired will fit into a single packet, subject
572          * to the MTU constraint for the path corresponding to the
573          * destination transport address to which the retransmission
574          * is being sent (this may be different from the address for
575          * which the timer expires [see Section 6.4]). Call this value
576          * K. Bundle and retransmit those K DATA chunks in a single
577          * packet to the destination endpoint.
578          *
579          * [Just to be painfully clear, if we are retransmitting
580          * because a timeout just happened, we should send only ONE
581          * packet of retransmitted data.]
582          *
583          * For fast retransmissions we also send only ONE packet.  However,
584          * if we are just flushing the queue due to open window, we'll
585          * try to send as much as possible.
586          */
587         list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
588
589                 /* Make sure that Gap Acked TSNs are not retransmitted.  A
590                  * simple approach is just to move such TSNs out of the
591                  * way and into a 'transmitted' queue and skip to the
592                  * next chunk.
593                  */
594                 if (chunk->tsn_gap_acked) {
595                         list_del(&chunk->transmitted_list);
596                         list_add_tail(&chunk->transmitted_list,
597                                         &transport->transmitted);
598                         continue;
599                 }
600
601                 /* If we are doing fast retransmit, ignore non-fast_rtransmit
602                  * chunks
603                  */
604                 if (fast_rtx && !chunk->fast_retransmit)
605                         continue;
606
607                 /* Attempt to append this chunk to the packet. */
608                 status = sctp_packet_append_chunk(pkt, chunk);
609
610                 switch (status) {
611                 case SCTP_XMIT_PMTU_FULL:
612                         /* Send this packet.  */
613                         error = sctp_packet_transmit(pkt);
614
615                         /* If we are retransmitting, we should only
616                          * send a single packet.
617                          */
618                         if (rtx_timeout || fast_rtx)
619                                 done = 1;
620
621                         /* Bundle next chunk in the next round.  */
622                         break;
623
624                 case SCTP_XMIT_RWND_FULL:
625                         /* Send this packet. */
626                         error = sctp_packet_transmit(pkt);
627
628                         /* Stop sending DATA as there is no more room
629                          * at the receiver.
630                          */
631                         done = 1;
632                         break;
633
634                 case SCTP_XMIT_NAGLE_DELAY:
635                         /* Send this packet. */
636                         error = sctp_packet_transmit(pkt);
637
638                         /* Stop sending DATA because of nagle delay. */
639                         done = 1;
640                         break;
641
642                 default:
643                         /* The append was successful, so add this chunk to
644                          * the transmitted list.
645                          */
646                         list_del(&chunk->transmitted_list);
647                         list_add_tail(&chunk->transmitted_list,
648                                         &transport->transmitted);
649
650                         /* Mark the chunk as ineligible for fast retransmit
651                          * after it is retransmitted.
652                          */
653                         if (chunk->fast_retransmit > 0)
654                                 chunk->fast_retransmit = -1;
655
656                         /* Force start T3-rtx timer when fast retransmitting
657                          * the earliest outstanding TSN
658                          */
659                         if (!timer && fast_rtx &&
660                             ntohl(chunk->subh.data_hdr->tsn) ==
661                                              asoc->ctsn_ack_point + 1)
662                                 timer = 2;
663
664                         q->empty = 0;
665                         break;
666                 }
667
668                 /* Set the timer if there were no errors */
669                 if (!error && !timer)
670                         timer = 1;
671
672                 if (done)
673                         break;
674         }
675
676         /* If we are here due to a retransmit timeout or a fast
677          * retransmit and if there are any chunks left in the retransmit
678          * queue that could not fit in the PMTU sized packet, they need
679          * to be marked as ineligible for a subsequent fast retransmit.
680          */
681         if (rtx_timeout || fast_rtx) {
682                 list_for_each_entry(chunk1, lqueue, transmitted_list) {
683                         if (chunk1->fast_retransmit > 0)
684                                 chunk1->fast_retransmit = -1;
685                 }
686         }
687
688         *start_timer = timer;
689
690         /* Clear fast retransmit hint */
691         if (fast_rtx)
692                 q->fast_rtx = 0;
693
694         return error;
695 }
696
697 /* Cork the outqueue so queued chunks are really queued. */
698 int sctp_outq_uncork(struct sctp_outq *q)
699 {
700         int error = 0;
701         if (q->cork)
702                 q->cork = 0;
703         error = sctp_outq_flush(q, 0);
704         return error;
705 }
706
707
708 /*
709  * Try to flush an outqueue.
710  *
711  * Description: Send everything in q which we legally can, subject to
712  * congestion limitations.
713  * * Note: This function can be called from multiple contexts so appropriate
714  * locking concerns must be made.  Today we use the sock lock to protect
715  * this function.
716  */
717 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
718 {
719         struct sctp_packet *packet;
720         struct sctp_packet singleton;
721         struct sctp_association *asoc = q->asoc;
722         __u16 sport = asoc->base.bind_addr.port;
723         __u16 dport = asoc->peer.port;
724         __u32 vtag = asoc->peer.i.init_tag;
725         struct sctp_transport *transport = NULL;
726         struct sctp_transport *new_transport;
727         struct sctp_chunk *chunk, *tmp;
728         sctp_xmit_t status;
729         int error = 0;
730         int start_timer = 0;
731         int one_packet = 0;
732
733         /* These transports have chunks to send. */
734         struct list_head transport_list;
735         struct list_head *ltransport;
736
737         INIT_LIST_HEAD(&transport_list);
738         packet = NULL;
739
740         /*
741          * 6.10 Bundling
742          *   ...
743          *   When bundling control chunks with DATA chunks, an
744          *   endpoint MUST place control chunks first in the outbound
745          *   SCTP packet.  The transmitter MUST transmit DATA chunks
746          *   within a SCTP packet in increasing order of TSN.
747          *   ...
748          */
749
750         list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
751                 list_del_init(&chunk->list);
752
753                 /* Pick the right transport to use. */
754                 new_transport = chunk->transport;
755
756                 if (!new_transport) {
757                         /*
758                          * If we have a prior transport pointer, see if
759                          * the destination address of the chunk
760                          * matches the destination address of the
761                          * current transport.  If not a match, then
762                          * try to look up the transport with a given
763                          * destination address.  We do this because
764                          * after processing ASCONFs, we may have new
765                          * transports created.
766                          */
767                         if (transport &&
768                             sctp_cmp_addr_exact(&chunk->dest,
769                                                 &transport->ipaddr))
770                                         new_transport = transport;
771                         else
772                                 new_transport = sctp_assoc_lookup_paddr(asoc,
773                                                                 &chunk->dest);
774
775                         /* if we still don't have a new transport, then
776                          * use the current active path.
777                          */
778                         if (!new_transport)
779                                 new_transport = asoc->peer.active_path;
780                 } else if ((new_transport->state == SCTP_INACTIVE) ||
781                            (new_transport->state == SCTP_UNCONFIRMED)) {
782                         /* If the chunk is Heartbeat or Heartbeat Ack,
783                          * send it to chunk->transport, even if it's
784                          * inactive.
785                          *
786                          * 3.3.6 Heartbeat Acknowledgement:
787                          * ...
788                          * A HEARTBEAT ACK is always sent to the source IP
789                          * address of the IP datagram containing the
790                          * HEARTBEAT chunk to which this ack is responding.
791                          * ...
792                          *
793                          * ASCONF_ACKs also must be sent to the source.
794                          */
795                         if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
796                             chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK &&
797                             chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK)
798                                 new_transport = asoc->peer.active_path;
799                 }
800
801                 /* Are we switching transports?
802                  * Take care of transport locks.
803                  */
804                 if (new_transport != transport) {
805                         transport = new_transport;
806                         if (list_empty(&transport->send_ready)) {
807                                 list_add_tail(&transport->send_ready,
808                                               &transport_list);
809                         }
810                         packet = &transport->packet;
811                         sctp_packet_config(packet, vtag,
812                                            asoc->peer.ecn_capable);
813                 }
814
815                 switch (chunk->chunk_hdr->type) {
816                 /*
817                  * 6.10 Bundling
818                  *   ...
819                  *   An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
820                  *   COMPLETE with any other chunks.  [Send them immediately.]
821                  */
822                 case SCTP_CID_INIT:
823                 case SCTP_CID_INIT_ACK:
824                 case SCTP_CID_SHUTDOWN_COMPLETE:
825                         sctp_packet_init(&singleton, transport, sport, dport);
826                         sctp_packet_config(&singleton, vtag, 0);
827                         sctp_packet_append_chunk(&singleton, chunk);
828                         error = sctp_packet_transmit(&singleton);
829                         if (error < 0)
830                                 return error;
831                         break;
832
833                 case SCTP_CID_ABORT:
834                         if (sctp_test_T_bit(chunk)) {
835                                 packet->vtag = asoc->c.my_vtag;
836                         }
837                 /* The following chunks are "response" chunks, i.e.
838                  * they are generated in response to something we
839                  * received.  If we are sending these, then we can
840                  * send only 1 packet containing these chunks.
841                  */
842                 case SCTP_CID_HEARTBEAT_ACK:
843                 case SCTP_CID_SHUTDOWN_ACK:
844                 case SCTP_CID_COOKIE_ACK:
845                 case SCTP_CID_COOKIE_ECHO:
846                 case SCTP_CID_ERROR:
847                 case SCTP_CID_ECN_CWR:
848                 case SCTP_CID_ASCONF_ACK:
849                         one_packet = 1;
850                         /* Fall throught */
851
852                 case SCTP_CID_SACK:
853                 case SCTP_CID_HEARTBEAT:
854                 case SCTP_CID_SHUTDOWN:
855                 case SCTP_CID_ECN_ECNE:
856                 case SCTP_CID_ASCONF:
857                 case SCTP_CID_FWD_TSN:
858                         status = sctp_packet_transmit_chunk(packet, chunk,
859                                                             one_packet);
860                         if (status  != SCTP_XMIT_OK) {
861                                 /* put the chunk back */
862                                 list_add(&chunk->list, &q->control_chunk_list);
863                         }
864                         break;
865
866                 default:
867                         /* We built a chunk with an illegal type! */
868                         BUG();
869                 }
870         }
871
872         /* Is it OK to send data chunks?  */
873         switch (asoc->state) {
874         case SCTP_STATE_COOKIE_ECHOED:
875                 /* Only allow bundling when this packet has a COOKIE-ECHO
876                  * chunk.
877                  */
878                 if (!packet || !packet->has_cookie_echo)
879                         break;
880
881                 /* fallthru */
882         case SCTP_STATE_ESTABLISHED:
883         case SCTP_STATE_SHUTDOWN_PENDING:
884         case SCTP_STATE_SHUTDOWN_RECEIVED:
885                 /*
886                  * RFC 2960 6.1  Transmission of DATA Chunks
887                  *
888                  * C) When the time comes for the sender to transmit,
889                  * before sending new DATA chunks, the sender MUST
890                  * first transmit any outstanding DATA chunks which
891                  * are marked for retransmission (limited by the
892                  * current cwnd).
893                  */
894                 if (!list_empty(&q->retransmit)) {
895                         if (transport == asoc->peer.retran_path)
896                                 goto retran;
897
898                         /* Switch transports & prepare the packet.  */
899
900                         transport = asoc->peer.retran_path;
901
902                         if (list_empty(&transport->send_ready)) {
903                                 list_add_tail(&transport->send_ready,
904                                               &transport_list);
905                         }
906
907                         packet = &transport->packet;
908                         sctp_packet_config(packet, vtag,
909                                            asoc->peer.ecn_capable);
910                 retran:
911                         error = sctp_outq_flush_rtx(q, packet,
912                                                     rtx_timeout, &start_timer);
913
914                         if (start_timer)
915                                 sctp_transport_reset_timers(transport,
916                                                             start_timer-1);
917
918                         /* This can happen on COOKIE-ECHO resend.  Only
919                          * one chunk can get bundled with a COOKIE-ECHO.
920                          */
921                         if (packet->has_cookie_echo)
922                                 goto sctp_flush_out;
923
924                         /* Don't send new data if there is still data
925                          * waiting to retransmit.
926                          */
927                         if (!list_empty(&q->retransmit))
928                                 goto sctp_flush_out;
929                 }
930
931                 /* Finally, transmit new packets.  */
932                 start_timer = 0;
933                 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
934                         /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
935                          * stream identifier.
936                          */
937                         if (chunk->sinfo.sinfo_stream >=
938                             asoc->c.sinit_num_ostreams) {
939
940                                 /* Mark as failed send. */
941                                 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
942                                 sctp_chunk_free(chunk);
943                                 continue;
944                         }
945
946                         /* Has this chunk expired? */
947                         if (sctp_chunk_abandoned(chunk)) {
948                                 sctp_chunk_fail(chunk, 0);
949                                 sctp_chunk_free(chunk);
950                                 continue;
951                         }
952
953                         /* If there is a specified transport, use it.
954                          * Otherwise, we want to use the active path.
955                          */
956                         new_transport = chunk->transport;
957                         if (!new_transport ||
958                             ((new_transport->state == SCTP_INACTIVE) ||
959                              (new_transport->state == SCTP_UNCONFIRMED)))
960                                 new_transport = asoc->peer.active_path;
961
962                         /* Change packets if necessary.  */
963                         if (new_transport != transport) {
964                                 transport = new_transport;
965
966                                 /* Schedule to have this transport's
967                                  * packet flushed.
968                                  */
969                                 if (list_empty(&transport->send_ready)) {
970                                         list_add_tail(&transport->send_ready,
971                                                       &transport_list);
972                                 }
973
974                                 packet = &transport->packet;
975                                 sctp_packet_config(packet, vtag,
976                                                    asoc->peer.ecn_capable);
977                         }
978
979                         SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
980                                           q, chunk,
981                                           chunk && chunk->chunk_hdr ?
982                                           sctp_cname(SCTP_ST_CHUNK(
983                                                   chunk->chunk_hdr->type))
984                                           : "Illegal Chunk");
985
986                         SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
987                                         "%p skb->users %d.\n",
988                                         ntohl(chunk->subh.data_hdr->tsn),
989                                         chunk->skb ?chunk->skb->head : NULL,
990                                         chunk->skb ?
991                                         atomic_read(&chunk->skb->users) : -1);
992
993                         /* Add the chunk to the packet.  */
994                         status = sctp_packet_transmit_chunk(packet, chunk, 0);
995
996                         switch (status) {
997                         case SCTP_XMIT_PMTU_FULL:
998                         case SCTP_XMIT_RWND_FULL:
999                         case SCTP_XMIT_NAGLE_DELAY:
1000                                 /* We could not append this chunk, so put
1001                                  * the chunk back on the output queue.
1002                                  */
1003                                 SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
1004                                         "not transmit TSN: 0x%x, status: %d\n",
1005                                         ntohl(chunk->subh.data_hdr->tsn),
1006                                         status);
1007                                 sctp_outq_head_data(q, chunk);
1008                                 goto sctp_flush_out;
1009                                 break;
1010
1011                         case SCTP_XMIT_OK:
1012                                 break;
1013
1014                         default:
1015                                 BUG();
1016                         }
1017
1018                         /* BUG: We assume that the sctp_packet_transmit()
1019                          * call below will succeed all the time and add the
1020                          * chunk to the transmitted list and restart the
1021                          * timers.
1022                          * It is possible that the call can fail under OOM
1023                          * conditions.
1024                          *
1025                          * Is this really a problem?  Won't this behave
1026                          * like a lost TSN?
1027                          */
1028                         list_add_tail(&chunk->transmitted_list,
1029                                       &transport->transmitted);
1030
1031                         sctp_transport_reset_timers(transport, start_timer-1);
1032
1033                         q->empty = 0;
1034
1035                         /* Only let one DATA chunk get bundled with a
1036                          * COOKIE-ECHO chunk.
1037                          */
1038                         if (packet->has_cookie_echo)
1039                                 goto sctp_flush_out;
1040                 }
1041                 break;
1042
1043         default:
1044                 /* Do nothing.  */
1045                 break;
1046         }
1047
1048 sctp_flush_out:
1049
1050         /* Before returning, examine all the transports touched in
1051          * this call.  Right now, we bluntly force clear all the
1052          * transports.  Things might change after we implement Nagle.
1053          * But such an examination is still required.
1054          *
1055          * --xguo
1056          */
1057         while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
1058                 struct sctp_transport *t = list_entry(ltransport,
1059                                                       struct sctp_transport,
1060                                                       send_ready);
1061                 packet = &t->packet;
1062                 if (!sctp_packet_empty(packet))
1063                         error = sctp_packet_transmit(packet);
1064         }
1065
1066         return error;
1067 }
1068
1069 /* Update unack_data based on the incoming SACK chunk */
1070 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1071                                         struct sctp_sackhdr *sack)
1072 {
1073         sctp_sack_variable_t *frags;
1074         __u16 unack_data;
1075         int i;
1076
1077         unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1078
1079         frags = sack->variable;
1080         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1081                 unack_data -= ((ntohs(frags[i].gab.end) -
1082                                 ntohs(frags[i].gab.start) + 1));
1083         }
1084
1085         assoc->unack_data = unack_data;
1086 }
1087
1088 /* Return the highest new tsn that is acknowledged by the given SACK chunk. */
1089 static __u32 sctp_highest_new_tsn(struct sctp_sackhdr *sack,
1090                                   struct sctp_association *asoc)
1091 {
1092         struct sctp_transport *transport;
1093         struct sctp_chunk *chunk;
1094         __u32 highest_new_tsn, tsn;
1095         struct list_head *transport_list = &asoc->peer.transport_addr_list;
1096
1097         highest_new_tsn = ntohl(sack->cum_tsn_ack);
1098
1099         list_for_each_entry(transport, transport_list, transports) {
1100                 list_for_each_entry(chunk, &transport->transmitted,
1101                                 transmitted_list) {
1102                         tsn = ntohl(chunk->subh.data_hdr->tsn);
1103
1104                         if (!chunk->tsn_gap_acked &&
1105                             TSN_lt(highest_new_tsn, tsn) &&
1106                             sctp_acked(sack, tsn))
1107                                 highest_new_tsn = tsn;
1108                 }
1109         }
1110
1111         return highest_new_tsn;
1112 }
1113
1114 /* This is where we REALLY process a SACK.
1115  *
1116  * Process the SACK against the outqueue.  Mostly, this just frees
1117  * things off the transmitted queue.
1118  */
1119 int sctp_outq_sack(struct sctp_outq *q, struct sctp_sackhdr *sack)
1120 {
1121         struct sctp_association *asoc = q->asoc;
1122         struct sctp_transport *transport;
1123         struct sctp_chunk *tchunk = NULL;
1124         struct list_head *lchunk, *transport_list, *temp;
1125         sctp_sack_variable_t *frags = sack->variable;
1126         __u32 sack_ctsn, ctsn, tsn;
1127         __u32 highest_tsn, highest_new_tsn;
1128         __u32 sack_a_rwnd;
1129         unsigned outstanding;
1130         struct sctp_transport *primary = asoc->peer.primary_path;
1131         int count_of_newacks = 0;
1132         int gap_ack_blocks;
1133
1134         /* Grab the association's destination address list. */
1135         transport_list = &asoc->peer.transport_addr_list;
1136
1137         sack_ctsn = ntohl(sack->cum_tsn_ack);
1138         gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1139         /*
1140          * SFR-CACC algorithm:
1141          * On receipt of a SACK the sender SHOULD execute the
1142          * following statements.
1143          *
1144          * 1) If the cumulative ack in the SACK passes next tsn_at_change
1145          * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1146          * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1147          * all destinations.
1148          * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1149          * is set the receiver of the SACK MUST take the following actions:
1150          *
1151          * A) Initialize the cacc_saw_newack to 0 for all destination
1152          * addresses.
1153          *
1154          * Only bother if changeover_active is set. Otherwise, this is
1155          * totally suboptimal to do on every SACK.
1156          */
1157         if (primary->cacc.changeover_active) {
1158                 u8 clear_cycling = 0;
1159
1160                 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1161                         primary->cacc.changeover_active = 0;
1162                         clear_cycling = 1;
1163                 }
1164
1165                 if (clear_cycling || gap_ack_blocks) {
1166                         list_for_each_entry(transport, transport_list,
1167                                         transports) {
1168                                 if (clear_cycling)
1169                                         transport->cacc.cycling_changeover = 0;
1170                                 if (gap_ack_blocks)
1171                                         transport->cacc.cacc_saw_newack = 0;
1172                         }
1173                 }
1174         }
1175
1176         /* Get the highest TSN in the sack. */
1177         highest_tsn = sack_ctsn;
1178         if (gap_ack_blocks)
1179                 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1180
1181         if (TSN_lt(asoc->highest_sacked, highest_tsn)) {
1182                 highest_new_tsn = highest_tsn;
1183                 asoc->highest_sacked = highest_tsn;
1184         } else {
1185                 highest_new_tsn = sctp_highest_new_tsn(sack, asoc);
1186         }
1187
1188
1189         /* Run through the retransmit queue.  Credit bytes received
1190          * and free those chunks that we can.
1191          */
1192         sctp_check_transmitted(q, &q->retransmit, NULL, sack, highest_new_tsn);
1193
1194         /* Run through the transmitted queue.
1195          * Credit bytes received and free those chunks which we can.
1196          *
1197          * This is a MASSIVE candidate for optimization.
1198          */
1199         list_for_each_entry(transport, transport_list, transports) {
1200                 sctp_check_transmitted(q, &transport->transmitted,
1201                                        transport, sack, highest_new_tsn);
1202                 /*
1203                  * SFR-CACC algorithm:
1204                  * C) Let count_of_newacks be the number of
1205                  * destinations for which cacc_saw_newack is set.
1206                  */
1207                 if (transport->cacc.cacc_saw_newack)
1208                         count_of_newacks ++;
1209         }
1210
1211         if (gap_ack_blocks) {
1212                 sctp_mark_missing(q, &q->retransmit, NULL, highest_new_tsn, 0);
1213
1214                 list_for_each_entry(transport, transport_list, transports)
1215                         sctp_mark_missing(q, &transport->transmitted, transport,
1216                                           highest_new_tsn, count_of_newacks);
1217         }
1218
1219         /* Move the Cumulative TSN Ack Point if appropriate.  */
1220         if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn))
1221                 asoc->ctsn_ack_point = sack_ctsn;
1222
1223         /* Update unack_data field in the assoc. */
1224         sctp_sack_update_unack_data(asoc, sack);
1225
1226         ctsn = asoc->ctsn_ack_point;
1227
1228         /* Throw away stuff rotting on the sack queue.  */
1229         list_for_each_safe(lchunk, temp, &q->sacked) {
1230                 tchunk = list_entry(lchunk, struct sctp_chunk,
1231                                     transmitted_list);
1232                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1233                 if (TSN_lte(tsn, ctsn)) {
1234                         list_del_init(&tchunk->transmitted_list);
1235                         sctp_chunk_free(tchunk);
1236                 }
1237         }
1238
1239         /* ii) Set rwnd equal to the newly received a_rwnd minus the
1240          *     number of bytes still outstanding after processing the
1241          *     Cumulative TSN Ack and the Gap Ack Blocks.
1242          */
1243
1244         sack_a_rwnd = ntohl(sack->a_rwnd);
1245         outstanding = q->outstanding_bytes;
1246
1247         if (outstanding < sack_a_rwnd)
1248                 sack_a_rwnd -= outstanding;
1249         else
1250                 sack_a_rwnd = 0;
1251
1252         asoc->peer.rwnd = sack_a_rwnd;
1253
1254         sctp_generate_fwdtsn(q, sack_ctsn);
1255
1256         SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n",
1257                           __func__, sack_ctsn);
1258         SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, "
1259                           "%p is 0x%x. Adv peer ack point: 0x%x\n",
1260                           __func__, asoc, ctsn, asoc->adv_peer_ack_point);
1261
1262         /* See if all chunks are acked.
1263          * Make sure the empty queue handler will get run later.
1264          */
1265         q->empty = (list_empty(&q->out_chunk_list) &&
1266                     list_empty(&q->retransmit));
1267         if (!q->empty)
1268                 goto finish;
1269
1270         list_for_each_entry(transport, transport_list, transports) {
1271                 q->empty = q->empty && list_empty(&transport->transmitted);
1272                 if (!q->empty)
1273                         goto finish;
1274         }
1275
1276         SCTP_DEBUG_PRINTK("sack queue is empty.\n");
1277 finish:
1278         return q->empty;
1279 }
1280
1281 /* Is the outqueue empty?  */
1282 int sctp_outq_is_empty(const struct sctp_outq *q)
1283 {
1284         return q->empty;
1285 }
1286
1287 /********************************************************************
1288  * 2nd Level Abstractions
1289  ********************************************************************/
1290
1291 /* Go through a transport's transmitted list or the association's retransmit
1292  * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1293  * The retransmit list will not have an associated transport.
1294  *
1295  * I added coherent debug information output.   --xguo
1296  *
1297  * Instead of printing 'sacked' or 'kept' for each TSN on the
1298  * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1299  * KEPT TSN6-TSN7, etc.
1300  */
1301 static void sctp_check_transmitted(struct sctp_outq *q,
1302                                    struct list_head *transmitted_queue,
1303                                    struct sctp_transport *transport,
1304                                    struct sctp_sackhdr *sack,
1305                                    __u32 highest_new_tsn_in_sack)
1306 {
1307         struct list_head *lchunk;
1308         struct sctp_chunk *tchunk;
1309         struct list_head tlist;
1310         __u32 tsn;
1311         __u32 sack_ctsn;
1312         __u32 rtt;
1313         __u8 restart_timer = 0;
1314         int bytes_acked = 0;
1315
1316         /* These state variables are for coherent debug output. --xguo */
1317
1318 #if SCTP_DEBUG
1319         __u32 dbg_ack_tsn = 0;  /* An ACKed TSN range starts here... */
1320         __u32 dbg_last_ack_tsn = 0;  /* ...and finishes here.        */
1321         __u32 dbg_kept_tsn = 0; /* An un-ACKed range starts here...  */
1322         __u32 dbg_last_kept_tsn = 0; /* ...and finishes here.        */
1323
1324         /* 0 : The last TSN was ACKed.
1325          * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1326          * -1: We need to initialize.
1327          */
1328         int dbg_prt_state = -1;
1329 #endif /* SCTP_DEBUG */
1330
1331         sack_ctsn = ntohl(sack->cum_tsn_ack);
1332
1333         INIT_LIST_HEAD(&tlist);
1334
1335         /* The while loop will skip empty transmitted queues. */
1336         while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1337                 tchunk = list_entry(lchunk, struct sctp_chunk,
1338                                     transmitted_list);
1339
1340                 if (sctp_chunk_abandoned(tchunk)) {
1341                         /* Move the chunk to abandoned list. */
1342                         sctp_insert_list(&q->abandoned, lchunk);
1343
1344                         /* If this chunk has not been acked, stop
1345                          * considering it as 'outstanding'.
1346                          */
1347                         if (!tchunk->tsn_gap_acked) {
1348                                 tchunk->transport->flight_size -=
1349                                                 sctp_data_size(tchunk);
1350                                 q->outstanding_bytes -= sctp_data_size(tchunk);
1351                         }
1352                         continue;
1353                 }
1354
1355                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1356                 if (sctp_acked(sack, tsn)) {
1357                         /* If this queue is the retransmit queue, the
1358                          * retransmit timer has already reclaimed
1359                          * the outstanding bytes for this chunk, so only
1360                          * count bytes associated with a transport.
1361                          */
1362                         if (transport) {
1363                                 /* If this chunk is being used for RTT
1364                                  * measurement, calculate the RTT and update
1365                                  * the RTO using this value.
1366                                  *
1367                                  * 6.3.1 C5) Karn's algorithm: RTT measurements
1368                                  * MUST NOT be made using packets that were
1369                                  * retransmitted (and thus for which it is
1370                                  * ambiguous whether the reply was for the
1371                                  * first instance of the packet or a later
1372                                  * instance).
1373                                  */
1374                                 if (!tchunk->tsn_gap_acked &&
1375                                     !tchunk->resent &&
1376                                     tchunk->rtt_in_progress) {
1377                                         tchunk->rtt_in_progress = 0;
1378                                         rtt = jiffies - tchunk->sent_at;
1379                                         sctp_transport_update_rto(transport,
1380                                                                   rtt);
1381                                 }
1382                         }
1383                         if (TSN_lte(tsn, sack_ctsn)) {
1384                                 /* RFC 2960  6.3.2 Retransmission Timer Rules
1385                                  *
1386                                  * R3) Whenever a SACK is received
1387                                  * that acknowledges the DATA chunk
1388                                  * with the earliest outstanding TSN
1389                                  * for that address, restart T3-rtx
1390                                  * timer for that address with its
1391                                  * current RTO.
1392                                  */
1393                                 restart_timer = 1;
1394
1395                                 if (!tchunk->tsn_gap_acked) {
1396                                         tchunk->tsn_gap_acked = 1;
1397                                         bytes_acked += sctp_data_size(tchunk);
1398                                         /*
1399                                          * SFR-CACC algorithm:
1400                                          * 2) If the SACK contains gap acks
1401                                          * and the flag CHANGEOVER_ACTIVE is
1402                                          * set the receiver of the SACK MUST
1403                                          * take the following action:
1404                                          *
1405                                          * B) For each TSN t being acked that
1406                                          * has not been acked in any SACK so
1407                                          * far, set cacc_saw_newack to 1 for
1408                                          * the destination that the TSN was
1409                                          * sent to.
1410                                          */
1411                                         if (transport &&
1412                                             sack->num_gap_ack_blocks &&
1413                                             q->asoc->peer.primary_path->cacc.
1414                                             changeover_active)
1415                                                 transport->cacc.cacc_saw_newack
1416                                                         = 1;
1417                                 }
1418
1419                                 list_add_tail(&tchunk->transmitted_list,
1420                                               &q->sacked);
1421                         } else {
1422                                 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1423                                  * M2) Each time a SACK arrives reporting
1424                                  * 'Stray DATA chunk(s)' record the highest TSN
1425                                  * reported as newly acknowledged, call this
1426                                  * value 'HighestTSNinSack'. A newly
1427                                  * acknowledged DATA chunk is one not
1428                                  * previously acknowledged in a SACK.
1429                                  *
1430                                  * When the SCTP sender of data receives a SACK
1431                                  * chunk that acknowledges, for the first time,
1432                                  * the receipt of a DATA chunk, all the still
1433                                  * unacknowledged DATA chunks whose TSN is
1434                                  * older than that newly acknowledged DATA
1435                                  * chunk, are qualified as 'Stray DATA chunks'.
1436                                  */
1437                                 if (!tchunk->tsn_gap_acked) {
1438                                         tchunk->tsn_gap_acked = 1;
1439                                         bytes_acked += sctp_data_size(tchunk);
1440                                 }
1441                                 list_add_tail(lchunk, &tlist);
1442                         }
1443
1444 #if SCTP_DEBUG
1445                         switch (dbg_prt_state) {
1446                         case 0: /* last TSN was ACKed */
1447                                 if (dbg_last_ack_tsn + 1 == tsn) {
1448                                         /* This TSN belongs to the
1449                                          * current ACK range.
1450                                          */
1451                                         break;
1452                                 }
1453
1454                                 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1455                                         /* Display the end of the
1456                                          * current range.
1457                                          */
1458                                         SCTP_DEBUG_PRINTK("-%08x",
1459                                                           dbg_last_ack_tsn);
1460                                 }
1461
1462                                 /* Start a new range.  */
1463                                 SCTP_DEBUG_PRINTK(",%08x", tsn);
1464                                 dbg_ack_tsn = tsn;
1465                                 break;
1466
1467                         case 1: /* The last TSN was NOT ACKed. */
1468                                 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1469                                         /* Display the end of current range. */
1470                                         SCTP_DEBUG_PRINTK("-%08x",
1471                                                           dbg_last_kept_tsn);
1472                                 }
1473
1474                                 SCTP_DEBUG_PRINTK("\n");
1475
1476                                 /* FALL THROUGH... */
1477                         default:
1478                                 /* This is the first-ever TSN we examined.  */
1479                                 /* Start a new range of ACK-ed TSNs.  */
1480                                 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn);
1481                                 dbg_prt_state = 0;
1482                                 dbg_ack_tsn = tsn;
1483                         }
1484
1485                         dbg_last_ack_tsn = tsn;
1486 #endif /* SCTP_DEBUG */
1487
1488                 } else {
1489                         if (tchunk->tsn_gap_acked) {
1490                                 SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1491                                                   "data TSN: 0x%x\n",
1492                                                   __func__,
1493                                                   tsn);
1494                                 tchunk->tsn_gap_acked = 0;
1495
1496                                 bytes_acked -= sctp_data_size(tchunk);
1497
1498                                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1499                                  *
1500                                  * R4) Whenever a SACK is received missing a
1501                                  * TSN that was previously acknowledged via a
1502                                  * Gap Ack Block, start T3-rtx for the
1503                                  * destination address to which the DATA
1504                                  * chunk was originally
1505                                  * transmitted if it is not already running.
1506                                  */
1507                                 restart_timer = 1;
1508                         }
1509
1510                         list_add_tail(lchunk, &tlist);
1511
1512 #if SCTP_DEBUG
1513                         /* See the above comments on ACK-ed TSNs. */
1514                         switch (dbg_prt_state) {
1515                         case 1:
1516                                 if (dbg_last_kept_tsn + 1 == tsn)
1517                                         break;
1518
1519                                 if (dbg_last_kept_tsn != dbg_kept_tsn)
1520                                         SCTP_DEBUG_PRINTK("-%08x",
1521                                                           dbg_last_kept_tsn);
1522
1523                                 SCTP_DEBUG_PRINTK(",%08x", tsn);
1524                                 dbg_kept_tsn = tsn;
1525                                 break;
1526
1527                         case 0:
1528                                 if (dbg_last_ack_tsn != dbg_ack_tsn)
1529                                         SCTP_DEBUG_PRINTK("-%08x",
1530                                                           dbg_last_ack_tsn);
1531                                 SCTP_DEBUG_PRINTK("\n");
1532
1533                                 /* FALL THROUGH... */
1534                         default:
1535                                 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn);
1536                                 dbg_prt_state = 1;
1537                                 dbg_kept_tsn = tsn;
1538                         }
1539
1540                         dbg_last_kept_tsn = tsn;
1541 #endif /* SCTP_DEBUG */
1542                 }
1543         }
1544
1545 #if SCTP_DEBUG
1546         /* Finish off the last range, displaying its ending TSN.  */
1547         switch (dbg_prt_state) {
1548         case 0:
1549                 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1550                         SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_ack_tsn);
1551                 } else {
1552                         SCTP_DEBUG_PRINTK("\n");
1553                 }
1554         break;
1555
1556         case 1:
1557                 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1558                         SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_kept_tsn);
1559                 } else {
1560                         SCTP_DEBUG_PRINTK("\n");
1561                 }
1562         }
1563 #endif /* SCTP_DEBUG */
1564         if (transport) {
1565                 if (bytes_acked) {
1566                         /* 8.2. When an outstanding TSN is acknowledged,
1567                          * the endpoint shall clear the error counter of
1568                          * the destination transport address to which the
1569                          * DATA chunk was last sent.
1570                          * The association's overall error counter is
1571                          * also cleared.
1572                          */
1573                         transport->error_count = 0;
1574                         transport->asoc->overall_error_count = 0;
1575
1576                         /* Mark the destination transport address as
1577                          * active if it is not so marked.
1578                          */
1579                         if ((transport->state == SCTP_INACTIVE) ||
1580                             (transport->state == SCTP_UNCONFIRMED)) {
1581                                 sctp_assoc_control_transport(
1582                                         transport->asoc,
1583                                         transport,
1584                                         SCTP_TRANSPORT_UP,
1585                                         SCTP_RECEIVED_SACK);
1586                         }
1587
1588                         sctp_transport_raise_cwnd(transport, sack_ctsn,
1589                                                   bytes_acked);
1590
1591                         transport->flight_size -= bytes_acked;
1592                         if (transport->flight_size == 0)
1593                                 transport->partial_bytes_acked = 0;
1594                         q->outstanding_bytes -= bytes_acked;
1595                 } else {
1596                         /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1597                          * When a sender is doing zero window probing, it
1598                          * should not timeout the association if it continues
1599                          * to receive new packets from the receiver. The
1600                          * reason is that the receiver MAY keep its window
1601                          * closed for an indefinite time.
1602                          * A sender is doing zero window probing when the
1603                          * receiver's advertised window is zero, and there is
1604                          * only one data chunk in flight to the receiver.
1605                          */
1606                         if (!q->asoc->peer.rwnd &&
1607                             !list_empty(&tlist) &&
1608                             (sack_ctsn+2 == q->asoc->next_tsn)) {
1609                                 SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1610                                                   "window probe: %u\n",
1611                                                   __func__, sack_ctsn);
1612                                 q->asoc->overall_error_count = 0;
1613                                 transport->error_count = 0;
1614                         }
1615                 }
1616
1617                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1618                  *
1619                  * R2) Whenever all outstanding data sent to an address have
1620                  * been acknowledged, turn off the T3-rtx timer of that
1621                  * address.
1622                  */
1623                 if (!transport->flight_size) {
1624                         if (timer_pending(&transport->T3_rtx_timer) &&
1625                             del_timer(&transport->T3_rtx_timer)) {
1626                                 sctp_transport_put(transport);
1627                         }
1628                 } else if (restart_timer) {
1629                         if (!mod_timer(&transport->T3_rtx_timer,
1630                                        jiffies + transport->rto))
1631                                 sctp_transport_hold(transport);
1632                 }
1633         }
1634
1635         list_splice(&tlist, transmitted_queue);
1636 }
1637
1638 /* Mark chunks as missing and consequently may get retransmitted. */
1639 static void sctp_mark_missing(struct sctp_outq *q,
1640                               struct list_head *transmitted_queue,
1641                               struct sctp_transport *transport,
1642                               __u32 highest_new_tsn_in_sack,
1643                               int count_of_newacks)
1644 {
1645         struct sctp_chunk *chunk;
1646         __u32 tsn;
1647         char do_fast_retransmit = 0;
1648         struct sctp_transport *primary = q->asoc->peer.primary_path;
1649
1650         list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1651
1652                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1653
1654                 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1655                  * 'Unacknowledged TSN's', if the TSN number of an
1656                  * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1657                  * value, increment the 'TSN.Missing.Report' count on that
1658                  * chunk if it has NOT been fast retransmitted or marked for
1659                  * fast retransmit already.
1660                  */
1661                 if (!chunk->fast_retransmit &&
1662                     !chunk->tsn_gap_acked &&
1663                     TSN_lt(tsn, highest_new_tsn_in_sack)) {
1664
1665                         /* SFR-CACC may require us to skip marking
1666                          * this chunk as missing.
1667                          */
1668                         if (!transport || !sctp_cacc_skip(primary, transport,
1669                                             count_of_newacks, tsn)) {
1670                                 chunk->tsn_missing_report++;
1671
1672                                 SCTP_DEBUG_PRINTK(
1673                                         "%s: TSN 0x%x missing counter: %d\n",
1674                                         __func__, tsn,
1675                                         chunk->tsn_missing_report);
1676                         }
1677                 }
1678                 /*
1679                  * M4) If any DATA chunk is found to have a
1680                  * 'TSN.Missing.Report'
1681                  * value larger than or equal to 3, mark that chunk for
1682                  * retransmission and start the fast retransmit procedure.
1683                  */
1684
1685                 if (chunk->tsn_missing_report >= 3) {
1686                         chunk->fast_retransmit = 1;
1687                         do_fast_retransmit = 1;
1688                 }
1689         }
1690
1691         if (transport) {
1692                 if (do_fast_retransmit)
1693                         sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1694
1695                 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1696                                   "ssthresh: %d, flight_size: %d, pba: %d\n",
1697                                   __func__, transport, transport->cwnd,
1698                                   transport->ssthresh, transport->flight_size,
1699                                   transport->partial_bytes_acked);
1700         }
1701 }
1702
1703 /* Is the given TSN acked by this packet?  */
1704 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1705 {
1706         int i;
1707         sctp_sack_variable_t *frags;
1708         __u16 gap;
1709         __u32 ctsn = ntohl(sack->cum_tsn_ack);
1710
1711         if (TSN_lte(tsn, ctsn))
1712                 goto pass;
1713
1714         /* 3.3.4 Selective Acknowledgement (SACK) (3):
1715          *
1716          * Gap Ack Blocks:
1717          *  These fields contain the Gap Ack Blocks. They are repeated
1718          *  for each Gap Ack Block up to the number of Gap Ack Blocks
1719          *  defined in the Number of Gap Ack Blocks field. All DATA
1720          *  chunks with TSNs greater than or equal to (Cumulative TSN
1721          *  Ack + Gap Ack Block Start) and less than or equal to
1722          *  (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1723          *  Block are assumed to have been received correctly.
1724          */
1725
1726         frags = sack->variable;
1727         gap = tsn - ctsn;
1728         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1729                 if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1730                     TSN_lte(gap, ntohs(frags[i].gab.end)))
1731                         goto pass;
1732         }
1733
1734         return 0;
1735 pass:
1736         return 1;
1737 }
1738
1739 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1740                                     int nskips, __be16 stream)
1741 {
1742         int i;
1743
1744         for (i = 0; i < nskips; i++) {
1745                 if (skiplist[i].stream == stream)
1746                         return i;
1747         }
1748         return i;
1749 }
1750
1751 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1752 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1753 {
1754         struct sctp_association *asoc = q->asoc;
1755         struct sctp_chunk *ftsn_chunk = NULL;
1756         struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1757         int nskips = 0;
1758         int skip_pos = 0;
1759         __u32 tsn;
1760         struct sctp_chunk *chunk;
1761         struct list_head *lchunk, *temp;
1762
1763         /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1764          * received SACK.
1765          *
1766          * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1767          * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1768          */
1769         if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1770                 asoc->adv_peer_ack_point = ctsn;
1771
1772         /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1773          * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1774          * the chunk next in the out-queue space is marked as "abandoned" as
1775          * shown in the following example:
1776          *
1777          * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1778          * and the Advanced.Peer.Ack.Point is updated to this value:
1779          *
1780          *   out-queue at the end of  ==>   out-queue after Adv.Ack.Point
1781          *   normal SACK processing           local advancement
1782          *                ...                           ...
1783          *   Adv.Ack.Pt-> 102 acked                     102 acked
1784          *                103 abandoned                 103 abandoned
1785          *                104 abandoned     Adv.Ack.P-> 104 abandoned
1786          *                105                           105
1787          *                106 acked                     106 acked
1788          *                ...                           ...
1789          *
1790          * In this example, the data sender successfully advanced the
1791          * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1792          */
1793         list_for_each_safe(lchunk, temp, &q->abandoned) {
1794                 chunk = list_entry(lchunk, struct sctp_chunk,
1795                                         transmitted_list);
1796                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1797
1798                 /* Remove any chunks in the abandoned queue that are acked by
1799                  * the ctsn.
1800                  */
1801                 if (TSN_lte(tsn, ctsn)) {
1802                         list_del_init(lchunk);
1803                         sctp_chunk_free(chunk);
1804                 } else {
1805                         if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1806                                 asoc->adv_peer_ack_point = tsn;
1807                                 if (chunk->chunk_hdr->flags &
1808                                          SCTP_DATA_UNORDERED)
1809                                         continue;
1810                                 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1811                                                 nskips,
1812                                                 chunk->subh.data_hdr->stream);
1813                                 ftsn_skip_arr[skip_pos].stream =
1814                                         chunk->subh.data_hdr->stream;
1815                                 ftsn_skip_arr[skip_pos].ssn =
1816                                          chunk->subh.data_hdr->ssn;
1817                                 if (skip_pos == nskips)
1818                                         nskips++;
1819                                 if (nskips == 10)
1820                                         break;
1821                         } else
1822                                 break;
1823                 }
1824         }
1825
1826         /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1827          * is greater than the Cumulative TSN ACK carried in the received
1828          * SACK, the data sender MUST send the data receiver a FORWARD TSN
1829          * chunk containing the latest value of the
1830          * "Advanced.Peer.Ack.Point".
1831          *
1832          * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1833          * list each stream and sequence number in the forwarded TSN. This
1834          * information will enable the receiver to easily find any
1835          * stranded TSN's waiting on stream reorder queues. Each stream
1836          * SHOULD only be reported once; this means that if multiple
1837          * abandoned messages occur in the same stream then only the
1838          * highest abandoned stream sequence number is reported. If the
1839          * total size of the FORWARD TSN does NOT fit in a single MTU then
1840          * the sender of the FORWARD TSN SHOULD lower the
1841          * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1842          * single MTU.
1843          */
1844         if (asoc->adv_peer_ack_point > ctsn)
1845                 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1846                                               nskips, &ftsn_skip_arr[0]);
1847
1848         if (ftsn_chunk) {
1849                 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1850                 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
1851         }
1852 }