Merge branches 'acpica', 'aml-custom', 'bugzilla-16548', 'bugzilla-20242', 'd3-cold...
[pandora-kernel.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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
50
51 #include <linux/types.h>
52 #include <linux/list.h>   /* For struct list_head */
53 #include <linux/socket.h>
54 #include <linux/ip.h>
55 #include <linux/slab.h>
56 #include <net/sock.h>     /* For skb_set_owner_w */
57
58 #include <net/sctp/sctp.h>
59 #include <net/sctp/sm.h>
60
61 /* Declare internal functions here.  */
62 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn);
63 static void sctp_check_transmitted(struct sctp_outq *q,
64                                    struct list_head *transmitted_queue,
65                                    struct sctp_transport *transport,
66                                    struct sctp_sackhdr *sack,
67                                    __u32 *highest_new_tsn);
68
69 static void sctp_mark_missing(struct sctp_outq *q,
70                               struct list_head *transmitted_queue,
71                               struct sctp_transport *transport,
72                               __u32 highest_new_tsn,
73                               int count_of_newacks);
74
75 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn);
76
77 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout);
78
79 /* Add data to the front of the queue. */
80 static inline void sctp_outq_head_data(struct sctp_outq *q,
81                                         struct sctp_chunk *ch)
82 {
83         list_add(&ch->list, &q->out_chunk_list);
84         q->out_qlen += ch->skb->len;
85 }
86
87 /* Take data from the front of the queue. */
88 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
89 {
90         struct sctp_chunk *ch = NULL;
91
92         if (!list_empty(&q->out_chunk_list)) {
93                 struct list_head *entry = q->out_chunk_list.next;
94
95                 ch = list_entry(entry, struct sctp_chunk, list);
96                 list_del_init(entry);
97                 q->out_qlen -= ch->skb->len;
98         }
99         return ch;
100 }
101 /* Add data chunk to the end of the queue. */
102 static inline void sctp_outq_tail_data(struct sctp_outq *q,
103                                        struct sctp_chunk *ch)
104 {
105         list_add_tail(&ch->list, &q->out_chunk_list);
106         q->out_qlen += ch->skb->len;
107 }
108
109 /*
110  * SFR-CACC algorithm:
111  * D) If count_of_newacks is greater than or equal to 2
112  * and t was not sent to the current primary then the
113  * sender MUST NOT increment missing report count for t.
114  */
115 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary,
116                                        struct sctp_transport *transport,
117                                        int count_of_newacks)
118 {
119         if (count_of_newacks >=2 && transport != primary)
120                 return 1;
121         return 0;
122 }
123
124 /*
125  * SFR-CACC algorithm:
126  * F) If count_of_newacks is less than 2, let d be the
127  * destination to which t was sent. If cacc_saw_newack
128  * is 0 for destination d, then the sender MUST NOT
129  * increment missing report count for t.
130  */
131 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport,
132                                        int count_of_newacks)
133 {
134         if (count_of_newacks < 2 && !transport->cacc.cacc_saw_newack)
135                 return 1;
136         return 0;
137 }
138
139 /*
140  * SFR-CACC algorithm:
141  * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
142  * execute steps C, D, F.
143  *
144  * C has been implemented in sctp_outq_sack
145  */
146 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
147                                      struct sctp_transport *transport,
148                                      int count_of_newacks)
149 {
150         if (!primary->cacc.cycling_changeover) {
151                 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
152                         return 1;
153                 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
154                         return 1;
155                 return 0;
156         }
157         return 0;
158 }
159
160 /*
161  * SFR-CACC algorithm:
162  * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
163  * than next_tsn_at_change of the current primary, then
164  * the sender MUST NOT increment missing report count
165  * for t.
166  */
167 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
168 {
169         if (primary->cacc.cycling_changeover &&
170             TSN_lt(tsn, primary->cacc.next_tsn_at_change))
171                 return 1;
172         return 0;
173 }
174
175 /*
176  * SFR-CACC algorithm:
177  * 3) If the missing report count for TSN t is to be
178  * incremented according to [RFC2960] and
179  * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
180  * then the sender MUST further execute steps 3.1 and
181  * 3.2 to determine if the missing report count for
182  * TSN t SHOULD NOT be incremented.
183  *
184  * 3.3) If 3.1 and 3.2 do not dictate that the missing
185  * report count for t should not be incremented, then
186  * the sender SHOULD increment missing report count for
187  * t (according to [RFC2960] and [SCTP_STEWART_2002]).
188  */
189 static inline int sctp_cacc_skip(struct sctp_transport *primary,
190                                  struct sctp_transport *transport,
191                                  int count_of_newacks,
192                                  __u32 tsn)
193 {
194         if (primary->cacc.changeover_active &&
195             (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) ||
196              sctp_cacc_skip_3_2(primary, tsn)))
197                 return 1;
198         return 0;
199 }
200
201 /* Initialize an existing sctp_outq.  This does the boring stuff.
202  * You still need to define handlers if you really want to DO
203  * something with this structure...
204  */
205 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
206 {
207         q->asoc = asoc;
208         INIT_LIST_HEAD(&q->out_chunk_list);
209         INIT_LIST_HEAD(&q->control_chunk_list);
210         INIT_LIST_HEAD(&q->retransmit);
211         INIT_LIST_HEAD(&q->sacked);
212         INIT_LIST_HEAD(&q->abandoned);
213
214         q->fast_rtx = 0;
215         q->outstanding_bytes = 0;
216         q->empty = 1;
217         q->cork  = 0;
218
219         q->malloced = 0;
220         q->out_qlen = 0;
221 }
222
223 /* Free the outqueue structure and any related pending chunks.
224  */
225 void sctp_outq_teardown(struct sctp_outq *q)
226 {
227         struct sctp_transport *transport;
228         struct list_head *lchunk, *temp;
229         struct sctp_chunk *chunk, *tmp;
230
231         /* Throw away unacknowledged chunks. */
232         list_for_each_entry(transport, &q->asoc->peer.transport_addr_list,
233                         transports) {
234                 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
235                         chunk = list_entry(lchunk, struct sctp_chunk,
236                                            transmitted_list);
237                         /* Mark as part of a failed message. */
238                         sctp_chunk_fail(chunk, q->error);
239                         sctp_chunk_free(chunk);
240                 }
241         }
242
243         /* Throw away chunks that have been gap ACKed.  */
244         list_for_each_safe(lchunk, temp, &q->sacked) {
245                 list_del_init(lchunk);
246                 chunk = list_entry(lchunk, struct sctp_chunk,
247                                    transmitted_list);
248                 sctp_chunk_fail(chunk, q->error);
249                 sctp_chunk_free(chunk);
250         }
251
252         /* Throw away any chunks in the retransmit queue. */
253         list_for_each_safe(lchunk, temp, &q->retransmit) {
254                 list_del_init(lchunk);
255                 chunk = list_entry(lchunk, struct sctp_chunk,
256                                    transmitted_list);
257                 sctp_chunk_fail(chunk, q->error);
258                 sctp_chunk_free(chunk);
259         }
260
261         /* Throw away any chunks that are in the abandoned queue. */
262         list_for_each_safe(lchunk, temp, &q->abandoned) {
263                 list_del_init(lchunk);
264                 chunk = list_entry(lchunk, struct sctp_chunk,
265                                    transmitted_list);
266                 sctp_chunk_fail(chunk, q->error);
267                 sctp_chunk_free(chunk);
268         }
269
270         /* Throw away any leftover data chunks. */
271         while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
272
273                 /* Mark as send failure. */
274                 sctp_chunk_fail(chunk, q->error);
275                 sctp_chunk_free(chunk);
276         }
277
278         q->error = 0;
279
280         /* Throw away any leftover control chunks. */
281         list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
282                 list_del_init(&chunk->list);
283                 sctp_chunk_free(chunk);
284         }
285 }
286
287 /* Free the outqueue structure and any related pending chunks.  */
288 void sctp_outq_free(struct sctp_outq *q)
289 {
290         /* Throw away leftover chunks. */
291         sctp_outq_teardown(q);
292
293         /* If we were kmalloc()'d, free the memory.  */
294         if (q->malloced)
295                 kfree(q);
296 }
297
298 /* Put a new chunk in an sctp_outq.  */
299 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk)
300 {
301         int error = 0;
302
303         SCTP_DEBUG_PRINTK("sctp_outq_tail(%p, %p[%s])\n",
304                           q, chunk, chunk && chunk->chunk_hdr ?
305                           sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
306                           : "Illegal Chunk");
307
308         /* If it is data, queue it up, otherwise, send it
309          * immediately.
310          */
311         if (sctp_chunk_is_data(chunk)) {
312                 /* Is it OK to queue data chunks?  */
313                 /* From 9. Termination of Association
314                  *
315                  * When either endpoint performs a shutdown, the
316                  * association on each peer will stop accepting new
317                  * data from its user and only deliver data in queue
318                  * at the time of sending or receiving the SHUTDOWN
319                  * chunk.
320                  */
321                 switch (q->asoc->state) {
322                 case SCTP_STATE_EMPTY:
323                 case SCTP_STATE_CLOSED:
324                 case SCTP_STATE_SHUTDOWN_PENDING:
325                 case SCTP_STATE_SHUTDOWN_SENT:
326                 case SCTP_STATE_SHUTDOWN_RECEIVED:
327                 case SCTP_STATE_SHUTDOWN_ACK_SENT:
328                         /* Cannot send after transport endpoint shutdown */
329                         error = -ESHUTDOWN;
330                         break;
331
332                 default:
333                         SCTP_DEBUG_PRINTK("outqueueing (%p, %p[%s])\n",
334                           q, chunk, chunk && chunk->chunk_hdr ?
335                           sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
336                           : "Illegal Chunk");
337
338                         sctp_outq_tail_data(q, chunk);
339                         if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
340                                 SCTP_INC_STATS(SCTP_MIB_OUTUNORDERCHUNKS);
341                         else
342                                 SCTP_INC_STATS(SCTP_MIB_OUTORDERCHUNKS);
343                         q->empty = 0;
344                         break;
345                 }
346         } else {
347                 list_add_tail(&chunk->list, &q->control_chunk_list);
348                 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
349         }
350
351         if (error < 0)
352                 return error;
353
354         if (!q->cork)
355                 error = sctp_outq_flush(q, 0);
356
357         return error;
358 }
359
360 /* Insert a chunk into the sorted list based on the TSNs.  The retransmit list
361  * and the abandoned list are in ascending order.
362  */
363 static void sctp_insert_list(struct list_head *head, struct list_head *new)
364 {
365         struct list_head *pos;
366         struct sctp_chunk *nchunk, *lchunk;
367         __u32 ntsn, ltsn;
368         int done = 0;
369
370         nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
371         ntsn = ntohl(nchunk->subh.data_hdr->tsn);
372
373         list_for_each(pos, head) {
374                 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
375                 ltsn = ntohl(lchunk->subh.data_hdr->tsn);
376                 if (TSN_lt(ntsn, ltsn)) {
377                         list_add(new, pos->prev);
378                         done = 1;
379                         break;
380                 }
381         }
382         if (!done)
383                 list_add_tail(new, head);
384 }
385
386 /* Mark all the eligible packets on a transport for retransmission.  */
387 void sctp_retransmit_mark(struct sctp_outq *q,
388                           struct sctp_transport *transport,
389                           __u8 reason)
390 {
391         struct list_head *lchunk, *ltemp;
392         struct sctp_chunk *chunk;
393
394         /* Walk through the specified transmitted queue.  */
395         list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
396                 chunk = list_entry(lchunk, struct sctp_chunk,
397                                    transmitted_list);
398
399                 /* If the chunk is abandoned, move it to abandoned list. */
400                 if (sctp_chunk_abandoned(chunk)) {
401                         list_del_init(lchunk);
402                         sctp_insert_list(&q->abandoned, lchunk);
403
404                         /* If this chunk has not been previousely acked,
405                          * stop considering it 'outstanding'.  Our peer
406                          * will most likely never see it since it will
407                          * not be retransmitted
408                          */
409                         if (!chunk->tsn_gap_acked) {
410                                 if (chunk->transport)
411                                         chunk->transport->flight_size -=
412                                                         sctp_data_size(chunk);
413                                 q->outstanding_bytes -= sctp_data_size(chunk);
414                                 q->asoc->peer.rwnd += (sctp_data_size(chunk) +
415                                                         sizeof(struct sk_buff));
416                         }
417                         continue;
418                 }
419
420                 /* If we are doing  retransmission due to a timeout or pmtu
421                  * discovery, only the  chunks that are not yet acked should
422                  * be added to the retransmit queue.
423                  */
424                 if ((reason == SCTP_RTXR_FAST_RTX  &&
425                             (chunk->fast_retransmit == SCTP_NEED_FRTX)) ||
426                     (reason != SCTP_RTXR_FAST_RTX  && !chunk->tsn_gap_acked)) {
427                         /* RFC 2960 6.2.1 Processing a Received SACK
428                          *
429                          * C) Any time a DATA chunk is marked for
430                          * retransmission (via either T3-rtx timer expiration
431                          * (Section 6.3.3) or via fast retransmit
432                          * (Section 7.2.4)), add the data size of those
433                          * chunks to the rwnd.
434                          */
435                         q->asoc->peer.rwnd += (sctp_data_size(chunk) +
436                                                 sizeof(struct sk_buff));
437                         q->outstanding_bytes -= sctp_data_size(chunk);
438                         if (chunk->transport)
439                                 transport->flight_size -= sctp_data_size(chunk);
440
441                         /* sctpimpguide-05 Section 2.8.2
442                          * M5) If a T3-rtx timer expires, the
443                          * 'TSN.Missing.Report' of all affected TSNs is set
444                          * to 0.
445                          */
446                         chunk->tsn_missing_report = 0;
447
448                         /* If a chunk that is being used for RTT measurement
449                          * has to be retransmitted, we cannot use this chunk
450                          * anymore for RTT measurements. Reset rto_pending so
451                          * that a new RTT measurement is started when a new
452                          * data chunk is sent.
453                          */
454                         if (chunk->rtt_in_progress) {
455                                 chunk->rtt_in_progress = 0;
456                                 transport->rto_pending = 0;
457                         }
458
459                         /* Move the chunk to the retransmit queue. The chunks
460                          * on the retransmit queue are always kept in order.
461                          */
462                         list_del_init(lchunk);
463                         sctp_insert_list(&q->retransmit, lchunk);
464                 }
465         }
466
467         SCTP_DEBUG_PRINTK("%s: transport: %p, reason: %d, "
468                           "cwnd: %d, ssthresh: %d, flight_size: %d, "
469                           "pba: %d\n", __func__,
470                           transport, reason,
471                           transport->cwnd, transport->ssthresh,
472                           transport->flight_size,
473                           transport->partial_bytes_acked);
474
475 }
476
477 /* Mark all the eligible packets on a transport for retransmission and force
478  * one packet out.
479  */
480 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
481                      sctp_retransmit_reason_t reason)
482 {
483         int error = 0;
484
485         switch(reason) {
486         case SCTP_RTXR_T3_RTX:
487                 SCTP_INC_STATS(SCTP_MIB_T3_RETRANSMITS);
488                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
489                 /* Update the retran path if the T3-rtx timer has expired for
490                  * the current retran path.
491                  */
492                 if (transport == transport->asoc->peer.retran_path)
493                         sctp_assoc_update_retran_path(transport->asoc);
494                 transport->asoc->rtx_data_chunks +=
495                         transport->asoc->unack_data;
496                 break;
497         case SCTP_RTXR_FAST_RTX:
498                 SCTP_INC_STATS(SCTP_MIB_FAST_RETRANSMITS);
499                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
500                 q->fast_rtx = 1;
501                 break;
502         case SCTP_RTXR_PMTUD:
503                 SCTP_INC_STATS(SCTP_MIB_PMTUD_RETRANSMITS);
504                 break;
505         case SCTP_RTXR_T1_RTX:
506                 SCTP_INC_STATS(SCTP_MIB_T1_RETRANSMITS);
507                 transport->asoc->init_retries++;
508                 break;
509         default:
510                 BUG();
511         }
512
513         sctp_retransmit_mark(q, transport, reason);
514
515         /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
516          * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
517          * following the procedures outlined in C1 - C5.
518          */
519         if (reason == SCTP_RTXR_T3_RTX)
520                 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
521
522         /* Flush the queues only on timeout, since fast_rtx is only
523          * triggered during sack processing and the queue
524          * will be flushed at the end.
525          */
526         if (reason != SCTP_RTXR_FAST_RTX)
527                 error = sctp_outq_flush(q, /* rtx_timeout */ 1);
528
529         if (error)
530                 q->asoc->base.sk->sk_err = -error;
531 }
532
533 /*
534  * Transmit DATA chunks on the retransmit queue.  Upon return from
535  * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
536  * need to be transmitted by the caller.
537  * We assume that pkt->transport has already been set.
538  *
539  * The return value is a normal kernel error return value.
540  */
541 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
542                                int rtx_timeout, int *start_timer)
543 {
544         struct list_head *lqueue;
545         struct sctp_transport *transport = pkt->transport;
546         sctp_xmit_t status;
547         struct sctp_chunk *chunk, *chunk1;
548         int fast_rtx;
549         int error = 0;
550         int timer = 0;
551         int done = 0;
552
553         lqueue = &q->retransmit;
554         fast_rtx = q->fast_rtx;
555
556         /* This loop handles time-out retransmissions, fast retransmissions,
557          * and retransmissions due to opening of whindow.
558          *
559          * RFC 2960 6.3.3 Handle T3-rtx Expiration
560          *
561          * E3) Determine how many of the earliest (i.e., lowest TSN)
562          * outstanding DATA chunks for the address for which the
563          * T3-rtx has expired will fit into a single packet, subject
564          * to the MTU constraint for the path corresponding to the
565          * destination transport address to which the retransmission
566          * is being sent (this may be different from the address for
567          * which the timer expires [see Section 6.4]). Call this value
568          * K. Bundle and retransmit those K DATA chunks in a single
569          * packet to the destination endpoint.
570          *
571          * [Just to be painfully clear, if we are retransmitting
572          * because a timeout just happened, we should send only ONE
573          * packet of retransmitted data.]
574          *
575          * For fast retransmissions we also send only ONE packet.  However,
576          * if we are just flushing the queue due to open window, we'll
577          * try to send as much as possible.
578          */
579         list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
580
581                 /* Make sure that Gap Acked TSNs are not retransmitted.  A
582                  * simple approach is just to move such TSNs out of the
583                  * way and into a 'transmitted' queue and skip to the
584                  * next chunk.
585                  */
586                 if (chunk->tsn_gap_acked) {
587                         list_del(&chunk->transmitted_list);
588                         list_add_tail(&chunk->transmitted_list,
589                                         &transport->transmitted);
590                         continue;
591                 }
592
593                 /* If we are doing fast retransmit, ignore non-fast_rtransmit
594                  * chunks
595                  */
596                 if (fast_rtx && !chunk->fast_retransmit)
597                         continue;
598
599 redo:
600                 /* Attempt to append this chunk to the packet. */
601                 status = sctp_packet_append_chunk(pkt, chunk);
602
603                 switch (status) {
604                 case SCTP_XMIT_PMTU_FULL:
605                         if (!pkt->has_data && !pkt->has_cookie_echo) {
606                                 /* If this packet did not contain DATA then
607                                  * retransmission did not happen, so do it
608                                  * again.  We'll ignore the error here since
609                                  * control chunks are already freed so there
610                                  * is nothing we can do.
611                                  */
612                                 sctp_packet_transmit(pkt);
613                                 goto redo;
614                         }
615
616                         /* Send this packet.  */
617                         error = sctp_packet_transmit(pkt);
618
619                         /* If we are retransmitting, we should only
620                          * send a single packet.
621                          */
622                         if (rtx_timeout || fast_rtx)
623                                 done = 1;
624
625                         /* Bundle next chunk in the next round.  */
626                         break;
627
628                 case SCTP_XMIT_RWND_FULL:
629                         /* Send this packet. */
630                         error = sctp_packet_transmit(pkt);
631
632                         /* Stop sending DATA as there is no more room
633                          * at the receiver.
634                          */
635                         done = 1;
636                         break;
637
638                 case SCTP_XMIT_NAGLE_DELAY:
639                         /* Send this packet. */
640                         error = sctp_packet_transmit(pkt);
641
642                         /* Stop sending DATA because of nagle delay. */
643                         done = 1;
644                         break;
645
646                 default:
647                         /* The append was successful, so add this chunk to
648                          * the transmitted list.
649                          */
650                         list_del(&chunk->transmitted_list);
651                         list_add_tail(&chunk->transmitted_list,
652                                         &transport->transmitted);
653
654                         /* Mark the chunk as ineligible for fast retransmit
655                          * after it is retransmitted.
656                          */
657                         if (chunk->fast_retransmit == SCTP_NEED_FRTX)
658                                 chunk->fast_retransmit = SCTP_DONT_FRTX;
659
660                         q->empty = 0;
661                         break;
662                 }
663
664                 /* Set the timer if there were no errors */
665                 if (!error && !timer)
666                         timer = 1;
667
668                 if (done)
669                         break;
670         }
671
672         /* If we are here due to a retransmit timeout or a fast
673          * retransmit and if there are any chunks left in the retransmit
674          * queue that could not fit in the PMTU sized packet, they need
675          * to be marked as ineligible for a subsequent fast retransmit.
676          */
677         if (rtx_timeout || fast_rtx) {
678                 list_for_each_entry(chunk1, lqueue, transmitted_list) {
679                         if (chunk1->fast_retransmit == SCTP_NEED_FRTX)
680                                 chunk1->fast_retransmit = SCTP_DONT_FRTX;
681                 }
682         }
683
684         *start_timer = timer;
685
686         /* Clear fast retransmit hint */
687         if (fast_rtx)
688                 q->fast_rtx = 0;
689
690         return error;
691 }
692
693 /* Cork the outqueue so queued chunks are really queued. */
694 int sctp_outq_uncork(struct sctp_outq *q)
695 {
696         int error = 0;
697         if (q->cork)
698                 q->cork = 0;
699         error = sctp_outq_flush(q, 0);
700         return error;
701 }
702
703
704 /*
705  * Try to flush an outqueue.
706  *
707  * Description: Send everything in q which we legally can, subject to
708  * congestion limitations.
709  * * Note: This function can be called from multiple contexts so appropriate
710  * locking concerns must be made.  Today we use the sock lock to protect
711  * this function.
712  */
713 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
714 {
715         struct sctp_packet *packet;
716         struct sctp_packet singleton;
717         struct sctp_association *asoc = q->asoc;
718         __u16 sport = asoc->base.bind_addr.port;
719         __u16 dport = asoc->peer.port;
720         __u32 vtag = asoc->peer.i.init_tag;
721         struct sctp_transport *transport = NULL;
722         struct sctp_transport *new_transport;
723         struct sctp_chunk *chunk, *tmp;
724         sctp_xmit_t status;
725         int error = 0;
726         int start_timer = 0;
727         int one_packet = 0;
728
729         /* These transports have chunks to send. */
730         struct list_head transport_list;
731         struct list_head *ltransport;
732
733         INIT_LIST_HEAD(&transport_list);
734         packet = NULL;
735
736         /*
737          * 6.10 Bundling
738          *   ...
739          *   When bundling control chunks with DATA chunks, an
740          *   endpoint MUST place control chunks first in the outbound
741          *   SCTP packet.  The transmitter MUST transmit DATA chunks
742          *   within a SCTP packet in increasing order of TSN.
743          *   ...
744          */
745
746         list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
747                 list_del_init(&chunk->list);
748
749                 /* Pick the right transport to use. */
750                 new_transport = chunk->transport;
751
752                 if (!new_transport) {
753                         /*
754                          * If we have a prior transport pointer, see if
755                          * the destination address of the chunk
756                          * matches the destination address of the
757                          * current transport.  If not a match, then
758                          * try to look up the transport with a given
759                          * destination address.  We do this because
760                          * after processing ASCONFs, we may have new
761                          * transports created.
762                          */
763                         if (transport &&
764                             sctp_cmp_addr_exact(&chunk->dest,
765                                                 &transport->ipaddr))
766                                         new_transport = transport;
767                         else
768                                 new_transport = sctp_assoc_lookup_paddr(asoc,
769                                                                 &chunk->dest);
770
771                         /* if we still don't have a new transport, then
772                          * use the current active path.
773                          */
774                         if (!new_transport)
775                                 new_transport = asoc->peer.active_path;
776                 } else if ((new_transport->state == SCTP_INACTIVE) ||
777                            (new_transport->state == SCTP_UNCONFIRMED)) {
778                         /* If the chunk is Heartbeat or Heartbeat Ack,
779                          * send it to chunk->transport, even if it's
780                          * inactive.
781                          *
782                          * 3.3.6 Heartbeat Acknowledgement:
783                          * ...
784                          * A HEARTBEAT ACK is always sent to the source IP
785                          * address of the IP datagram containing the
786                          * HEARTBEAT chunk to which this ack is responding.
787                          * ...
788                          *
789                          * ASCONF_ACKs also must be sent to the source.
790                          */
791                         if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
792                             chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK &&
793                             chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK)
794                                 new_transport = asoc->peer.active_path;
795                 }
796
797                 /* Are we switching transports?
798                  * Take care of transport locks.
799                  */
800                 if (new_transport != transport) {
801                         transport = new_transport;
802                         if (list_empty(&transport->send_ready)) {
803                                 list_add_tail(&transport->send_ready,
804                                               &transport_list);
805                         }
806                         packet = &transport->packet;
807                         sctp_packet_config(packet, vtag,
808                                            asoc->peer.ecn_capable);
809                 }
810
811                 switch (chunk->chunk_hdr->type) {
812                 /*
813                  * 6.10 Bundling
814                  *   ...
815                  *   An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
816                  *   COMPLETE with any other chunks.  [Send them immediately.]
817                  */
818                 case SCTP_CID_INIT:
819                 case SCTP_CID_INIT_ACK:
820                 case SCTP_CID_SHUTDOWN_COMPLETE:
821                         sctp_packet_init(&singleton, transport, sport, dport);
822                         sctp_packet_config(&singleton, vtag, 0);
823                         sctp_packet_append_chunk(&singleton, chunk);
824                         error = sctp_packet_transmit(&singleton);
825                         if (error < 0)
826                                 return error;
827                         break;
828
829                 case SCTP_CID_ABORT:
830                         if (sctp_test_T_bit(chunk)) {
831                                 packet->vtag = asoc->c.my_vtag;
832                         }
833                 /* The following chunks are "response" chunks, i.e.
834                  * they are generated in response to something we
835                  * received.  If we are sending these, then we can
836                  * send only 1 packet containing these chunks.
837                  */
838                 case SCTP_CID_HEARTBEAT_ACK:
839                 case SCTP_CID_SHUTDOWN_ACK:
840                 case SCTP_CID_COOKIE_ACK:
841                 case SCTP_CID_COOKIE_ECHO:
842                 case SCTP_CID_ERROR:
843                 case SCTP_CID_ECN_CWR:
844                 case SCTP_CID_ASCONF_ACK:
845                         one_packet = 1;
846                         /* Fall through */
847
848                 case SCTP_CID_SACK:
849                 case SCTP_CID_HEARTBEAT:
850                 case SCTP_CID_SHUTDOWN:
851                 case SCTP_CID_ECN_ECNE:
852                 case SCTP_CID_ASCONF:
853                 case SCTP_CID_FWD_TSN:
854                         status = sctp_packet_transmit_chunk(packet, chunk,
855                                                             one_packet);
856                         if (status  != SCTP_XMIT_OK) {
857                                 /* put the chunk back */
858                                 list_add(&chunk->list, &q->control_chunk_list);
859                         } else if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN) {
860                                 /* PR-SCTP C5) If a FORWARD TSN is sent, the
861                                  * sender MUST assure that at least one T3-rtx
862                                  * timer is running.
863                                  */
864                                 sctp_transport_reset_timers(transport);
865                         }
866                         break;
867
868                 default:
869                         /* We built a chunk with an illegal type! */
870                         BUG();
871                 }
872         }
873
874         /* Is it OK to send data chunks?  */
875         switch (asoc->state) {
876         case SCTP_STATE_COOKIE_ECHOED:
877                 /* Only allow bundling when this packet has a COOKIE-ECHO
878                  * chunk.
879                  */
880                 if (!packet || !packet->has_cookie_echo)
881                         break;
882
883                 /* fallthru */
884         case SCTP_STATE_ESTABLISHED:
885         case SCTP_STATE_SHUTDOWN_PENDING:
886         case SCTP_STATE_SHUTDOWN_RECEIVED:
887                 /*
888                  * RFC 2960 6.1  Transmission of DATA Chunks
889                  *
890                  * C) When the time comes for the sender to transmit,
891                  * before sending new DATA chunks, the sender MUST
892                  * first transmit any outstanding DATA chunks which
893                  * are marked for retransmission (limited by the
894                  * current cwnd).
895                  */
896                 if (!list_empty(&q->retransmit)) {
897                         if (transport == asoc->peer.retran_path)
898                                 goto retran;
899
900                         /* Switch transports & prepare the packet.  */
901
902                         transport = asoc->peer.retran_path;
903
904                         if (list_empty(&transport->send_ready)) {
905                                 list_add_tail(&transport->send_ready,
906                                               &transport_list);
907                         }
908
909                         packet = &transport->packet;
910                         sctp_packet_config(packet, vtag,
911                                            asoc->peer.ecn_capable);
912                 retran:
913                         error = sctp_outq_flush_rtx(q, packet,
914                                                     rtx_timeout, &start_timer);
915
916                         if (start_timer)
917                                 sctp_transport_reset_timers(transport);
918
919                         /* This can happen on COOKIE-ECHO resend.  Only
920                          * one chunk can get bundled with a COOKIE-ECHO.
921                          */
922                         if (packet->has_cookie_echo)
923                                 goto sctp_flush_out;
924
925                         /* Don't send new data if there is still data
926                          * waiting to retransmit.
927                          */
928                         if (!list_empty(&q->retransmit))
929                                 goto sctp_flush_out;
930                 }
931
932                 /* Apply Max.Burst limitation to the current transport in
933                  * case it will be used for new data.  We are going to
934                  * rest it before we return, but we want to apply the limit
935                  * to the currently queued data.
936                  */
937                 if (transport)
938                         sctp_transport_burst_limited(transport);
939
940                 /* Finally, transmit new packets.  */
941                 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
942                         /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
943                          * stream identifier.
944                          */
945                         if (chunk->sinfo.sinfo_stream >=
946                             asoc->c.sinit_num_ostreams) {
947
948                                 /* Mark as failed send. */
949                                 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
950                                 sctp_chunk_free(chunk);
951                                 continue;
952                         }
953
954                         /* Has this chunk expired? */
955                         if (sctp_chunk_abandoned(chunk)) {
956                                 sctp_chunk_fail(chunk, 0);
957                                 sctp_chunk_free(chunk);
958                                 continue;
959                         }
960
961                         /* If there is a specified transport, use it.
962                          * Otherwise, we want to use the active path.
963                          */
964                         new_transport = chunk->transport;
965                         if (!new_transport ||
966                             ((new_transport->state == SCTP_INACTIVE) ||
967                              (new_transport->state == SCTP_UNCONFIRMED)))
968                                 new_transport = asoc->peer.active_path;
969
970                         /* Change packets if necessary.  */
971                         if (new_transport != transport) {
972                                 transport = new_transport;
973
974                                 /* Schedule to have this transport's
975                                  * packet flushed.
976                                  */
977                                 if (list_empty(&transport->send_ready)) {
978                                         list_add_tail(&transport->send_ready,
979                                                       &transport_list);
980                                 }
981
982                                 packet = &transport->packet;
983                                 sctp_packet_config(packet, vtag,
984                                                    asoc->peer.ecn_capable);
985                                 /* We've switched transports, so apply the
986                                  * Burst limit to the new transport.
987                                  */
988                                 sctp_transport_burst_limited(transport);
989                         }
990
991                         SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
992                                           q, chunk,
993                                           chunk && chunk->chunk_hdr ?
994                                           sctp_cname(SCTP_ST_CHUNK(
995                                                   chunk->chunk_hdr->type))
996                                           : "Illegal Chunk");
997
998                         SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
999                                         "%p skb->users %d.\n",
1000                                         ntohl(chunk->subh.data_hdr->tsn),
1001                                         chunk->skb ?chunk->skb->head : NULL,
1002                                         chunk->skb ?
1003                                         atomic_read(&chunk->skb->users) : -1);
1004
1005                         /* Add the chunk to the packet.  */
1006                         status = sctp_packet_transmit_chunk(packet, chunk, 0);
1007
1008                         switch (status) {
1009                         case SCTP_XMIT_PMTU_FULL:
1010                         case SCTP_XMIT_RWND_FULL:
1011                         case SCTP_XMIT_NAGLE_DELAY:
1012                                 /* We could not append this chunk, so put
1013                                  * the chunk back on the output queue.
1014                                  */
1015                                 SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
1016                                         "not transmit TSN: 0x%x, status: %d\n",
1017                                         ntohl(chunk->subh.data_hdr->tsn),
1018                                         status);
1019                                 sctp_outq_head_data(q, chunk);
1020                                 goto sctp_flush_out;
1021                                 break;
1022
1023                         case SCTP_XMIT_OK:
1024                                 /* The sender is in the SHUTDOWN-PENDING state,
1025                                  * The sender MAY set the I-bit in the DATA
1026                                  * chunk header.
1027                                  */
1028                                 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING)
1029                                         chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM;
1030
1031                                 break;
1032
1033                         default:
1034                                 BUG();
1035                         }
1036
1037                         /* BUG: We assume that the sctp_packet_transmit()
1038                          * call below will succeed all the time and add the
1039                          * chunk to the transmitted list and restart the
1040                          * timers.
1041                          * It is possible that the call can fail under OOM
1042                          * conditions.
1043                          *
1044                          * Is this really a problem?  Won't this behave
1045                          * like a lost TSN?
1046                          */
1047                         list_add_tail(&chunk->transmitted_list,
1048                                       &transport->transmitted);
1049
1050                         sctp_transport_reset_timers(transport);
1051
1052                         q->empty = 0;
1053
1054                         /* Only let one DATA chunk get bundled with a
1055                          * COOKIE-ECHO chunk.
1056                          */
1057                         if (packet->has_cookie_echo)
1058                                 goto sctp_flush_out;
1059                 }
1060                 break;
1061
1062         default:
1063                 /* Do nothing.  */
1064                 break;
1065         }
1066
1067 sctp_flush_out:
1068
1069         /* Before returning, examine all the transports touched in
1070          * this call.  Right now, we bluntly force clear all the
1071          * transports.  Things might change after we implement Nagle.
1072          * But such an examination is still required.
1073          *
1074          * --xguo
1075          */
1076         while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
1077                 struct sctp_transport *t = list_entry(ltransport,
1078                                                       struct sctp_transport,
1079                                                       send_ready);
1080                 packet = &t->packet;
1081                 if (!sctp_packet_empty(packet))
1082                         error = sctp_packet_transmit(packet);
1083
1084                 /* Clear the burst limited state, if any */
1085                 sctp_transport_burst_reset(t);
1086         }
1087
1088         return error;
1089 }
1090
1091 /* Update unack_data based on the incoming SACK chunk */
1092 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1093                                         struct sctp_sackhdr *sack)
1094 {
1095         sctp_sack_variable_t *frags;
1096         __u16 unack_data;
1097         int i;
1098
1099         unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1100
1101         frags = sack->variable;
1102         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1103                 unack_data -= ((ntohs(frags[i].gab.end) -
1104                                 ntohs(frags[i].gab.start) + 1));
1105         }
1106
1107         assoc->unack_data = unack_data;
1108 }
1109
1110 /* This is where we REALLY process a SACK.
1111  *
1112  * Process the SACK against the outqueue.  Mostly, this just frees
1113  * things off the transmitted queue.
1114  */
1115 int sctp_outq_sack(struct sctp_outq *q, struct sctp_sackhdr *sack)
1116 {
1117         struct sctp_association *asoc = q->asoc;
1118         struct sctp_transport *transport;
1119         struct sctp_chunk *tchunk = NULL;
1120         struct list_head *lchunk, *transport_list, *temp;
1121         sctp_sack_variable_t *frags = sack->variable;
1122         __u32 sack_ctsn, ctsn, tsn;
1123         __u32 highest_tsn, highest_new_tsn;
1124         __u32 sack_a_rwnd;
1125         unsigned outstanding;
1126         struct sctp_transport *primary = asoc->peer.primary_path;
1127         int count_of_newacks = 0;
1128         int gap_ack_blocks;
1129         u8 accum_moved = 0;
1130
1131         /* Grab the association's destination address list. */
1132         transport_list = &asoc->peer.transport_addr_list;
1133
1134         sack_ctsn = ntohl(sack->cum_tsn_ack);
1135         gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1136         /*
1137          * SFR-CACC algorithm:
1138          * On receipt of a SACK the sender SHOULD execute the
1139          * following statements.
1140          *
1141          * 1) If the cumulative ack in the SACK passes next tsn_at_change
1142          * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1143          * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1144          * all destinations.
1145          * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1146          * is set the receiver of the SACK MUST take the following actions:
1147          *
1148          * A) Initialize the cacc_saw_newack to 0 for all destination
1149          * addresses.
1150          *
1151          * Only bother if changeover_active is set. Otherwise, this is
1152          * totally suboptimal to do on every SACK.
1153          */
1154         if (primary->cacc.changeover_active) {
1155                 u8 clear_cycling = 0;
1156
1157                 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1158                         primary->cacc.changeover_active = 0;
1159                         clear_cycling = 1;
1160                 }
1161
1162                 if (clear_cycling || gap_ack_blocks) {
1163                         list_for_each_entry(transport, transport_list,
1164                                         transports) {
1165                                 if (clear_cycling)
1166                                         transport->cacc.cycling_changeover = 0;
1167                                 if (gap_ack_blocks)
1168                                         transport->cacc.cacc_saw_newack = 0;
1169                         }
1170                 }
1171         }
1172
1173         /* Get the highest TSN in the sack. */
1174         highest_tsn = sack_ctsn;
1175         if (gap_ack_blocks)
1176                 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1177
1178         if (TSN_lt(asoc->highest_sacked, highest_tsn))
1179                 asoc->highest_sacked = highest_tsn;
1180
1181         highest_new_tsn = sack_ctsn;
1182
1183         /* Run through the retransmit queue.  Credit bytes received
1184          * and free those chunks that we can.
1185          */
1186         sctp_check_transmitted(q, &q->retransmit, NULL, sack, &highest_new_tsn);
1187
1188         /* Run through the transmitted queue.
1189          * Credit bytes received and free those chunks which we can.
1190          *
1191          * This is a MASSIVE candidate for optimization.
1192          */
1193         list_for_each_entry(transport, transport_list, transports) {
1194                 sctp_check_transmitted(q, &transport->transmitted,
1195                                        transport, sack, &highest_new_tsn);
1196                 /*
1197                  * SFR-CACC algorithm:
1198                  * C) Let count_of_newacks be the number of
1199                  * destinations for which cacc_saw_newack is set.
1200                  */
1201                 if (transport->cacc.cacc_saw_newack)
1202                         count_of_newacks ++;
1203         }
1204
1205         /* Move the Cumulative TSN Ack Point if appropriate.  */
1206         if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) {
1207                 asoc->ctsn_ack_point = sack_ctsn;
1208                 accum_moved = 1;
1209         }
1210
1211         if (gap_ack_blocks) {
1212
1213                 if (asoc->fast_recovery && accum_moved)
1214                         highest_new_tsn = highest_tsn;
1215
1216                 list_for_each_entry(transport, transport_list, transports)
1217                         sctp_mark_missing(q, &transport->transmitted, transport,
1218                                           highest_new_tsn, count_of_newacks);
1219         }
1220
1221         /* Update unack_data field in the assoc. */
1222         sctp_sack_update_unack_data(asoc, sack);
1223
1224         ctsn = asoc->ctsn_ack_point;
1225
1226         /* Throw away stuff rotting on the sack queue.  */
1227         list_for_each_safe(lchunk, temp, &q->sacked) {
1228                 tchunk = list_entry(lchunk, struct sctp_chunk,
1229                                     transmitted_list);
1230                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1231                 if (TSN_lte(tsn, ctsn)) {
1232                         list_del_init(&tchunk->transmitted_list);
1233                         sctp_chunk_free(tchunk);
1234                 }
1235         }
1236
1237         /* ii) Set rwnd equal to the newly received a_rwnd minus the
1238          *     number of bytes still outstanding after processing the
1239          *     Cumulative TSN Ack and the Gap Ack Blocks.
1240          */
1241
1242         sack_a_rwnd = ntohl(sack->a_rwnd);
1243         outstanding = q->outstanding_bytes;
1244
1245         if (outstanding < sack_a_rwnd)
1246                 sack_a_rwnd -= outstanding;
1247         else
1248                 sack_a_rwnd = 0;
1249
1250         asoc->peer.rwnd = sack_a_rwnd;
1251
1252         sctp_generate_fwdtsn(q, sack_ctsn);
1253
1254         SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n",
1255                           __func__, sack_ctsn);
1256         SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, "
1257                           "%p is 0x%x. Adv peer ack point: 0x%x\n",
1258                           __func__, asoc, ctsn, asoc->adv_peer_ack_point);
1259
1260         /* See if all chunks are acked.
1261          * Make sure the empty queue handler will get run later.
1262          */
1263         q->empty = (list_empty(&q->out_chunk_list) &&
1264                     list_empty(&q->retransmit));
1265         if (!q->empty)
1266                 goto finish;
1267
1268         list_for_each_entry(transport, transport_list, transports) {
1269                 q->empty = q->empty && list_empty(&transport->transmitted);
1270                 if (!q->empty)
1271                         goto finish;
1272         }
1273
1274         SCTP_DEBUG_PRINTK("sack queue is empty.\n");
1275 finish:
1276         return q->empty;
1277 }
1278
1279 /* Is the outqueue empty?  */
1280 int sctp_outq_is_empty(const struct sctp_outq *q)
1281 {
1282         return q->empty;
1283 }
1284
1285 /********************************************************************
1286  * 2nd Level Abstractions
1287  ********************************************************************/
1288
1289 /* Go through a transport's transmitted list or the association's retransmit
1290  * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1291  * The retransmit list will not have an associated transport.
1292  *
1293  * I added coherent debug information output.   --xguo
1294  *
1295  * Instead of printing 'sacked' or 'kept' for each TSN on the
1296  * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1297  * KEPT TSN6-TSN7, etc.
1298  */
1299 static void sctp_check_transmitted(struct sctp_outq *q,
1300                                    struct list_head *transmitted_queue,
1301                                    struct sctp_transport *transport,
1302                                    struct sctp_sackhdr *sack,
1303                                    __u32 *highest_new_tsn_in_sack)
1304 {
1305         struct list_head *lchunk;
1306         struct sctp_chunk *tchunk;
1307         struct list_head tlist;
1308         __u32 tsn;
1309         __u32 sack_ctsn;
1310         __u32 rtt;
1311         __u8 restart_timer = 0;
1312         int bytes_acked = 0;
1313         int migrate_bytes = 0;
1314
1315         /* These state variables are for coherent debug output. --xguo */
1316
1317 #if SCTP_DEBUG
1318         __u32 dbg_ack_tsn = 0;  /* An ACKed TSN range starts here... */
1319         __u32 dbg_last_ack_tsn = 0;  /* ...and finishes here.        */
1320         __u32 dbg_kept_tsn = 0; /* An un-ACKed range starts here...  */
1321         __u32 dbg_last_kept_tsn = 0; /* ...and finishes here.        */
1322
1323         /* 0 : The last TSN was ACKed.
1324          * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1325          * -1: We need to initialize.
1326          */
1327         int dbg_prt_state = -1;
1328 #endif /* SCTP_DEBUG */
1329
1330         sack_ctsn = ntohl(sack->cum_tsn_ack);
1331
1332         INIT_LIST_HEAD(&tlist);
1333
1334         /* The while loop will skip empty transmitted queues. */
1335         while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1336                 tchunk = list_entry(lchunk, struct sctp_chunk,
1337                                     transmitted_list);
1338
1339                 if (sctp_chunk_abandoned(tchunk)) {
1340                         /* Move the chunk to abandoned list. */
1341                         sctp_insert_list(&q->abandoned, lchunk);
1342
1343                         /* If this chunk has not been acked, stop
1344                          * considering it as 'outstanding'.
1345                          */
1346                         if (!tchunk->tsn_gap_acked) {
1347                                 if (tchunk->transport)
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->rtt_in_progress) {
1376                                         tchunk->rtt_in_progress = 0;
1377                                         rtt = jiffies - tchunk->sent_at;
1378                                         sctp_transport_update_rto(transport,
1379                                                                   rtt);
1380                                 }
1381                         }
1382
1383                         /* If the chunk hasn't been marked as ACKED,
1384                          * mark it and account bytes_acked if the
1385                          * chunk had a valid transport (it will not
1386                          * have a transport if ASCONF had deleted it
1387                          * while DATA was outstanding).
1388                          */
1389                         if (!tchunk->tsn_gap_acked) {
1390                                 tchunk->tsn_gap_acked = 1;
1391                                 *highest_new_tsn_in_sack = tsn;
1392                                 bytes_acked += sctp_data_size(tchunk);
1393                                 if (!tchunk->transport)
1394                                         migrate_bytes += sctp_data_size(tchunk);
1395                         }
1396
1397                         if (TSN_lte(tsn, sack_ctsn)) {
1398                                 /* RFC 2960  6.3.2 Retransmission Timer Rules
1399                                  *
1400                                  * R3) Whenever a SACK is received
1401                                  * that acknowledges the DATA chunk
1402                                  * with the earliest outstanding TSN
1403                                  * for that address, restart T3-rtx
1404                                  * timer for that address with its
1405                                  * current RTO.
1406                                  */
1407                                 restart_timer = 1;
1408
1409                                 if (!tchunk->tsn_gap_acked) {
1410                                         /*
1411                                          * SFR-CACC algorithm:
1412                                          * 2) If the SACK contains gap acks
1413                                          * and the flag CHANGEOVER_ACTIVE is
1414                                          * set the receiver of the SACK MUST
1415                                          * take the following action:
1416                                          *
1417                                          * B) For each TSN t being acked that
1418                                          * has not been acked in any SACK so
1419                                          * far, set cacc_saw_newack to 1 for
1420                                          * the destination that the TSN was
1421                                          * sent to.
1422                                          */
1423                                         if (transport &&
1424                                             sack->num_gap_ack_blocks &&
1425                                             q->asoc->peer.primary_path->cacc.
1426                                             changeover_active)
1427                                                 transport->cacc.cacc_saw_newack
1428                                                         = 1;
1429                                 }
1430
1431                                 list_add_tail(&tchunk->transmitted_list,
1432                                               &q->sacked);
1433                         } else {
1434                                 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1435                                  * M2) Each time a SACK arrives reporting
1436                                  * 'Stray DATA chunk(s)' record the highest TSN
1437                                  * reported as newly acknowledged, call this
1438                                  * value 'HighestTSNinSack'. A newly
1439                                  * acknowledged DATA chunk is one not
1440                                  * previously acknowledged in a SACK.
1441                                  *
1442                                  * When the SCTP sender of data receives a SACK
1443                                  * chunk that acknowledges, for the first time,
1444                                  * the receipt of a DATA chunk, all the still
1445                                  * unacknowledged DATA chunks whose TSN is
1446                                  * older than that newly acknowledged DATA
1447                                  * chunk, are qualified as 'Stray DATA chunks'.
1448                                  */
1449                                 list_add_tail(lchunk, &tlist);
1450                         }
1451
1452 #if SCTP_DEBUG
1453                         switch (dbg_prt_state) {
1454                         case 0: /* last TSN was ACKed */
1455                                 if (dbg_last_ack_tsn + 1 == tsn) {
1456                                         /* This TSN belongs to the
1457                                          * current ACK range.
1458                                          */
1459                                         break;
1460                                 }
1461
1462                                 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1463                                         /* Display the end of the
1464                                          * current range.
1465                                          */
1466                                         SCTP_DEBUG_PRINTK_CONT("-%08x",
1467                                                                dbg_last_ack_tsn);
1468                                 }
1469
1470                                 /* Start a new range.  */
1471                                 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn);
1472                                 dbg_ack_tsn = tsn;
1473                                 break;
1474
1475                         case 1: /* The last TSN was NOT ACKed. */
1476                                 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1477                                         /* Display the end of current range. */
1478                                         SCTP_DEBUG_PRINTK_CONT("-%08x",
1479                                                                dbg_last_kept_tsn);
1480                                 }
1481
1482                                 SCTP_DEBUG_PRINTK_CONT("\n");
1483
1484                                 /* FALL THROUGH... */
1485                         default:
1486                                 /* This is the first-ever TSN we examined.  */
1487                                 /* Start a new range of ACK-ed TSNs.  */
1488                                 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn);
1489                                 dbg_prt_state = 0;
1490                                 dbg_ack_tsn = tsn;
1491                         }
1492
1493                         dbg_last_ack_tsn = tsn;
1494 #endif /* SCTP_DEBUG */
1495
1496                 } else {
1497                         if (tchunk->tsn_gap_acked) {
1498                                 SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1499                                                   "data TSN: 0x%x\n",
1500                                                   __func__,
1501                                                   tsn);
1502                                 tchunk->tsn_gap_acked = 0;
1503
1504                                 if (tchunk->transport)
1505                                         bytes_acked -= sctp_data_size(tchunk);
1506
1507                                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1508                                  *
1509                                  * R4) Whenever a SACK is received missing a
1510                                  * TSN that was previously acknowledged via a
1511                                  * Gap Ack Block, start T3-rtx for the
1512                                  * destination address to which the DATA
1513                                  * chunk was originally
1514                                  * transmitted if it is not already running.
1515                                  */
1516                                 restart_timer = 1;
1517                         }
1518
1519                         list_add_tail(lchunk, &tlist);
1520
1521 #if SCTP_DEBUG
1522                         /* See the above comments on ACK-ed TSNs. */
1523                         switch (dbg_prt_state) {
1524                         case 1:
1525                                 if (dbg_last_kept_tsn + 1 == tsn)
1526                                         break;
1527
1528                                 if (dbg_last_kept_tsn != dbg_kept_tsn)
1529                                         SCTP_DEBUG_PRINTK_CONT("-%08x",
1530                                                                dbg_last_kept_tsn);
1531
1532                                 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn);
1533                                 dbg_kept_tsn = tsn;
1534                                 break;
1535
1536                         case 0:
1537                                 if (dbg_last_ack_tsn != dbg_ack_tsn)
1538                                         SCTP_DEBUG_PRINTK_CONT("-%08x",
1539                                                                dbg_last_ack_tsn);
1540                                 SCTP_DEBUG_PRINTK_CONT("\n");
1541
1542                                 /* FALL THROUGH... */
1543                         default:
1544                                 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn);
1545                                 dbg_prt_state = 1;
1546                                 dbg_kept_tsn = tsn;
1547                         }
1548
1549                         dbg_last_kept_tsn = tsn;
1550 #endif /* SCTP_DEBUG */
1551                 }
1552         }
1553
1554 #if SCTP_DEBUG
1555         /* Finish off the last range, displaying its ending TSN.  */
1556         switch (dbg_prt_state) {
1557         case 0:
1558                 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1559                         SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_ack_tsn);
1560                 } else {
1561                         SCTP_DEBUG_PRINTK_CONT("\n");
1562                 }
1563         break;
1564
1565         case 1:
1566                 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1567                         SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_kept_tsn);
1568                 } else {
1569                         SCTP_DEBUG_PRINTK_CONT("\n");
1570                 }
1571         }
1572 #endif /* SCTP_DEBUG */
1573         if (transport) {
1574                 if (bytes_acked) {
1575                         /* We may have counted DATA that was migrated
1576                          * to this transport due to DEL-IP operation.
1577                          * Subtract those bytes, since the were never
1578                          * send on this transport and shouldn't be
1579                          * credited to this transport.
1580                          */
1581                         bytes_acked -= migrate_bytes;
1582
1583                         /* 8.2. When an outstanding TSN is acknowledged,
1584                          * the endpoint shall clear the error counter of
1585                          * the destination transport address to which the
1586                          * DATA chunk was last sent.
1587                          * The association's overall error counter is
1588                          * also cleared.
1589                          */
1590                         transport->error_count = 0;
1591                         transport->asoc->overall_error_count = 0;
1592
1593                         /* Mark the destination transport address as
1594                          * active if it is not so marked.
1595                          */
1596                         if ((transport->state == SCTP_INACTIVE) ||
1597                             (transport->state == SCTP_UNCONFIRMED)) {
1598                                 sctp_assoc_control_transport(
1599                                         transport->asoc,
1600                                         transport,
1601                                         SCTP_TRANSPORT_UP,
1602                                         SCTP_RECEIVED_SACK);
1603                         }
1604
1605                         sctp_transport_raise_cwnd(transport, sack_ctsn,
1606                                                   bytes_acked);
1607
1608                         transport->flight_size -= bytes_acked;
1609                         if (transport->flight_size == 0)
1610                                 transport->partial_bytes_acked = 0;
1611                         q->outstanding_bytes -= bytes_acked + migrate_bytes;
1612                 } else {
1613                         /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1614                          * When a sender is doing zero window probing, it
1615                          * should not timeout the association if it continues
1616                          * to receive new packets from the receiver. The
1617                          * reason is that the receiver MAY keep its window
1618                          * closed for an indefinite time.
1619                          * A sender is doing zero window probing when the
1620                          * receiver's advertised window is zero, and there is
1621                          * only one data chunk in flight to the receiver.
1622                          */
1623                         if (!q->asoc->peer.rwnd &&
1624                             !list_empty(&tlist) &&
1625                             (sack_ctsn+2 == q->asoc->next_tsn)) {
1626                                 SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1627                                                   "window probe: %u\n",
1628                                                   __func__, sack_ctsn);
1629                                 q->asoc->overall_error_count = 0;
1630                                 transport->error_count = 0;
1631                         }
1632                 }
1633
1634                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1635                  *
1636                  * R2) Whenever all outstanding data sent to an address have
1637                  * been acknowledged, turn off the T3-rtx timer of that
1638                  * address.
1639                  */
1640                 if (!transport->flight_size) {
1641                         if (timer_pending(&transport->T3_rtx_timer) &&
1642                             del_timer(&transport->T3_rtx_timer)) {
1643                                 sctp_transport_put(transport);
1644                         }
1645                 } else if (restart_timer) {
1646                         if (!mod_timer(&transport->T3_rtx_timer,
1647                                        jiffies + transport->rto))
1648                                 sctp_transport_hold(transport);
1649                 }
1650         }
1651
1652         list_splice(&tlist, transmitted_queue);
1653 }
1654
1655 /* Mark chunks as missing and consequently may get retransmitted. */
1656 static void sctp_mark_missing(struct sctp_outq *q,
1657                               struct list_head *transmitted_queue,
1658                               struct sctp_transport *transport,
1659                               __u32 highest_new_tsn_in_sack,
1660                               int count_of_newacks)
1661 {
1662         struct sctp_chunk *chunk;
1663         __u32 tsn;
1664         char do_fast_retransmit = 0;
1665         struct sctp_association *asoc = q->asoc;
1666         struct sctp_transport *primary = asoc->peer.primary_path;
1667
1668         list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1669
1670                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1671
1672                 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1673                  * 'Unacknowledged TSN's', if the TSN number of an
1674                  * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1675                  * value, increment the 'TSN.Missing.Report' count on that
1676                  * chunk if it has NOT been fast retransmitted or marked for
1677                  * fast retransmit already.
1678                  */
1679                 if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
1680                     !chunk->tsn_gap_acked &&
1681                     TSN_lt(tsn, highest_new_tsn_in_sack)) {
1682
1683                         /* SFR-CACC may require us to skip marking
1684                          * this chunk as missing.
1685                          */
1686                         if (!transport || !sctp_cacc_skip(primary, transport,
1687                                             count_of_newacks, tsn)) {
1688                                 chunk->tsn_missing_report++;
1689
1690                                 SCTP_DEBUG_PRINTK(
1691                                         "%s: TSN 0x%x missing counter: %d\n",
1692                                         __func__, tsn,
1693                                         chunk->tsn_missing_report);
1694                         }
1695                 }
1696                 /*
1697                  * M4) If any DATA chunk is found to have a
1698                  * 'TSN.Missing.Report'
1699                  * value larger than or equal to 3, mark that chunk for
1700                  * retransmission and start the fast retransmit procedure.
1701                  */
1702
1703                 if (chunk->tsn_missing_report >= 3) {
1704                         chunk->fast_retransmit = SCTP_NEED_FRTX;
1705                         do_fast_retransmit = 1;
1706                 }
1707         }
1708
1709         if (transport) {
1710                 if (do_fast_retransmit)
1711                         sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1712
1713                 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1714                                   "ssthresh: %d, flight_size: %d, pba: %d\n",
1715                                   __func__, transport, transport->cwnd,
1716                                   transport->ssthresh, transport->flight_size,
1717                                   transport->partial_bytes_acked);
1718         }
1719 }
1720
1721 /* Is the given TSN acked by this packet?  */
1722 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1723 {
1724         int i;
1725         sctp_sack_variable_t *frags;
1726         __u16 gap;
1727         __u32 ctsn = ntohl(sack->cum_tsn_ack);
1728
1729         if (TSN_lte(tsn, ctsn))
1730                 goto pass;
1731
1732         /* 3.3.4 Selective Acknowledgement (SACK) (3):
1733          *
1734          * Gap Ack Blocks:
1735          *  These fields contain the Gap Ack Blocks. They are repeated
1736          *  for each Gap Ack Block up to the number of Gap Ack Blocks
1737          *  defined in the Number of Gap Ack Blocks field. All DATA
1738          *  chunks with TSNs greater than or equal to (Cumulative TSN
1739          *  Ack + Gap Ack Block Start) and less than or equal to
1740          *  (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1741          *  Block are assumed to have been received correctly.
1742          */
1743
1744         frags = sack->variable;
1745         gap = tsn - ctsn;
1746         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1747                 if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1748                     TSN_lte(gap, ntohs(frags[i].gab.end)))
1749                         goto pass;
1750         }
1751
1752         return 0;
1753 pass:
1754         return 1;
1755 }
1756
1757 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1758                                     int nskips, __be16 stream)
1759 {
1760         int i;
1761
1762         for (i = 0; i < nskips; i++) {
1763                 if (skiplist[i].stream == stream)
1764                         return i;
1765         }
1766         return i;
1767 }
1768
1769 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1770 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1771 {
1772         struct sctp_association *asoc = q->asoc;
1773         struct sctp_chunk *ftsn_chunk = NULL;
1774         struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1775         int nskips = 0;
1776         int skip_pos = 0;
1777         __u32 tsn;
1778         struct sctp_chunk *chunk;
1779         struct list_head *lchunk, *temp;
1780
1781         if (!asoc->peer.prsctp_capable)
1782                 return;
1783
1784         /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1785          * received SACK.
1786          *
1787          * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1788          * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1789          */
1790         if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1791                 asoc->adv_peer_ack_point = ctsn;
1792
1793         /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1794          * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1795          * the chunk next in the out-queue space is marked as "abandoned" as
1796          * shown in the following example:
1797          *
1798          * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1799          * and the Advanced.Peer.Ack.Point is updated to this value:
1800          *
1801          *   out-queue at the end of  ==>   out-queue after Adv.Ack.Point
1802          *   normal SACK processing           local advancement
1803          *                ...                           ...
1804          *   Adv.Ack.Pt-> 102 acked                     102 acked
1805          *                103 abandoned                 103 abandoned
1806          *                104 abandoned     Adv.Ack.P-> 104 abandoned
1807          *                105                           105
1808          *                106 acked                     106 acked
1809          *                ...                           ...
1810          *
1811          * In this example, the data sender successfully advanced the
1812          * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1813          */
1814         list_for_each_safe(lchunk, temp, &q->abandoned) {
1815                 chunk = list_entry(lchunk, struct sctp_chunk,
1816                                         transmitted_list);
1817                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1818
1819                 /* Remove any chunks in the abandoned queue that are acked by
1820                  * the ctsn.
1821                  */
1822                 if (TSN_lte(tsn, ctsn)) {
1823                         list_del_init(lchunk);
1824                         sctp_chunk_free(chunk);
1825                 } else {
1826                         if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1827                                 asoc->adv_peer_ack_point = tsn;
1828                                 if (chunk->chunk_hdr->flags &
1829                                          SCTP_DATA_UNORDERED)
1830                                         continue;
1831                                 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1832                                                 nskips,
1833                                                 chunk->subh.data_hdr->stream);
1834                                 ftsn_skip_arr[skip_pos].stream =
1835                                         chunk->subh.data_hdr->stream;
1836                                 ftsn_skip_arr[skip_pos].ssn =
1837                                          chunk->subh.data_hdr->ssn;
1838                                 if (skip_pos == nskips)
1839                                         nskips++;
1840                                 if (nskips == 10)
1841                                         break;
1842                         } else
1843                                 break;
1844                 }
1845         }
1846
1847         /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1848          * is greater than the Cumulative TSN ACK carried in the received
1849          * SACK, the data sender MUST send the data receiver a FORWARD TSN
1850          * chunk containing the latest value of the
1851          * "Advanced.Peer.Ack.Point".
1852          *
1853          * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1854          * list each stream and sequence number in the forwarded TSN. This
1855          * information will enable the receiver to easily find any
1856          * stranded TSN's waiting on stream reorder queues. Each stream
1857          * SHOULD only be reported once; this means that if multiple
1858          * abandoned messages occur in the same stream then only the
1859          * highest abandoned stream sequence number is reported. If the
1860          * total size of the FORWARD TSN does NOT fit in a single MTU then
1861          * the sender of the FORWARD TSN SHOULD lower the
1862          * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1863          * single MTU.
1864          */
1865         if (asoc->adv_peer_ack_point > ctsn)
1866                 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1867                                               nskips, &ftsn_skip_arr[0]);
1868
1869         if (ftsn_chunk) {
1870                 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1871                 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
1872         }
1873 }