nfsd4: don't try to map gid's in generic rpc code
[linux-2.6.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                                 if (chunk->transport)
410                                         chunk->transport->flight_size -=
411                                                         sctp_data_size(chunk);
412                                 q->outstanding_bytes -= sctp_data_size(chunk);
413                                 q->asoc->peer.rwnd += (sctp_data_size(chunk) +
414                                                         sizeof(struct sk_buff));
415                         }
416                         continue;
417                 }
418
419                 /* If we are doing  retransmission due to a timeout or pmtu
420                  * discovery, only the  chunks that are not yet acked should
421                  * be added to the retransmit queue.
422                  */
423                 if ((reason == SCTP_RTXR_FAST_RTX  &&
424                             (chunk->fast_retransmit == SCTP_NEED_FRTX)) ||
425                     (reason != SCTP_RTXR_FAST_RTX  && !chunk->tsn_gap_acked)) {
426                         /* If this chunk was sent less then 1 rto ago, do not
427                          * retransmit this chunk, but give the peer time
428                          * to acknowlege it.  Do this only when
429                          * retransmitting due to T3 timeout.
430                          */
431                         if (reason == SCTP_RTXR_T3_RTX &&
432                             time_before(jiffies, chunk->sent_at +
433                                                  transport->last_rto))
434                                 continue;
435
436                         /* RFC 2960 6.2.1 Processing a Received SACK
437                          *
438                          * C) Any time a DATA chunk is marked for
439                          * retransmission (via either T3-rtx timer expiration
440                          * (Section 6.3.3) or via fast retransmit
441                          * (Section 7.2.4)), add the data size of those
442                          * chunks to the rwnd.
443                          */
444                         q->asoc->peer.rwnd += (sctp_data_size(chunk) +
445                                                 sizeof(struct sk_buff));
446                         q->outstanding_bytes -= sctp_data_size(chunk);
447                         if (chunk->transport)
448                                 transport->flight_size -= sctp_data_size(chunk);
449
450                         /* sctpimpguide-05 Section 2.8.2
451                          * M5) If a T3-rtx timer expires, the
452                          * 'TSN.Missing.Report' of all affected TSNs is set
453                          * to 0.
454                          */
455                         chunk->tsn_missing_report = 0;
456
457                         /* If a chunk that is being used for RTT measurement
458                          * has to be retransmitted, we cannot use this chunk
459                          * anymore for RTT measurements. Reset rto_pending so
460                          * that a new RTT measurement is started when a new
461                          * data chunk is sent.
462                          */
463                         if (chunk->rtt_in_progress) {
464                                 chunk->rtt_in_progress = 0;
465                                 transport->rto_pending = 0;
466                         }
467
468                         /* Move the chunk to the retransmit queue. The chunks
469                          * on the retransmit queue are always kept in order.
470                          */
471                         list_del_init(lchunk);
472                         sctp_insert_list(&q->retransmit, lchunk);
473                 }
474         }
475
476         SCTP_DEBUG_PRINTK("%s: transport: %p, reason: %d, "
477                           "cwnd: %d, ssthresh: %d, flight_size: %d, "
478                           "pba: %d\n", __func__,
479                           transport, reason,
480                           transport->cwnd, transport->ssthresh,
481                           transport->flight_size,
482                           transport->partial_bytes_acked);
483
484 }
485
486 /* Mark all the eligible packets on a transport for retransmission and force
487  * one packet out.
488  */
489 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
490                      sctp_retransmit_reason_t reason)
491 {
492         int error = 0;
493
494         switch(reason) {
495         case SCTP_RTXR_T3_RTX:
496                 SCTP_INC_STATS(SCTP_MIB_T3_RETRANSMITS);
497                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
498                 /* Update the retran path if the T3-rtx timer has expired for
499                  * the current retran path.
500                  */
501                 if (transport == transport->asoc->peer.retran_path)
502                         sctp_assoc_update_retran_path(transport->asoc);
503                 transport->asoc->rtx_data_chunks +=
504                         transport->asoc->unack_data;
505                 break;
506         case SCTP_RTXR_FAST_RTX:
507                 SCTP_INC_STATS(SCTP_MIB_FAST_RETRANSMITS);
508                 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
509                 q->fast_rtx = 1;
510                 break;
511         case SCTP_RTXR_PMTUD:
512                 SCTP_INC_STATS(SCTP_MIB_PMTUD_RETRANSMITS);
513                 break;
514         case SCTP_RTXR_T1_RTX:
515                 SCTP_INC_STATS(SCTP_MIB_T1_RETRANSMITS);
516                 transport->asoc->init_retries++;
517                 break;
518         default:
519                 BUG();
520         }
521
522         sctp_retransmit_mark(q, transport, reason);
523
524         /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
525          * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
526          * following the procedures outlined in C1 - C5.
527          */
528         if (reason == SCTP_RTXR_T3_RTX)
529                 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
530
531         /* Flush the queues only on timeout, since fast_rtx is only
532          * triggered during sack processing and the queue
533          * will be flushed at the end.
534          */
535         if (reason != SCTP_RTXR_FAST_RTX)
536                 error = sctp_outq_flush(q, /* rtx_timeout */ 1);
537
538         if (error)
539                 q->asoc->base.sk->sk_err = -error;
540 }
541
542 /*
543  * Transmit DATA chunks on the retransmit queue.  Upon return from
544  * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
545  * need to be transmitted by the caller.
546  * We assume that pkt->transport has already been set.
547  *
548  * The return value is a normal kernel error return value.
549  */
550 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
551                                int rtx_timeout, int *start_timer)
552 {
553         struct list_head *lqueue;
554         struct sctp_transport *transport = pkt->transport;
555         sctp_xmit_t status;
556         struct sctp_chunk *chunk, *chunk1;
557         struct sctp_association *asoc;
558         int fast_rtx;
559         int error = 0;
560         int timer = 0;
561         int done = 0;
562
563         asoc = q->asoc;
564         lqueue = &q->retransmit;
565         fast_rtx = q->fast_rtx;
566
567         /* This loop handles time-out retransmissions, fast retransmissions,
568          * and retransmissions due to opening of whindow.
569          *
570          * RFC 2960 6.3.3 Handle T3-rtx Expiration
571          *
572          * E3) Determine how many of the earliest (i.e., lowest TSN)
573          * outstanding DATA chunks for the address for which the
574          * T3-rtx has expired will fit into a single packet, subject
575          * to the MTU constraint for the path corresponding to the
576          * destination transport address to which the retransmission
577          * is being sent (this may be different from the address for
578          * which the timer expires [see Section 6.4]). Call this value
579          * K. Bundle and retransmit those K DATA chunks in a single
580          * packet to the destination endpoint.
581          *
582          * [Just to be painfully clear, if we are retransmitting
583          * because a timeout just happened, we should send only ONE
584          * packet of retransmitted data.]
585          *
586          * For fast retransmissions we also send only ONE packet.  However,
587          * if we are just flushing the queue due to open window, we'll
588          * try to send as much as possible.
589          */
590         list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
591
592                 /* Make sure that Gap Acked TSNs are not retransmitted.  A
593                  * simple approach is just to move such TSNs out of the
594                  * way and into a 'transmitted' queue and skip to the
595                  * next chunk.
596                  */
597                 if (chunk->tsn_gap_acked) {
598                         list_del(&chunk->transmitted_list);
599                         list_add_tail(&chunk->transmitted_list,
600                                         &transport->transmitted);
601                         continue;
602                 }
603
604                 /* If we are doing fast retransmit, ignore non-fast_rtransmit
605                  * chunks
606                  */
607                 if (fast_rtx && !chunk->fast_retransmit)
608                         continue;
609
610                 /* Attempt to append this chunk to the packet. */
611                 status = sctp_packet_append_chunk(pkt, chunk);
612
613                 switch (status) {
614                 case SCTP_XMIT_PMTU_FULL:
615                         /* Send this packet.  */
616                         error = sctp_packet_transmit(pkt);
617
618                         /* If we are retransmitting, we should only
619                          * send a single packet.
620                          */
621                         if (rtx_timeout || fast_rtx)
622                                 done = 1;
623
624                         /* Bundle next chunk in the next round.  */
625                         break;
626
627                 case SCTP_XMIT_RWND_FULL:
628                         /* Send this packet. */
629                         error = sctp_packet_transmit(pkt);
630
631                         /* Stop sending DATA as there is no more room
632                          * at the receiver.
633                          */
634                         done = 1;
635                         break;
636
637                 case SCTP_XMIT_NAGLE_DELAY:
638                         /* Send this packet. */
639                         error = sctp_packet_transmit(pkt);
640
641                         /* Stop sending DATA because of nagle delay. */
642                         done = 1;
643                         break;
644
645                 default:
646                         /* The append was successful, so add this chunk to
647                          * the transmitted list.
648                          */
649                         list_del(&chunk->transmitted_list);
650                         list_add_tail(&chunk->transmitted_list,
651                                         &transport->transmitted);
652
653                         /* Mark the chunk as ineligible for fast retransmit
654                          * after it is retransmitted.
655                          */
656                         if (chunk->fast_retransmit == SCTP_NEED_FRTX)
657                                 chunk->fast_retransmit = SCTP_DONT_FRTX;
658
659                         /* Force start T3-rtx timer when fast retransmitting
660                          * the earliest outstanding TSN
661                          */
662                         if (!timer && fast_rtx &&
663                             ntohl(chunk->subh.data_hdr->tsn) ==
664                                              asoc->ctsn_ack_point + 1)
665                                 timer = 2;
666
667                         q->empty = 0;
668                         break;
669                 }
670
671                 /* Set the timer if there were no errors */
672                 if (!error && !timer)
673                         timer = 1;
674
675                 if (done)
676                         break;
677         }
678
679         /* If we are here due to a retransmit timeout or a fast
680          * retransmit and if there are any chunks left in the retransmit
681          * queue that could not fit in the PMTU sized packet, they need
682          * to be marked as ineligible for a subsequent fast retransmit.
683          */
684         if (rtx_timeout || fast_rtx) {
685                 list_for_each_entry(chunk1, lqueue, transmitted_list) {
686                         if (chunk1->fast_retransmit == SCTP_NEED_FRTX)
687                                 chunk1->fast_retransmit = SCTP_DONT_FRTX;
688                 }
689         }
690
691         *start_timer = timer;
692
693         /* Clear fast retransmit hint */
694         if (fast_rtx)
695                 q->fast_rtx = 0;
696
697         return error;
698 }
699
700 /* Cork the outqueue so queued chunks are really queued. */
701 int sctp_outq_uncork(struct sctp_outq *q)
702 {
703         int error = 0;
704         if (q->cork)
705                 q->cork = 0;
706         error = sctp_outq_flush(q, 0);
707         return error;
708 }
709
710
711 /*
712  * Try to flush an outqueue.
713  *
714  * Description: Send everything in q which we legally can, subject to
715  * congestion limitations.
716  * * Note: This function can be called from multiple contexts so appropriate
717  * locking concerns must be made.  Today we use the sock lock to protect
718  * this function.
719  */
720 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
721 {
722         struct sctp_packet *packet;
723         struct sctp_packet singleton;
724         struct sctp_association *asoc = q->asoc;
725         __u16 sport = asoc->base.bind_addr.port;
726         __u16 dport = asoc->peer.port;
727         __u32 vtag = asoc->peer.i.init_tag;
728         struct sctp_transport *transport = NULL;
729         struct sctp_transport *new_transport;
730         struct sctp_chunk *chunk, *tmp;
731         sctp_xmit_t status;
732         int error = 0;
733         int start_timer = 0;
734         int one_packet = 0;
735
736         /* These transports have chunks to send. */
737         struct list_head transport_list;
738         struct list_head *ltransport;
739
740         INIT_LIST_HEAD(&transport_list);
741         packet = NULL;
742
743         /*
744          * 6.10 Bundling
745          *   ...
746          *   When bundling control chunks with DATA chunks, an
747          *   endpoint MUST place control chunks first in the outbound
748          *   SCTP packet.  The transmitter MUST transmit DATA chunks
749          *   within a SCTP packet in increasing order of TSN.
750          *   ...
751          */
752
753         list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
754                 list_del_init(&chunk->list);
755
756                 /* Pick the right transport to use. */
757                 new_transport = chunk->transport;
758
759                 if (!new_transport) {
760                         /*
761                          * If we have a prior transport pointer, see if
762                          * the destination address of the chunk
763                          * matches the destination address of the
764                          * current transport.  If not a match, then
765                          * try to look up the transport with a given
766                          * destination address.  We do this because
767                          * after processing ASCONFs, we may have new
768                          * transports created.
769                          */
770                         if (transport &&
771                             sctp_cmp_addr_exact(&chunk->dest,
772                                                 &transport->ipaddr))
773                                         new_transport = transport;
774                         else
775                                 new_transport = sctp_assoc_lookup_paddr(asoc,
776                                                                 &chunk->dest);
777
778                         /* if we still don't have a new transport, then
779                          * use the current active path.
780                          */
781                         if (!new_transport)
782                                 new_transport = asoc->peer.active_path;
783                 } else if ((new_transport->state == SCTP_INACTIVE) ||
784                            (new_transport->state == SCTP_UNCONFIRMED)) {
785                         /* If the chunk is Heartbeat or Heartbeat Ack,
786                          * send it to chunk->transport, even if it's
787                          * inactive.
788                          *
789                          * 3.3.6 Heartbeat Acknowledgement:
790                          * ...
791                          * A HEARTBEAT ACK is always sent to the source IP
792                          * address of the IP datagram containing the
793                          * HEARTBEAT chunk to which this ack is responding.
794                          * ...
795                          *
796                          * ASCONF_ACKs also must be sent to the source.
797                          */
798                         if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
799                             chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK &&
800                             chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK)
801                                 new_transport = asoc->peer.active_path;
802                 }
803
804                 /* Are we switching transports?
805                  * Take care of transport locks.
806                  */
807                 if (new_transport != transport) {
808                         transport = new_transport;
809                         if (list_empty(&transport->send_ready)) {
810                                 list_add_tail(&transport->send_ready,
811                                               &transport_list);
812                         }
813                         packet = &transport->packet;
814                         sctp_packet_config(packet, vtag,
815                                            asoc->peer.ecn_capable);
816                 }
817
818                 switch (chunk->chunk_hdr->type) {
819                 /*
820                  * 6.10 Bundling
821                  *   ...
822                  *   An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
823                  *   COMPLETE with any other chunks.  [Send them immediately.]
824                  */
825                 case SCTP_CID_INIT:
826                 case SCTP_CID_INIT_ACK:
827                 case SCTP_CID_SHUTDOWN_COMPLETE:
828                         sctp_packet_init(&singleton, transport, sport, dport);
829                         sctp_packet_config(&singleton, vtag, 0);
830                         sctp_packet_append_chunk(&singleton, chunk);
831                         error = sctp_packet_transmit(&singleton);
832                         if (error < 0)
833                                 return error;
834                         break;
835
836                 case SCTP_CID_ABORT:
837                         if (sctp_test_T_bit(chunk)) {
838                                 packet->vtag = asoc->c.my_vtag;
839                         }
840                 /* The following chunks are "response" chunks, i.e.
841                  * they are generated in response to something we
842                  * received.  If we are sending these, then we can
843                  * send only 1 packet containing these chunks.
844                  */
845                 case SCTP_CID_HEARTBEAT_ACK:
846                 case SCTP_CID_SHUTDOWN_ACK:
847                 case SCTP_CID_COOKIE_ACK:
848                 case SCTP_CID_COOKIE_ECHO:
849                 case SCTP_CID_ERROR:
850                 case SCTP_CID_ECN_CWR:
851                 case SCTP_CID_ASCONF_ACK:
852                         one_packet = 1;
853                         /* Fall throught */
854
855                 case SCTP_CID_SACK:
856                 case SCTP_CID_HEARTBEAT:
857                 case SCTP_CID_SHUTDOWN:
858                 case SCTP_CID_ECN_ECNE:
859                 case SCTP_CID_ASCONF:
860                 case SCTP_CID_FWD_TSN:
861                         status = sctp_packet_transmit_chunk(packet, chunk,
862                                                             one_packet);
863                         if (status  != SCTP_XMIT_OK) {
864                                 /* put the chunk back */
865                                 list_add(&chunk->list, &q->control_chunk_list);
866                         }
867                         break;
868
869                 default:
870                         /* We built a chunk with an illegal type! */
871                         BUG();
872                 }
873         }
874
875         /* Is it OK to send data chunks?  */
876         switch (asoc->state) {
877         case SCTP_STATE_COOKIE_ECHOED:
878                 /* Only allow bundling when this packet has a COOKIE-ECHO
879                  * chunk.
880                  */
881                 if (!packet || !packet->has_cookie_echo)
882                         break;
883
884                 /* fallthru */
885         case SCTP_STATE_ESTABLISHED:
886         case SCTP_STATE_SHUTDOWN_PENDING:
887         case SCTP_STATE_SHUTDOWN_RECEIVED:
888                 /*
889                  * RFC 2960 6.1  Transmission of DATA Chunks
890                  *
891                  * C) When the time comes for the sender to transmit,
892                  * before sending new DATA chunks, the sender MUST
893                  * first transmit any outstanding DATA chunks which
894                  * are marked for retransmission (limited by the
895                  * current cwnd).
896                  */
897                 if (!list_empty(&q->retransmit)) {
898                         if (transport == asoc->peer.retran_path)
899                                 goto retran;
900
901                         /* Switch transports & prepare the packet.  */
902
903                         transport = asoc->peer.retran_path;
904
905                         if (list_empty(&transport->send_ready)) {
906                                 list_add_tail(&transport->send_ready,
907                                               &transport_list);
908                         }
909
910                         packet = &transport->packet;
911                         sctp_packet_config(packet, vtag,
912                                            asoc->peer.ecn_capable);
913                 retran:
914                         error = sctp_outq_flush_rtx(q, packet,
915                                                     rtx_timeout, &start_timer);
916
917                         if (start_timer)
918                                 sctp_transport_reset_timers(transport,
919                                                             start_timer-1);
920
921                         /* This can happen on COOKIE-ECHO resend.  Only
922                          * one chunk can get bundled with a COOKIE-ECHO.
923                          */
924                         if (packet->has_cookie_echo)
925                                 goto sctp_flush_out;
926
927                         /* Don't send new data if there is still data
928                          * waiting to retransmit.
929                          */
930                         if (!list_empty(&q->retransmit))
931                                 goto sctp_flush_out;
932                 }
933
934                 /* Finally, transmit new packets.  */
935                 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
936                         /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
937                          * stream identifier.
938                          */
939                         if (chunk->sinfo.sinfo_stream >=
940                             asoc->c.sinit_num_ostreams) {
941
942                                 /* Mark as failed send. */
943                                 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
944                                 sctp_chunk_free(chunk);
945                                 continue;
946                         }
947
948                         /* Has this chunk expired? */
949                         if (sctp_chunk_abandoned(chunk)) {
950                                 sctp_chunk_fail(chunk, 0);
951                                 sctp_chunk_free(chunk);
952                                 continue;
953                         }
954
955                         /* If there is a specified transport, use it.
956                          * Otherwise, we want to use the active path.
957                          */
958                         new_transport = chunk->transport;
959                         if (!new_transport ||
960                             ((new_transport->state == SCTP_INACTIVE) ||
961                              (new_transport->state == SCTP_UNCONFIRMED)))
962                                 new_transport = asoc->peer.active_path;
963
964                         /* Change packets if necessary.  */
965                         if (new_transport != transport) {
966                                 transport = new_transport;
967
968                                 /* Schedule to have this transport's
969                                  * packet flushed.
970                                  */
971                                 if (list_empty(&transport->send_ready)) {
972                                         list_add_tail(&transport->send_ready,
973                                                       &transport_list);
974                                 }
975
976                                 packet = &transport->packet;
977                                 sctp_packet_config(packet, vtag,
978                                                    asoc->peer.ecn_capable);
979                         }
980
981                         SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
982                                           q, chunk,
983                                           chunk && chunk->chunk_hdr ?
984                                           sctp_cname(SCTP_ST_CHUNK(
985                                                   chunk->chunk_hdr->type))
986                                           : "Illegal Chunk");
987
988                         SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
989                                         "%p skb->users %d.\n",
990                                         ntohl(chunk->subh.data_hdr->tsn),
991                                         chunk->skb ?chunk->skb->head : NULL,
992                                         chunk->skb ?
993                                         atomic_read(&chunk->skb->users) : -1);
994
995                         /* Add the chunk to the packet.  */
996                         status = sctp_packet_transmit_chunk(packet, chunk, 0);
997
998                         switch (status) {
999                         case SCTP_XMIT_PMTU_FULL:
1000                         case SCTP_XMIT_RWND_FULL:
1001                         case SCTP_XMIT_NAGLE_DELAY:
1002                                 /* We could not append this chunk, so put
1003                                  * the chunk back on the output queue.
1004                                  */
1005                                 SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
1006                                         "not transmit TSN: 0x%x, status: %d\n",
1007                                         ntohl(chunk->subh.data_hdr->tsn),
1008                                         status);
1009                                 sctp_outq_head_data(q, chunk);
1010                                 goto sctp_flush_out;
1011                                 break;
1012
1013                         case SCTP_XMIT_OK:
1014                                 break;
1015
1016                         default:
1017                                 BUG();
1018                         }
1019
1020                         /* BUG: We assume that the sctp_packet_transmit()
1021                          * call below will succeed all the time and add the
1022                          * chunk to the transmitted list and restart the
1023                          * timers.
1024                          * It is possible that the call can fail under OOM
1025                          * conditions.
1026                          *
1027                          * Is this really a problem?  Won't this behave
1028                          * like a lost TSN?
1029                          */
1030                         list_add_tail(&chunk->transmitted_list,
1031                                       &transport->transmitted);
1032
1033                         sctp_transport_reset_timers(transport, 0);
1034
1035                         q->empty = 0;
1036
1037                         /* Only let one DATA chunk get bundled with a
1038                          * COOKIE-ECHO chunk.
1039                          */
1040                         if (packet->has_cookie_echo)
1041                                 goto sctp_flush_out;
1042                 }
1043                 break;
1044
1045         default:
1046                 /* Do nothing.  */
1047                 break;
1048         }
1049
1050 sctp_flush_out:
1051
1052         /* Before returning, examine all the transports touched in
1053          * this call.  Right now, we bluntly force clear all the
1054          * transports.  Things might change after we implement Nagle.
1055          * But such an examination is still required.
1056          *
1057          * --xguo
1058          */
1059         while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
1060                 struct sctp_transport *t = list_entry(ltransport,
1061                                                       struct sctp_transport,
1062                                                       send_ready);
1063                 packet = &t->packet;
1064                 if (!sctp_packet_empty(packet))
1065                         error = sctp_packet_transmit(packet);
1066         }
1067
1068         return error;
1069 }
1070
1071 /* Update unack_data based on the incoming SACK chunk */
1072 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1073                                         struct sctp_sackhdr *sack)
1074 {
1075         sctp_sack_variable_t *frags;
1076         __u16 unack_data;
1077         int i;
1078
1079         unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1080
1081         frags = sack->variable;
1082         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1083                 unack_data -= ((ntohs(frags[i].gab.end) -
1084                                 ntohs(frags[i].gab.start) + 1));
1085         }
1086
1087         assoc->unack_data = unack_data;
1088 }
1089
1090 /* Return the highest new tsn that is acknowledged by the given SACK chunk. */
1091 static __u32 sctp_highest_new_tsn(struct sctp_sackhdr *sack,
1092                                   struct sctp_association *asoc)
1093 {
1094         struct sctp_transport *transport;
1095         struct sctp_chunk *chunk;
1096         __u32 highest_new_tsn, tsn;
1097         struct list_head *transport_list = &asoc->peer.transport_addr_list;
1098
1099         highest_new_tsn = ntohl(sack->cum_tsn_ack);
1100
1101         list_for_each_entry(transport, transport_list, transports) {
1102                 list_for_each_entry(chunk, &transport->transmitted,
1103                                 transmitted_list) {
1104                         tsn = ntohl(chunk->subh.data_hdr->tsn);
1105
1106                         if (!chunk->tsn_gap_acked &&
1107                             TSN_lt(highest_new_tsn, tsn) &&
1108                             sctp_acked(sack, tsn))
1109                                 highest_new_tsn = tsn;
1110                 }
1111         }
1112
1113         return highest_new_tsn;
1114 }
1115
1116 /* This is where we REALLY process a SACK.
1117  *
1118  * Process the SACK against the outqueue.  Mostly, this just frees
1119  * things off the transmitted queue.
1120  */
1121 int sctp_outq_sack(struct sctp_outq *q, struct sctp_sackhdr *sack)
1122 {
1123         struct sctp_association *asoc = q->asoc;
1124         struct sctp_transport *transport;
1125         struct sctp_chunk *tchunk = NULL;
1126         struct list_head *lchunk, *transport_list, *temp;
1127         sctp_sack_variable_t *frags = sack->variable;
1128         __u32 sack_ctsn, ctsn, tsn;
1129         __u32 highest_tsn, highest_new_tsn;
1130         __u32 sack_a_rwnd;
1131         unsigned outstanding;
1132         struct sctp_transport *primary = asoc->peer.primary_path;
1133         int count_of_newacks = 0;
1134         int gap_ack_blocks;
1135
1136         /* Grab the association's destination address list. */
1137         transport_list = &asoc->peer.transport_addr_list;
1138
1139         sack_ctsn = ntohl(sack->cum_tsn_ack);
1140         gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1141         /*
1142          * SFR-CACC algorithm:
1143          * On receipt of a SACK the sender SHOULD execute the
1144          * following statements.
1145          *
1146          * 1) If the cumulative ack in the SACK passes next tsn_at_change
1147          * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1148          * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1149          * all destinations.
1150          * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1151          * is set the receiver of the SACK MUST take the following actions:
1152          *
1153          * A) Initialize the cacc_saw_newack to 0 for all destination
1154          * addresses.
1155          *
1156          * Only bother if changeover_active is set. Otherwise, this is
1157          * totally suboptimal to do on every SACK.
1158          */
1159         if (primary->cacc.changeover_active) {
1160                 u8 clear_cycling = 0;
1161
1162                 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1163                         primary->cacc.changeover_active = 0;
1164                         clear_cycling = 1;
1165                 }
1166
1167                 if (clear_cycling || gap_ack_blocks) {
1168                         list_for_each_entry(transport, transport_list,
1169                                         transports) {
1170                                 if (clear_cycling)
1171                                         transport->cacc.cycling_changeover = 0;
1172                                 if (gap_ack_blocks)
1173                                         transport->cacc.cacc_saw_newack = 0;
1174                         }
1175                 }
1176         }
1177
1178         /* Get the highest TSN in the sack. */
1179         highest_tsn = sack_ctsn;
1180         if (gap_ack_blocks)
1181                 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1182
1183         if (TSN_lt(asoc->highest_sacked, highest_tsn)) {
1184                 highest_new_tsn = highest_tsn;
1185                 asoc->highest_sacked = highest_tsn;
1186         } else {
1187                 highest_new_tsn = sctp_highest_new_tsn(sack, asoc);
1188         }
1189
1190
1191         /* Run through the retransmit queue.  Credit bytes received
1192          * and free those chunks that we can.
1193          */
1194         sctp_check_transmitted(q, &q->retransmit, NULL, sack, highest_new_tsn);
1195
1196         /* Run through the transmitted queue.
1197          * Credit bytes received and free those chunks which we can.
1198          *
1199          * This is a MASSIVE candidate for optimization.
1200          */
1201         list_for_each_entry(transport, transport_list, transports) {
1202                 sctp_check_transmitted(q, &transport->transmitted,
1203                                        transport, sack, highest_new_tsn);
1204                 /*
1205                  * SFR-CACC algorithm:
1206                  * C) Let count_of_newacks be the number of
1207                  * destinations for which cacc_saw_newack is set.
1208                  */
1209                 if (transport->cacc.cacc_saw_newack)
1210                         count_of_newacks ++;
1211         }
1212
1213         if (gap_ack_blocks) {
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         int migrate_bytes = 0;
1316
1317         /* These state variables are for coherent debug output. --xguo */
1318
1319 #if SCTP_DEBUG
1320         __u32 dbg_ack_tsn = 0;  /* An ACKed TSN range starts here... */
1321         __u32 dbg_last_ack_tsn = 0;  /* ...and finishes here.        */
1322         __u32 dbg_kept_tsn = 0; /* An un-ACKed range starts here...  */
1323         __u32 dbg_last_kept_tsn = 0; /* ...and finishes here.        */
1324
1325         /* 0 : The last TSN was ACKed.
1326          * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1327          * -1: We need to initialize.
1328          */
1329         int dbg_prt_state = -1;
1330 #endif /* SCTP_DEBUG */
1331
1332         sack_ctsn = ntohl(sack->cum_tsn_ack);
1333
1334         INIT_LIST_HEAD(&tlist);
1335
1336         /* The while loop will skip empty transmitted queues. */
1337         while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1338                 tchunk = list_entry(lchunk, struct sctp_chunk,
1339                                     transmitted_list);
1340
1341                 if (sctp_chunk_abandoned(tchunk)) {
1342                         /* Move the chunk to abandoned list. */
1343                         sctp_insert_list(&q->abandoned, lchunk);
1344
1345                         /* If this chunk has not been acked, stop
1346                          * considering it as 'outstanding'.
1347                          */
1348                         if (!tchunk->tsn_gap_acked) {
1349                                 if (tchunk->transport)
1350                                         tchunk->transport->flight_size -=
1351                                                         sctp_data_size(tchunk);
1352                                 q->outstanding_bytes -= sctp_data_size(tchunk);
1353                         }
1354                         continue;
1355                 }
1356
1357                 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1358                 if (sctp_acked(sack, tsn)) {
1359                         /* If this queue is the retransmit queue, the
1360                          * retransmit timer has already reclaimed
1361                          * the outstanding bytes for this chunk, so only
1362                          * count bytes associated with a transport.
1363                          */
1364                         if (transport) {
1365                                 /* If this chunk is being used for RTT
1366                                  * measurement, calculate the RTT and update
1367                                  * the RTO using this value.
1368                                  *
1369                                  * 6.3.1 C5) Karn's algorithm: RTT measurements
1370                                  * MUST NOT be made using packets that were
1371                                  * retransmitted (and thus for which it is
1372                                  * ambiguous whether the reply was for the
1373                                  * first instance of the packet or a later
1374                                  * instance).
1375                                  */
1376                                 if (!tchunk->tsn_gap_acked &&
1377                                     !tchunk->resent &&
1378                                     tchunk->rtt_in_progress) {
1379                                         tchunk->rtt_in_progress = 0;
1380                                         rtt = jiffies - tchunk->sent_at;
1381                                         sctp_transport_update_rto(transport,
1382                                                                   rtt);
1383                                 }
1384                         }
1385
1386                         /* If the chunk hasn't been marked as ACKED,
1387                          * mark it and account bytes_acked if the
1388                          * chunk had a valid transport (it will not
1389                          * have a transport if ASCONF had deleted it
1390                          * while DATA was outstanding).
1391                          */
1392                         if (!tchunk->tsn_gap_acked) {
1393                                 tchunk->tsn_gap_acked = 1;
1394                                 bytes_acked += sctp_data_size(tchunk);
1395                                 if (!tchunk->transport)
1396                                         migrate_bytes += sctp_data_size(tchunk);
1397                         }
1398
1399                         if (TSN_lte(tsn, sack_ctsn)) {
1400                                 /* RFC 2960  6.3.2 Retransmission Timer Rules
1401                                  *
1402                                  * R3) Whenever a SACK is received
1403                                  * that acknowledges the DATA chunk
1404                                  * with the earliest outstanding TSN
1405                                  * for that address, restart T3-rtx
1406                                  * timer for that address with its
1407                                  * current RTO.
1408                                  */
1409                                 restart_timer = 1;
1410
1411                                 if (!tchunk->tsn_gap_acked) {
1412                                         /*
1413                                          * SFR-CACC algorithm:
1414                                          * 2) If the SACK contains gap acks
1415                                          * and the flag CHANGEOVER_ACTIVE is
1416                                          * set the receiver of the SACK MUST
1417                                          * take the following action:
1418                                          *
1419                                          * B) For each TSN t being acked that
1420                                          * has not been acked in any SACK so
1421                                          * far, set cacc_saw_newack to 1 for
1422                                          * the destination that the TSN was
1423                                          * sent to.
1424                                          */
1425                                         if (transport &&
1426                                             sack->num_gap_ack_blocks &&
1427                                             q->asoc->peer.primary_path->cacc.
1428                                             changeover_active)
1429                                                 transport->cacc.cacc_saw_newack
1430                                                         = 1;
1431                                 }
1432
1433                                 list_add_tail(&tchunk->transmitted_list,
1434                                               &q->sacked);
1435                         } else {
1436                                 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1437                                  * M2) Each time a SACK arrives reporting
1438                                  * 'Stray DATA chunk(s)' record the highest TSN
1439                                  * reported as newly acknowledged, call this
1440                                  * value 'HighestTSNinSack'. A newly
1441                                  * acknowledged DATA chunk is one not
1442                                  * previously acknowledged in a SACK.
1443                                  *
1444                                  * When the SCTP sender of data receives a SACK
1445                                  * chunk that acknowledges, for the first time,
1446                                  * the receipt of a DATA chunk, all the still
1447                                  * unacknowledged DATA chunks whose TSN is
1448                                  * older than that newly acknowledged DATA
1449                                  * chunk, are qualified as 'Stray DATA chunks'.
1450                                  */
1451                                 list_add_tail(lchunk, &tlist);
1452                         }
1453
1454 #if SCTP_DEBUG
1455                         switch (dbg_prt_state) {
1456                         case 0: /* last TSN was ACKed */
1457                                 if (dbg_last_ack_tsn + 1 == tsn) {
1458                                         /* This TSN belongs to the
1459                                          * current ACK range.
1460                                          */
1461                                         break;
1462                                 }
1463
1464                                 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1465                                         /* Display the end of the
1466                                          * current range.
1467                                          */
1468                                         SCTP_DEBUG_PRINTK("-%08x",
1469                                                           dbg_last_ack_tsn);
1470                                 }
1471
1472                                 /* Start a new range.  */
1473                                 SCTP_DEBUG_PRINTK(",%08x", tsn);
1474                                 dbg_ack_tsn = tsn;
1475                                 break;
1476
1477                         case 1: /* The last TSN was NOT ACKed. */
1478                                 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1479                                         /* Display the end of current range. */
1480                                         SCTP_DEBUG_PRINTK("-%08x",
1481                                                           dbg_last_kept_tsn);
1482                                 }
1483
1484                                 SCTP_DEBUG_PRINTK("\n");
1485
1486                                 /* FALL THROUGH... */
1487                         default:
1488                                 /* This is the first-ever TSN we examined.  */
1489                                 /* Start a new range of ACK-ed TSNs.  */
1490                                 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn);
1491                                 dbg_prt_state = 0;
1492                                 dbg_ack_tsn = tsn;
1493                         }
1494
1495                         dbg_last_ack_tsn = tsn;
1496 #endif /* SCTP_DEBUG */
1497
1498                 } else {
1499                         if (tchunk->tsn_gap_acked) {
1500                                 SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1501                                                   "data TSN: 0x%x\n",
1502                                                   __func__,
1503                                                   tsn);
1504                                 tchunk->tsn_gap_acked = 0;
1505
1506                                 if (tchunk->transport)
1507                                         bytes_acked -= sctp_data_size(tchunk);
1508
1509                                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1510                                  *
1511                                  * R4) Whenever a SACK is received missing a
1512                                  * TSN that was previously acknowledged via a
1513                                  * Gap Ack Block, start T3-rtx for the
1514                                  * destination address to which the DATA
1515                                  * chunk was originally
1516                                  * transmitted if it is not already running.
1517                                  */
1518                                 restart_timer = 1;
1519                         }
1520
1521                         list_add_tail(lchunk, &tlist);
1522
1523 #if SCTP_DEBUG
1524                         /* See the above comments on ACK-ed TSNs. */
1525                         switch (dbg_prt_state) {
1526                         case 1:
1527                                 if (dbg_last_kept_tsn + 1 == tsn)
1528                                         break;
1529
1530                                 if (dbg_last_kept_tsn != dbg_kept_tsn)
1531                                         SCTP_DEBUG_PRINTK("-%08x",
1532                                                           dbg_last_kept_tsn);
1533
1534                                 SCTP_DEBUG_PRINTK(",%08x", tsn);
1535                                 dbg_kept_tsn = tsn;
1536                                 break;
1537
1538                         case 0:
1539                                 if (dbg_last_ack_tsn != dbg_ack_tsn)
1540                                         SCTP_DEBUG_PRINTK("-%08x",
1541                                                           dbg_last_ack_tsn);
1542                                 SCTP_DEBUG_PRINTK("\n");
1543
1544                                 /* FALL THROUGH... */
1545                         default:
1546                                 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn);
1547                                 dbg_prt_state = 1;
1548                                 dbg_kept_tsn = tsn;
1549                         }
1550
1551                         dbg_last_kept_tsn = tsn;
1552 #endif /* SCTP_DEBUG */
1553                 }
1554         }
1555
1556 #if SCTP_DEBUG
1557         /* Finish off the last range, displaying its ending TSN.  */
1558         switch (dbg_prt_state) {
1559         case 0:
1560                 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1561                         SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_ack_tsn);
1562                 } else {
1563                         SCTP_DEBUG_PRINTK("\n");
1564                 }
1565         break;
1566
1567         case 1:
1568                 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1569                         SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_kept_tsn);
1570                 } else {
1571                         SCTP_DEBUG_PRINTK("\n");
1572                 }
1573         }
1574 #endif /* SCTP_DEBUG */
1575         if (transport) {
1576                 if (bytes_acked) {
1577                         /* We may have counted DATA that was migrated
1578                          * to this transport due to DEL-IP operation.
1579                          * Subtract those bytes, since the were never
1580                          * send on this transport and shouldn't be
1581                          * credited to this transport.
1582                          */
1583                         bytes_acked -= migrate_bytes;
1584
1585                         /* 8.2. When an outstanding TSN is acknowledged,
1586                          * the endpoint shall clear the error counter of
1587                          * the destination transport address to which the
1588                          * DATA chunk was last sent.
1589                          * The association's overall error counter is
1590                          * also cleared.
1591                          */
1592                         transport->error_count = 0;
1593                         transport->asoc->overall_error_count = 0;
1594
1595                         /* Mark the destination transport address as
1596                          * active if it is not so marked.
1597                          */
1598                         if ((transport->state == SCTP_INACTIVE) ||
1599                             (transport->state == SCTP_UNCONFIRMED)) {
1600                                 sctp_assoc_control_transport(
1601                                         transport->asoc,
1602                                         transport,
1603                                         SCTP_TRANSPORT_UP,
1604                                         SCTP_RECEIVED_SACK);
1605                         }
1606
1607                         sctp_transport_raise_cwnd(transport, sack_ctsn,
1608                                                   bytes_acked);
1609
1610                         transport->flight_size -= bytes_acked;
1611                         if (transport->flight_size == 0)
1612                                 transport->partial_bytes_acked = 0;
1613                         q->outstanding_bytes -= bytes_acked + migrate_bytes;
1614                 } else {
1615                         /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1616                          * When a sender is doing zero window probing, it
1617                          * should not timeout the association if it continues
1618                          * to receive new packets from the receiver. The
1619                          * reason is that the receiver MAY keep its window
1620                          * closed for an indefinite time.
1621                          * A sender is doing zero window probing when the
1622                          * receiver's advertised window is zero, and there is
1623                          * only one data chunk in flight to the receiver.
1624                          */
1625                         if (!q->asoc->peer.rwnd &&
1626                             !list_empty(&tlist) &&
1627                             (sack_ctsn+2 == q->asoc->next_tsn)) {
1628                                 SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1629                                                   "window probe: %u\n",
1630                                                   __func__, sack_ctsn);
1631                                 q->asoc->overall_error_count = 0;
1632                                 transport->error_count = 0;
1633                         }
1634                 }
1635
1636                 /* RFC 2960 6.3.2 Retransmission Timer Rules
1637                  *
1638                  * R2) Whenever all outstanding data sent to an address have
1639                  * been acknowledged, turn off the T3-rtx timer of that
1640                  * address.
1641                  */
1642                 if (!transport->flight_size) {
1643                         if (timer_pending(&transport->T3_rtx_timer) &&
1644                             del_timer(&transport->T3_rtx_timer)) {
1645                                 sctp_transport_put(transport);
1646                         }
1647                 } else if (restart_timer) {
1648                         if (!mod_timer(&transport->T3_rtx_timer,
1649                                        jiffies + transport->rto))
1650                                 sctp_transport_hold(transport);
1651                 }
1652         }
1653
1654         list_splice(&tlist, transmitted_queue);
1655 }
1656
1657 /* Mark chunks as missing and consequently may get retransmitted. */
1658 static void sctp_mark_missing(struct sctp_outq *q,
1659                               struct list_head *transmitted_queue,
1660                               struct sctp_transport *transport,
1661                               __u32 highest_new_tsn_in_sack,
1662                               int count_of_newacks)
1663 {
1664         struct sctp_chunk *chunk;
1665         __u32 tsn;
1666         char do_fast_retransmit = 0;
1667         struct sctp_transport *primary = q->asoc->peer.primary_path;
1668
1669         list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1670
1671                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1672
1673                 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1674                  * 'Unacknowledged TSN's', if the TSN number of an
1675                  * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1676                  * value, increment the 'TSN.Missing.Report' count on that
1677                  * chunk if it has NOT been fast retransmitted or marked for
1678                  * fast retransmit already.
1679                  */
1680                 if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
1681                     !chunk->tsn_gap_acked &&
1682                     TSN_lt(tsn, highest_new_tsn_in_sack)) {
1683
1684                         /* SFR-CACC may require us to skip marking
1685                          * this chunk as missing.
1686                          */
1687                         if (!transport || !sctp_cacc_skip(primary, transport,
1688                                             count_of_newacks, tsn)) {
1689                                 chunk->tsn_missing_report++;
1690
1691                                 SCTP_DEBUG_PRINTK(
1692                                         "%s: TSN 0x%x missing counter: %d\n",
1693                                         __func__, tsn,
1694                                         chunk->tsn_missing_report);
1695                         }
1696                 }
1697                 /*
1698                  * M4) If any DATA chunk is found to have a
1699                  * 'TSN.Missing.Report'
1700                  * value larger than or equal to 3, mark that chunk for
1701                  * retransmission and start the fast retransmit procedure.
1702                  */
1703
1704                 if (chunk->tsn_missing_report >= 3) {
1705                         chunk->fast_retransmit = SCTP_NEED_FRTX;
1706                         do_fast_retransmit = 1;
1707                 }
1708         }
1709
1710         if (transport) {
1711                 if (do_fast_retransmit)
1712                         sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1713
1714                 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1715                                   "ssthresh: %d, flight_size: %d, pba: %d\n",
1716                                   __func__, transport, transport->cwnd,
1717                                   transport->ssthresh, transport->flight_size,
1718                                   transport->partial_bytes_acked);
1719         }
1720 }
1721
1722 /* Is the given TSN acked by this packet?  */
1723 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1724 {
1725         int i;
1726         sctp_sack_variable_t *frags;
1727         __u16 gap;
1728         __u32 ctsn = ntohl(sack->cum_tsn_ack);
1729
1730         if (TSN_lte(tsn, ctsn))
1731                 goto pass;
1732
1733         /* 3.3.4 Selective Acknowledgement (SACK) (3):
1734          *
1735          * Gap Ack Blocks:
1736          *  These fields contain the Gap Ack Blocks. They are repeated
1737          *  for each Gap Ack Block up to the number of Gap Ack Blocks
1738          *  defined in the Number of Gap Ack Blocks field. All DATA
1739          *  chunks with TSNs greater than or equal to (Cumulative TSN
1740          *  Ack + Gap Ack Block Start) and less than or equal to
1741          *  (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1742          *  Block are assumed to have been received correctly.
1743          */
1744
1745         frags = sack->variable;
1746         gap = tsn - ctsn;
1747         for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1748                 if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1749                     TSN_lte(gap, ntohs(frags[i].gab.end)))
1750                         goto pass;
1751         }
1752
1753         return 0;
1754 pass:
1755         return 1;
1756 }
1757
1758 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1759                                     int nskips, __be16 stream)
1760 {
1761         int i;
1762
1763         for (i = 0; i < nskips; i++) {
1764                 if (skiplist[i].stream == stream)
1765                         return i;
1766         }
1767         return i;
1768 }
1769
1770 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1771 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1772 {
1773         struct sctp_association *asoc = q->asoc;
1774         struct sctp_chunk *ftsn_chunk = NULL;
1775         struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1776         int nskips = 0;
1777         int skip_pos = 0;
1778         __u32 tsn;
1779         struct sctp_chunk *chunk;
1780         struct list_head *lchunk, *temp;
1781
1782         if (!asoc->peer.prsctp_capable)
1783                 return;
1784
1785         /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1786          * received SACK.
1787          *
1788          * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1789          * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1790          */
1791         if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1792                 asoc->adv_peer_ack_point = ctsn;
1793
1794         /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1795          * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1796          * the chunk next in the out-queue space is marked as "abandoned" as
1797          * shown in the following example:
1798          *
1799          * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1800          * and the Advanced.Peer.Ack.Point is updated to this value:
1801          *
1802          *   out-queue at the end of  ==>   out-queue after Adv.Ack.Point
1803          *   normal SACK processing           local advancement
1804          *                ...                           ...
1805          *   Adv.Ack.Pt-> 102 acked                     102 acked
1806          *                103 abandoned                 103 abandoned
1807          *                104 abandoned     Adv.Ack.P-> 104 abandoned
1808          *                105                           105
1809          *                106 acked                     106 acked
1810          *                ...                           ...
1811          *
1812          * In this example, the data sender successfully advanced the
1813          * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1814          */
1815         list_for_each_safe(lchunk, temp, &q->abandoned) {
1816                 chunk = list_entry(lchunk, struct sctp_chunk,
1817                                         transmitted_list);
1818                 tsn = ntohl(chunk->subh.data_hdr->tsn);
1819
1820                 /* Remove any chunks in the abandoned queue that are acked by
1821                  * the ctsn.
1822                  */
1823                 if (TSN_lte(tsn, ctsn)) {
1824                         list_del_init(lchunk);
1825                         sctp_chunk_free(chunk);
1826                 } else {
1827                         if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1828                                 asoc->adv_peer_ack_point = tsn;
1829                                 if (chunk->chunk_hdr->flags &
1830                                          SCTP_DATA_UNORDERED)
1831                                         continue;
1832                                 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1833                                                 nskips,
1834                                                 chunk->subh.data_hdr->stream);
1835                                 ftsn_skip_arr[skip_pos].stream =
1836                                         chunk->subh.data_hdr->stream;
1837                                 ftsn_skip_arr[skip_pos].ssn =
1838                                          chunk->subh.data_hdr->ssn;
1839                                 if (skip_pos == nskips)
1840                                         nskips++;
1841                                 if (nskips == 10)
1842                                         break;
1843                         } else
1844                                 break;
1845                 }
1846         }
1847
1848         /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1849          * is greater than the Cumulative TSN ACK carried in the received
1850          * SACK, the data sender MUST send the data receiver a FORWARD TSN
1851          * chunk containing the latest value of the
1852          * "Advanced.Peer.Ack.Point".
1853          *
1854          * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1855          * list each stream and sequence number in the forwarded TSN. This
1856          * information will enable the receiver to easily find any
1857          * stranded TSN's waiting on stream reorder queues. Each stream
1858          * SHOULD only be reported once; this means that if multiple
1859          * abandoned messages occur in the same stream then only the
1860          * highest abandoned stream sequence number is reported. If the
1861          * total size of the FORWARD TSN does NOT fit in a single MTU then
1862          * the sender of the FORWARD TSN SHOULD lower the
1863          * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1864          * single MTU.
1865          */
1866         if (asoc->adv_peer_ack_point > ctsn)
1867                 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1868                                               nskips, &ftsn_skip_arr[0]);
1869
1870         if (ftsn_chunk) {
1871                 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1872                 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
1873         }
1874 }