blob: d765fc53e74dd63447d5d47a96664ecedbde45a8 [file] [log] [blame]
Vlad Yasevich60c778b2008-01-11 09:57:09 -05001/* SCTP kernel implementation
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * (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 *
Vlad Yasevich60c778b2008-01-11 09:57:09 -05007 * This file is part of the SCTP kernel implementation
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 *
9 * These functions implement the sctp_outq class. The outqueue handles
10 * bundling and queueing of outgoing SCTP chunks.
11 *
Vlad Yasevich60c778b2008-01-11 09:57:09 -050012 * This SCTP implementation is free software;
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 * 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 *
Vlad Yasevich60c778b2008-01-11 09:57:09 -050018 * This SCTP implementation is distributed in the hope that it
Linus Torvalds1da177e2005-04-16 15:20:36 -070019 * 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. */
59static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn);
60static 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
66static 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
72static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn);
73
Adrian Bunkabd0b1982008-07-22 14:20:45 -070074static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout);
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076/* Add data to the front of the queue. */
77static inline void sctp_outq_head_data(struct sctp_outq *q,
78 struct sctp_chunk *ch)
79{
David S. Miller79af02c2005-07-08 21:47:49 -070080 list_add(&ch->list, &q->out_chunk_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 q->out_qlen += ch->skb->len;
82 return;
83}
84
85/* Take data from the front of the queue. */
86static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
87{
David S. Miller79af02c2005-07-08 21:47:49 -070088 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 q->out_qlen -= ch->skb->len;
David S. Miller79af02c2005-07-08 21:47:49 -070096 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 return ch;
98}
99/* Add data chunk to the end of the queue. */
100static inline void sctp_outq_tail_data(struct sctp_outq *q,
101 struct sctp_chunk *ch)
102{
David S. Miller79af02c2005-07-08 21:47:49 -0700103 list_add_tail(&ch->list, &q->out_chunk_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 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 */
114static 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 */
130static 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 */
145static 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 */
166static 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 */
188static 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 */
204void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
205{
206 q->asoc = asoc;
David S. Miller79af02c2005-07-08 21:47:49 -0700207 INIT_LIST_HEAD(&q->out_chunk_list);
208 INIT_LIST_HEAD(&q->control_chunk_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 INIT_LIST_HEAD(&q->retransmit);
210 INIT_LIST_HEAD(&q->sacked);
211 INIT_LIST_HEAD(&q->abandoned);
212
Vlad Yasevich62aeaff2008-06-04 12:39:11 -0700213 q->fast_rtx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 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 */
224void sctp_outq_teardown(struct sctp_outq *q)
225{
226 struct sctp_transport *transport;
Robert P. J. Day9dbc15f2008-04-12 18:54:24 -0700227 struct list_head *lchunk, *temp;
David S. Miller79af02c2005-07-08 21:47:49 -0700228 struct sctp_chunk *chunk, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
230 /* Throw away unacknowledged chunks. */
Robert P. J. Day9dbc15f2008-04-12 18:54:24 -0700231 list_for_each_entry(transport, &q->asoc->peer.transport_addr_list,
232 transports) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 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. */
David S. Miller79af02c2005-07-08 21:47:49 -0700280 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
281 list_del_init(&chunk->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 sctp_chunk_free(chunk);
David S. Miller79af02c2005-07-08 21:47:49 -0700283 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284}
285
286/* Free the outqueue structure and any related pending chunks. */
287void 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. */
298int 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;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700344 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 } else {
David S. Miller79af02c2005-07-08 21:47:49 -0700346 list_add_tail(&chunk->list, &q->control_chunk_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 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 */
362static 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)
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900382 list_add_tail(new, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383}
384
385/* Mark all the eligible packets on a transport for retransmission. */
386void sctp_retransmit_mark(struct sctp_outq *q,
387 struct sctp_transport *transport,
Vlad Yasevichb6157d82007-10-24 15:59:16 -0400388 __u8 reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389{
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);
Vlad Yasevich8c4a2d42007-02-21 02:06:04 -0800402
403 /* If this chunk has not been previousely acked,
404 * stop considering it 'outstanding'. Our peer
405 * will most likely never see it since it will
406 * not be retransmitted
407 */
408 if (!chunk->tsn_gap_acked) {
409 chunk->transport->flight_size -=
410 sctp_data_size(chunk);
411 q->outstanding_bytes -= sctp_data_size(chunk);
412 q->asoc->peer.rwnd += (sctp_data_size(chunk) +
413 sizeof(struct sk_buff));
414 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 continue;
416 }
417
Vlad Yasevichb6157d82007-10-24 15:59:16 -0400418 /* If we are doing retransmission due to a timeout or pmtu
419 * discovery, only the chunks that are not yet acked should
420 * be added to the retransmit queue.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 */
Vlad Yasevichb6157d82007-10-24 15:59:16 -0400422 if ((reason == SCTP_RTXR_FAST_RTX &&
Neil Hormanc226ef92008-07-25 12:44:09 -0400423 (chunk->fast_retransmit == SCTP_NEED_FRTX)) ||
Vlad Yasevichb6157d82007-10-24 15:59:16 -0400424 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) {
Vlad Yasevichd0ce9292007-08-24 19:37:46 +0900425 /* If this chunk was sent less then 1 rto ago, do not
426 * retransmit this chunk, but give the peer time
Vlad Yasevichb6157d82007-10-24 15:59:16 -0400427 * to acknowlege it. Do this only when
428 * retransmitting due to T3 timeout.
Vlad Yasevichd0ce9292007-08-24 19:37:46 +0900429 */
Vlad Yasevichb6157d82007-10-24 15:59:16 -0400430 if (reason == SCTP_RTXR_T3_RTX &&
Wei Yongjunf61f6f82009-03-02 09:46:13 +0000431 time_before(jiffies, chunk->sent_at +
432 transport->last_rto))
Vlad Yasevichd0ce9292007-08-24 19:37:46 +0900433 continue;
434
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 /* RFC 2960 6.2.1 Processing a Received SACK
436 *
437 * C) Any time a DATA chunk is marked for
438 * retransmission (via either T3-rtx timer expiration
439 * (Section 6.3.3) or via fast retransmit
440 * (Section 7.2.4)), add the data size of those
441 * chunks to the rwnd.
442 */
Sridhar Samudralacd497882006-09-29 17:09:05 -0700443 q->asoc->peer.rwnd += (sctp_data_size(chunk) +
444 sizeof(struct sk_buff));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 q->outstanding_bytes -= sctp_data_size(chunk);
446 transport->flight_size -= sctp_data_size(chunk);
447
448 /* sctpimpguide-05 Section 2.8.2
449 * M5) If a T3-rtx timer expires, the
450 * 'TSN.Missing.Report' of all affected TSNs is set
451 * to 0.
452 */
453 chunk->tsn_missing_report = 0;
454
455 /* If a chunk that is being used for RTT measurement
456 * has to be retransmitted, we cannot use this chunk
457 * anymore for RTT measurements. Reset rto_pending so
458 * that a new RTT measurement is started when a new
459 * data chunk is sent.
460 */
461 if (chunk->rtt_in_progress) {
462 chunk->rtt_in_progress = 0;
463 transport->rto_pending = 0;
464 }
465
466 /* Move the chunk to the retransmit queue. The chunks
467 * on the retransmit queue are always kept in order.
468 */
469 list_del_init(lchunk);
470 sctp_insert_list(&q->retransmit, lchunk);
471 }
472 }
473
Vlad Yasevichb6157d82007-10-24 15:59:16 -0400474 SCTP_DEBUG_PRINTK("%s: transport: %p, reason: %d, "
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 "cwnd: %d, ssthresh: %d, flight_size: %d, "
Harvey Harrison0dc47872008-03-05 20:47:47 -0800476 "pba: %d\n", __func__,
Vlad Yasevichb6157d82007-10-24 15:59:16 -0400477 transport, reason,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 transport->cwnd, transport->ssthresh,
479 transport->flight_size,
480 transport->partial_bytes_acked);
481
482}
483
484/* Mark all the eligible packets on a transport for retransmission and force
485 * one packet out.
486 */
487void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
488 sctp_retransmit_reason_t reason)
489{
490 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
492 switch(reason) {
493 case SCTP_RTXR_T3_RTX:
Sridhar Samudralaac0b0462006-08-22 00:15:33 -0700494 SCTP_INC_STATS(SCTP_MIB_T3_RETRANSMITS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
496 /* Update the retran path if the T3-rtx timer has expired for
497 * the current retran path.
498 */
499 if (transport == transport->asoc->peer.retran_path)
500 sctp_assoc_update_retran_path(transport->asoc);
Neil Horman58fbbed2008-02-29 11:40:56 -0800501 transport->asoc->rtx_data_chunks +=
502 transport->asoc->unack_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 break;
504 case SCTP_RTXR_FAST_RTX:
Sridhar Samudralaac0b0462006-08-22 00:15:33 -0700505 SCTP_INC_STATS(SCTP_MIB_FAST_RETRANSMITS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
Vlad Yasevich62aeaff2008-06-04 12:39:11 -0700507 q->fast_rtx = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 break;
509 case SCTP_RTXR_PMTUD:
Sridhar Samudralaac0b0462006-08-22 00:15:33 -0700510 SCTP_INC_STATS(SCTP_MIB_PMTUD_RETRANSMITS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 break;
Vlad Yasevichb6157d82007-10-24 15:59:16 -0400512 case SCTP_RTXR_T1_RTX:
513 SCTP_INC_STATS(SCTP_MIB_T1_RETRANSMITS);
Neil Horman58fbbed2008-02-29 11:40:56 -0800514 transport->asoc->init_retries++;
Vlad Yasevichb6157d82007-10-24 15:59:16 -0400515 break;
Sridhar Samudralaac0b0462006-08-22 00:15:33 -0700516 default:
517 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 }
519
Vlad Yasevichb6157d82007-10-24 15:59:16 -0400520 sctp_retransmit_mark(q, transport, reason);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
522 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
523 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
524 * following the procedures outlined in C1 - C5.
525 */
Vlad Yasevich8b750ce2008-06-04 12:39:36 -0700526 if (reason == SCTP_RTXR_T3_RTX)
527 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528
Vlad Yasevich8b750ce2008-06-04 12:39:36 -0700529 /* Flush the queues only on timeout, since fast_rtx is only
530 * triggered during sack processing and the queue
531 * will be flushed at the end.
532 */
533 if (reason != SCTP_RTXR_FAST_RTX)
534 error = sctp_outq_flush(q, /* rtx_timeout */ 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
536 if (error)
537 q->asoc->base.sk->sk_err = -error;
538}
539
540/*
541 * Transmit DATA chunks on the retransmit queue. Upon return from
542 * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
543 * need to be transmitted by the caller.
544 * We assume that pkt->transport has already been set.
545 *
546 * The return value is a normal kernel error return value.
547 */
548static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
549 int rtx_timeout, int *start_timer)
550{
551 struct list_head *lqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 struct sctp_transport *transport = pkt->transport;
553 sctp_xmit_t status;
554 struct sctp_chunk *chunk, *chunk1;
555 struct sctp_association *asoc;
Vlad Yasevich62aeaff2008-06-04 12:39:11 -0700556 int fast_rtx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 int error = 0;
Vlad Yasevich62aeaff2008-06-04 12:39:11 -0700558 int timer = 0;
Vlad Yasevich8b750ce2008-06-04 12:39:36 -0700559 int done = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
561 asoc = q->asoc;
562 lqueue = &q->retransmit;
Vlad Yasevich62aeaff2008-06-04 12:39:11 -0700563 fast_rtx = q->fast_rtx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
Vlad Yasevich8b750ce2008-06-04 12:39:36 -0700565 /* This loop handles time-out retransmissions, fast retransmissions,
566 * and retransmissions due to opening of whindow.
567 *
568 * RFC 2960 6.3.3 Handle T3-rtx Expiration
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 *
570 * E3) Determine how many of the earliest (i.e., lowest TSN)
571 * outstanding DATA chunks for the address for which the
572 * T3-rtx has expired will fit into a single packet, subject
573 * to the MTU constraint for the path corresponding to the
574 * destination transport address to which the retransmission
575 * is being sent (this may be different from the address for
576 * which the timer expires [see Section 6.4]). Call this value
577 * K. Bundle and retransmit those K DATA chunks in a single
578 * packet to the destination endpoint.
579 *
580 * [Just to be painfully clear, if we are retransmitting
581 * because a timeout just happened, we should send only ONE
582 * packet of retransmitted data.]
Vlad Yasevich8b750ce2008-06-04 12:39:36 -0700583 *
584 * For fast retransmissions we also send only ONE packet. However,
585 * if we are just flushing the queue due to open window, we'll
586 * try to send as much as possible.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 */
Vlad Yasevich8b750ce2008-06-04 12:39:36 -0700588 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
590 /* Make sure that Gap Acked TSNs are not retransmitted. A
591 * simple approach is just to move such TSNs out of the
592 * way and into a 'transmitted' queue and skip to the
593 * next chunk.
594 */
595 if (chunk->tsn_gap_acked) {
Vlad Yasevich8b750ce2008-06-04 12:39:36 -0700596 list_del(&chunk->transmitted_list);
597 list_add_tail(&chunk->transmitted_list,
598 &transport->transmitted);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 continue;
600 }
601
Vlad Yasevich8b750ce2008-06-04 12:39:36 -0700602 /* If we are doing fast retransmit, ignore non-fast_rtransmit
603 * chunks
604 */
605 if (fast_rtx && !chunk->fast_retransmit)
606 continue;
607
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 /* Attempt to append this chunk to the packet. */
609 status = sctp_packet_append_chunk(pkt, chunk);
610
611 switch (status) {
612 case SCTP_XMIT_PMTU_FULL:
613 /* Send this packet. */
Vlad Yasevich62aeaff2008-06-04 12:39:11 -0700614 error = sctp_packet_transmit(pkt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
616 /* If we are retransmitting, we should only
617 * send a single packet.
618 */
Vlad Yasevich8b750ce2008-06-04 12:39:36 -0700619 if (rtx_timeout || fast_rtx)
620 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
Vlad Yasevich8b750ce2008-06-04 12:39:36 -0700622 /* Bundle next chunk in the next round. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 break;
624
625 case SCTP_XMIT_RWND_FULL:
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900626 /* Send this packet. */
Vlad Yasevich62aeaff2008-06-04 12:39:11 -0700627 error = sctp_packet_transmit(pkt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
629 /* Stop sending DATA as there is no more room
630 * at the receiver.
631 */
Vlad Yasevich8b750ce2008-06-04 12:39:36 -0700632 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 break;
634
635 case SCTP_XMIT_NAGLE_DELAY:
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900636 /* Send this packet. */
Vlad Yasevich62aeaff2008-06-04 12:39:11 -0700637 error = sctp_packet_transmit(pkt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
639 /* Stop sending DATA because of nagle delay. */
Vlad Yasevich8b750ce2008-06-04 12:39:36 -0700640 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 break;
642
643 default:
644 /* The append was successful, so add this chunk to
645 * the transmitted list.
646 */
Vlad Yasevich8b750ce2008-06-04 12:39:36 -0700647 list_del(&chunk->transmitted_list);
648 list_add_tail(&chunk->transmitted_list,
649 &transport->transmitted);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900651 /* Mark the chunk as ineligible for fast retransmit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 * after it is retransmitted.
653 */
Neil Hormanc226ef92008-07-25 12:44:09 -0400654 if (chunk->fast_retransmit == SCTP_NEED_FRTX)
655 chunk->fast_retransmit = SCTP_DONT_FRTX;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Vlad Yasevich62aeaff2008-06-04 12:39:11 -0700657 /* Force start T3-rtx timer when fast retransmitting
658 * the earliest outstanding TSN
659 */
660 if (!timer && fast_rtx &&
661 ntohl(chunk->subh.data_hdr->tsn) ==
662 asoc->ctsn_ack_point + 1)
663 timer = 2;
664
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 q->empty = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700667 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
Vlad Yasevich62aeaff2008-06-04 12:39:11 -0700669 /* Set the timer if there were no errors */
670 if (!error && !timer)
671 timer = 1;
672
Vlad Yasevich8b750ce2008-06-04 12:39:36 -0700673 if (done)
674 break;
675 }
676
677 /* If we are here due to a retransmit timeout or a fast
678 * retransmit and if there are any chunks left in the retransmit
679 * queue that could not fit in the PMTU sized packet, they need
680 * to be marked as ineligible for a subsequent fast retransmit.
681 */
682 if (rtx_timeout || fast_rtx) {
683 list_for_each_entry(chunk1, lqueue, transmitted_list) {
Neil Hormanc226ef92008-07-25 12:44:09 -0400684 if (chunk1->fast_retransmit == SCTP_NEED_FRTX)
685 chunk1->fast_retransmit = SCTP_DONT_FRTX;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 }
687 }
688
Vlad Yasevich62aeaff2008-06-04 12:39:11 -0700689 *start_timer = timer;
690
691 /* Clear fast retransmit hint */
692 if (fast_rtx)
693 q->fast_rtx = 0;
694
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 return error;
696}
697
698/* Cork the outqueue so queued chunks are really queued. */
699int sctp_outq_uncork(struct sctp_outq *q)
700{
701 int error = 0;
Vlad Yasevich7d54dc62007-11-09 11:43:41 -0500702 if (q->cork)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 q->cork = 0;
Vlad Yasevich7d54dc62007-11-09 11:43:41 -0500704 error = sctp_outq_flush(q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 return error;
706}
707
Vlad Yasevich2e3216c2008-06-19 16:08:18 -0700708
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709/*
710 * Try to flush an outqueue.
711 *
712 * Description: Send everything in q which we legally can, subject to
713 * congestion limitations.
714 * * Note: This function can be called from multiple contexts so appropriate
715 * locking concerns must be made. Today we use the sock lock to protect
716 * this function.
717 */
Adrian Bunkabd0b1982008-07-22 14:20:45 -0700718static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719{
720 struct sctp_packet *packet;
721 struct sctp_packet singleton;
722 struct sctp_association *asoc = q->asoc;
723 __u16 sport = asoc->base.bind_addr.port;
724 __u16 dport = asoc->peer.port;
725 __u32 vtag = asoc->peer.i.init_tag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 struct sctp_transport *transport = NULL;
727 struct sctp_transport *new_transport;
David S. Miller79af02c2005-07-08 21:47:49 -0700728 struct sctp_chunk *chunk, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 sctp_xmit_t status;
730 int error = 0;
731 int start_timer = 0;
Vlad Yasevich2e3216c2008-06-19 16:08:18 -0700732 int one_packet = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
734 /* These transports have chunks to send. */
735 struct list_head transport_list;
736 struct list_head *ltransport;
737
738 INIT_LIST_HEAD(&transport_list);
739 packet = NULL;
740
741 /*
742 * 6.10 Bundling
743 * ...
744 * When bundling control chunks with DATA chunks, an
745 * endpoint MUST place control chunks first in the outbound
746 * SCTP packet. The transmitter MUST transmit DATA chunks
747 * within a SCTP packet in increasing order of TSN.
748 * ...
749 */
750
David S. Miller79af02c2005-07-08 21:47:49 -0700751 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
752 list_del_init(&chunk->list);
753
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 /* Pick the right transport to use. */
755 new_transport = chunk->transport;
756
757 if (!new_transport) {
Vlad Yasevicha08de642007-12-20 14:11:47 -0800758 /*
759 * If we have a prior transport pointer, see if
760 * the destination address of the chunk
761 * matches the destination address of the
762 * current transport. If not a match, then
763 * try to look up the transport with a given
764 * destination address. We do this because
765 * after processing ASCONFs, we may have new
766 * transports created.
767 */
768 if (transport &&
769 sctp_cmp_addr_exact(&chunk->dest,
770 &transport->ipaddr))
771 new_transport = transport;
772 else
773 new_transport = sctp_assoc_lookup_paddr(asoc,
774 &chunk->dest);
775
776 /* if we still don't have a new transport, then
777 * use the current active path.
778 */
779 if (!new_transport)
780 new_transport = asoc->peer.active_path;
Sridhar Samudralaad8fec12006-07-21 14:48:50 -0700781 } else if ((new_transport->state == SCTP_INACTIVE) ||
782 (new_transport->state == SCTP_UNCONFIRMED)) {
Frank Filz3f7a87d2005-06-20 13:14:57 -0700783 /* If the chunk is Heartbeat or Heartbeat Ack,
784 * send it to chunk->transport, even if it's
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 * inactive.
786 *
787 * 3.3.6 Heartbeat Acknowledgement:
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900788 * ...
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 * A HEARTBEAT ACK is always sent to the source IP
790 * address of the IP datagram containing the
791 * HEARTBEAT chunk to which this ack is responding.
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900792 * ...
Vlad Yasevicha08de642007-12-20 14:11:47 -0800793 *
794 * ASCONF_ACKs also must be sent to the source.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 */
796 if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
Vlad Yasevicha08de642007-12-20 14:11:47 -0800797 chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK &&
798 chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 new_transport = asoc->peer.active_path;
800 }
801
802 /* Are we switching transports?
803 * Take care of transport locks.
804 */
805 if (new_transport != transport) {
806 transport = new_transport;
807 if (list_empty(&transport->send_ready)) {
808 list_add_tail(&transport->send_ready,
809 &transport_list);
810 }
811 packet = &transport->packet;
812 sctp_packet_config(packet, vtag,
813 asoc->peer.ecn_capable);
814 }
815
816 switch (chunk->chunk_hdr->type) {
817 /*
818 * 6.10 Bundling
819 * ...
820 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
821 * COMPLETE with any other chunks. [Send them immediately.]
822 */
823 case SCTP_CID_INIT:
824 case SCTP_CID_INIT_ACK:
825 case SCTP_CID_SHUTDOWN_COMPLETE:
826 sctp_packet_init(&singleton, transport, sport, dport);
827 sctp_packet_config(&singleton, vtag, 0);
828 sctp_packet_append_chunk(&singleton, chunk);
829 error = sctp_packet_transmit(&singleton);
830 if (error < 0)
831 return error;
832 break;
833
834 case SCTP_CID_ABORT:
Gui Jianfengf4ad85c2008-04-12 18:39:34 -0700835 if (sctp_test_T_bit(chunk)) {
836 packet->vtag = asoc->c.my_vtag;
837 }
Vlad Yasevich2e3216c2008-06-19 16:08:18 -0700838 /* The following chunks are "response" chunks, i.e.
839 * they are generated in response to something we
840 * received. If we are sending these, then we can
841 * send only 1 packet containing these chunks.
842 */
843 case SCTP_CID_HEARTBEAT_ACK:
844 case SCTP_CID_SHUTDOWN_ACK:
845 case SCTP_CID_COOKIE_ACK:
846 case SCTP_CID_COOKIE_ECHO:
847 case SCTP_CID_ERROR:
848 case SCTP_CID_ECN_CWR:
849 case SCTP_CID_ASCONF_ACK:
850 one_packet = 1;
851 /* Fall throught */
852
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 case SCTP_CID_SACK:
854 case SCTP_CID_HEARTBEAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 case SCTP_CID_SHUTDOWN:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 case SCTP_CID_ECN_ECNE:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 case SCTP_CID_ASCONF:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 case SCTP_CID_FWD_TSN:
Vlad Yasevich2e3216c2008-06-19 16:08:18 -0700859 status = sctp_packet_transmit_chunk(packet, chunk,
860 one_packet);
861 if (status != SCTP_XMIT_OK) {
862 /* put the chunk back */
863 list_add(&chunk->list, &q->control_chunk_list);
864 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 break;
866
867 default:
868 /* We built a chunk with an illegal type! */
869 BUG();
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700870 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 }
872
873 /* Is it OK to send data chunks? */
874 switch (asoc->state) {
875 case SCTP_STATE_COOKIE_ECHOED:
876 /* Only allow bundling when this packet has a COOKIE-ECHO
877 * chunk.
878 */
879 if (!packet || !packet->has_cookie_echo)
880 break;
881
882 /* fallthru */
883 case SCTP_STATE_ESTABLISHED:
884 case SCTP_STATE_SHUTDOWN_PENDING:
885 case SCTP_STATE_SHUTDOWN_RECEIVED:
886 /*
887 * RFC 2960 6.1 Transmission of DATA Chunks
888 *
889 * C) When the time comes for the sender to transmit,
890 * before sending new DATA chunks, the sender MUST
891 * first transmit any outstanding DATA chunks which
892 * are marked for retransmission (limited by the
893 * current cwnd).
894 */
895 if (!list_empty(&q->retransmit)) {
896 if (transport == asoc->peer.retran_path)
897 goto retran;
898
899 /* Switch transports & prepare the packet. */
900
901 transport = asoc->peer.retran_path;
902
903 if (list_empty(&transport->send_ready)) {
904 list_add_tail(&transport->send_ready,
905 &transport_list);
906 }
907
908 packet = &transport->packet;
909 sctp_packet_config(packet, vtag,
910 asoc->peer.ecn_capable);
911 retran:
912 error = sctp_outq_flush_rtx(q, packet,
913 rtx_timeout, &start_timer);
914
915 if (start_timer)
Vlad Yasevich62aeaff2008-06-04 12:39:11 -0700916 sctp_transport_reset_timers(transport,
917 start_timer-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918
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 /* Finally, transmit new packets. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
934 /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
935 * stream identifier.
936 */
937 if (chunk->sinfo.sinfo_stream >=
938 asoc->c.sinit_num_ostreams) {
939
940 /* Mark as failed send. */
941 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
942 sctp_chunk_free(chunk);
943 continue;
944 }
945
946 /* Has this chunk expired? */
947 if (sctp_chunk_abandoned(chunk)) {
948 sctp_chunk_fail(chunk, 0);
949 sctp_chunk_free(chunk);
950 continue;
951 }
952
953 /* If there is a specified transport, use it.
954 * Otherwise, we want to use the active path.
955 */
956 new_transport = chunk->transport;
Frank Filz3f7a87d2005-06-20 13:14:57 -0700957 if (!new_transport ||
Sridhar Samudralaad8fec12006-07-21 14:48:50 -0700958 ((new_transport->state == SCTP_INACTIVE) ||
959 (new_transport->state == SCTP_UNCONFIRMED)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 new_transport = asoc->peer.active_path;
961
962 /* Change packets if necessary. */
963 if (new_transport != transport) {
964 transport = new_transport;
965
966 /* Schedule to have this transport's
967 * packet flushed.
968 */
969 if (list_empty(&transport->send_ready)) {
970 list_add_tail(&transport->send_ready,
971 &transport_list);
972 }
973
974 packet = &transport->packet;
975 sctp_packet_config(packet, vtag,
976 asoc->peer.ecn_capable);
977 }
978
979 SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
980 q, chunk,
981 chunk && chunk->chunk_hdr ?
982 sctp_cname(SCTP_ST_CHUNK(
983 chunk->chunk_hdr->type))
984 : "Illegal Chunk");
985
986 SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
987 "%p skb->users %d.\n",
988 ntohl(chunk->subh.data_hdr->tsn),
989 chunk->skb ?chunk->skb->head : NULL,
990 chunk->skb ?
991 atomic_read(&chunk->skb->users) : -1);
992
993 /* Add the chunk to the packet. */
Vlad Yasevich2e3216c2008-06-19 16:08:18 -0700994 status = sctp_packet_transmit_chunk(packet, chunk, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
996 switch (status) {
997 case SCTP_XMIT_PMTU_FULL:
998 case SCTP_XMIT_RWND_FULL:
999 case SCTP_XMIT_NAGLE_DELAY:
1000 /* We could not append this chunk, so put
1001 * the chunk back on the output queue.
1002 */
1003 SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
1004 "not transmit TSN: 0x%x, status: %d\n",
1005 ntohl(chunk->subh.data_hdr->tsn),
1006 status);
1007 sctp_outq_head_data(q, chunk);
1008 goto sctp_flush_out;
1009 break;
1010
1011 case SCTP_XMIT_OK:
1012 break;
1013
1014 default:
1015 BUG();
1016 }
1017
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +09001018 /* BUG: We assume that the sctp_packet_transmit()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 * call below will succeed all the time and add the
1020 * chunk to the transmitted list and restart the
1021 * timers.
1022 * It is possible that the call can fail under OOM
1023 * conditions.
1024 *
1025 * Is this really a problem? Won't this behave
1026 * like a lost TSN?
1027 */
1028 list_add_tail(&chunk->transmitted_list,
1029 &transport->transmitted);
1030
Vlad Yasevich6574df92009-01-22 14:52:43 -08001031 sctp_transport_reset_timers(transport, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
1033 q->empty = 0;
1034
1035 /* Only let one DATA chunk get bundled with a
1036 * COOKIE-ECHO chunk.
1037 */
1038 if (packet->has_cookie_echo)
1039 goto sctp_flush_out;
1040 }
1041 break;
1042
1043 default:
1044 /* Do nothing. */
1045 break;
1046 }
1047
1048sctp_flush_out:
1049
1050 /* Before returning, examine all the transports touched in
1051 * this call. Right now, we bluntly force clear all the
1052 * transports. Things might change after we implement Nagle.
1053 * But such an examination is still required.
1054 *
1055 * --xguo
1056 */
1057 while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
1058 struct sctp_transport *t = list_entry(ltransport,
1059 struct sctp_transport,
1060 send_ready);
1061 packet = &t->packet;
1062 if (!sctp_packet_empty(packet))
1063 error = sctp_packet_transmit(packet);
1064 }
1065
1066 return error;
1067}
1068
1069/* Update unack_data based on the incoming SACK chunk */
1070static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1071 struct sctp_sackhdr *sack)
1072{
1073 sctp_sack_variable_t *frags;
1074 __u16 unack_data;
1075 int i;
1076
1077 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1078
1079 frags = sack->variable;
1080 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1081 unack_data -= ((ntohs(frags[i].gab.end) -
1082 ntohs(frags[i].gab.start) + 1));
1083 }
1084
1085 assoc->unack_data = unack_data;
1086}
1087
1088/* Return the highest new tsn that is acknowledged by the given SACK chunk. */
1089static __u32 sctp_highest_new_tsn(struct sctp_sackhdr *sack,
1090 struct sctp_association *asoc)
1091{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 struct sctp_transport *transport;
1093 struct sctp_chunk *chunk;
1094 __u32 highest_new_tsn, tsn;
1095 struct list_head *transport_list = &asoc->peer.transport_addr_list;
1096
1097 highest_new_tsn = ntohl(sack->cum_tsn_ack);
1098
Robert P. J. Day9dbc15f2008-04-12 18:54:24 -07001099 list_for_each_entry(transport, transport_list, transports) {
1100 list_for_each_entry(chunk, &transport->transmitted,
1101 transmitted_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 tsn = ntohl(chunk->subh.data_hdr->tsn);
1103
1104 if (!chunk->tsn_gap_acked &&
1105 TSN_lt(highest_new_tsn, tsn) &&
1106 sctp_acked(sack, tsn))
1107 highest_new_tsn = tsn;
1108 }
1109 }
1110
1111 return highest_new_tsn;
1112}
1113
1114/* This is where we REALLY process a SACK.
1115 *
1116 * Process the SACK against the outqueue. Mostly, this just frees
1117 * things off the transmitted queue.
1118 */
1119int sctp_outq_sack(struct sctp_outq *q, struct sctp_sackhdr *sack)
1120{
1121 struct sctp_association *asoc = q->asoc;
1122 struct sctp_transport *transport;
1123 struct sctp_chunk *tchunk = NULL;
Robert P. J. Day9dbc15f2008-04-12 18:54:24 -07001124 struct list_head *lchunk, *transport_list, *temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 sctp_sack_variable_t *frags = sack->variable;
1126 __u32 sack_ctsn, ctsn, tsn;
1127 __u32 highest_tsn, highest_new_tsn;
1128 __u32 sack_a_rwnd;
1129 unsigned outstanding;
1130 struct sctp_transport *primary = asoc->peer.primary_path;
1131 int count_of_newacks = 0;
Vlad Yasevich2cd9b822008-06-19 17:59:13 -04001132 int gap_ack_blocks;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133
1134 /* Grab the association's destination address list. */
1135 transport_list = &asoc->peer.transport_addr_list;
1136
1137 sack_ctsn = ntohl(sack->cum_tsn_ack);
Vlad Yasevich2cd9b822008-06-19 17:59:13 -04001138 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 /*
1140 * SFR-CACC algorithm:
1141 * On receipt of a SACK the sender SHOULD execute the
1142 * following statements.
1143 *
1144 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1145 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1146 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1147 * all destinations.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1149 * is set the receiver of the SACK MUST take the following actions:
1150 *
1151 * A) Initialize the cacc_saw_newack to 0 for all destination
1152 * addresses.
Vlad Yasevichab5216a2008-06-19 18:17:24 -04001153 *
1154 * Only bother if changeover_active is set. Otherwise, this is
1155 * totally suboptimal to do on every SACK.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 */
Vlad Yasevichab5216a2008-06-19 18:17:24 -04001157 if (primary->cacc.changeover_active) {
1158 u8 clear_cycling = 0;
1159
1160 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1161 primary->cacc.changeover_active = 0;
1162 clear_cycling = 1;
1163 }
1164
1165 if (clear_cycling || gap_ack_blocks) {
1166 list_for_each_entry(transport, transport_list,
1167 transports) {
1168 if (clear_cycling)
1169 transport->cacc.cycling_changeover = 0;
1170 if (gap_ack_blocks)
1171 transport->cacc.cacc_saw_newack = 0;
1172 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 }
1174 }
1175
1176 /* Get the highest TSN in the sack. */
1177 highest_tsn = sack_ctsn;
Vlad Yasevich2cd9b822008-06-19 17:59:13 -04001178 if (gap_ack_blocks)
1179 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
1181 if (TSN_lt(asoc->highest_sacked, highest_tsn)) {
1182 highest_new_tsn = highest_tsn;
1183 asoc->highest_sacked = highest_tsn;
1184 } else {
1185 highest_new_tsn = sctp_highest_new_tsn(sack, asoc);
1186 }
1187
Vlad Yasevich2cd9b822008-06-19 17:59:13 -04001188
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 /* Run through the retransmit queue. Credit bytes received
1190 * and free those chunks that we can.
1191 */
1192 sctp_check_transmitted(q, &q->retransmit, NULL, sack, highest_new_tsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
1194 /* Run through the transmitted queue.
1195 * Credit bytes received and free those chunks which we can.
1196 *
1197 * This is a MASSIVE candidate for optimization.
1198 */
Robert P. J. Day9dbc15f2008-04-12 18:54:24 -07001199 list_for_each_entry(transport, transport_list, transports) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 sctp_check_transmitted(q, &transport->transmitted,
1201 transport, sack, highest_new_tsn);
1202 /*
1203 * SFR-CACC algorithm:
1204 * C) Let count_of_newacks be the number of
1205 * destinations for which cacc_saw_newack is set.
1206 */
1207 if (transport->cacc.cacc_saw_newack)
1208 count_of_newacks ++;
1209 }
1210
Vlad Yasevich2cd9b822008-06-19 17:59:13 -04001211 if (gap_ack_blocks) {
Vlad Yasevich2cd9b822008-06-19 17:59:13 -04001212 list_for_each_entry(transport, transport_list, transports)
1213 sctp_mark_missing(q, &transport->transmitted, transport,
1214 highest_new_tsn, count_of_newacks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 }
1216
1217 /* Move the Cumulative TSN Ack Point if appropriate. */
1218 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn))
1219 asoc->ctsn_ack_point = sack_ctsn;
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);
Vlad Yasevich5f9646c2008-02-05 14:23:44 -05001231 if (TSN_lte(tsn, ctsn)) {
1232 list_del_init(&tchunk->transmitted_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 sctp_chunk_free(tchunk);
Vlad Yasevich5f9646c2008-02-05 14:23:44 -05001234 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 }
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",
Harvey Harrison0dc47872008-03-05 20:47:47 -08001255 __func__, sack_ctsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, "
1257 "%p is 0x%x. Adv peer ack point: 0x%x\n",
Harvey Harrison0dc47872008-03-05 20:47:47 -08001258 __func__, asoc, ctsn, asoc->adv_peer_ack_point);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
1260 /* See if all chunks are acked.
1261 * Make sure the empty queue handler will get run later.
1262 */
David S. Miller79af02c2005-07-08 21:47:49 -07001263 q->empty = (list_empty(&q->out_chunk_list) &&
David S. Miller79af02c2005-07-08 21:47:49 -07001264 list_empty(&q->retransmit));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 if (!q->empty)
1266 goto finish;
1267
Robert P. J. Day9dbc15f2008-04-12 18:54:24 -07001268 list_for_each_entry(transport, transport_list, transports) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 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");
1275finish:
1276 return q->empty;
1277}
1278
1279/* Is the outqueue empty? */
1280int 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 */
1299static 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
1314 /* These state variables are for coherent debug output. --xguo */
1315
1316#if SCTP_DEBUG
1317 __u32 dbg_ack_tsn = 0; /* An ACKed TSN range starts here... */
1318 __u32 dbg_last_ack_tsn = 0; /* ...and finishes here. */
1319 __u32 dbg_kept_tsn = 0; /* An un-ACKed range starts here... */
1320 __u32 dbg_last_kept_tsn = 0; /* ...and finishes here. */
1321
1322 /* 0 : The last TSN was ACKed.
1323 * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1324 * -1: We need to initialize.
1325 */
1326 int dbg_prt_state = -1;
1327#endif /* SCTP_DEBUG */
1328
1329 sack_ctsn = ntohl(sack->cum_tsn_ack);
1330
1331 INIT_LIST_HEAD(&tlist);
1332
1333 /* The while loop will skip empty transmitted queues. */
1334 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1335 tchunk = list_entry(lchunk, struct sctp_chunk,
1336 transmitted_list);
1337
1338 if (sctp_chunk_abandoned(tchunk)) {
1339 /* Move the chunk to abandoned list. */
1340 sctp_insert_list(&q->abandoned, lchunk);
Vlad Yasevich8c4a2d42007-02-21 02:06:04 -08001341
1342 /* If this chunk has not been acked, stop
1343 * considering it as 'outstanding'.
1344 */
1345 if (!tchunk->tsn_gap_acked) {
1346 tchunk->transport->flight_size -=
1347 sctp_data_size(tchunk);
1348 q->outstanding_bytes -= sctp_data_size(tchunk);
1349 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 continue;
1351 }
1352
1353 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1354 if (sctp_acked(sack, tsn)) {
1355 /* If this queue is the retransmit queue, the
1356 * retransmit timer has already reclaimed
1357 * the outstanding bytes for this chunk, so only
1358 * count bytes associated with a transport.
1359 */
1360 if (transport) {
1361 /* If this chunk is being used for RTT
1362 * measurement, calculate the RTT and update
1363 * the RTO using this value.
1364 *
1365 * 6.3.1 C5) Karn's algorithm: RTT measurements
1366 * MUST NOT be made using packets that were
1367 * retransmitted (and thus for which it is
1368 * ambiguous whether the reply was for the
1369 * first instance of the packet or a later
1370 * instance).
1371 */
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +09001372 if (!tchunk->tsn_gap_acked &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 !tchunk->resent &&
1374 tchunk->rtt_in_progress) {
Vlad Yasevich4c9f5d52006-06-17 22:56:08 -07001375 tchunk->rtt_in_progress = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 rtt = jiffies - tchunk->sent_at;
1377 sctp_transport_update_rto(transport,
1378 rtt);
1379 }
1380 }
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +09001381 if (TSN_lte(tsn, sack_ctsn)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 /* RFC 2960 6.3.2 Retransmission Timer Rules
1383 *
1384 * R3) Whenever a SACK is received
1385 * that acknowledges the DATA chunk
1386 * with the earliest outstanding TSN
1387 * for that address, restart T3-rtx
1388 * timer for that address with its
1389 * current RTO.
1390 */
1391 restart_timer = 1;
1392
1393 if (!tchunk->tsn_gap_acked) {
1394 tchunk->tsn_gap_acked = 1;
1395 bytes_acked += sctp_data_size(tchunk);
1396 /*
1397 * SFR-CACC algorithm:
1398 * 2) If the SACK contains gap acks
1399 * and the flag CHANGEOVER_ACTIVE is
1400 * set the receiver of the SACK MUST
1401 * take the following action:
1402 *
1403 * B) For each TSN t being acked that
1404 * has not been acked in any SACK so
1405 * far, set cacc_saw_newack to 1 for
1406 * the destination that the TSN was
1407 * sent to.
1408 */
1409 if (transport &&
1410 sack->num_gap_ack_blocks &&
1411 q->asoc->peer.primary_path->cacc.
1412 changeover_active)
1413 transport->cacc.cacc_saw_newack
1414 = 1;
1415 }
1416
1417 list_add_tail(&tchunk->transmitted_list,
1418 &q->sacked);
1419 } else {
1420 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1421 * M2) Each time a SACK arrives reporting
1422 * 'Stray DATA chunk(s)' record the highest TSN
1423 * reported as newly acknowledged, call this
1424 * value 'HighestTSNinSack'. A newly
1425 * acknowledged DATA chunk is one not
1426 * previously acknowledged in a SACK.
1427 *
1428 * When the SCTP sender of data receives a SACK
1429 * chunk that acknowledges, for the first time,
1430 * the receipt of a DATA chunk, all the still
1431 * unacknowledged DATA chunks whose TSN is
1432 * older than that newly acknowledged DATA
1433 * chunk, are qualified as 'Stray DATA chunks'.
1434 */
1435 if (!tchunk->tsn_gap_acked) {
1436 tchunk->tsn_gap_acked = 1;
1437 bytes_acked += sctp_data_size(tchunk);
1438 }
1439 list_add_tail(lchunk, &tlist);
1440 }
1441
1442#if SCTP_DEBUG
1443 switch (dbg_prt_state) {
1444 case 0: /* last TSN was ACKed */
1445 if (dbg_last_ack_tsn + 1 == tsn) {
1446 /* This TSN belongs to the
1447 * current ACK range.
1448 */
1449 break;
1450 }
1451
1452 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1453 /* Display the end of the
1454 * current range.
1455 */
1456 SCTP_DEBUG_PRINTK("-%08x",
1457 dbg_last_ack_tsn);
1458 }
1459
1460 /* Start a new range. */
1461 SCTP_DEBUG_PRINTK(",%08x", tsn);
1462 dbg_ack_tsn = tsn;
1463 break;
1464
1465 case 1: /* The last TSN was NOT ACKed. */
1466 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1467 /* Display the end of current range. */
1468 SCTP_DEBUG_PRINTK("-%08x",
1469 dbg_last_kept_tsn);
1470 }
1471
1472 SCTP_DEBUG_PRINTK("\n");
1473
1474 /* FALL THROUGH... */
1475 default:
1476 /* This is the first-ever TSN we examined. */
1477 /* Start a new range of ACK-ed TSNs. */
1478 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn);
1479 dbg_prt_state = 0;
1480 dbg_ack_tsn = tsn;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07001481 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482
1483 dbg_last_ack_tsn = tsn;
1484#endif /* SCTP_DEBUG */
1485
1486 } else {
1487 if (tchunk->tsn_gap_acked) {
1488 SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1489 "data TSN: 0x%x\n",
Harvey Harrison0dc47872008-03-05 20:47:47 -08001490 __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 tsn);
1492 tchunk->tsn_gap_acked = 0;
1493
1494 bytes_acked -= sctp_data_size(tchunk);
1495
1496 /* RFC 2960 6.3.2 Retransmission Timer Rules
1497 *
1498 * R4) Whenever a SACK is received missing a
1499 * TSN that was previously acknowledged via a
1500 * Gap Ack Block, start T3-rtx for the
1501 * destination address to which the DATA
1502 * chunk was originally
1503 * transmitted if it is not already running.
1504 */
1505 restart_timer = 1;
1506 }
1507
1508 list_add_tail(lchunk, &tlist);
1509
1510#if SCTP_DEBUG
1511 /* See the above comments on ACK-ed TSNs. */
1512 switch (dbg_prt_state) {
1513 case 1:
1514 if (dbg_last_kept_tsn + 1 == tsn)
1515 break;
1516
1517 if (dbg_last_kept_tsn != dbg_kept_tsn)
1518 SCTP_DEBUG_PRINTK("-%08x",
1519 dbg_last_kept_tsn);
1520
1521 SCTP_DEBUG_PRINTK(",%08x", tsn);
1522 dbg_kept_tsn = tsn;
1523 break;
1524
1525 case 0:
1526 if (dbg_last_ack_tsn != dbg_ack_tsn)
1527 SCTP_DEBUG_PRINTK("-%08x",
1528 dbg_last_ack_tsn);
1529 SCTP_DEBUG_PRINTK("\n");
1530
1531 /* FALL THROUGH... */
1532 default:
1533 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn);
1534 dbg_prt_state = 1;
1535 dbg_kept_tsn = tsn;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07001536 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
1538 dbg_last_kept_tsn = tsn;
1539#endif /* SCTP_DEBUG */
1540 }
1541 }
1542
1543#if SCTP_DEBUG
1544 /* Finish off the last range, displaying its ending TSN. */
1545 switch (dbg_prt_state) {
1546 case 0:
1547 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1548 SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_ack_tsn);
1549 } else {
1550 SCTP_DEBUG_PRINTK("\n");
1551 }
1552 break;
1553
1554 case 1:
1555 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1556 SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_kept_tsn);
1557 } else {
1558 SCTP_DEBUG_PRINTK("\n");
1559 }
Stephen Hemminger3ff50b72007-04-20 17:09:22 -07001560 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561#endif /* SCTP_DEBUG */
1562 if (transport) {
1563 if (bytes_acked) {
1564 /* 8.2. When an outstanding TSN is acknowledged,
1565 * the endpoint shall clear the error counter of
1566 * the destination transport address to which the
1567 * DATA chunk was last sent.
1568 * The association's overall error counter is
1569 * also cleared.
1570 */
1571 transport->error_count = 0;
1572 transport->asoc->overall_error_count = 0;
1573
1574 /* Mark the destination transport address as
1575 * active if it is not so marked.
1576 */
Sridhar Samudralaad8fec12006-07-21 14:48:50 -07001577 if ((transport->state == SCTP_INACTIVE) ||
1578 (transport->state == SCTP_UNCONFIRMED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 sctp_assoc_control_transport(
1580 transport->asoc,
1581 transport,
1582 SCTP_TRANSPORT_UP,
1583 SCTP_RECEIVED_SACK);
1584 }
1585
1586 sctp_transport_raise_cwnd(transport, sack_ctsn,
1587 bytes_acked);
1588
1589 transport->flight_size -= bytes_acked;
Gui Jianfeng8b73a072008-04-17 14:22:18 -07001590 if (transport->flight_size == 0)
1591 transport->partial_bytes_acked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 q->outstanding_bytes -= bytes_acked;
1593 } else {
1594 /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1595 * When a sender is doing zero window probing, it
1596 * should not timeout the association if it continues
1597 * to receive new packets from the receiver. The
1598 * reason is that the receiver MAY keep its window
1599 * closed for an indefinite time.
1600 * A sender is doing zero window probing when the
1601 * receiver's advertised window is zero, and there is
1602 * only one data chunk in flight to the receiver.
1603 */
1604 if (!q->asoc->peer.rwnd &&
1605 !list_empty(&tlist) &&
1606 (sack_ctsn+2 == q->asoc->next_tsn)) {
1607 SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1608 "window probe: %u\n",
Harvey Harrison0dc47872008-03-05 20:47:47 -08001609 __func__, sack_ctsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 q->asoc->overall_error_count = 0;
1611 transport->error_count = 0;
1612 }
1613 }
1614
1615 /* RFC 2960 6.3.2 Retransmission Timer Rules
1616 *
1617 * R2) Whenever all outstanding data sent to an address have
1618 * been acknowledged, turn off the T3-rtx timer of that
1619 * address.
1620 */
1621 if (!transport->flight_size) {
1622 if (timer_pending(&transport->T3_rtx_timer) &&
1623 del_timer(&transport->T3_rtx_timer)) {
1624 sctp_transport_put(transport);
1625 }
1626 } else if (restart_timer) {
1627 if (!mod_timer(&transport->T3_rtx_timer,
1628 jiffies + transport->rto))
1629 sctp_transport_hold(transport);
1630 }
1631 }
1632
1633 list_splice(&tlist, transmitted_queue);
1634}
1635
1636/* Mark chunks as missing and consequently may get retransmitted. */
1637static void sctp_mark_missing(struct sctp_outq *q,
1638 struct list_head *transmitted_queue,
1639 struct sctp_transport *transport,
1640 __u32 highest_new_tsn_in_sack,
1641 int count_of_newacks)
1642{
1643 struct sctp_chunk *chunk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 __u32 tsn;
1645 char do_fast_retransmit = 0;
1646 struct sctp_transport *primary = q->asoc->peer.primary_path;
1647
Robert P. J. Day9dbc15f2008-04-12 18:54:24 -07001648 list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 tsn = ntohl(chunk->subh.data_hdr->tsn);
1651
1652 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1653 * 'Unacknowledged TSN's', if the TSN number of an
1654 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1655 * value, increment the 'TSN.Missing.Report' count on that
1656 * chunk if it has NOT been fast retransmitted or marked for
1657 * fast retransmit already.
1658 */
Neil Hormanc226ef92008-07-25 12:44:09 -04001659 if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 !chunk->tsn_gap_acked &&
1661 TSN_lt(tsn, highest_new_tsn_in_sack)) {
1662
1663 /* SFR-CACC may require us to skip marking
1664 * this chunk as missing.
1665 */
1666 if (!transport || !sctp_cacc_skip(primary, transport,
1667 count_of_newacks, tsn)) {
1668 chunk->tsn_missing_report++;
1669
1670 SCTP_DEBUG_PRINTK(
1671 "%s: TSN 0x%x missing counter: %d\n",
Harvey Harrison0dc47872008-03-05 20:47:47 -08001672 __func__, tsn,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 chunk->tsn_missing_report);
1674 }
1675 }
1676 /*
1677 * M4) If any DATA chunk is found to have a
1678 * 'TSN.Missing.Report'
Vlad Yasevich27852c22006-02-02 16:57:31 -08001679 * value larger than or equal to 3, mark that chunk for
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 * retransmission and start the fast retransmit procedure.
1681 */
1682
Vlad Yasevich27852c22006-02-02 16:57:31 -08001683 if (chunk->tsn_missing_report >= 3) {
Neil Hormanc226ef92008-07-25 12:44:09 -04001684 chunk->fast_retransmit = SCTP_NEED_FRTX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 do_fast_retransmit = 1;
1686 }
1687 }
1688
1689 if (transport) {
1690 if (do_fast_retransmit)
1691 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1692
1693 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1694 "ssthresh: %d, flight_size: %d, pba: %d\n",
Harvey Harrison0dc47872008-03-05 20:47:47 -08001695 __func__, transport, transport->cwnd,
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +09001696 transport->ssthresh, transport->flight_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 transport->partial_bytes_acked);
1698 }
1699}
1700
1701/* Is the given TSN acked by this packet? */
1702static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1703{
1704 int i;
1705 sctp_sack_variable_t *frags;
1706 __u16 gap;
1707 __u32 ctsn = ntohl(sack->cum_tsn_ack);
1708
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +09001709 if (TSN_lte(tsn, ctsn))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 goto pass;
1711
1712 /* 3.3.4 Selective Acknowledgement (SACK) (3):
1713 *
1714 * Gap Ack Blocks:
1715 * These fields contain the Gap Ack Blocks. They are repeated
1716 * for each Gap Ack Block up to the number of Gap Ack Blocks
1717 * defined in the Number of Gap Ack Blocks field. All DATA
1718 * chunks with TSNs greater than or equal to (Cumulative TSN
1719 * Ack + Gap Ack Block Start) and less than or equal to
1720 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1721 * Block are assumed to have been received correctly.
1722 */
1723
1724 frags = sack->variable;
1725 gap = tsn - ctsn;
1726 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1727 if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1728 TSN_lte(gap, ntohs(frags[i].gab.end)))
1729 goto pass;
1730 }
1731
1732 return 0;
1733pass:
1734 return 1;
1735}
1736
1737static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
Al Viro9f81bcd2006-11-20 17:26:34 -08001738 int nskips, __be16 stream)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739{
1740 int i;
1741
1742 for (i = 0; i < nskips; i++) {
1743 if (skiplist[i].stream == stream)
1744 return i;
1745 }
1746 return i;
1747}
1748
1749/* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1750static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1751{
1752 struct sctp_association *asoc = q->asoc;
1753 struct sctp_chunk *ftsn_chunk = NULL;
1754 struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1755 int nskips = 0;
1756 int skip_pos = 0;
1757 __u32 tsn;
1758 struct sctp_chunk *chunk;
1759 struct list_head *lchunk, *temp;
1760
Wei Yongjun76595022009-03-12 09:49:19 +00001761 if (!asoc->peer.prsctp_capable)
1762 return;
1763
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1765 * received SACK.
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +09001766 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1768 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1769 */
1770 if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1771 asoc->adv_peer_ack_point = ctsn;
1772
1773 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1774 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1775 * the chunk next in the out-queue space is marked as "abandoned" as
1776 * shown in the following example:
1777 *
1778 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1779 * and the Advanced.Peer.Ack.Point is updated to this value:
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +09001780 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 * out-queue at the end of ==> out-queue after Adv.Ack.Point
1782 * normal SACK processing local advancement
1783 * ... ...
1784 * Adv.Ack.Pt-> 102 acked 102 acked
1785 * 103 abandoned 103 abandoned
1786 * 104 abandoned Adv.Ack.P-> 104 abandoned
1787 * 105 105
1788 * 106 acked 106 acked
1789 * ... ...
1790 *
1791 * In this example, the data sender successfully advanced the
1792 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1793 */
1794 list_for_each_safe(lchunk, temp, &q->abandoned) {
1795 chunk = list_entry(lchunk, struct sctp_chunk,
1796 transmitted_list);
1797 tsn = ntohl(chunk->subh.data_hdr->tsn);
1798
1799 /* Remove any chunks in the abandoned queue that are acked by
1800 * the ctsn.
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +09001801 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 if (TSN_lte(tsn, ctsn)) {
1803 list_del_init(lchunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 sctp_chunk_free(chunk);
1805 } else {
1806 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1807 asoc->adv_peer_ack_point = tsn;
1808 if (chunk->chunk_hdr->flags &
1809 SCTP_DATA_UNORDERED)
1810 continue;
1811 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1812 nskips,
1813 chunk->subh.data_hdr->stream);
1814 ftsn_skip_arr[skip_pos].stream =
1815 chunk->subh.data_hdr->stream;
1816 ftsn_skip_arr[skip_pos].ssn =
1817 chunk->subh.data_hdr->ssn;
1818 if (skip_pos == nskips)
1819 nskips++;
1820 if (nskips == 10)
1821 break;
1822 } else
1823 break;
1824 }
1825 }
1826
1827 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1828 * is greater than the Cumulative TSN ACK carried in the received
1829 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1830 * chunk containing the latest value of the
1831 * "Advanced.Peer.Ack.Point".
1832 *
1833 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1834 * list each stream and sequence number in the forwarded TSN. This
1835 * information will enable the receiver to easily find any
1836 * stranded TSN's waiting on stream reorder queues. Each stream
1837 * SHOULD only be reported once; this means that if multiple
1838 * abandoned messages occur in the same stream then only the
1839 * highest abandoned stream sequence number is reported. If the
1840 * total size of the FORWARD TSN does NOT fit in a single MTU then
1841 * the sender of the FORWARD TSN SHOULD lower the
1842 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1843 * single MTU.
1844 */
1845 if (asoc->adv_peer_ack_point > ctsn)
1846 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +09001847 nskips, &ftsn_skip_arr[0]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848
1849 if (ftsn_chunk) {
David S. Miller79af02c2005-07-08 21:47:49 -07001850 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
1852 }
1853}