Merge branch 'v4l_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab...
[linux-2.6.git] / net / sctp / outqueue.c
blobbe35e2dbcc9aed6dc12fd5c2abc02d584d1d8bc5
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.
7 * This file is part of the SCTP kernel implementation
9 * These functions implement the sctp_outq class. The outqueue handles
10 * bundling and queueing of outgoing SCTP chunks.
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.
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.
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.
29 * Please send any bug reports or fixes you make to the
30 * email address(es):
31 * lksctp developers <lksctp-developers@lists.sourceforge.net>
33 * Or submit a bug report through the following website:
34 * http://www.sf.net/projects/lksctp
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>
45 * Any bugs reported given to us we will try to fix... any fixes shared will
46 * be incorporated into the next SCTP release.
49 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
51 #include <linux/types.h>
52 #include <linux/list.h> /* For struct list_head */
53 #include <linux/socket.h>
54 #include <linux/ip.h>
55 #include <linux/slab.h>
56 #include <net/sock.h> /* For skb_set_owner_w */
58 #include <net/sctp/sctp.h>
59 #include <net/sctp/sm.h>
61 /* Declare internal functions here. */
62 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn);
63 static void sctp_check_transmitted(struct sctp_outq *q,
64 struct list_head *transmitted_queue,
65 struct sctp_transport *transport,
66 union sctp_addr *saddr,
67 struct sctp_sackhdr *sack,
68 __u32 *highest_new_tsn);
70 static void sctp_mark_missing(struct sctp_outq *q,
71 struct list_head *transmitted_queue,
72 struct sctp_transport *transport,
73 __u32 highest_new_tsn,
74 int count_of_newacks);
76 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn);
78 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout);
80 /* Add data to the front of the queue. */
81 static inline void sctp_outq_head_data(struct sctp_outq *q,
82 struct sctp_chunk *ch)
84 list_add(&ch->list, &q->out_chunk_list);
85 q->out_qlen += ch->skb->len;
88 /* Take data from the front of the queue. */
89 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
91 struct sctp_chunk *ch = NULL;
93 if (!list_empty(&q->out_chunk_list)) {
94 struct list_head *entry = q->out_chunk_list.next;
96 ch = list_entry(entry, struct sctp_chunk, list);
97 list_del_init(entry);
98 q->out_qlen -= ch->skb->len;
100 return ch;
102 /* Add data chunk to the end of the queue. */
103 static inline void sctp_outq_tail_data(struct sctp_outq *q,
104 struct sctp_chunk *ch)
106 list_add_tail(&ch->list, &q->out_chunk_list);
107 q->out_qlen += ch->skb->len;
111 * SFR-CACC algorithm:
112 * D) If count_of_newacks is greater than or equal to 2
113 * and t was not sent to the current primary then the
114 * sender MUST NOT increment missing report count for t.
116 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary,
117 struct sctp_transport *transport,
118 int count_of_newacks)
120 if (count_of_newacks >=2 && transport != primary)
121 return 1;
122 return 0;
126 * SFR-CACC algorithm:
127 * F) If count_of_newacks is less than 2, let d be the
128 * destination to which t was sent. If cacc_saw_newack
129 * is 0 for destination d, then the sender MUST NOT
130 * increment missing report count for t.
132 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport,
133 int count_of_newacks)
135 if (count_of_newacks < 2 &&
136 (transport && !transport->cacc.cacc_saw_newack))
137 return 1;
138 return 0;
142 * SFR-CACC algorithm:
143 * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
144 * execute steps C, D, F.
146 * C has been implemented in sctp_outq_sack
148 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
149 struct sctp_transport *transport,
150 int count_of_newacks)
152 if (!primary->cacc.cycling_changeover) {
153 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
154 return 1;
155 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
156 return 1;
157 return 0;
159 return 0;
163 * SFR-CACC algorithm:
164 * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
165 * than next_tsn_at_change of the current primary, then
166 * the sender MUST NOT increment missing report count
167 * for t.
169 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
171 if (primary->cacc.cycling_changeover &&
172 TSN_lt(tsn, primary->cacc.next_tsn_at_change))
173 return 1;
174 return 0;
178 * SFR-CACC algorithm:
179 * 3) If the missing report count for TSN t is to be
180 * incremented according to [RFC2960] and
181 * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
182 * then the sender MUST further execute steps 3.1 and
183 * 3.2 to determine if the missing report count for
184 * TSN t SHOULD NOT be incremented.
186 * 3.3) If 3.1 and 3.2 do not dictate that the missing
187 * report count for t should not be incremented, then
188 * the sender SHOULD increment missing report count for
189 * t (according to [RFC2960] and [SCTP_STEWART_2002]).
191 static inline int sctp_cacc_skip(struct sctp_transport *primary,
192 struct sctp_transport *transport,
193 int count_of_newacks,
194 __u32 tsn)
196 if (primary->cacc.changeover_active &&
197 (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) ||
198 sctp_cacc_skip_3_2(primary, tsn)))
199 return 1;
200 return 0;
203 /* Initialize an existing sctp_outq. This does the boring stuff.
204 * You still need to define handlers if you really want to DO
205 * something with this structure...
207 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
209 memset(q, 0, sizeof(struct sctp_outq));
211 q->asoc = asoc;
212 INIT_LIST_HEAD(&q->out_chunk_list);
213 INIT_LIST_HEAD(&q->control_chunk_list);
214 INIT_LIST_HEAD(&q->retransmit);
215 INIT_LIST_HEAD(&q->sacked);
216 INIT_LIST_HEAD(&q->abandoned);
218 q->empty = 1;
221 /* Free the outqueue structure and any related pending chunks.
223 static void __sctp_outq_teardown(struct sctp_outq *q)
225 struct sctp_transport *transport;
226 struct list_head *lchunk, *temp;
227 struct sctp_chunk *chunk, *tmp;
229 /* Throw away unacknowledged chunks. */
230 list_for_each_entry(transport, &q->asoc->peer.transport_addr_list,
231 transports) {
232 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
233 chunk = list_entry(lchunk, struct sctp_chunk,
234 transmitted_list);
235 /* Mark as part of a failed message. */
236 sctp_chunk_fail(chunk, q->error);
237 sctp_chunk_free(chunk);
241 /* Throw away chunks that have been gap ACKed. */
242 list_for_each_safe(lchunk, temp, &q->sacked) {
243 list_del_init(lchunk);
244 chunk = list_entry(lchunk, struct sctp_chunk,
245 transmitted_list);
246 sctp_chunk_fail(chunk, q->error);
247 sctp_chunk_free(chunk);
250 /* Throw away any chunks in the retransmit queue. */
251 list_for_each_safe(lchunk, temp, &q->retransmit) {
252 list_del_init(lchunk);
253 chunk = list_entry(lchunk, struct sctp_chunk,
254 transmitted_list);
255 sctp_chunk_fail(chunk, q->error);
256 sctp_chunk_free(chunk);
259 /* Throw away any chunks that are in the abandoned queue. */
260 list_for_each_safe(lchunk, temp, &q->abandoned) {
261 list_del_init(lchunk);
262 chunk = list_entry(lchunk, struct sctp_chunk,
263 transmitted_list);
264 sctp_chunk_fail(chunk, q->error);
265 sctp_chunk_free(chunk);
268 /* Throw away any leftover data chunks. */
269 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
271 /* Mark as send failure. */
272 sctp_chunk_fail(chunk, q->error);
273 sctp_chunk_free(chunk);
276 /* Throw away any leftover control chunks. */
277 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
278 list_del_init(&chunk->list);
279 sctp_chunk_free(chunk);
283 void sctp_outq_teardown(struct sctp_outq *q)
285 __sctp_outq_teardown(q);
286 sctp_outq_init(q->asoc, q);
289 /* Free the outqueue structure and any related pending chunks. */
290 void sctp_outq_free(struct sctp_outq *q)
292 /* Throw away leftover chunks. */
293 __sctp_outq_teardown(q);
296 /* Put a new chunk in an sctp_outq. */
297 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk)
299 struct net *net = sock_net(q->asoc->base.sk);
300 int error = 0;
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");
307 /* If it is data, queue it up, otherwise, send it
308 * immediately.
310 if (sctp_chunk_is_data(chunk)) {
311 /* Is it OK to queue data chunks? */
312 /* From 9. Termination of Association
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.
320 switch (q->asoc->state) {
321 case SCTP_STATE_CLOSED:
322 case SCTP_STATE_SHUTDOWN_PENDING:
323 case SCTP_STATE_SHUTDOWN_SENT:
324 case SCTP_STATE_SHUTDOWN_RECEIVED:
325 case SCTP_STATE_SHUTDOWN_ACK_SENT:
326 /* Cannot send after transport endpoint shutdown */
327 error = -ESHUTDOWN;
328 break;
330 default:
331 SCTP_DEBUG_PRINTK("outqueueing (%p, %p[%s])\n",
332 q, chunk, chunk && chunk->chunk_hdr ?
333 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
334 : "Illegal Chunk");
336 sctp_outq_tail_data(q, chunk);
337 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
338 SCTP_INC_STATS(net, SCTP_MIB_OUTUNORDERCHUNKS);
339 else
340 SCTP_INC_STATS(net, SCTP_MIB_OUTORDERCHUNKS);
341 q->empty = 0;
342 break;
344 } else {
345 list_add_tail(&chunk->list, &q->control_chunk_list);
346 SCTP_INC_STATS(net, SCTP_MIB_OUTCTRLCHUNKS);
349 if (error < 0)
350 return error;
352 if (!q->cork)
353 error = sctp_outq_flush(q, 0);
355 return error;
358 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list
359 * and the abandoned list are in ascending order.
361 static void sctp_insert_list(struct list_head *head, struct list_head *new)
363 struct list_head *pos;
364 struct sctp_chunk *nchunk, *lchunk;
365 __u32 ntsn, ltsn;
366 int done = 0;
368 nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
369 ntsn = ntohl(nchunk->subh.data_hdr->tsn);
371 list_for_each(pos, head) {
372 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
373 ltsn = ntohl(lchunk->subh.data_hdr->tsn);
374 if (TSN_lt(ntsn, ltsn)) {
375 list_add(new, pos->prev);
376 done = 1;
377 break;
380 if (!done)
381 list_add_tail(new, head);
384 /* Mark all the eligible packets on a transport for retransmission. */
385 void sctp_retransmit_mark(struct sctp_outq *q,
386 struct sctp_transport *transport,
387 __u8 reason)
389 struct list_head *lchunk, *ltemp;
390 struct sctp_chunk *chunk;
392 /* Walk through the specified transmitted queue. */
393 list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
394 chunk = list_entry(lchunk, struct sctp_chunk,
395 transmitted_list);
397 /* If the chunk is abandoned, move it to abandoned list. */
398 if (sctp_chunk_abandoned(chunk)) {
399 list_del_init(lchunk);
400 sctp_insert_list(&q->abandoned, lchunk);
402 /* If this chunk has not been previousely acked,
403 * stop considering it 'outstanding'. Our peer
404 * will most likely never see it since it will
405 * not be retransmitted
407 if (!chunk->tsn_gap_acked) {
408 if (chunk->transport)
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);
414 continue;
417 /* If we are doing retransmission due to a timeout or pmtu
418 * discovery, only the chunks that are not yet acked should
419 * be added to the retransmit queue.
421 if ((reason == SCTP_RTXR_FAST_RTX &&
422 (chunk->fast_retransmit == SCTP_NEED_FRTX)) ||
423 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) {
424 /* RFC 2960 6.2.1 Processing a Received SACK
426 * C) Any time a DATA chunk is marked for
427 * retransmission (via either T3-rtx timer expiration
428 * (Section 6.3.3) or via fast retransmit
429 * (Section 7.2.4)), add the data size of those
430 * chunks to the rwnd.
432 q->asoc->peer.rwnd += sctp_data_size(chunk);
433 q->outstanding_bytes -= sctp_data_size(chunk);
434 if (chunk->transport)
435 transport->flight_size -= sctp_data_size(chunk);
437 /* sctpimpguide-05 Section 2.8.2
438 * M5) If a T3-rtx timer expires, the
439 * 'TSN.Missing.Report' of all affected TSNs is set
440 * to 0.
442 chunk->tsn_missing_report = 0;
444 /* If a chunk that is being used for RTT measurement
445 * has to be retransmitted, we cannot use this chunk
446 * anymore for RTT measurements. Reset rto_pending so
447 * that a new RTT measurement is started when a new
448 * data chunk is sent.
450 if (chunk->rtt_in_progress) {
451 chunk->rtt_in_progress = 0;
452 transport->rto_pending = 0;
455 /* Move the chunk to the retransmit queue. The chunks
456 * on the retransmit queue are always kept in order.
458 list_del_init(lchunk);
459 sctp_insert_list(&q->retransmit, lchunk);
463 SCTP_DEBUG_PRINTK("%s: transport: %p, reason: %d, "
464 "cwnd: %d, ssthresh: %d, flight_size: %d, "
465 "pba: %d\n", __func__,
466 transport, reason,
467 transport->cwnd, transport->ssthresh,
468 transport->flight_size,
469 transport->partial_bytes_acked);
473 /* Mark all the eligible packets on a transport for retransmission and force
474 * one packet out.
476 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
477 sctp_retransmit_reason_t reason)
479 struct net *net = sock_net(q->asoc->base.sk);
480 int error = 0;
482 switch(reason) {
483 case SCTP_RTXR_T3_RTX:
484 SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS);
485 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
486 /* Update the retran path if the T3-rtx timer has expired for
487 * the current retran path.
489 if (transport == transport->asoc->peer.retran_path)
490 sctp_assoc_update_retran_path(transport->asoc);
491 transport->asoc->rtx_data_chunks +=
492 transport->asoc->unack_data;
493 break;
494 case SCTP_RTXR_FAST_RTX:
495 SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS);
496 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
497 q->fast_rtx = 1;
498 break;
499 case SCTP_RTXR_PMTUD:
500 SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS);
501 break;
502 case SCTP_RTXR_T1_RTX:
503 SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS);
504 transport->asoc->init_retries++;
505 break;
506 default:
507 BUG();
510 sctp_retransmit_mark(q, transport, reason);
512 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
513 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
514 * following the procedures outlined in C1 - C5.
516 if (reason == SCTP_RTXR_T3_RTX)
517 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
519 /* Flush the queues only on timeout, since fast_rtx is only
520 * triggered during sack processing and the queue
521 * will be flushed at the end.
523 if (reason != SCTP_RTXR_FAST_RTX)
524 error = sctp_outq_flush(q, /* rtx_timeout */ 1);
526 if (error)
527 q->asoc->base.sk->sk_err = -error;
531 * Transmit DATA chunks on the retransmit queue. Upon return from
532 * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
533 * need to be transmitted by the caller.
534 * We assume that pkt->transport has already been set.
536 * The return value is a normal kernel error return value.
538 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
539 int rtx_timeout, int *start_timer)
541 struct list_head *lqueue;
542 struct sctp_transport *transport = pkt->transport;
543 sctp_xmit_t status;
544 struct sctp_chunk *chunk, *chunk1;
545 int fast_rtx;
546 int error = 0;
547 int timer = 0;
548 int done = 0;
550 lqueue = &q->retransmit;
551 fast_rtx = q->fast_rtx;
553 /* This loop handles time-out retransmissions, fast retransmissions,
554 * and retransmissions due to opening of whindow.
556 * RFC 2960 6.3.3 Handle T3-rtx Expiration
558 * E3) Determine how many of the earliest (i.e., lowest TSN)
559 * outstanding DATA chunks for the address for which the
560 * T3-rtx has expired will fit into a single packet, subject
561 * to the MTU constraint for the path corresponding to the
562 * destination transport address to which the retransmission
563 * is being sent (this may be different from the address for
564 * which the timer expires [see Section 6.4]). Call this value
565 * K. Bundle and retransmit those K DATA chunks in a single
566 * packet to the destination endpoint.
568 * [Just to be painfully clear, if we are retransmitting
569 * because a timeout just happened, we should send only ONE
570 * packet of retransmitted data.]
572 * For fast retransmissions we also send only ONE packet. However,
573 * if we are just flushing the queue due to open window, we'll
574 * try to send as much as possible.
576 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
577 /* If the chunk is abandoned, move it to abandoned list. */
578 if (sctp_chunk_abandoned(chunk)) {
579 list_del_init(&chunk->transmitted_list);
580 sctp_insert_list(&q->abandoned,
581 &chunk->transmitted_list);
582 continue;
585 /* Make sure that Gap Acked TSNs are not retransmitted. A
586 * simple approach is just to move such TSNs out of the
587 * way and into a 'transmitted' queue and skip to the
588 * next chunk.
590 if (chunk->tsn_gap_acked) {
591 list_move_tail(&chunk->transmitted_list,
592 &transport->transmitted);
593 continue;
596 /* If we are doing fast retransmit, ignore non-fast_rtransmit
597 * chunks
599 if (fast_rtx && !chunk->fast_retransmit)
600 continue;
602 redo:
603 /* Attempt to append this chunk to the packet. */
604 status = sctp_packet_append_chunk(pkt, chunk);
606 switch (status) {
607 case SCTP_XMIT_PMTU_FULL:
608 if (!pkt->has_data && !pkt->has_cookie_echo) {
609 /* If this packet did not contain DATA then
610 * retransmission did not happen, so do it
611 * again. We'll ignore the error here since
612 * control chunks are already freed so there
613 * is nothing we can do.
615 sctp_packet_transmit(pkt);
616 goto redo;
619 /* Send this packet. */
620 error = sctp_packet_transmit(pkt);
622 /* If we are retransmitting, we should only
623 * send a single packet.
624 * Otherwise, try appending this chunk again.
626 if (rtx_timeout || fast_rtx)
627 done = 1;
628 else
629 goto redo;
631 /* Bundle next chunk in the next round. */
632 break;
634 case SCTP_XMIT_RWND_FULL:
635 /* Send this packet. */
636 error = sctp_packet_transmit(pkt);
638 /* Stop sending DATA as there is no more room
639 * at the receiver.
641 done = 1;
642 break;
644 case SCTP_XMIT_NAGLE_DELAY:
645 /* Send this packet. */
646 error = sctp_packet_transmit(pkt);
648 /* Stop sending DATA because of nagle delay. */
649 done = 1;
650 break;
652 default:
653 /* The append was successful, so add this chunk to
654 * the transmitted list.
656 list_move_tail(&chunk->transmitted_list,
657 &transport->transmitted);
659 /* Mark the chunk as ineligible for fast retransmit
660 * after it is retransmitted.
662 if (chunk->fast_retransmit == SCTP_NEED_FRTX)
663 chunk->fast_retransmit = SCTP_DONT_FRTX;
665 q->empty = 0;
666 q->asoc->stats.rtxchunks++;
667 break;
670 /* Set the timer if there were no errors */
671 if (!error && !timer)
672 timer = 1;
674 if (done)
675 break;
678 /* If we are here due to a retransmit timeout or a fast
679 * retransmit and if there are any chunks left in the retransmit
680 * queue that could not fit in the PMTU sized packet, they need
681 * to be marked as ineligible for a subsequent fast retransmit.
683 if (rtx_timeout || fast_rtx) {
684 list_for_each_entry(chunk1, lqueue, transmitted_list) {
685 if (chunk1->fast_retransmit == SCTP_NEED_FRTX)
686 chunk1->fast_retransmit = SCTP_DONT_FRTX;
690 *start_timer = timer;
692 /* Clear fast retransmit hint */
693 if (fast_rtx)
694 q->fast_rtx = 0;
696 return error;
699 /* Cork the outqueue so queued chunks are really queued. */
700 int sctp_outq_uncork(struct sctp_outq *q)
702 if (q->cork)
703 q->cork = 0;
705 return sctp_outq_flush(q, 0);
710 * Try to flush an outqueue.
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.
718 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
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;
726 struct sctp_transport *transport = NULL;
727 struct sctp_transport *new_transport;
728 struct sctp_chunk *chunk, *tmp;
729 sctp_xmit_t status;
730 int error = 0;
731 int start_timer = 0;
732 int one_packet = 0;
734 /* These transports have chunks to send. */
735 struct list_head transport_list;
736 struct list_head *ltransport;
738 INIT_LIST_HEAD(&transport_list);
739 packet = NULL;
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 * ...
751 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
752 /* RFC 5061, 5.3
753 * F1) This means that until such time as the ASCONF
754 * containing the add is acknowledged, the sender MUST
755 * NOT use the new IP address as a source for ANY SCTP
756 * packet except on carrying an ASCONF Chunk.
758 if (asoc->src_out_of_asoc_ok &&
759 chunk->chunk_hdr->type != SCTP_CID_ASCONF)
760 continue;
762 list_del_init(&chunk->list);
764 /* Pick the right transport to use. */
765 new_transport = chunk->transport;
767 if (!new_transport) {
769 * If we have a prior transport pointer, see if
770 * the destination address of the chunk
771 * matches the destination address of the
772 * current transport. If not a match, then
773 * try to look up the transport with a given
774 * destination address. We do this because
775 * after processing ASCONFs, we may have new
776 * transports created.
778 if (transport &&
779 sctp_cmp_addr_exact(&chunk->dest,
780 &transport->ipaddr))
781 new_transport = transport;
782 else
783 new_transport = sctp_assoc_lookup_paddr(asoc,
784 &chunk->dest);
786 /* if we still don't have a new transport, then
787 * use the current active path.
789 if (!new_transport)
790 new_transport = asoc->peer.active_path;
791 } else if ((new_transport->state == SCTP_INACTIVE) ||
792 (new_transport->state == SCTP_UNCONFIRMED) ||
793 (new_transport->state == SCTP_PF)) {
794 /* If the chunk is Heartbeat or Heartbeat Ack,
795 * send it to chunk->transport, even if it's
796 * inactive.
798 * 3.3.6 Heartbeat Acknowledgement:
799 * ...
800 * A HEARTBEAT ACK is always sent to the source IP
801 * address of the IP datagram containing the
802 * HEARTBEAT chunk to which this ack is responding.
803 * ...
805 * ASCONF_ACKs also must be sent to the source.
807 if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
808 chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK &&
809 chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK)
810 new_transport = asoc->peer.active_path;
813 /* Are we switching transports?
814 * Take care of transport locks.
816 if (new_transport != transport) {
817 transport = new_transport;
818 if (list_empty(&transport->send_ready)) {
819 list_add_tail(&transport->send_ready,
820 &transport_list);
822 packet = &transport->packet;
823 sctp_packet_config(packet, vtag,
824 asoc->peer.ecn_capable);
827 switch (chunk->chunk_hdr->type) {
829 * 6.10 Bundling
830 * ...
831 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
832 * COMPLETE with any other chunks. [Send them immediately.]
834 case SCTP_CID_INIT:
835 case SCTP_CID_INIT_ACK:
836 case SCTP_CID_SHUTDOWN_COMPLETE:
837 sctp_packet_init(&singleton, transport, sport, dport);
838 sctp_packet_config(&singleton, vtag, 0);
839 sctp_packet_append_chunk(&singleton, chunk);
840 error = sctp_packet_transmit(&singleton);
841 if (error < 0)
842 return error;
843 break;
845 case SCTP_CID_ABORT:
846 if (sctp_test_T_bit(chunk)) {
847 packet->vtag = asoc->c.my_vtag;
849 /* The following chunks are "response" chunks, i.e.
850 * they are generated in response to something we
851 * received. If we are sending these, then we can
852 * send only 1 packet containing these chunks.
854 case SCTP_CID_HEARTBEAT_ACK:
855 case SCTP_CID_SHUTDOWN_ACK:
856 case SCTP_CID_COOKIE_ACK:
857 case SCTP_CID_COOKIE_ECHO:
858 case SCTP_CID_ERROR:
859 case SCTP_CID_ECN_CWR:
860 case SCTP_CID_ASCONF_ACK:
861 one_packet = 1;
862 /* Fall through */
864 case SCTP_CID_SACK:
865 case SCTP_CID_HEARTBEAT:
866 case SCTP_CID_SHUTDOWN:
867 case SCTP_CID_ECN_ECNE:
868 case SCTP_CID_ASCONF:
869 case SCTP_CID_FWD_TSN:
870 status = sctp_packet_transmit_chunk(packet, chunk,
871 one_packet);
872 if (status != SCTP_XMIT_OK) {
873 /* put the chunk back */
874 list_add(&chunk->list, &q->control_chunk_list);
875 } else {
876 asoc->stats.octrlchunks++;
877 /* PR-SCTP C5) If a FORWARD TSN is sent, the
878 * sender MUST assure that at least one T3-rtx
879 * timer is running.
881 if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN)
882 sctp_transport_reset_timers(transport);
884 break;
886 default:
887 /* We built a chunk with an illegal type! */
888 BUG();
892 if (q->asoc->src_out_of_asoc_ok)
893 goto sctp_flush_out;
895 /* Is it OK to send data chunks? */
896 switch (asoc->state) {
897 case SCTP_STATE_COOKIE_ECHOED:
898 /* Only allow bundling when this packet has a COOKIE-ECHO
899 * chunk.
901 if (!packet || !packet->has_cookie_echo)
902 break;
904 /* fallthru */
905 case SCTP_STATE_ESTABLISHED:
906 case SCTP_STATE_SHUTDOWN_PENDING:
907 case SCTP_STATE_SHUTDOWN_RECEIVED:
909 * RFC 2960 6.1 Transmission of DATA Chunks
911 * C) When the time comes for the sender to transmit,
912 * before sending new DATA chunks, the sender MUST
913 * first transmit any outstanding DATA chunks which
914 * are marked for retransmission (limited by the
915 * current cwnd).
917 if (!list_empty(&q->retransmit)) {
918 if (asoc->peer.retran_path->state == SCTP_UNCONFIRMED)
919 goto sctp_flush_out;
920 if (transport == asoc->peer.retran_path)
921 goto retran;
923 /* Switch transports & prepare the packet. */
925 transport = asoc->peer.retran_path;
927 if (list_empty(&transport->send_ready)) {
928 list_add_tail(&transport->send_ready,
929 &transport_list);
932 packet = &transport->packet;
933 sctp_packet_config(packet, vtag,
934 asoc->peer.ecn_capable);
935 retran:
936 error = sctp_outq_flush_rtx(q, packet,
937 rtx_timeout, &start_timer);
939 if (start_timer)
940 sctp_transport_reset_timers(transport);
942 /* This can happen on COOKIE-ECHO resend. Only
943 * one chunk can get bundled with a COOKIE-ECHO.
945 if (packet->has_cookie_echo)
946 goto sctp_flush_out;
948 /* Don't send new data if there is still data
949 * waiting to retransmit.
951 if (!list_empty(&q->retransmit))
952 goto sctp_flush_out;
955 /* Apply Max.Burst limitation to the current transport in
956 * case it will be used for new data. We are going to
957 * rest it before we return, but we want to apply the limit
958 * to the currently queued data.
960 if (transport)
961 sctp_transport_burst_limited(transport);
963 /* Finally, transmit new packets. */
964 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
965 /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
966 * stream identifier.
968 if (chunk->sinfo.sinfo_stream >=
969 asoc->c.sinit_num_ostreams) {
971 /* Mark as failed send. */
972 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
973 sctp_chunk_free(chunk);
974 continue;
977 /* Has this chunk expired? */
978 if (sctp_chunk_abandoned(chunk)) {
979 sctp_chunk_fail(chunk, 0);
980 sctp_chunk_free(chunk);
981 continue;
984 /* If there is a specified transport, use it.
985 * Otherwise, we want to use the active path.
987 new_transport = chunk->transport;
988 if (!new_transport ||
989 ((new_transport->state == SCTP_INACTIVE) ||
990 (new_transport->state == SCTP_UNCONFIRMED) ||
991 (new_transport->state == SCTP_PF)))
992 new_transport = asoc->peer.active_path;
993 if (new_transport->state == SCTP_UNCONFIRMED)
994 continue;
996 /* Change packets if necessary. */
997 if (new_transport != transport) {
998 transport = new_transport;
1000 /* Schedule to have this transport's
1001 * packet flushed.
1003 if (list_empty(&transport->send_ready)) {
1004 list_add_tail(&transport->send_ready,
1005 &transport_list);
1008 packet = &transport->packet;
1009 sctp_packet_config(packet, vtag,
1010 asoc->peer.ecn_capable);
1011 /* We've switched transports, so apply the
1012 * Burst limit to the new transport.
1014 sctp_transport_burst_limited(transport);
1017 SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
1018 q, chunk,
1019 chunk && chunk->chunk_hdr ?
1020 sctp_cname(SCTP_ST_CHUNK(
1021 chunk->chunk_hdr->type))
1022 : "Illegal Chunk");
1024 SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
1025 "%p skb->users %d.\n",
1026 ntohl(chunk->subh.data_hdr->tsn),
1027 chunk->skb ?chunk->skb->head : NULL,
1028 chunk->skb ?
1029 atomic_read(&chunk->skb->users) : -1);
1031 /* Add the chunk to the packet. */
1032 status = sctp_packet_transmit_chunk(packet, chunk, 0);
1034 switch (status) {
1035 case SCTP_XMIT_PMTU_FULL:
1036 case SCTP_XMIT_RWND_FULL:
1037 case SCTP_XMIT_NAGLE_DELAY:
1038 /* We could not append this chunk, so put
1039 * the chunk back on the output queue.
1041 SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
1042 "not transmit TSN: 0x%x, status: %d\n",
1043 ntohl(chunk->subh.data_hdr->tsn),
1044 status);
1045 sctp_outq_head_data(q, chunk);
1046 goto sctp_flush_out;
1047 break;
1049 case SCTP_XMIT_OK:
1050 /* The sender is in the SHUTDOWN-PENDING state,
1051 * The sender MAY set the I-bit in the DATA
1052 * chunk header.
1054 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING)
1055 chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM;
1056 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
1057 asoc->stats.ouodchunks++;
1058 else
1059 asoc->stats.oodchunks++;
1061 break;
1063 default:
1064 BUG();
1067 /* BUG: We assume that the sctp_packet_transmit()
1068 * call below will succeed all the time and add the
1069 * chunk to the transmitted list and restart the
1070 * timers.
1071 * It is possible that the call can fail under OOM
1072 * conditions.
1074 * Is this really a problem? Won't this behave
1075 * like a lost TSN?
1077 list_add_tail(&chunk->transmitted_list,
1078 &transport->transmitted);
1080 sctp_transport_reset_timers(transport);
1082 q->empty = 0;
1084 /* Only let one DATA chunk get bundled with a
1085 * COOKIE-ECHO chunk.
1087 if (packet->has_cookie_echo)
1088 goto sctp_flush_out;
1090 break;
1092 default:
1093 /* Do nothing. */
1094 break;
1097 sctp_flush_out:
1099 /* Before returning, examine all the transports touched in
1100 * this call. Right now, we bluntly force clear all the
1101 * transports. Things might change after we implement Nagle.
1102 * But such an examination is still required.
1104 * --xguo
1106 while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
1107 struct sctp_transport *t = list_entry(ltransport,
1108 struct sctp_transport,
1109 send_ready);
1110 packet = &t->packet;
1111 if (!sctp_packet_empty(packet))
1112 error = sctp_packet_transmit(packet);
1114 /* Clear the burst limited state, if any */
1115 sctp_transport_burst_reset(t);
1118 return error;
1121 /* Update unack_data based on the incoming SACK chunk */
1122 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1123 struct sctp_sackhdr *sack)
1125 sctp_sack_variable_t *frags;
1126 __u16 unack_data;
1127 int i;
1129 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1131 frags = sack->variable;
1132 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1133 unack_data -= ((ntohs(frags[i].gab.end) -
1134 ntohs(frags[i].gab.start) + 1));
1137 assoc->unack_data = unack_data;
1140 /* This is where we REALLY process a SACK.
1142 * Process the SACK against the outqueue. Mostly, this just frees
1143 * things off the transmitted queue.
1145 int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk)
1147 struct sctp_association *asoc = q->asoc;
1148 struct sctp_sackhdr *sack = chunk->subh.sack_hdr;
1149 struct sctp_transport *transport;
1150 struct sctp_chunk *tchunk = NULL;
1151 struct list_head *lchunk, *transport_list, *temp;
1152 sctp_sack_variable_t *frags = sack->variable;
1153 __u32 sack_ctsn, ctsn, tsn;
1154 __u32 highest_tsn, highest_new_tsn;
1155 __u32 sack_a_rwnd;
1156 unsigned int outstanding;
1157 struct sctp_transport *primary = asoc->peer.primary_path;
1158 int count_of_newacks = 0;
1159 int gap_ack_blocks;
1160 u8 accum_moved = 0;
1162 /* Grab the association's destination address list. */
1163 transport_list = &asoc->peer.transport_addr_list;
1165 sack_ctsn = ntohl(sack->cum_tsn_ack);
1166 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1167 asoc->stats.gapcnt += gap_ack_blocks;
1169 * SFR-CACC algorithm:
1170 * On receipt of a SACK the sender SHOULD execute the
1171 * following statements.
1173 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1174 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1175 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1176 * all destinations.
1177 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1178 * is set the receiver of the SACK MUST take the following actions:
1180 * A) Initialize the cacc_saw_newack to 0 for all destination
1181 * addresses.
1183 * Only bother if changeover_active is set. Otherwise, this is
1184 * totally suboptimal to do on every SACK.
1186 if (primary->cacc.changeover_active) {
1187 u8 clear_cycling = 0;
1189 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1190 primary->cacc.changeover_active = 0;
1191 clear_cycling = 1;
1194 if (clear_cycling || gap_ack_blocks) {
1195 list_for_each_entry(transport, transport_list,
1196 transports) {
1197 if (clear_cycling)
1198 transport->cacc.cycling_changeover = 0;
1199 if (gap_ack_blocks)
1200 transport->cacc.cacc_saw_newack = 0;
1205 /* Get the highest TSN in the sack. */
1206 highest_tsn = sack_ctsn;
1207 if (gap_ack_blocks)
1208 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1210 if (TSN_lt(asoc->highest_sacked, highest_tsn))
1211 asoc->highest_sacked = highest_tsn;
1213 highest_new_tsn = sack_ctsn;
1215 /* Run through the retransmit queue. Credit bytes received
1216 * and free those chunks that we can.
1218 sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn);
1220 /* Run through the transmitted queue.
1221 * Credit bytes received and free those chunks which we can.
1223 * This is a MASSIVE candidate for optimization.
1225 list_for_each_entry(transport, transport_list, transports) {
1226 sctp_check_transmitted(q, &transport->transmitted,
1227 transport, &chunk->source, sack,
1228 &highest_new_tsn);
1230 * SFR-CACC algorithm:
1231 * C) Let count_of_newacks be the number of
1232 * destinations for which cacc_saw_newack is set.
1234 if (transport->cacc.cacc_saw_newack)
1235 count_of_newacks ++;
1238 /* Move the Cumulative TSN Ack Point if appropriate. */
1239 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) {
1240 asoc->ctsn_ack_point = sack_ctsn;
1241 accum_moved = 1;
1244 if (gap_ack_blocks) {
1246 if (asoc->fast_recovery && accum_moved)
1247 highest_new_tsn = highest_tsn;
1249 list_for_each_entry(transport, transport_list, transports)
1250 sctp_mark_missing(q, &transport->transmitted, transport,
1251 highest_new_tsn, count_of_newacks);
1254 /* Update unack_data field in the assoc. */
1255 sctp_sack_update_unack_data(asoc, sack);
1257 ctsn = asoc->ctsn_ack_point;
1259 /* Throw away stuff rotting on the sack queue. */
1260 list_for_each_safe(lchunk, temp, &q->sacked) {
1261 tchunk = list_entry(lchunk, struct sctp_chunk,
1262 transmitted_list);
1263 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1264 if (TSN_lte(tsn, ctsn)) {
1265 list_del_init(&tchunk->transmitted_list);
1266 sctp_chunk_free(tchunk);
1270 /* ii) Set rwnd equal to the newly received a_rwnd minus the
1271 * number of bytes still outstanding after processing the
1272 * Cumulative TSN Ack and the Gap Ack Blocks.
1275 sack_a_rwnd = ntohl(sack->a_rwnd);
1276 outstanding = q->outstanding_bytes;
1278 if (outstanding < sack_a_rwnd)
1279 sack_a_rwnd -= outstanding;
1280 else
1281 sack_a_rwnd = 0;
1283 asoc->peer.rwnd = sack_a_rwnd;
1285 sctp_generate_fwdtsn(q, sack_ctsn);
1287 SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n",
1288 __func__, sack_ctsn);
1289 SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, "
1290 "%p is 0x%x. Adv peer ack point: 0x%x\n",
1291 __func__, asoc, ctsn, asoc->adv_peer_ack_point);
1293 /* See if all chunks are acked.
1294 * Make sure the empty queue handler will get run later.
1296 q->empty = (list_empty(&q->out_chunk_list) &&
1297 list_empty(&q->retransmit));
1298 if (!q->empty)
1299 goto finish;
1301 list_for_each_entry(transport, transport_list, transports) {
1302 q->empty = q->empty && list_empty(&transport->transmitted);
1303 if (!q->empty)
1304 goto finish;
1307 SCTP_DEBUG_PRINTK("sack queue is empty.\n");
1308 finish:
1309 return q->empty;
1312 /* Is the outqueue empty? */
1313 int sctp_outq_is_empty(const struct sctp_outq *q)
1315 return q->empty;
1318 /********************************************************************
1319 * 2nd Level Abstractions
1320 ********************************************************************/
1322 /* Go through a transport's transmitted list or the association's retransmit
1323 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1324 * The retransmit list will not have an associated transport.
1326 * I added coherent debug information output. --xguo
1328 * Instead of printing 'sacked' or 'kept' for each TSN on the
1329 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1330 * KEPT TSN6-TSN7, etc.
1332 static void sctp_check_transmitted(struct sctp_outq *q,
1333 struct list_head *transmitted_queue,
1334 struct sctp_transport *transport,
1335 union sctp_addr *saddr,
1336 struct sctp_sackhdr *sack,
1337 __u32 *highest_new_tsn_in_sack)
1339 struct list_head *lchunk;
1340 struct sctp_chunk *tchunk;
1341 struct list_head tlist;
1342 __u32 tsn;
1343 __u32 sack_ctsn;
1344 __u32 rtt;
1345 __u8 restart_timer = 0;
1346 int bytes_acked = 0;
1347 int migrate_bytes = 0;
1349 /* These state variables are for coherent debug output. --xguo */
1351 #if SCTP_DEBUG
1352 __u32 dbg_ack_tsn = 0; /* An ACKed TSN range starts here... */
1353 __u32 dbg_last_ack_tsn = 0; /* ...and finishes here. */
1354 __u32 dbg_kept_tsn = 0; /* An un-ACKed range starts here... */
1355 __u32 dbg_last_kept_tsn = 0; /* ...and finishes here. */
1357 /* 0 : The last TSN was ACKed.
1358 * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1359 * -1: We need to initialize.
1361 int dbg_prt_state = -1;
1362 #endif /* SCTP_DEBUG */
1364 sack_ctsn = ntohl(sack->cum_tsn_ack);
1366 INIT_LIST_HEAD(&tlist);
1368 /* The while loop will skip empty transmitted queues. */
1369 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1370 tchunk = list_entry(lchunk, struct sctp_chunk,
1371 transmitted_list);
1373 if (sctp_chunk_abandoned(tchunk)) {
1374 /* Move the chunk to abandoned list. */
1375 sctp_insert_list(&q->abandoned, lchunk);
1377 /* If this chunk has not been acked, stop
1378 * considering it as 'outstanding'.
1380 if (!tchunk->tsn_gap_acked) {
1381 if (tchunk->transport)
1382 tchunk->transport->flight_size -=
1383 sctp_data_size(tchunk);
1384 q->outstanding_bytes -= sctp_data_size(tchunk);
1386 continue;
1389 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1390 if (sctp_acked(sack, tsn)) {
1391 /* If this queue is the retransmit queue, the
1392 * retransmit timer has already reclaimed
1393 * the outstanding bytes for this chunk, so only
1394 * count bytes associated with a transport.
1396 if (transport) {
1397 /* If this chunk is being used for RTT
1398 * measurement, calculate the RTT and update
1399 * the RTO using this value.
1401 * 6.3.1 C5) Karn's algorithm: RTT measurements
1402 * MUST NOT be made using packets that were
1403 * retransmitted (and thus for which it is
1404 * ambiguous whether the reply was for the
1405 * first instance of the packet or a later
1406 * instance).
1408 if (!tchunk->tsn_gap_acked &&
1409 tchunk->rtt_in_progress) {
1410 tchunk->rtt_in_progress = 0;
1411 rtt = jiffies - tchunk->sent_at;
1412 sctp_transport_update_rto(transport,
1413 rtt);
1417 /* If the chunk hasn't been marked as ACKED,
1418 * mark it and account bytes_acked if the
1419 * chunk had a valid transport (it will not
1420 * have a transport if ASCONF had deleted it
1421 * while DATA was outstanding).
1423 if (!tchunk->tsn_gap_acked) {
1424 tchunk->tsn_gap_acked = 1;
1425 *highest_new_tsn_in_sack = tsn;
1426 bytes_acked += sctp_data_size(tchunk);
1427 if (!tchunk->transport)
1428 migrate_bytes += sctp_data_size(tchunk);
1431 if (TSN_lte(tsn, sack_ctsn)) {
1432 /* RFC 2960 6.3.2 Retransmission Timer Rules
1434 * R3) Whenever a SACK is received
1435 * that acknowledges the DATA chunk
1436 * with the earliest outstanding TSN
1437 * for that address, restart T3-rtx
1438 * timer for that address with its
1439 * current RTO.
1441 restart_timer = 1;
1443 if (!tchunk->tsn_gap_acked) {
1445 * SFR-CACC algorithm:
1446 * 2) If the SACK contains gap acks
1447 * and the flag CHANGEOVER_ACTIVE is
1448 * set the receiver of the SACK MUST
1449 * take the following action:
1451 * B) For each TSN t being acked that
1452 * has not been acked in any SACK so
1453 * far, set cacc_saw_newack to 1 for
1454 * the destination that the TSN was
1455 * sent to.
1457 if (transport &&
1458 sack->num_gap_ack_blocks &&
1459 q->asoc->peer.primary_path->cacc.
1460 changeover_active)
1461 transport->cacc.cacc_saw_newack
1462 = 1;
1465 list_add_tail(&tchunk->transmitted_list,
1466 &q->sacked);
1467 } else {
1468 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1469 * M2) Each time a SACK arrives reporting
1470 * 'Stray DATA chunk(s)' record the highest TSN
1471 * reported as newly acknowledged, call this
1472 * value 'HighestTSNinSack'. A newly
1473 * acknowledged DATA chunk is one not
1474 * previously acknowledged in a SACK.
1476 * When the SCTP sender of data receives a SACK
1477 * chunk that acknowledges, for the first time,
1478 * the receipt of a DATA chunk, all the still
1479 * unacknowledged DATA chunks whose TSN is
1480 * older than that newly acknowledged DATA
1481 * chunk, are qualified as 'Stray DATA chunks'.
1483 list_add_tail(lchunk, &tlist);
1486 #if SCTP_DEBUG
1487 switch (dbg_prt_state) {
1488 case 0: /* last TSN was ACKed */
1489 if (dbg_last_ack_tsn + 1 == tsn) {
1490 /* This TSN belongs to the
1491 * current ACK range.
1493 break;
1496 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1497 /* Display the end of the
1498 * current range.
1500 SCTP_DEBUG_PRINTK_CONT("-%08x",
1501 dbg_last_ack_tsn);
1504 /* Start a new range. */
1505 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn);
1506 dbg_ack_tsn = tsn;
1507 break;
1509 case 1: /* The last TSN was NOT ACKed. */
1510 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1511 /* Display the end of current range. */
1512 SCTP_DEBUG_PRINTK_CONT("-%08x",
1513 dbg_last_kept_tsn);
1516 SCTP_DEBUG_PRINTK_CONT("\n");
1518 /* FALL THROUGH... */
1519 default:
1520 /* This is the first-ever TSN we examined. */
1521 /* Start a new range of ACK-ed TSNs. */
1522 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn);
1523 dbg_prt_state = 0;
1524 dbg_ack_tsn = tsn;
1527 dbg_last_ack_tsn = tsn;
1528 #endif /* SCTP_DEBUG */
1530 } else {
1531 if (tchunk->tsn_gap_acked) {
1532 SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1533 "data TSN: 0x%x\n",
1534 __func__,
1535 tsn);
1536 tchunk->tsn_gap_acked = 0;
1538 if (tchunk->transport)
1539 bytes_acked -= sctp_data_size(tchunk);
1541 /* RFC 2960 6.3.2 Retransmission Timer Rules
1543 * R4) Whenever a SACK is received missing a
1544 * TSN that was previously acknowledged via a
1545 * Gap Ack Block, start T3-rtx for the
1546 * destination address to which the DATA
1547 * chunk was originally
1548 * transmitted if it is not already running.
1550 restart_timer = 1;
1553 list_add_tail(lchunk, &tlist);
1555 #if SCTP_DEBUG
1556 /* See the above comments on ACK-ed TSNs. */
1557 switch (dbg_prt_state) {
1558 case 1:
1559 if (dbg_last_kept_tsn + 1 == tsn)
1560 break;
1562 if (dbg_last_kept_tsn != dbg_kept_tsn)
1563 SCTP_DEBUG_PRINTK_CONT("-%08x",
1564 dbg_last_kept_tsn);
1566 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn);
1567 dbg_kept_tsn = tsn;
1568 break;
1570 case 0:
1571 if (dbg_last_ack_tsn != dbg_ack_tsn)
1572 SCTP_DEBUG_PRINTK_CONT("-%08x",
1573 dbg_last_ack_tsn);
1574 SCTP_DEBUG_PRINTK_CONT("\n");
1576 /* FALL THROUGH... */
1577 default:
1578 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn);
1579 dbg_prt_state = 1;
1580 dbg_kept_tsn = tsn;
1583 dbg_last_kept_tsn = tsn;
1584 #endif /* SCTP_DEBUG */
1588 #if SCTP_DEBUG
1589 /* Finish off the last range, displaying its ending TSN. */
1590 switch (dbg_prt_state) {
1591 case 0:
1592 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1593 SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_ack_tsn);
1594 } else {
1595 SCTP_DEBUG_PRINTK_CONT("\n");
1597 break;
1599 case 1:
1600 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1601 SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_kept_tsn);
1602 } else {
1603 SCTP_DEBUG_PRINTK_CONT("\n");
1606 #endif /* SCTP_DEBUG */
1607 if (transport) {
1608 if (bytes_acked) {
1609 struct sctp_association *asoc = transport->asoc;
1611 /* We may have counted DATA that was migrated
1612 * to this transport due to DEL-IP operation.
1613 * Subtract those bytes, since the were never
1614 * send on this transport and shouldn't be
1615 * credited to this transport.
1617 bytes_acked -= migrate_bytes;
1619 /* 8.2. When an outstanding TSN is acknowledged,
1620 * the endpoint shall clear the error counter of
1621 * the destination transport address to which the
1622 * DATA chunk was last sent.
1623 * The association's overall error counter is
1624 * also cleared.
1626 transport->error_count = 0;
1627 transport->asoc->overall_error_count = 0;
1630 * While in SHUTDOWN PENDING, we may have started
1631 * the T5 shutdown guard timer after reaching the
1632 * retransmission limit. Stop that timer as soon
1633 * as the receiver acknowledged any data.
1635 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING &&
1636 del_timer(&asoc->timers
1637 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]))
1638 sctp_association_put(asoc);
1640 /* Mark the destination transport address as
1641 * active if it is not so marked.
1643 if ((transport->state == SCTP_INACTIVE ||
1644 transport->state == SCTP_UNCONFIRMED) &&
1645 sctp_cmp_addr_exact(&transport->ipaddr, saddr)) {
1646 sctp_assoc_control_transport(
1647 transport->asoc,
1648 transport,
1649 SCTP_TRANSPORT_UP,
1650 SCTP_RECEIVED_SACK);
1653 sctp_transport_raise_cwnd(transport, sack_ctsn,
1654 bytes_acked);
1656 transport->flight_size -= bytes_acked;
1657 if (transport->flight_size == 0)
1658 transport->partial_bytes_acked = 0;
1659 q->outstanding_bytes -= bytes_acked + migrate_bytes;
1660 } else {
1661 /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1662 * When a sender is doing zero window probing, it
1663 * should not timeout the association if it continues
1664 * to receive new packets from the receiver. The
1665 * reason is that the receiver MAY keep its window
1666 * closed for an indefinite time.
1667 * A sender is doing zero window probing when the
1668 * receiver's advertised window is zero, and there is
1669 * only one data chunk in flight to the receiver.
1671 * Allow the association to timeout while in SHUTDOWN
1672 * PENDING or SHUTDOWN RECEIVED in case the receiver
1673 * stays in zero window mode forever.
1675 if (!q->asoc->peer.rwnd &&
1676 !list_empty(&tlist) &&
1677 (sack_ctsn+2 == q->asoc->next_tsn) &&
1678 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) {
1679 SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1680 "window probe: %u\n",
1681 __func__, sack_ctsn);
1682 q->asoc->overall_error_count = 0;
1683 transport->error_count = 0;
1687 /* RFC 2960 6.3.2 Retransmission Timer Rules
1689 * R2) Whenever all outstanding data sent to an address have
1690 * been acknowledged, turn off the T3-rtx timer of that
1691 * address.
1693 if (!transport->flight_size) {
1694 if (del_timer(&transport->T3_rtx_timer))
1695 sctp_transport_put(transport);
1696 } else if (restart_timer) {
1697 if (!mod_timer(&transport->T3_rtx_timer,
1698 jiffies + transport->rto))
1699 sctp_transport_hold(transport);
1703 list_splice(&tlist, transmitted_queue);
1706 /* Mark chunks as missing and consequently may get retransmitted. */
1707 static void sctp_mark_missing(struct sctp_outq *q,
1708 struct list_head *transmitted_queue,
1709 struct sctp_transport *transport,
1710 __u32 highest_new_tsn_in_sack,
1711 int count_of_newacks)
1713 struct sctp_chunk *chunk;
1714 __u32 tsn;
1715 char do_fast_retransmit = 0;
1716 struct sctp_association *asoc = q->asoc;
1717 struct sctp_transport *primary = asoc->peer.primary_path;
1719 list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1721 tsn = ntohl(chunk->subh.data_hdr->tsn);
1723 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1724 * 'Unacknowledged TSN's', if the TSN number of an
1725 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1726 * value, increment the 'TSN.Missing.Report' count on that
1727 * chunk if it has NOT been fast retransmitted or marked for
1728 * fast retransmit already.
1730 if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
1731 !chunk->tsn_gap_acked &&
1732 TSN_lt(tsn, highest_new_tsn_in_sack)) {
1734 /* SFR-CACC may require us to skip marking
1735 * this chunk as missing.
1737 if (!transport || !sctp_cacc_skip(primary,
1738 chunk->transport,
1739 count_of_newacks, tsn)) {
1740 chunk->tsn_missing_report++;
1742 SCTP_DEBUG_PRINTK(
1743 "%s: TSN 0x%x missing counter: %d\n",
1744 __func__, tsn,
1745 chunk->tsn_missing_report);
1749 * M4) If any DATA chunk is found to have a
1750 * 'TSN.Missing.Report'
1751 * value larger than or equal to 3, mark that chunk for
1752 * retransmission and start the fast retransmit procedure.
1755 if (chunk->tsn_missing_report >= 3) {
1756 chunk->fast_retransmit = SCTP_NEED_FRTX;
1757 do_fast_retransmit = 1;
1761 if (transport) {
1762 if (do_fast_retransmit)
1763 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1765 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1766 "ssthresh: %d, flight_size: %d, pba: %d\n",
1767 __func__, transport, transport->cwnd,
1768 transport->ssthresh, transport->flight_size,
1769 transport->partial_bytes_acked);
1773 /* Is the given TSN acked by this packet? */
1774 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1776 int i;
1777 sctp_sack_variable_t *frags;
1778 __u16 gap;
1779 __u32 ctsn = ntohl(sack->cum_tsn_ack);
1781 if (TSN_lte(tsn, ctsn))
1782 goto pass;
1784 /* 3.3.4 Selective Acknowledgement (SACK) (3):
1786 * Gap Ack Blocks:
1787 * These fields contain the Gap Ack Blocks. They are repeated
1788 * for each Gap Ack Block up to the number of Gap Ack Blocks
1789 * defined in the Number of Gap Ack Blocks field. All DATA
1790 * chunks with TSNs greater than or equal to (Cumulative TSN
1791 * Ack + Gap Ack Block Start) and less than or equal to
1792 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1793 * Block are assumed to have been received correctly.
1796 frags = sack->variable;
1797 gap = tsn - ctsn;
1798 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1799 if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1800 TSN_lte(gap, ntohs(frags[i].gab.end)))
1801 goto pass;
1804 return 0;
1805 pass:
1806 return 1;
1809 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1810 int nskips, __be16 stream)
1812 int i;
1814 for (i = 0; i < nskips; i++) {
1815 if (skiplist[i].stream == stream)
1816 return i;
1818 return i;
1821 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1822 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1824 struct sctp_association *asoc = q->asoc;
1825 struct sctp_chunk *ftsn_chunk = NULL;
1826 struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1827 int nskips = 0;
1828 int skip_pos = 0;
1829 __u32 tsn;
1830 struct sctp_chunk *chunk;
1831 struct list_head *lchunk, *temp;
1833 if (!asoc->peer.prsctp_capable)
1834 return;
1836 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1837 * received SACK.
1839 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1840 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1842 if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1843 asoc->adv_peer_ack_point = ctsn;
1845 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1846 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1847 * the chunk next in the out-queue space is marked as "abandoned" as
1848 * shown in the following example:
1850 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1851 * and the Advanced.Peer.Ack.Point is updated to this value:
1853 * out-queue at the end of ==> out-queue after Adv.Ack.Point
1854 * normal SACK processing local advancement
1855 * ... ...
1856 * Adv.Ack.Pt-> 102 acked 102 acked
1857 * 103 abandoned 103 abandoned
1858 * 104 abandoned Adv.Ack.P-> 104 abandoned
1859 * 105 105
1860 * 106 acked 106 acked
1861 * ... ...
1863 * In this example, the data sender successfully advanced the
1864 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1866 list_for_each_safe(lchunk, temp, &q->abandoned) {
1867 chunk = list_entry(lchunk, struct sctp_chunk,
1868 transmitted_list);
1869 tsn = ntohl(chunk->subh.data_hdr->tsn);
1871 /* Remove any chunks in the abandoned queue that are acked by
1872 * the ctsn.
1874 if (TSN_lte(tsn, ctsn)) {
1875 list_del_init(lchunk);
1876 sctp_chunk_free(chunk);
1877 } else {
1878 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1879 asoc->adv_peer_ack_point = tsn;
1880 if (chunk->chunk_hdr->flags &
1881 SCTP_DATA_UNORDERED)
1882 continue;
1883 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1884 nskips,
1885 chunk->subh.data_hdr->stream);
1886 ftsn_skip_arr[skip_pos].stream =
1887 chunk->subh.data_hdr->stream;
1888 ftsn_skip_arr[skip_pos].ssn =
1889 chunk->subh.data_hdr->ssn;
1890 if (skip_pos == nskips)
1891 nskips++;
1892 if (nskips == 10)
1893 break;
1894 } else
1895 break;
1899 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1900 * is greater than the Cumulative TSN ACK carried in the received
1901 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1902 * chunk containing the latest value of the
1903 * "Advanced.Peer.Ack.Point".
1905 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1906 * list each stream and sequence number in the forwarded TSN. This
1907 * information will enable the receiver to easily find any
1908 * stranded TSN's waiting on stream reorder queues. Each stream
1909 * SHOULD only be reported once; this means that if multiple
1910 * abandoned messages occur in the same stream then only the
1911 * highest abandoned stream sequence number is reported. If the
1912 * total size of the FORWARD TSN does NOT fit in a single MTU then
1913 * the sender of the FORWARD TSN SHOULD lower the
1914 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1915 * single MTU.
1917 if (asoc->adv_peer_ack_point > ctsn)
1918 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1919 nskips, &ftsn_skip_arr[0]);
1921 if (ftsn_chunk) {
1922 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1923 SCTP_INC_STATS(sock_net(asoc->base.sk), SCTP_MIB_OUTCTRLCHUNKS);