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)
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, see
26 * <http://www.gnu.org/licenses/>.
28 * Please send any bug reports or fixes you make to the
30 * lksctp developers <linux-sctp@vger.kernel.org>
32 * Written or modified by:
33 * La Monte H.P. Yarroll <piggy@acm.org>
34 * Karl Knutson <karl@athena.chicago.il.us>
35 * Perry Melange <pmelange@null.cc.uic.edu>
36 * Xingang Guo <xingang.guo@intel.com>
37 * Hui Huang <hui.huang@nokia.com>
38 * Sridhar Samudrala <sri@us.ibm.com>
39 * Jon Grimm <jgrimm@us.ibm.com>
42 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
44 #include <linux/types.h>
45 #include <linux/list.h> /* For struct list_head */
46 #include <linux/socket.h>
48 #include <linux/slab.h>
49 #include <net/sock.h> /* For skb_set_owner_w */
51 #include <net/sctp/sctp.h>
52 #include <net/sctp/sm.h>
53 #include <net/sctp/stream_sched.h>
55 /* Declare internal functions here. */
56 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn);
57 static void sctp_check_transmitted(struct sctp_outq *q,
58 struct list_head *transmitted_queue,
59 struct sctp_transport *transport,
60 union sctp_addr *saddr,
61 struct sctp_sackhdr *sack,
62 __u32 *highest_new_tsn);
64 static void sctp_mark_missing(struct sctp_outq *q,
65 struct list_head *transmitted_queue,
66 struct sctp_transport *transport,
67 __u32 highest_new_tsn,
68 int count_of_newacks);
70 static void sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp);
72 /* Add data to the front of the queue. */
73 static inline void sctp_outq_head_data(struct sctp_outq *q,
74 struct sctp_chunk *ch)
76 struct sctp_stream_out_ext *oute;
79 list_add(&ch->list, &q->out_chunk_list);
80 q->out_qlen += ch->skb->len;
82 stream = sctp_chunk_stream_no(ch);
83 oute = q->asoc->stream.out[stream].ext;
84 list_add(&ch->stream_list, &oute->outq);
87 /* Take data from the front of the queue. */
88 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
90 return q->sched->dequeue(q);
93 /* Add data chunk to the end of the queue. */
94 static inline void sctp_outq_tail_data(struct sctp_outq *q,
95 struct sctp_chunk *ch)
97 struct sctp_stream_out_ext *oute;
100 list_add_tail(&ch->list, &q->out_chunk_list);
101 q->out_qlen += ch->skb->len;
103 stream = sctp_chunk_stream_no(ch);
104 oute = q->asoc->stream.out[stream].ext;
105 list_add_tail(&ch->stream_list, &oute->outq);
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.
114 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary,
115 struct sctp_transport *transport,
116 int count_of_newacks)
118 if (count_of_newacks >= 2 && transport != primary)
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.
130 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport,
131 int count_of_newacks)
133 if (count_of_newacks < 2 &&
134 (transport && !transport->cacc.cacc_saw_newack))
140 * SFR-CACC algorithm:
141 * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
142 * execute steps C, D, F.
144 * C has been implemented in sctp_outq_sack
146 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
147 struct sctp_transport *transport,
148 int count_of_newacks)
150 if (!primary->cacc.cycling_changeover) {
151 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
153 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
161 * SFR-CACC algorithm:
162 * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
163 * than next_tsn_at_change of the current primary, then
164 * the sender MUST NOT increment missing report count
167 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
169 if (primary->cacc.cycling_changeover &&
170 TSN_lt(tsn, primary->cacc.next_tsn_at_change))
176 * SFR-CACC algorithm:
177 * 3) If the missing report count for TSN t is to be
178 * incremented according to [RFC2960] and
179 * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
180 * then the sender MUST further execute steps 3.1 and
181 * 3.2 to determine if the missing report count for
182 * TSN t SHOULD NOT be incremented.
184 * 3.3) If 3.1 and 3.2 do not dictate that the missing
185 * report count for t should not be incremented, then
186 * the sender SHOULD increment missing report count for
187 * t (according to [RFC2960] and [SCTP_STEWART_2002]).
189 static inline int sctp_cacc_skip(struct sctp_transport *primary,
190 struct sctp_transport *transport,
191 int count_of_newacks,
194 if (primary->cacc.changeover_active &&
195 (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) ||
196 sctp_cacc_skip_3_2(primary, tsn)))
201 /* Initialize an existing sctp_outq. This does the boring stuff.
202 * You still need to define handlers if you really want to DO
203 * something with this structure...
205 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
207 memset(q, 0, sizeof(struct sctp_outq));
210 INIT_LIST_HEAD(&q->out_chunk_list);
211 INIT_LIST_HEAD(&q->control_chunk_list);
212 INIT_LIST_HEAD(&q->retransmit);
213 INIT_LIST_HEAD(&q->sacked);
214 INIT_LIST_HEAD(&q->abandoned);
215 sctp_sched_set_sched(asoc, SCTP_SS_FCFS);
218 /* Free the outqueue structure and any related pending chunks.
220 static void __sctp_outq_teardown(struct sctp_outq *q)
222 struct sctp_transport *transport;
223 struct list_head *lchunk, *temp;
224 struct sctp_chunk *chunk, *tmp;
226 /* Throw away unacknowledged chunks. */
227 list_for_each_entry(transport, &q->asoc->peer.transport_addr_list,
229 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
230 chunk = list_entry(lchunk, struct sctp_chunk,
232 /* Mark as part of a failed message. */
233 sctp_chunk_fail(chunk, q->error);
234 sctp_chunk_free(chunk);
238 /* Throw away chunks that have been gap ACKed. */
239 list_for_each_safe(lchunk, temp, &q->sacked) {
240 list_del_init(lchunk);
241 chunk = list_entry(lchunk, struct sctp_chunk,
243 sctp_chunk_fail(chunk, q->error);
244 sctp_chunk_free(chunk);
247 /* Throw away any chunks in the retransmit queue. */
248 list_for_each_safe(lchunk, temp, &q->retransmit) {
249 list_del_init(lchunk);
250 chunk = list_entry(lchunk, struct sctp_chunk,
252 sctp_chunk_fail(chunk, q->error);
253 sctp_chunk_free(chunk);
256 /* Throw away any chunks that are in the abandoned queue. */
257 list_for_each_safe(lchunk, temp, &q->abandoned) {
258 list_del_init(lchunk);
259 chunk = list_entry(lchunk, struct sctp_chunk,
261 sctp_chunk_fail(chunk, q->error);
262 sctp_chunk_free(chunk);
265 /* Throw away any leftover data chunks. */
266 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
267 sctp_sched_dequeue_done(q, chunk);
269 /* Mark as send failure. */
270 sctp_chunk_fail(chunk, q->error);
271 sctp_chunk_free(chunk);
274 /* Throw away any leftover control chunks. */
275 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
276 list_del_init(&chunk->list);
277 sctp_chunk_free(chunk);
281 void sctp_outq_teardown(struct sctp_outq *q)
283 __sctp_outq_teardown(q);
284 sctp_outq_init(q->asoc, q);
287 /* Free the outqueue structure and any related pending chunks. */
288 void sctp_outq_free(struct sctp_outq *q)
290 /* Throw away leftover chunks. */
291 __sctp_outq_teardown(q);
294 /* Put a new chunk in an sctp_outq. */
295 void sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk, gfp_t gfp)
297 struct net *net = sock_net(q->asoc->base.sk);
299 pr_debug("%s: outq:%p, chunk:%p[%s]\n", __func__, q, chunk,
300 chunk && chunk->chunk_hdr ?
301 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
304 /* If it is data, queue it up, otherwise, send it
307 if (sctp_chunk_is_data(chunk)) {
308 pr_debug("%s: outqueueing: outq:%p, chunk:%p[%s])\n",
309 __func__, q, chunk, chunk && chunk->chunk_hdr ?
310 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
313 sctp_outq_tail_data(q, chunk);
314 if (chunk->asoc->peer.prsctp_capable &&
315 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags))
316 chunk->asoc->sent_cnt_removable++;
317 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
318 SCTP_INC_STATS(net, SCTP_MIB_OUTUNORDERCHUNKS);
320 SCTP_INC_STATS(net, SCTP_MIB_OUTORDERCHUNKS);
322 list_add_tail(&chunk->list, &q->control_chunk_list);
323 SCTP_INC_STATS(net, SCTP_MIB_OUTCTRLCHUNKS);
327 sctp_outq_flush(q, 0, gfp);
330 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list
331 * and the abandoned list are in ascending order.
333 static void sctp_insert_list(struct list_head *head, struct list_head *new)
335 struct list_head *pos;
336 struct sctp_chunk *nchunk, *lchunk;
340 nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
341 ntsn = ntohl(nchunk->subh.data_hdr->tsn);
343 list_for_each(pos, head) {
344 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
345 ltsn = ntohl(lchunk->subh.data_hdr->tsn);
346 if (TSN_lt(ntsn, ltsn)) {
347 list_add(new, pos->prev);
353 list_add_tail(new, head);
356 static int sctp_prsctp_prune_sent(struct sctp_association *asoc,
357 struct sctp_sndrcvinfo *sinfo,
358 struct list_head *queue, int msg_len)
360 struct sctp_chunk *chk, *temp;
362 list_for_each_entry_safe(chk, temp, queue, transmitted_list) {
363 struct sctp_stream_out *streamout;
365 if (!chk->msg->abandoned &&
366 (!SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) ||
367 chk->sinfo.sinfo_timetolive <= sinfo->sinfo_timetolive))
370 chk->msg->abandoned = 1;
371 list_del_init(&chk->transmitted_list);
372 sctp_insert_list(&asoc->outqueue.abandoned,
373 &chk->transmitted_list);
375 streamout = &asoc->stream.out[chk->sinfo.sinfo_stream];
376 asoc->sent_cnt_removable--;
377 asoc->abandoned_sent[SCTP_PR_INDEX(PRIO)]++;
378 streamout->ext->abandoned_sent[SCTP_PR_INDEX(PRIO)]++;
380 if (queue != &asoc->outqueue.retransmit &&
381 !chk->tsn_gap_acked) {
383 chk->transport->flight_size -=
385 asoc->outqueue.outstanding_bytes -= sctp_data_size(chk);
388 msg_len -= SCTP_DATA_SNDSIZE(chk) +
389 sizeof(struct sk_buff) +
390 sizeof(struct sctp_chunk);
398 static int sctp_prsctp_prune_unsent(struct sctp_association *asoc,
399 struct sctp_sndrcvinfo *sinfo, int msg_len)
401 struct sctp_outq *q = &asoc->outqueue;
402 struct sctp_chunk *chk, *temp;
404 q->sched->unsched_all(&asoc->stream);
406 list_for_each_entry_safe(chk, temp, &q->out_chunk_list, list) {
407 if (!chk->msg->abandoned &&
408 (!(chk->chunk_hdr->flags & SCTP_DATA_FIRST_FRAG) ||
409 !SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) ||
410 chk->sinfo.sinfo_timetolive <= sinfo->sinfo_timetolive))
413 chk->msg->abandoned = 1;
414 sctp_sched_dequeue_common(q, chk);
415 asoc->sent_cnt_removable--;
416 asoc->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++;
417 if (chk->sinfo.sinfo_stream < asoc->stream.outcnt) {
418 struct sctp_stream_out *streamout =
419 &asoc->stream.out[chk->sinfo.sinfo_stream];
421 streamout->ext->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++;
424 msg_len -= SCTP_DATA_SNDSIZE(chk) +
425 sizeof(struct sk_buff) +
426 sizeof(struct sctp_chunk);
427 sctp_chunk_free(chk);
432 q->sched->sched_all(&asoc->stream);
437 /* Abandon the chunks according their priorities */
438 void sctp_prsctp_prune(struct sctp_association *asoc,
439 struct sctp_sndrcvinfo *sinfo, int msg_len)
441 struct sctp_transport *transport;
443 if (!asoc->peer.prsctp_capable || !asoc->sent_cnt_removable)
446 msg_len = sctp_prsctp_prune_sent(asoc, sinfo,
447 &asoc->outqueue.retransmit,
452 list_for_each_entry(transport, &asoc->peer.transport_addr_list,
454 msg_len = sctp_prsctp_prune_sent(asoc, sinfo,
455 &transport->transmitted,
461 sctp_prsctp_prune_unsent(asoc, sinfo, msg_len);
464 /* Mark all the eligible packets on a transport for retransmission. */
465 void sctp_retransmit_mark(struct sctp_outq *q,
466 struct sctp_transport *transport,
469 struct list_head *lchunk, *ltemp;
470 struct sctp_chunk *chunk;
472 /* Walk through the specified transmitted queue. */
473 list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
474 chunk = list_entry(lchunk, struct sctp_chunk,
477 /* If the chunk is abandoned, move it to abandoned list. */
478 if (sctp_chunk_abandoned(chunk)) {
479 list_del_init(lchunk);
480 sctp_insert_list(&q->abandoned, lchunk);
482 /* If this chunk has not been previousely acked,
483 * stop considering it 'outstanding'. Our peer
484 * will most likely never see it since it will
485 * not be retransmitted
487 if (!chunk->tsn_gap_acked) {
488 if (chunk->transport)
489 chunk->transport->flight_size -=
490 sctp_data_size(chunk);
491 q->outstanding_bytes -= sctp_data_size(chunk);
492 q->asoc->peer.rwnd += sctp_data_size(chunk);
497 /* If we are doing retransmission due to a timeout or pmtu
498 * discovery, only the chunks that are not yet acked should
499 * be added to the retransmit queue.
501 if ((reason == SCTP_RTXR_FAST_RTX &&
502 (chunk->fast_retransmit == SCTP_NEED_FRTX)) ||
503 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) {
504 /* RFC 2960 6.2.1 Processing a Received SACK
506 * C) Any time a DATA chunk is marked for
507 * retransmission (via either T3-rtx timer expiration
508 * (Section 6.3.3) or via fast retransmit
509 * (Section 7.2.4)), add the data size of those
510 * chunks to the rwnd.
512 q->asoc->peer.rwnd += sctp_data_size(chunk);
513 q->outstanding_bytes -= sctp_data_size(chunk);
514 if (chunk->transport)
515 transport->flight_size -= sctp_data_size(chunk);
517 /* sctpimpguide-05 Section 2.8.2
518 * M5) If a T3-rtx timer expires, the
519 * 'TSN.Missing.Report' of all affected TSNs is set
522 chunk->tsn_missing_report = 0;
524 /* If a chunk that is being used for RTT measurement
525 * has to be retransmitted, we cannot use this chunk
526 * anymore for RTT measurements. Reset rto_pending so
527 * that a new RTT measurement is started when a new
528 * data chunk is sent.
530 if (chunk->rtt_in_progress) {
531 chunk->rtt_in_progress = 0;
532 transport->rto_pending = 0;
535 /* Move the chunk to the retransmit queue. The chunks
536 * on the retransmit queue are always kept in order.
538 list_del_init(lchunk);
539 sctp_insert_list(&q->retransmit, lchunk);
543 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, "
544 "flight_size:%d, pba:%d\n", __func__, transport, reason,
545 transport->cwnd, transport->ssthresh, transport->flight_size,
546 transport->partial_bytes_acked);
549 /* Mark all the eligible packets on a transport for retransmission and force
552 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
553 enum sctp_retransmit_reason reason)
555 struct net *net = sock_net(q->asoc->base.sk);
558 case SCTP_RTXR_T3_RTX:
559 SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS);
560 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
561 /* Update the retran path if the T3-rtx timer has expired for
562 * the current retran path.
564 if (transport == transport->asoc->peer.retran_path)
565 sctp_assoc_update_retran_path(transport->asoc);
566 transport->asoc->rtx_data_chunks +=
567 transport->asoc->unack_data;
569 case SCTP_RTXR_FAST_RTX:
570 SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS);
571 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
574 case SCTP_RTXR_PMTUD:
575 SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS);
577 case SCTP_RTXR_T1_RTX:
578 SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS);
579 transport->asoc->init_retries++;
585 sctp_retransmit_mark(q, transport, reason);
587 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
588 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
589 * following the procedures outlined in C1 - C5.
591 if (reason == SCTP_RTXR_T3_RTX)
592 q->asoc->stream.si->generate_ftsn(q, q->asoc->ctsn_ack_point);
594 /* Flush the queues only on timeout, since fast_rtx is only
595 * triggered during sack processing and the queue
596 * will be flushed at the end.
598 if (reason != SCTP_RTXR_FAST_RTX)
599 sctp_outq_flush(q, /* rtx_timeout */ 1, GFP_ATOMIC);
603 * Transmit DATA chunks on the retransmit queue. Upon return from
604 * __sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
605 * need to be transmitted by the caller.
606 * We assume that pkt->transport has already been set.
608 * The return value is a normal kernel error return value.
610 static int __sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
611 int rtx_timeout, int *start_timer)
613 struct sctp_transport *transport = pkt->transport;
614 struct sctp_chunk *chunk, *chunk1;
615 struct list_head *lqueue;
616 enum sctp_xmit status;
622 lqueue = &q->retransmit;
623 fast_rtx = q->fast_rtx;
625 /* This loop handles time-out retransmissions, fast retransmissions,
626 * and retransmissions due to opening of whindow.
628 * RFC 2960 6.3.3 Handle T3-rtx Expiration
630 * E3) Determine how many of the earliest (i.e., lowest TSN)
631 * outstanding DATA chunks for the address for which the
632 * T3-rtx has expired will fit into a single packet, subject
633 * to the MTU constraint for the path corresponding to the
634 * destination transport address to which the retransmission
635 * is being sent (this may be different from the address for
636 * which the timer expires [see Section 6.4]). Call this value
637 * K. Bundle and retransmit those K DATA chunks in a single
638 * packet to the destination endpoint.
640 * [Just to be painfully clear, if we are retransmitting
641 * because a timeout just happened, we should send only ONE
642 * packet of retransmitted data.]
644 * For fast retransmissions we also send only ONE packet. However,
645 * if we are just flushing the queue due to open window, we'll
646 * try to send as much as possible.
648 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
649 /* If the chunk is abandoned, move it to abandoned list. */
650 if (sctp_chunk_abandoned(chunk)) {
651 list_del_init(&chunk->transmitted_list);
652 sctp_insert_list(&q->abandoned,
653 &chunk->transmitted_list);
657 /* Make sure that Gap Acked TSNs are not retransmitted. A
658 * simple approach is just to move such TSNs out of the
659 * way and into a 'transmitted' queue and skip to the
662 if (chunk->tsn_gap_acked) {
663 list_move_tail(&chunk->transmitted_list,
664 &transport->transmitted);
668 /* If we are doing fast retransmit, ignore non-fast_rtransmit
671 if (fast_rtx && !chunk->fast_retransmit)
675 /* Attempt to append this chunk to the packet. */
676 status = sctp_packet_append_chunk(pkt, chunk);
679 case SCTP_XMIT_PMTU_FULL:
680 if (!pkt->has_data && !pkt->has_cookie_echo) {
681 /* If this packet did not contain DATA then
682 * retransmission did not happen, so do it
683 * again. We'll ignore the error here since
684 * control chunks are already freed so there
685 * is nothing we can do.
687 sctp_packet_transmit(pkt, GFP_ATOMIC);
691 /* Send this packet. */
692 error = sctp_packet_transmit(pkt, GFP_ATOMIC);
694 /* If we are retransmitting, we should only
695 * send a single packet.
696 * Otherwise, try appending this chunk again.
698 if (rtx_timeout || fast_rtx)
703 /* Bundle next chunk in the next round. */
706 case SCTP_XMIT_RWND_FULL:
707 /* Send this packet. */
708 error = sctp_packet_transmit(pkt, GFP_ATOMIC);
710 /* Stop sending DATA as there is no more room
716 case SCTP_XMIT_DELAY:
717 /* Send this packet. */
718 error = sctp_packet_transmit(pkt, GFP_ATOMIC);
720 /* Stop sending DATA because of nagle delay. */
725 /* The append was successful, so add this chunk to
726 * the transmitted list.
728 list_move_tail(&chunk->transmitted_list,
729 &transport->transmitted);
731 /* Mark the chunk as ineligible for fast retransmit
732 * after it is retransmitted.
734 if (chunk->fast_retransmit == SCTP_NEED_FRTX)
735 chunk->fast_retransmit = SCTP_DONT_FRTX;
737 q->asoc->stats.rtxchunks++;
741 /* Set the timer if there were no errors */
742 if (!error && !timer)
749 /* If we are here due to a retransmit timeout or a fast
750 * retransmit and if there are any chunks left in the retransmit
751 * queue that could not fit in the PMTU sized packet, they need
752 * to be marked as ineligible for a subsequent fast retransmit.
754 if (rtx_timeout || fast_rtx) {
755 list_for_each_entry(chunk1, lqueue, transmitted_list) {
756 if (chunk1->fast_retransmit == SCTP_NEED_FRTX)
757 chunk1->fast_retransmit = SCTP_DONT_FRTX;
761 *start_timer = timer;
763 /* Clear fast retransmit hint */
770 /* Cork the outqueue so queued chunks are really queued. */
771 void sctp_outq_uncork(struct sctp_outq *q, gfp_t gfp)
776 sctp_outq_flush(q, 0, gfp);
779 static int sctp_packet_singleton(struct sctp_transport *transport,
780 struct sctp_chunk *chunk, gfp_t gfp)
782 const struct sctp_association *asoc = transport->asoc;
783 const __u16 sport = asoc->base.bind_addr.port;
784 const __u16 dport = asoc->peer.port;
785 const __u32 vtag = asoc->peer.i.init_tag;
786 struct sctp_packet singleton;
788 sctp_packet_init(&singleton, transport, sport, dport);
789 sctp_packet_config(&singleton, vtag, 0);
790 sctp_packet_append_chunk(&singleton, chunk);
791 return sctp_packet_transmit(&singleton, gfp);
794 static bool sctp_outq_select_transport(struct sctp_chunk *chunk,
795 struct sctp_association *asoc,
796 struct sctp_transport **transport,
797 struct list_head *transport_list)
799 struct sctp_transport *new_transport = chunk->transport;
800 struct sctp_transport *curr = *transport;
801 bool changed = false;
803 if (!new_transport) {
804 if (!sctp_chunk_is_data(chunk)) {
806 * If we have a prior transport pointer, see if
807 * the destination address of the chunk
808 * matches the destination address of the
809 * current transport. If not a match, then
810 * try to look up the transport with a given
811 * destination address. We do this because
812 * after processing ASCONFs, we may have new
813 * transports created.
815 if (curr && sctp_cmp_addr_exact(&chunk->dest,
817 new_transport = curr;
819 new_transport = sctp_assoc_lookup_paddr(asoc,
823 /* if we still don't have a new transport, then
824 * use the current active path.
827 new_transport = asoc->peer.active_path;
831 switch (new_transport->state) {
833 case SCTP_UNCONFIRMED:
835 /* If the chunk is Heartbeat or Heartbeat Ack,
836 * send it to chunk->transport, even if it's
839 * 3.3.6 Heartbeat Acknowledgement:
841 * A HEARTBEAT ACK is always sent to the source IP
842 * address of the IP datagram containing the
843 * HEARTBEAT chunk to which this ack is responding.
846 * ASCONF_ACKs also must be sent to the source.
848 type = chunk->chunk_hdr->type;
849 if (type != SCTP_CID_HEARTBEAT &&
850 type != SCTP_CID_HEARTBEAT_ACK &&
851 type != SCTP_CID_ASCONF_ACK)
852 new_transport = asoc->peer.active_path;
859 /* Are we switching transports? Take care of transport locks. */
860 if (new_transport != curr) {
862 curr = new_transport;
864 if (list_empty(&curr->send_ready))
865 list_add_tail(&curr->send_ready, transport_list);
867 sctp_packet_config(&curr->packet, asoc->peer.i.init_tag,
868 asoc->peer.ecn_capable);
869 /* We've switched transports, so apply the
870 * Burst limit to the new transport.
872 sctp_transport_burst_limited(curr);
878 static void sctp_outq_flush_ctrl(struct sctp_outq *q,
879 struct sctp_transport **_transport,
880 struct list_head *transport_list,
883 struct sctp_transport *transport = *_transport;
884 struct sctp_association *asoc = q->asoc;
885 struct sctp_packet *packet = NULL;
886 struct sctp_chunk *chunk, *tmp;
887 enum sctp_xmit status;
888 int one_packet, error;
890 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
894 * F1) This means that until such time as the ASCONF
895 * containing the add is acknowledged, the sender MUST
896 * NOT use the new IP address as a source for ANY SCTP
897 * packet except on carrying an ASCONF Chunk.
899 if (asoc->src_out_of_asoc_ok &&
900 chunk->chunk_hdr->type != SCTP_CID_ASCONF)
903 list_del_init(&chunk->list);
905 /* Pick the right transport to use. Should always be true for
906 * the first chunk as we don't have a transport by then.
908 if (sctp_outq_select_transport(chunk, asoc, _transport,
910 transport = *_transport;
911 packet = &transport->packet;
914 switch (chunk->chunk_hdr->type) {
918 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
919 * COMPLETE with any other chunks. [Send them immediately.]
922 case SCTP_CID_INIT_ACK:
923 case SCTP_CID_SHUTDOWN_COMPLETE:
924 error = sctp_packet_singleton(transport, chunk, gfp);
926 asoc->base.sk->sk_err = -error;
932 if (sctp_test_T_bit(chunk))
933 packet->vtag = asoc->c.my_vtag;
936 /* The following chunks are "response" chunks, i.e.
937 * they are generated in response to something we
938 * received. If we are sending these, then we can
939 * send only 1 packet containing these chunks.
941 case SCTP_CID_HEARTBEAT_ACK:
942 case SCTP_CID_SHUTDOWN_ACK:
943 case SCTP_CID_COOKIE_ACK:
944 case SCTP_CID_COOKIE_ECHO:
946 case SCTP_CID_ECN_CWR:
947 case SCTP_CID_ASCONF_ACK:
952 case SCTP_CID_HEARTBEAT:
953 case SCTP_CID_SHUTDOWN:
954 case SCTP_CID_ECN_ECNE:
955 case SCTP_CID_ASCONF:
956 case SCTP_CID_FWD_TSN:
957 case SCTP_CID_I_FWD_TSN:
958 case SCTP_CID_RECONF:
959 status = sctp_packet_transmit_chunk(packet, chunk,
961 if (status != SCTP_XMIT_OK) {
962 /* put the chunk back */
963 list_add(&chunk->list, &q->control_chunk_list);
967 asoc->stats.octrlchunks++;
968 /* PR-SCTP C5) If a FORWARD TSN is sent, the
969 * sender MUST assure that at least one T3-rtx
972 if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN ||
973 chunk->chunk_hdr->type == SCTP_CID_I_FWD_TSN) {
974 sctp_transport_reset_t3_rtx(transport);
975 transport->last_time_sent = jiffies;
978 if (chunk == asoc->strreset_chunk)
979 sctp_transport_reset_reconf_timer(transport);
984 /* We built a chunk with an illegal type! */
990 /* Returns false if new data shouldn't be sent */
991 static bool sctp_outq_flush_rtx(struct sctp_outq *q,
992 struct sctp_transport **_transport,
993 struct list_head *transport_list,
996 struct sctp_transport *transport = *_transport;
997 struct sctp_packet *packet = transport ? &transport->packet : NULL;
998 struct sctp_association *asoc = q->asoc;
999 int error, start_timer = 0;
1001 if (asoc->peer.retran_path->state == SCTP_UNCONFIRMED)
1004 if (transport != asoc->peer.retran_path) {
1005 /* Switch transports & prepare the packet. */
1006 transport = asoc->peer.retran_path;
1007 *_transport = transport;
1009 if (list_empty(&transport->send_ready))
1010 list_add_tail(&transport->send_ready,
1013 packet = &transport->packet;
1014 sctp_packet_config(packet, asoc->peer.i.init_tag,
1015 asoc->peer.ecn_capable);
1018 error = __sctp_outq_flush_rtx(q, packet, rtx_timeout, &start_timer);
1020 asoc->base.sk->sk_err = -error;
1023 sctp_transport_reset_t3_rtx(transport);
1024 transport->last_time_sent = jiffies;
1027 /* This can happen on COOKIE-ECHO resend. Only
1028 * one chunk can get bundled with a COOKIE-ECHO.
1030 if (packet->has_cookie_echo)
1033 /* Don't send new data if there is still data
1034 * waiting to retransmit.
1036 if (!list_empty(&q->retransmit))
1042 static void sctp_outq_flush_data(struct sctp_outq *q,
1043 struct sctp_transport **_transport,
1044 struct list_head *transport_list,
1045 int rtx_timeout, gfp_t gfp)
1047 struct sctp_transport *transport = *_transport;
1048 struct sctp_packet *packet = transport ? &transport->packet : NULL;
1049 struct sctp_association *asoc = q->asoc;
1050 struct sctp_chunk *chunk;
1051 enum sctp_xmit status;
1053 /* Is it OK to send data chunks? */
1054 switch (asoc->state) {
1055 case SCTP_STATE_COOKIE_ECHOED:
1056 /* Only allow bundling when this packet has a COOKIE-ECHO
1059 if (!packet || !packet->has_cookie_echo)
1063 case SCTP_STATE_ESTABLISHED:
1064 case SCTP_STATE_SHUTDOWN_PENDING:
1065 case SCTP_STATE_SHUTDOWN_RECEIVED:
1067 * RFC 2960 6.1 Transmission of DATA Chunks
1069 * C) When the time comes for the sender to transmit,
1070 * before sending new DATA chunks, the sender MUST
1071 * first transmit any outstanding DATA chunks which
1072 * are marked for retransmission (limited by the
1075 if (!list_empty(&q->retransmit)) {
1076 if (!sctp_outq_flush_rtx(q, _transport, transport_list,
1079 /* We may have switched current transport */
1080 transport = *_transport;
1081 packet = &transport->packet;
1084 /* Apply Max.Burst limitation to the current transport in
1085 * case it will be used for new data. We are going to
1086 * rest it before we return, but we want to apply the limit
1087 * to the currently queued data.
1090 sctp_transport_burst_limited(transport);
1092 /* Finally, transmit new packets. */
1093 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
1094 __u32 sid = ntohs(chunk->subh.data_hdr->stream);
1096 /* Has this chunk expired? */
1097 if (sctp_chunk_abandoned(chunk)) {
1098 sctp_sched_dequeue_done(q, chunk);
1099 sctp_chunk_fail(chunk, 0);
1100 sctp_chunk_free(chunk);
1104 if (asoc->stream.out[sid].state == SCTP_STREAM_CLOSED) {
1105 sctp_outq_head_data(q, chunk);
1109 if (sctp_outq_select_transport(chunk, asoc, _transport,
1111 transport = *_transport;
1112 packet = &transport->packet;
1115 pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p "
1117 __func__, q, chunk, chunk && chunk->chunk_hdr ?
1118 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
1119 "illegal chunk", ntohl(chunk->subh.data_hdr->tsn),
1120 chunk->skb ? chunk->skb->head : NULL, chunk->skb ?
1121 refcount_read(&chunk->skb->users) : -1);
1123 /* Add the chunk to the packet. */
1124 status = sctp_packet_transmit_chunk(packet, chunk, 0, gfp);
1129 case SCTP_XMIT_PMTU_FULL:
1130 case SCTP_XMIT_RWND_FULL:
1131 case SCTP_XMIT_DELAY:
1132 /* We could not append this chunk, so put
1133 * the chunk back on the output queue.
1135 pr_debug("%s: could not transmit tsn:0x%x, status:%d\n",
1136 __func__, ntohl(chunk->subh.data_hdr->tsn),
1139 sctp_outq_head_data(q, chunk);
1143 /* The sender is in the SHUTDOWN-PENDING state,
1144 * The sender MAY set the I-bit in the DATA
1147 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING)
1148 chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM;
1149 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
1150 asoc->stats.ouodchunks++;
1152 asoc->stats.oodchunks++;
1154 /* Only now it's safe to consider this
1155 * chunk as sent, sched-wise.
1157 sctp_sched_dequeue_done(q, chunk);
1159 list_add_tail(&chunk->transmitted_list,
1160 &transport->transmitted);
1162 sctp_transport_reset_t3_rtx(transport);
1163 transport->last_time_sent = jiffies;
1165 /* Only let one DATA chunk get bundled with a
1166 * COOKIE-ECHO chunk.
1168 if (packet->has_cookie_echo)
1180 * Try to flush an outqueue.
1182 * Description: Send everything in q which we legally can, subject to
1183 * congestion limitations.
1184 * * Note: This function can be called from multiple contexts so appropriate
1185 * locking concerns must be made. Today we use the sock lock to protect
1188 static void sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp)
1190 struct sctp_packet *packet;
1191 struct sctp_association *asoc = q->asoc;
1192 struct sctp_transport *transport = NULL;
1195 /* These transports have chunks to send. */
1196 struct list_head transport_list;
1197 struct list_head *ltransport;
1199 INIT_LIST_HEAD(&transport_list);
1205 * When bundling control chunks with DATA chunks, an
1206 * endpoint MUST place control chunks first in the outbound
1207 * SCTP packet. The transmitter MUST transmit DATA chunks
1208 * within a SCTP packet in increasing order of TSN.
1212 sctp_outq_flush_ctrl(q, &transport, &transport_list, gfp);
1214 if (q->asoc->src_out_of_asoc_ok)
1215 goto sctp_flush_out;
1217 sctp_outq_flush_data(q, &transport, &transport_list, rtx_timeout, gfp);
1221 /* Before returning, examine all the transports touched in
1222 * this call. Right now, we bluntly force clear all the
1223 * transports. Things might change after we implement Nagle.
1224 * But such an examination is still required.
1228 while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL) {
1229 struct sctp_transport *t = list_entry(ltransport,
1230 struct sctp_transport,
1232 packet = &t->packet;
1233 if (!sctp_packet_empty(packet)) {
1234 error = sctp_packet_transmit(packet, gfp);
1236 asoc->base.sk->sk_err = -error;
1239 /* Clear the burst limited state, if any */
1240 sctp_transport_burst_reset(t);
1244 /* Update unack_data based on the incoming SACK chunk */
1245 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1246 struct sctp_sackhdr *sack)
1248 union sctp_sack_variable *frags;
1252 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1254 frags = sack->variable;
1255 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1256 unack_data -= ((ntohs(frags[i].gab.end) -
1257 ntohs(frags[i].gab.start) + 1));
1260 assoc->unack_data = unack_data;
1263 /* This is where we REALLY process a SACK.
1265 * Process the SACK against the outqueue. Mostly, this just frees
1266 * things off the transmitted queue.
1268 int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk)
1270 struct sctp_association *asoc = q->asoc;
1271 struct sctp_sackhdr *sack = chunk->subh.sack_hdr;
1272 struct sctp_transport *transport;
1273 struct sctp_chunk *tchunk = NULL;
1274 struct list_head *lchunk, *transport_list, *temp;
1275 union sctp_sack_variable *frags = sack->variable;
1276 __u32 sack_ctsn, ctsn, tsn;
1277 __u32 highest_tsn, highest_new_tsn;
1279 unsigned int outstanding;
1280 struct sctp_transport *primary = asoc->peer.primary_path;
1281 int count_of_newacks = 0;
1285 /* Grab the association's destination address list. */
1286 transport_list = &asoc->peer.transport_addr_list;
1288 sack_ctsn = ntohl(sack->cum_tsn_ack);
1289 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1290 asoc->stats.gapcnt += gap_ack_blocks;
1292 * SFR-CACC algorithm:
1293 * On receipt of a SACK the sender SHOULD execute the
1294 * following statements.
1296 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1297 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1298 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1300 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1301 * is set the receiver of the SACK MUST take the following actions:
1303 * A) Initialize the cacc_saw_newack to 0 for all destination
1306 * Only bother if changeover_active is set. Otherwise, this is
1307 * totally suboptimal to do on every SACK.
1309 if (primary->cacc.changeover_active) {
1310 u8 clear_cycling = 0;
1312 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1313 primary->cacc.changeover_active = 0;
1317 if (clear_cycling || gap_ack_blocks) {
1318 list_for_each_entry(transport, transport_list,
1321 transport->cacc.cycling_changeover = 0;
1323 transport->cacc.cacc_saw_newack = 0;
1328 /* Get the highest TSN in the sack. */
1329 highest_tsn = sack_ctsn;
1331 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1333 if (TSN_lt(asoc->highest_sacked, highest_tsn))
1334 asoc->highest_sacked = highest_tsn;
1336 highest_new_tsn = sack_ctsn;
1338 /* Run through the retransmit queue. Credit bytes received
1339 * and free those chunks that we can.
1341 sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn);
1343 /* Run through the transmitted queue.
1344 * Credit bytes received and free those chunks which we can.
1346 * This is a MASSIVE candidate for optimization.
1348 list_for_each_entry(transport, transport_list, transports) {
1349 sctp_check_transmitted(q, &transport->transmitted,
1350 transport, &chunk->source, sack,
1353 * SFR-CACC algorithm:
1354 * C) Let count_of_newacks be the number of
1355 * destinations for which cacc_saw_newack is set.
1357 if (transport->cacc.cacc_saw_newack)
1361 /* Move the Cumulative TSN Ack Point if appropriate. */
1362 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) {
1363 asoc->ctsn_ack_point = sack_ctsn;
1367 if (gap_ack_blocks) {
1369 if (asoc->fast_recovery && accum_moved)
1370 highest_new_tsn = highest_tsn;
1372 list_for_each_entry(transport, transport_list, transports)
1373 sctp_mark_missing(q, &transport->transmitted, transport,
1374 highest_new_tsn, count_of_newacks);
1377 /* Update unack_data field in the assoc. */
1378 sctp_sack_update_unack_data(asoc, sack);
1380 ctsn = asoc->ctsn_ack_point;
1382 /* Throw away stuff rotting on the sack queue. */
1383 list_for_each_safe(lchunk, temp, &q->sacked) {
1384 tchunk = list_entry(lchunk, struct sctp_chunk,
1386 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1387 if (TSN_lte(tsn, ctsn)) {
1388 list_del_init(&tchunk->transmitted_list);
1389 if (asoc->peer.prsctp_capable &&
1390 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags))
1391 asoc->sent_cnt_removable--;
1392 sctp_chunk_free(tchunk);
1396 /* ii) Set rwnd equal to the newly received a_rwnd minus the
1397 * number of bytes still outstanding after processing the
1398 * Cumulative TSN Ack and the Gap Ack Blocks.
1401 sack_a_rwnd = ntohl(sack->a_rwnd);
1402 asoc->peer.zero_window_announced = !sack_a_rwnd;
1403 outstanding = q->outstanding_bytes;
1405 if (outstanding < sack_a_rwnd)
1406 sack_a_rwnd -= outstanding;
1410 asoc->peer.rwnd = sack_a_rwnd;
1412 asoc->stream.si->generate_ftsn(q, sack_ctsn);
1414 pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__, sack_ctsn);
1415 pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, "
1416 "advertised peer ack point:0x%x\n", __func__, asoc, ctsn,
1417 asoc->adv_peer_ack_point);
1419 return sctp_outq_is_empty(q);
1422 /* Is the outqueue empty?
1423 * The queue is empty when we have not pending data, no in-flight data
1424 * and nothing pending retransmissions.
1426 int sctp_outq_is_empty(const struct sctp_outq *q)
1428 return q->out_qlen == 0 && q->outstanding_bytes == 0 &&
1429 list_empty(&q->retransmit);
1432 /********************************************************************
1433 * 2nd Level Abstractions
1434 ********************************************************************/
1436 /* Go through a transport's transmitted list or the association's retransmit
1437 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1438 * The retransmit list will not have an associated transport.
1440 * I added coherent debug information output. --xguo
1442 * Instead of printing 'sacked' or 'kept' for each TSN on the
1443 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1444 * KEPT TSN6-TSN7, etc.
1446 static void sctp_check_transmitted(struct sctp_outq *q,
1447 struct list_head *transmitted_queue,
1448 struct sctp_transport *transport,
1449 union sctp_addr *saddr,
1450 struct sctp_sackhdr *sack,
1451 __u32 *highest_new_tsn_in_sack)
1453 struct list_head *lchunk;
1454 struct sctp_chunk *tchunk;
1455 struct list_head tlist;
1459 __u8 restart_timer = 0;
1460 int bytes_acked = 0;
1461 int migrate_bytes = 0;
1462 bool forward_progress = false;
1464 sack_ctsn = ntohl(sack->cum_tsn_ack);
1466 INIT_LIST_HEAD(&tlist);
1468 /* The while loop will skip empty transmitted queues. */
1469 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1470 tchunk = list_entry(lchunk, struct sctp_chunk,
1473 if (sctp_chunk_abandoned(tchunk)) {
1474 /* Move the chunk to abandoned list. */
1475 sctp_insert_list(&q->abandoned, lchunk);
1477 /* If this chunk has not been acked, stop
1478 * considering it as 'outstanding'.
1480 if (transmitted_queue != &q->retransmit &&
1481 !tchunk->tsn_gap_acked) {
1482 if (tchunk->transport)
1483 tchunk->transport->flight_size -=
1484 sctp_data_size(tchunk);
1485 q->outstanding_bytes -= sctp_data_size(tchunk);
1490 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1491 if (sctp_acked(sack, tsn)) {
1492 /* If this queue is the retransmit queue, the
1493 * retransmit timer has already reclaimed
1494 * the outstanding bytes for this chunk, so only
1495 * count bytes associated with a transport.
1497 if (transport && !tchunk->tsn_gap_acked) {
1498 /* If this chunk is being used for RTT
1499 * measurement, calculate the RTT and update
1500 * the RTO using this value.
1502 * 6.3.1 C5) Karn's algorithm: RTT measurements
1503 * MUST NOT be made using packets that were
1504 * retransmitted (and thus for which it is
1505 * ambiguous whether the reply was for the
1506 * first instance of the packet or a later
1509 if (!sctp_chunk_retransmitted(tchunk) &&
1510 tchunk->rtt_in_progress) {
1511 tchunk->rtt_in_progress = 0;
1512 rtt = jiffies - tchunk->sent_at;
1513 sctp_transport_update_rto(transport,
1517 if (TSN_lte(tsn, sack_ctsn)) {
1519 * SFR-CACC algorithm:
1520 * 2) If the SACK contains gap acks
1521 * and the flag CHANGEOVER_ACTIVE is
1522 * set the receiver of the SACK MUST
1523 * take the following action:
1525 * B) For each TSN t being acked that
1526 * has not been acked in any SACK so
1527 * far, set cacc_saw_newack to 1 for
1528 * the destination that the TSN was
1531 if (sack->num_gap_ack_blocks &&
1532 q->asoc->peer.primary_path->cacc.
1534 transport->cacc.cacc_saw_newack
1539 /* If the chunk hasn't been marked as ACKED,
1540 * mark it and account bytes_acked if the
1541 * chunk had a valid transport (it will not
1542 * have a transport if ASCONF had deleted it
1543 * while DATA was outstanding).
1545 if (!tchunk->tsn_gap_acked) {
1546 tchunk->tsn_gap_acked = 1;
1547 if (TSN_lt(*highest_new_tsn_in_sack, tsn))
1548 *highest_new_tsn_in_sack = tsn;
1549 bytes_acked += sctp_data_size(tchunk);
1550 if (!tchunk->transport)
1551 migrate_bytes += sctp_data_size(tchunk);
1552 forward_progress = true;
1555 if (TSN_lte(tsn, sack_ctsn)) {
1556 /* RFC 2960 6.3.2 Retransmission Timer Rules
1558 * R3) Whenever a SACK is received
1559 * that acknowledges the DATA chunk
1560 * with the earliest outstanding TSN
1561 * for that address, restart T3-rtx
1562 * timer for that address with its
1566 forward_progress = true;
1568 list_add_tail(&tchunk->transmitted_list,
1571 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1572 * M2) Each time a SACK arrives reporting
1573 * 'Stray DATA chunk(s)' record the highest TSN
1574 * reported as newly acknowledged, call this
1575 * value 'HighestTSNinSack'. A newly
1576 * acknowledged DATA chunk is one not
1577 * previously acknowledged in a SACK.
1579 * When the SCTP sender of data receives a SACK
1580 * chunk that acknowledges, for the first time,
1581 * the receipt of a DATA chunk, all the still
1582 * unacknowledged DATA chunks whose TSN is
1583 * older than that newly acknowledged DATA
1584 * chunk, are qualified as 'Stray DATA chunks'.
1586 list_add_tail(lchunk, &tlist);
1589 if (tchunk->tsn_gap_acked) {
1590 pr_debug("%s: receiver reneged on data TSN:0x%x\n",
1593 tchunk->tsn_gap_acked = 0;
1595 if (tchunk->transport)
1596 bytes_acked -= sctp_data_size(tchunk);
1598 /* RFC 2960 6.3.2 Retransmission Timer Rules
1600 * R4) Whenever a SACK is received missing a
1601 * TSN that was previously acknowledged via a
1602 * Gap Ack Block, start T3-rtx for the
1603 * destination address to which the DATA
1604 * chunk was originally
1605 * transmitted if it is not already running.
1610 list_add_tail(lchunk, &tlist);
1616 struct sctp_association *asoc = transport->asoc;
1618 /* We may have counted DATA that was migrated
1619 * to this transport due to DEL-IP operation.
1620 * Subtract those bytes, since the were never
1621 * send on this transport and shouldn't be
1622 * credited to this transport.
1624 bytes_acked -= migrate_bytes;
1626 /* 8.2. When an outstanding TSN is acknowledged,
1627 * the endpoint shall clear the error counter of
1628 * the destination transport address to which the
1629 * DATA chunk was last sent.
1630 * The association's overall error counter is
1633 transport->error_count = 0;
1634 transport->asoc->overall_error_count = 0;
1635 forward_progress = true;
1638 * While in SHUTDOWN PENDING, we may have started
1639 * the T5 shutdown guard timer after reaching the
1640 * retransmission limit. Stop that timer as soon
1641 * as the receiver acknowledged any data.
1643 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING &&
1644 del_timer(&asoc->timers
1645 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]))
1646 sctp_association_put(asoc);
1648 /* Mark the destination transport address as
1649 * active if it is not so marked.
1651 if ((transport->state == SCTP_INACTIVE ||
1652 transport->state == SCTP_UNCONFIRMED) &&
1653 sctp_cmp_addr_exact(&transport->ipaddr, saddr)) {
1654 sctp_assoc_control_transport(
1658 SCTP_RECEIVED_SACK);
1661 sctp_transport_raise_cwnd(transport, sack_ctsn,
1664 transport->flight_size -= bytes_acked;
1665 if (transport->flight_size == 0)
1666 transport->partial_bytes_acked = 0;
1667 q->outstanding_bytes -= bytes_acked + migrate_bytes;
1669 /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1670 * When a sender is doing zero window probing, it
1671 * should not timeout the association if it continues
1672 * to receive new packets from the receiver. The
1673 * reason is that the receiver MAY keep its window
1674 * closed for an indefinite time.
1675 * A sender is doing zero window probing when the
1676 * receiver's advertised window is zero, and there is
1677 * only one data chunk in flight to the receiver.
1679 * Allow the association to timeout while in SHUTDOWN
1680 * PENDING or SHUTDOWN RECEIVED in case the receiver
1681 * stays in zero window mode forever.
1683 if (!q->asoc->peer.rwnd &&
1684 !list_empty(&tlist) &&
1685 (sack_ctsn+2 == q->asoc->next_tsn) &&
1686 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) {
1687 pr_debug("%s: sack received for zero window "
1688 "probe:%u\n", __func__, sack_ctsn);
1690 q->asoc->overall_error_count = 0;
1691 transport->error_count = 0;
1695 /* RFC 2960 6.3.2 Retransmission Timer Rules
1697 * R2) Whenever all outstanding data sent to an address have
1698 * been acknowledged, turn off the T3-rtx timer of that
1701 if (!transport->flight_size) {
1702 if (del_timer(&transport->T3_rtx_timer))
1703 sctp_transport_put(transport);
1704 } else if (restart_timer) {
1705 if (!mod_timer(&transport->T3_rtx_timer,
1706 jiffies + transport->rto))
1707 sctp_transport_hold(transport);
1710 if (forward_progress) {
1712 sctp_transport_dst_confirm(transport);
1716 list_splice(&tlist, transmitted_queue);
1719 /* Mark chunks as missing and consequently may get retransmitted. */
1720 static void sctp_mark_missing(struct sctp_outq *q,
1721 struct list_head *transmitted_queue,
1722 struct sctp_transport *transport,
1723 __u32 highest_new_tsn_in_sack,
1724 int count_of_newacks)
1726 struct sctp_chunk *chunk;
1728 char do_fast_retransmit = 0;
1729 struct sctp_association *asoc = q->asoc;
1730 struct sctp_transport *primary = asoc->peer.primary_path;
1732 list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1734 tsn = ntohl(chunk->subh.data_hdr->tsn);
1736 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1737 * 'Unacknowledged TSN's', if the TSN number of an
1738 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1739 * value, increment the 'TSN.Missing.Report' count on that
1740 * chunk if it has NOT been fast retransmitted or marked for
1741 * fast retransmit already.
1743 if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
1744 !chunk->tsn_gap_acked &&
1745 TSN_lt(tsn, highest_new_tsn_in_sack)) {
1747 /* SFR-CACC may require us to skip marking
1748 * this chunk as missing.
1750 if (!transport || !sctp_cacc_skip(primary,
1752 count_of_newacks, tsn)) {
1753 chunk->tsn_missing_report++;
1755 pr_debug("%s: tsn:0x%x missing counter:%d\n",
1756 __func__, tsn, chunk->tsn_missing_report);
1760 * M4) If any DATA chunk is found to have a
1761 * 'TSN.Missing.Report'
1762 * value larger than or equal to 3, mark that chunk for
1763 * retransmission and start the fast retransmit procedure.
1766 if (chunk->tsn_missing_report >= 3) {
1767 chunk->fast_retransmit = SCTP_NEED_FRTX;
1768 do_fast_retransmit = 1;
1773 if (do_fast_retransmit)
1774 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1776 pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, "
1777 "flight_size:%d, pba:%d\n", __func__, transport,
1778 transport->cwnd, transport->ssthresh,
1779 transport->flight_size, transport->partial_bytes_acked);
1783 /* Is the given TSN acked by this packet? */
1784 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1786 __u32 ctsn = ntohl(sack->cum_tsn_ack);
1787 union sctp_sack_variable *frags;
1788 __u16 tsn_offset, blocks;
1791 if (TSN_lte(tsn, ctsn))
1794 /* 3.3.4 Selective Acknowledgement (SACK) (3):
1797 * These fields contain the Gap Ack Blocks. They are repeated
1798 * for each Gap Ack Block up to the number of Gap Ack Blocks
1799 * defined in the Number of Gap Ack Blocks field. All DATA
1800 * chunks with TSNs greater than or equal to (Cumulative TSN
1801 * Ack + Gap Ack Block Start) and less than or equal to
1802 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1803 * Block are assumed to have been received correctly.
1806 frags = sack->variable;
1807 blocks = ntohs(sack->num_gap_ack_blocks);
1808 tsn_offset = tsn - ctsn;
1809 for (i = 0; i < blocks; ++i) {
1810 if (tsn_offset >= ntohs(frags[i].gab.start) &&
1811 tsn_offset <= ntohs(frags[i].gab.end))
1820 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1821 int nskips, __be16 stream)
1825 for (i = 0; i < nskips; i++) {
1826 if (skiplist[i].stream == stream)
1832 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1833 void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1835 struct sctp_association *asoc = q->asoc;
1836 struct sctp_chunk *ftsn_chunk = NULL;
1837 struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1841 struct sctp_chunk *chunk;
1842 struct list_head *lchunk, *temp;
1844 if (!asoc->peer.prsctp_capable)
1847 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1850 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1851 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1853 if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1854 asoc->adv_peer_ack_point = ctsn;
1856 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1857 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1858 * the chunk next in the out-queue space is marked as "abandoned" as
1859 * shown in the following example:
1861 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1862 * and the Advanced.Peer.Ack.Point is updated to this value:
1864 * out-queue at the end of ==> out-queue after Adv.Ack.Point
1865 * normal SACK processing local advancement
1867 * Adv.Ack.Pt-> 102 acked 102 acked
1868 * 103 abandoned 103 abandoned
1869 * 104 abandoned Adv.Ack.P-> 104 abandoned
1871 * 106 acked 106 acked
1874 * In this example, the data sender successfully advanced the
1875 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1877 list_for_each_safe(lchunk, temp, &q->abandoned) {
1878 chunk = list_entry(lchunk, struct sctp_chunk,
1880 tsn = ntohl(chunk->subh.data_hdr->tsn);
1882 /* Remove any chunks in the abandoned queue that are acked by
1885 if (TSN_lte(tsn, ctsn)) {
1886 list_del_init(lchunk);
1887 sctp_chunk_free(chunk);
1889 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1890 asoc->adv_peer_ack_point = tsn;
1891 if (chunk->chunk_hdr->flags &
1892 SCTP_DATA_UNORDERED)
1894 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1896 chunk->subh.data_hdr->stream);
1897 ftsn_skip_arr[skip_pos].stream =
1898 chunk->subh.data_hdr->stream;
1899 ftsn_skip_arr[skip_pos].ssn =
1900 chunk->subh.data_hdr->ssn;
1901 if (skip_pos == nskips)
1910 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1911 * is greater than the Cumulative TSN ACK carried in the received
1912 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1913 * chunk containing the latest value of the
1914 * "Advanced.Peer.Ack.Point".
1916 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1917 * list each stream and sequence number in the forwarded TSN. This
1918 * information will enable the receiver to easily find any
1919 * stranded TSN's waiting on stream reorder queues. Each stream
1920 * SHOULD only be reported once; this means that if multiple
1921 * abandoned messages occur in the same stream then only the
1922 * highest abandoned stream sequence number is reported. If the
1923 * total size of the FORWARD TSN does NOT fit in a single MTU then
1924 * the sender of the FORWARD TSN SHOULD lower the
1925 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1928 if (asoc->adv_peer_ack_point > ctsn)
1929 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1930 nskips, &ftsn_skip_arr[0]);
1933 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1934 SCTP_INC_STATS(sock_net(asoc->base.sk), SCTP_MIB_OUTCTRLCHUNKS);