1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* SCTP kernel implementation
3 * (C) Copyright IBM Corp. 2001, 2004
4 * Copyright (c) 1999-2000 Cisco, Inc.
5 * Copyright (c) 1999-2001 Motorola, Inc.
6 * Copyright (c) 2001-2003 Intel Corp.
8 * This file is part of the SCTP kernel implementation
10 * These functions implement the sctp_outq class. The outqueue handles
11 * bundling and queueing of outgoing SCTP chunks.
13 * Please send any bug reports or fixes you make to the
15 * lksctp developers <linux-sctp@vger.kernel.org>
17 * Written or modified by:
18 * La Monte H.P. Yarroll <piggy@acm.org>
19 * Karl Knutson <karl@athena.chicago.il.us>
20 * Perry Melange <pmelange@null.cc.uic.edu>
21 * Xingang Guo <xingang.guo@intel.com>
22 * Hui Huang <hui.huang@nokia.com>
23 * Sridhar Samudrala <sri@us.ibm.com>
24 * Jon Grimm <jgrimm@us.ibm.com>
27 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29 #include <linux/types.h>
30 #include <linux/list.h> /* For struct list_head */
31 #include <linux/socket.h>
33 #include <linux/slab.h>
34 #include <net/sock.h> /* For skb_set_owner_w */
36 #include <net/sctp/sctp.h>
37 #include <net/sctp/sm.h>
38 #include <net/sctp/stream_sched.h>
39 #include <trace/events/sctp.h>
41 /* Declare internal functions here. */
42 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn);
43 static void sctp_check_transmitted(struct sctp_outq *q,
44 struct list_head *transmitted_queue,
45 struct sctp_transport *transport,
46 union sctp_addr *saddr,
47 struct sctp_sackhdr *sack,
48 __u32 *highest_new_tsn);
50 static void sctp_mark_missing(struct sctp_outq *q,
51 struct list_head *transmitted_queue,
52 struct sctp_transport *transport,
53 __u32 highest_new_tsn,
54 int count_of_newacks);
56 static void sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp);
58 /* Add data to the front of the queue. */
59 static inline void sctp_outq_head_data(struct sctp_outq *q,
60 struct sctp_chunk *ch)
62 struct sctp_stream_out_ext *oute;
65 list_add(&ch->list, &q->out_chunk_list);
66 q->out_qlen += ch->skb->len;
68 stream = sctp_chunk_stream_no(ch);
69 oute = SCTP_SO(&q->asoc->stream, stream)->ext;
70 list_add(&ch->stream_list, &oute->outq);
73 /* Take data from the front of the queue. */
74 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
76 return q->sched->dequeue(q);
79 /* Add data chunk to the end of the queue. */
80 static inline void sctp_outq_tail_data(struct sctp_outq *q,
81 struct sctp_chunk *ch)
83 struct sctp_stream_out_ext *oute;
86 list_add_tail(&ch->list, &q->out_chunk_list);
87 q->out_qlen += ch->skb->len;
89 stream = sctp_chunk_stream_no(ch);
90 oute = SCTP_SO(&q->asoc->stream, stream)->ext;
91 list_add_tail(&ch->stream_list, &oute->outq);
96 * D) If count_of_newacks is greater than or equal to 2
97 * and t was not sent to the current primary then the
98 * sender MUST NOT increment missing report count for t.
100 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary,
101 struct sctp_transport *transport,
102 int count_of_newacks)
104 if (count_of_newacks >= 2 && transport != primary)
110 * SFR-CACC algorithm:
111 * F) If count_of_newacks is less than 2, let d be the
112 * destination to which t was sent. If cacc_saw_newack
113 * is 0 for destination d, then the sender MUST NOT
114 * increment missing report count for t.
116 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport,
117 int count_of_newacks)
119 if (count_of_newacks < 2 &&
120 (transport && !transport->cacc.cacc_saw_newack))
126 * SFR-CACC algorithm:
127 * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
128 * execute steps C, D, F.
130 * C has been implemented in sctp_outq_sack
132 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
133 struct sctp_transport *transport,
134 int count_of_newacks)
136 if (!primary->cacc.cycling_changeover) {
137 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
139 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
147 * SFR-CACC algorithm:
148 * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
149 * than next_tsn_at_change of the current primary, then
150 * the sender MUST NOT increment missing report count
153 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
155 if (primary->cacc.cycling_changeover &&
156 TSN_lt(tsn, primary->cacc.next_tsn_at_change))
162 * SFR-CACC algorithm:
163 * 3) If the missing report count for TSN t is to be
164 * incremented according to [RFC2960] and
165 * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
166 * then the sender MUST further execute steps 3.1 and
167 * 3.2 to determine if the missing report count for
168 * TSN t SHOULD NOT be incremented.
170 * 3.3) If 3.1 and 3.2 do not dictate that the missing
171 * report count for t should not be incremented, then
172 * the sender SHOULD increment missing report count for
173 * t (according to [RFC2960] and [SCTP_STEWART_2002]).
175 static inline int sctp_cacc_skip(struct sctp_transport *primary,
176 struct sctp_transport *transport,
177 int count_of_newacks,
180 if (primary->cacc.changeover_active &&
181 (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) ||
182 sctp_cacc_skip_3_2(primary, tsn)))
187 /* Initialize an existing sctp_outq. This does the boring stuff.
188 * You still need to define handlers if you really want to DO
189 * something with this structure...
191 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
193 memset(q, 0, sizeof(struct sctp_outq));
196 INIT_LIST_HEAD(&q->out_chunk_list);
197 INIT_LIST_HEAD(&q->control_chunk_list);
198 INIT_LIST_HEAD(&q->retransmit);
199 INIT_LIST_HEAD(&q->sacked);
200 INIT_LIST_HEAD(&q->abandoned);
201 sctp_sched_set_sched(asoc, sctp_sk(asoc->base.sk)->default_ss);
204 /* Free the outqueue structure and any related pending chunks.
206 static void __sctp_outq_teardown(struct sctp_outq *q)
208 struct sctp_transport *transport;
209 struct list_head *lchunk, *temp;
210 struct sctp_chunk *chunk, *tmp;
212 /* Throw away unacknowledged chunks. */
213 list_for_each_entry(transport, &q->asoc->peer.transport_addr_list,
215 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
216 chunk = list_entry(lchunk, struct sctp_chunk,
218 /* Mark as part of a failed message. */
219 sctp_chunk_fail(chunk, q->error);
220 sctp_chunk_free(chunk);
224 /* Throw away chunks that have been gap ACKed. */
225 list_for_each_safe(lchunk, temp, &q->sacked) {
226 list_del_init(lchunk);
227 chunk = list_entry(lchunk, struct sctp_chunk,
229 sctp_chunk_fail(chunk, q->error);
230 sctp_chunk_free(chunk);
233 /* Throw away any chunks in the retransmit queue. */
234 list_for_each_safe(lchunk, temp, &q->retransmit) {
235 list_del_init(lchunk);
236 chunk = list_entry(lchunk, struct sctp_chunk,
238 sctp_chunk_fail(chunk, q->error);
239 sctp_chunk_free(chunk);
242 /* Throw away any chunks that are in the abandoned queue. */
243 list_for_each_safe(lchunk, temp, &q->abandoned) {
244 list_del_init(lchunk);
245 chunk = list_entry(lchunk, struct sctp_chunk,
247 sctp_chunk_fail(chunk, q->error);
248 sctp_chunk_free(chunk);
251 /* Throw away any leftover data chunks. */
252 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
253 sctp_sched_dequeue_done(q, chunk);
255 /* Mark as send failure. */
256 sctp_chunk_fail(chunk, q->error);
257 sctp_chunk_free(chunk);
260 /* Throw away any leftover control chunks. */
261 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
262 list_del_init(&chunk->list);
263 sctp_chunk_free(chunk);
267 void sctp_outq_teardown(struct sctp_outq *q)
269 __sctp_outq_teardown(q);
270 sctp_outq_init(q->asoc, q);
273 /* Free the outqueue structure and any related pending chunks. */
274 void sctp_outq_free(struct sctp_outq *q)
276 /* Throw away leftover chunks. */
277 __sctp_outq_teardown(q);
280 /* Put a new chunk in an sctp_outq. */
281 void sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk, gfp_t gfp)
283 struct net *net = q->asoc->base.net;
285 pr_debug("%s: outq:%p, chunk:%p[%s]\n", __func__, q, chunk,
286 chunk && chunk->chunk_hdr ?
287 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
290 /* If it is data, queue it up, otherwise, send it
293 if (sctp_chunk_is_data(chunk)) {
294 pr_debug("%s: outqueueing: outq:%p, chunk:%p[%s])\n",
295 __func__, q, chunk, chunk && chunk->chunk_hdr ?
296 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
299 sctp_outq_tail_data(q, chunk);
300 if (chunk->asoc->peer.prsctp_capable &&
301 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags))
302 chunk->asoc->sent_cnt_removable++;
303 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
304 SCTP_INC_STATS(net, SCTP_MIB_OUTUNORDERCHUNKS);
306 SCTP_INC_STATS(net, SCTP_MIB_OUTORDERCHUNKS);
308 list_add_tail(&chunk->list, &q->control_chunk_list);
309 SCTP_INC_STATS(net, SCTP_MIB_OUTCTRLCHUNKS);
313 sctp_outq_flush(q, 0, gfp);
316 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list
317 * and the abandoned list are in ascending order.
319 static void sctp_insert_list(struct list_head *head, struct list_head *new)
321 struct list_head *pos;
322 struct sctp_chunk *nchunk, *lchunk;
326 nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
327 ntsn = ntohl(nchunk->subh.data_hdr->tsn);
329 list_for_each(pos, head) {
330 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
331 ltsn = ntohl(lchunk->subh.data_hdr->tsn);
332 if (TSN_lt(ntsn, ltsn)) {
333 list_add(new, pos->prev);
339 list_add_tail(new, head);
342 static int sctp_prsctp_prune_sent(struct sctp_association *asoc,
343 struct sctp_sndrcvinfo *sinfo,
344 struct list_head *queue, int msg_len)
346 struct sctp_chunk *chk, *temp;
348 list_for_each_entry_safe(chk, temp, queue, transmitted_list) {
349 struct sctp_stream_out *streamout;
351 if (!chk->msg->abandoned &&
352 (!SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) ||
353 chk->sinfo.sinfo_timetolive <= sinfo->sinfo_timetolive))
356 chk->msg->abandoned = 1;
357 list_del_init(&chk->transmitted_list);
358 sctp_insert_list(&asoc->outqueue.abandoned,
359 &chk->transmitted_list);
361 streamout = SCTP_SO(&asoc->stream, chk->sinfo.sinfo_stream);
362 asoc->sent_cnt_removable--;
363 asoc->abandoned_sent[SCTP_PR_INDEX(PRIO)]++;
364 streamout->ext->abandoned_sent[SCTP_PR_INDEX(PRIO)]++;
366 if (queue != &asoc->outqueue.retransmit &&
367 !chk->tsn_gap_acked) {
369 chk->transport->flight_size -=
371 asoc->outqueue.outstanding_bytes -= sctp_data_size(chk);
374 msg_len -= chk->skb->truesize + sizeof(struct sctp_chunk);
382 static int sctp_prsctp_prune_unsent(struct sctp_association *asoc,
383 struct sctp_sndrcvinfo *sinfo, int msg_len)
385 struct sctp_outq *q = &asoc->outqueue;
386 struct sctp_chunk *chk, *temp;
388 q->sched->unsched_all(&asoc->stream);
390 list_for_each_entry_safe(chk, temp, &q->out_chunk_list, list) {
391 if (!chk->msg->abandoned &&
392 (!(chk->chunk_hdr->flags & SCTP_DATA_FIRST_FRAG) ||
393 !SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) ||
394 chk->sinfo.sinfo_timetolive <= sinfo->sinfo_timetolive))
397 chk->msg->abandoned = 1;
398 sctp_sched_dequeue_common(q, chk);
399 asoc->sent_cnt_removable--;
400 asoc->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++;
401 if (chk->sinfo.sinfo_stream < asoc->stream.outcnt) {
402 struct sctp_stream_out *streamout =
403 SCTP_SO(&asoc->stream, chk->sinfo.sinfo_stream);
405 streamout->ext->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++;
408 msg_len -= chk->skb->truesize + sizeof(struct sctp_chunk);
409 sctp_chunk_free(chk);
414 q->sched->sched_all(&asoc->stream);
419 /* Abandon the chunks according their priorities */
420 void sctp_prsctp_prune(struct sctp_association *asoc,
421 struct sctp_sndrcvinfo *sinfo, int msg_len)
423 struct sctp_transport *transport;
425 if (!asoc->peer.prsctp_capable || !asoc->sent_cnt_removable)
428 msg_len = sctp_prsctp_prune_sent(asoc, sinfo,
429 &asoc->outqueue.retransmit,
434 list_for_each_entry(transport, &asoc->peer.transport_addr_list,
436 msg_len = sctp_prsctp_prune_sent(asoc, sinfo,
437 &transport->transmitted,
443 sctp_prsctp_prune_unsent(asoc, sinfo, msg_len);
446 /* Mark all the eligible packets on a transport for retransmission. */
447 void sctp_retransmit_mark(struct sctp_outq *q,
448 struct sctp_transport *transport,
451 struct list_head *lchunk, *ltemp;
452 struct sctp_chunk *chunk;
454 /* Walk through the specified transmitted queue. */
455 list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
456 chunk = list_entry(lchunk, struct sctp_chunk,
459 /* If the chunk is abandoned, move it to abandoned list. */
460 if (sctp_chunk_abandoned(chunk)) {
461 list_del_init(lchunk);
462 sctp_insert_list(&q->abandoned, lchunk);
464 /* If this chunk has not been previousely acked,
465 * stop considering it 'outstanding'. Our peer
466 * will most likely never see it since it will
467 * not be retransmitted
469 if (!chunk->tsn_gap_acked) {
470 if (chunk->transport)
471 chunk->transport->flight_size -=
472 sctp_data_size(chunk);
473 q->outstanding_bytes -= sctp_data_size(chunk);
474 q->asoc->peer.rwnd += sctp_data_size(chunk);
479 /* If we are doing retransmission due to a timeout or pmtu
480 * discovery, only the chunks that are not yet acked should
481 * be added to the retransmit queue.
483 if ((reason == SCTP_RTXR_FAST_RTX &&
484 (chunk->fast_retransmit == SCTP_NEED_FRTX)) ||
485 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) {
486 /* RFC 2960 6.2.1 Processing a Received SACK
488 * C) Any time a DATA chunk is marked for
489 * retransmission (via either T3-rtx timer expiration
490 * (Section 6.3.3) or via fast retransmit
491 * (Section 7.2.4)), add the data size of those
492 * chunks to the rwnd.
494 q->asoc->peer.rwnd += sctp_data_size(chunk);
495 q->outstanding_bytes -= sctp_data_size(chunk);
496 if (chunk->transport)
497 transport->flight_size -= sctp_data_size(chunk);
499 /* sctpimpguide-05 Section 2.8.2
500 * M5) If a T3-rtx timer expires, the
501 * 'TSN.Missing.Report' of all affected TSNs is set
504 chunk->tsn_missing_report = 0;
506 /* If a chunk that is being used for RTT measurement
507 * has to be retransmitted, we cannot use this chunk
508 * anymore for RTT measurements. Reset rto_pending so
509 * that a new RTT measurement is started when a new
510 * data chunk is sent.
512 if (chunk->rtt_in_progress) {
513 chunk->rtt_in_progress = 0;
514 transport->rto_pending = 0;
517 /* Move the chunk to the retransmit queue. The chunks
518 * on the retransmit queue are always kept in order.
520 list_del_init(lchunk);
521 sctp_insert_list(&q->retransmit, lchunk);
525 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, "
526 "flight_size:%d, pba:%d\n", __func__, transport, reason,
527 transport->cwnd, transport->ssthresh, transport->flight_size,
528 transport->partial_bytes_acked);
531 /* Mark all the eligible packets on a transport for retransmission and force
534 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
535 enum sctp_retransmit_reason reason)
537 struct net *net = q->asoc->base.net;
540 case SCTP_RTXR_T3_RTX:
541 SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS);
542 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
543 /* Update the retran path if the T3-rtx timer has expired for
544 * the current retran path.
546 if (transport == transport->asoc->peer.retran_path)
547 sctp_assoc_update_retran_path(transport->asoc);
548 transport->asoc->rtx_data_chunks +=
549 transport->asoc->unack_data;
551 case SCTP_RTXR_FAST_RTX:
552 SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS);
553 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
556 case SCTP_RTXR_PMTUD:
557 SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS);
559 case SCTP_RTXR_T1_RTX:
560 SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS);
561 transport->asoc->init_retries++;
567 sctp_retransmit_mark(q, transport, reason);
569 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
570 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
571 * following the procedures outlined in C1 - C5.
573 if (reason == SCTP_RTXR_T3_RTX)
574 q->asoc->stream.si->generate_ftsn(q, q->asoc->ctsn_ack_point);
576 /* Flush the queues only on timeout, since fast_rtx is only
577 * triggered during sack processing and the queue
578 * will be flushed at the end.
580 if (reason != SCTP_RTXR_FAST_RTX)
581 sctp_outq_flush(q, /* rtx_timeout */ 1, GFP_ATOMIC);
585 * Transmit DATA chunks on the retransmit queue. Upon return from
586 * __sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
587 * need to be transmitted by the caller.
588 * We assume that pkt->transport has already been set.
590 * The return value is a normal kernel error return value.
592 static int __sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
593 int rtx_timeout, int *start_timer, gfp_t gfp)
595 struct sctp_transport *transport = pkt->transport;
596 struct sctp_chunk *chunk, *chunk1;
597 struct list_head *lqueue;
598 enum sctp_xmit status;
604 lqueue = &q->retransmit;
605 fast_rtx = q->fast_rtx;
607 /* This loop handles time-out retransmissions, fast retransmissions,
608 * and retransmissions due to opening of whindow.
610 * RFC 2960 6.3.3 Handle T3-rtx Expiration
612 * E3) Determine how many of the earliest (i.e., lowest TSN)
613 * outstanding DATA chunks for the address for which the
614 * T3-rtx has expired will fit into a single packet, subject
615 * to the MTU constraint for the path corresponding to the
616 * destination transport address to which the retransmission
617 * is being sent (this may be different from the address for
618 * which the timer expires [see Section 6.4]). Call this value
619 * K. Bundle and retransmit those K DATA chunks in a single
620 * packet to the destination endpoint.
622 * [Just to be painfully clear, if we are retransmitting
623 * because a timeout just happened, we should send only ONE
624 * packet of retransmitted data.]
626 * For fast retransmissions we also send only ONE packet. However,
627 * if we are just flushing the queue due to open window, we'll
628 * try to send as much as possible.
630 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
631 /* If the chunk is abandoned, move it to abandoned list. */
632 if (sctp_chunk_abandoned(chunk)) {
633 list_del_init(&chunk->transmitted_list);
634 sctp_insert_list(&q->abandoned,
635 &chunk->transmitted_list);
639 /* Make sure that Gap Acked TSNs are not retransmitted. A
640 * simple approach is just to move such TSNs out of the
641 * way and into a 'transmitted' queue and skip to the
644 if (chunk->tsn_gap_acked) {
645 list_move_tail(&chunk->transmitted_list,
646 &transport->transmitted);
650 /* If we are doing fast retransmit, ignore non-fast_rtransmit
653 if (fast_rtx && !chunk->fast_retransmit)
657 /* Attempt to append this chunk to the packet. */
658 status = sctp_packet_append_chunk(pkt, chunk);
661 case SCTP_XMIT_PMTU_FULL:
662 if (!pkt->has_data && !pkt->has_cookie_echo) {
663 /* If this packet did not contain DATA then
664 * retransmission did not happen, so do it
665 * again. We'll ignore the error here since
666 * control chunks are already freed so there
667 * is nothing we can do.
669 sctp_packet_transmit(pkt, gfp);
673 /* Send this packet. */
674 error = sctp_packet_transmit(pkt, gfp);
676 /* If we are retransmitting, we should only
677 * send a single packet.
678 * Otherwise, try appending this chunk again.
680 if (rtx_timeout || fast_rtx)
685 /* Bundle next chunk in the next round. */
688 case SCTP_XMIT_RWND_FULL:
689 /* Send this packet. */
690 error = sctp_packet_transmit(pkt, gfp);
692 /* Stop sending DATA as there is no more room
698 case SCTP_XMIT_DELAY:
699 /* Send this packet. */
700 error = sctp_packet_transmit(pkt, gfp);
702 /* Stop sending DATA because of nagle delay. */
707 /* The append was successful, so add this chunk to
708 * the transmitted list.
710 list_move_tail(&chunk->transmitted_list,
711 &transport->transmitted);
713 /* Mark the chunk as ineligible for fast retransmit
714 * after it is retransmitted.
716 if (chunk->fast_retransmit == SCTP_NEED_FRTX)
717 chunk->fast_retransmit = SCTP_DONT_FRTX;
719 q->asoc->stats.rtxchunks++;
723 /* Set the timer if there were no errors */
724 if (!error && !timer)
731 /* If we are here due to a retransmit timeout or a fast
732 * retransmit and if there are any chunks left in the retransmit
733 * queue that could not fit in the PMTU sized packet, they need
734 * to be marked as ineligible for a subsequent fast retransmit.
736 if (rtx_timeout || fast_rtx) {
737 list_for_each_entry(chunk1, lqueue, transmitted_list) {
738 if (chunk1->fast_retransmit == SCTP_NEED_FRTX)
739 chunk1->fast_retransmit = SCTP_DONT_FRTX;
743 *start_timer = timer;
745 /* Clear fast retransmit hint */
752 /* Cork the outqueue so queued chunks are really queued. */
753 void sctp_outq_uncork(struct sctp_outq *q, gfp_t gfp)
758 sctp_outq_flush(q, 0, gfp);
761 static int sctp_packet_singleton(struct sctp_transport *transport,
762 struct sctp_chunk *chunk, gfp_t gfp)
764 const struct sctp_association *asoc = transport->asoc;
765 const __u16 sport = asoc->base.bind_addr.port;
766 const __u16 dport = asoc->peer.port;
767 const __u32 vtag = asoc->peer.i.init_tag;
768 struct sctp_packet singleton;
770 sctp_packet_init(&singleton, transport, sport, dport);
771 sctp_packet_config(&singleton, vtag, 0);
772 if (sctp_packet_append_chunk(&singleton, chunk) != SCTP_XMIT_OK) {
773 list_del_init(&chunk->list);
774 sctp_chunk_free(chunk);
777 return sctp_packet_transmit(&singleton, gfp);
780 /* Struct to hold the context during sctp outq flush */
781 struct sctp_flush_ctx {
783 /* Current transport being used. It's NOT the same as curr active one */
784 struct sctp_transport *transport;
785 /* These transports have chunks to send. */
786 struct list_head transport_list;
787 struct sctp_association *asoc;
788 /* Packet on the current transport above */
789 struct sctp_packet *packet;
793 /* transport: current transport */
794 static void sctp_outq_select_transport(struct sctp_flush_ctx *ctx,
795 struct sctp_chunk *chunk)
797 struct sctp_transport *new_transport = chunk->transport;
799 if (!new_transport) {
800 if (!sctp_chunk_is_data(chunk)) {
801 /* If we have a prior transport pointer, see if
802 * the destination address of the chunk
803 * matches the destination address of the
804 * current transport. If not a match, then
805 * try to look up the transport with a given
806 * destination address. We do this because
807 * after processing ASCONFs, we may have new
808 * transports created.
810 if (ctx->transport && sctp_cmp_addr_exact(&chunk->dest,
811 &ctx->transport->ipaddr))
812 new_transport = ctx->transport;
814 new_transport = sctp_assoc_lookup_paddr(ctx->asoc,
818 /* if we still don't have a new transport, then
819 * use the current active path.
822 new_transport = ctx->asoc->peer.active_path;
826 switch (new_transport->state) {
828 case SCTP_UNCONFIRMED:
830 /* If the chunk is Heartbeat or Heartbeat Ack,
831 * send it to chunk->transport, even if it's
834 * 3.3.6 Heartbeat Acknowledgement:
836 * A HEARTBEAT ACK is always sent to the source IP
837 * address of the IP datagram containing the
838 * HEARTBEAT chunk to which this ack is responding.
841 * ASCONF_ACKs also must be sent to the source.
843 type = chunk->chunk_hdr->type;
844 if (type != SCTP_CID_HEARTBEAT &&
845 type != SCTP_CID_HEARTBEAT_ACK &&
846 type != SCTP_CID_ASCONF_ACK)
847 new_transport = ctx->asoc->peer.active_path;
854 /* Are we switching transports? Take care of transport locks. */
855 if (new_transport != ctx->transport) {
856 ctx->transport = new_transport;
857 ctx->packet = &ctx->transport->packet;
859 if (list_empty(&ctx->transport->send_ready))
860 list_add_tail(&ctx->transport->send_ready,
861 &ctx->transport_list);
863 sctp_packet_config(ctx->packet,
864 ctx->asoc->peer.i.init_tag,
865 ctx->asoc->peer.ecn_capable);
866 /* We've switched transports, so apply the
867 * Burst limit to the new transport.
869 sctp_transport_burst_limited(ctx->transport);
873 static void sctp_outq_flush_ctrl(struct sctp_flush_ctx *ctx)
875 struct sctp_chunk *chunk, *tmp;
876 enum sctp_xmit status;
877 int one_packet, error;
879 list_for_each_entry_safe(chunk, tmp, &ctx->q->control_chunk_list, list) {
883 * F1) This means that until such time as the ASCONF
884 * containing the add is acknowledged, the sender MUST
885 * NOT use the new IP address as a source for ANY SCTP
886 * packet except on carrying an ASCONF Chunk.
888 if (ctx->asoc->src_out_of_asoc_ok &&
889 chunk->chunk_hdr->type != SCTP_CID_ASCONF)
892 list_del_init(&chunk->list);
894 /* Pick the right transport to use. Should always be true for
895 * the first chunk as we don't have a transport by then.
897 sctp_outq_select_transport(ctx, chunk);
899 switch (chunk->chunk_hdr->type) {
902 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
903 * COMPLETE with any other chunks. [Send them immediately.]
906 case SCTP_CID_INIT_ACK:
907 case SCTP_CID_SHUTDOWN_COMPLETE:
908 error = sctp_packet_singleton(ctx->transport, chunk,
911 ctx->asoc->base.sk->sk_err = -error;
917 if (sctp_test_T_bit(chunk))
918 ctx->packet->vtag = ctx->asoc->c.my_vtag;
921 /* The following chunks are "response" chunks, i.e.
922 * they are generated in response to something we
923 * received. If we are sending these, then we can
924 * send only 1 packet containing these chunks.
926 case SCTP_CID_HEARTBEAT_ACK:
927 case SCTP_CID_SHUTDOWN_ACK:
928 case SCTP_CID_COOKIE_ACK:
929 case SCTP_CID_COOKIE_ECHO:
931 case SCTP_CID_ECN_CWR:
932 case SCTP_CID_ASCONF_ACK:
936 case SCTP_CID_HEARTBEAT:
937 if (chunk->pmtu_probe) {
938 sctp_packet_singleton(ctx->transport, chunk, ctx->gfp);
943 case SCTP_CID_SHUTDOWN:
944 case SCTP_CID_ECN_ECNE:
945 case SCTP_CID_ASCONF:
946 case SCTP_CID_FWD_TSN:
947 case SCTP_CID_I_FWD_TSN:
948 case SCTP_CID_RECONF:
949 status = sctp_packet_transmit_chunk(ctx->packet, chunk,
950 one_packet, ctx->gfp);
951 if (status != SCTP_XMIT_OK) {
952 /* put the chunk back */
953 list_add(&chunk->list, &ctx->q->control_chunk_list);
957 ctx->asoc->stats.octrlchunks++;
958 /* PR-SCTP C5) If a FORWARD TSN is sent, the
959 * sender MUST assure that at least one T3-rtx
962 if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN ||
963 chunk->chunk_hdr->type == SCTP_CID_I_FWD_TSN) {
964 sctp_transport_reset_t3_rtx(ctx->transport);
965 ctx->transport->last_time_sent = jiffies;
968 if (chunk == ctx->asoc->strreset_chunk)
969 sctp_transport_reset_reconf_timer(ctx->transport);
974 /* We built a chunk with an illegal type! */
980 /* Returns false if new data shouldn't be sent */
981 static bool sctp_outq_flush_rtx(struct sctp_flush_ctx *ctx,
984 int error, start_timer = 0;
986 if (ctx->asoc->peer.retran_path->state == SCTP_UNCONFIRMED)
989 if (ctx->transport != ctx->asoc->peer.retran_path) {
990 /* Switch transports & prepare the packet. */
991 ctx->transport = ctx->asoc->peer.retran_path;
992 ctx->packet = &ctx->transport->packet;
994 if (list_empty(&ctx->transport->send_ready))
995 list_add_tail(&ctx->transport->send_ready,
996 &ctx->transport_list);
998 sctp_packet_config(ctx->packet, ctx->asoc->peer.i.init_tag,
999 ctx->asoc->peer.ecn_capable);
1002 error = __sctp_outq_flush_rtx(ctx->q, ctx->packet, rtx_timeout,
1003 &start_timer, ctx->gfp);
1005 ctx->asoc->base.sk->sk_err = -error;
1008 sctp_transport_reset_t3_rtx(ctx->transport);
1009 ctx->transport->last_time_sent = jiffies;
1012 /* This can happen on COOKIE-ECHO resend. Only
1013 * one chunk can get bundled with a COOKIE-ECHO.
1015 if (ctx->packet->has_cookie_echo)
1018 /* Don't send new data if there is still data
1019 * waiting to retransmit.
1021 if (!list_empty(&ctx->q->retransmit))
1027 static void sctp_outq_flush_data(struct sctp_flush_ctx *ctx,
1030 struct sctp_chunk *chunk;
1031 enum sctp_xmit status;
1033 /* Is it OK to send data chunks? */
1034 switch (ctx->asoc->state) {
1035 case SCTP_STATE_COOKIE_ECHOED:
1036 /* Only allow bundling when this packet has a COOKIE-ECHO
1039 if (!ctx->packet || !ctx->packet->has_cookie_echo)
1043 case SCTP_STATE_ESTABLISHED:
1044 case SCTP_STATE_SHUTDOWN_PENDING:
1045 case SCTP_STATE_SHUTDOWN_RECEIVED:
1053 /* RFC 2960 6.1 Transmission of DATA Chunks
1055 * C) When the time comes for the sender to transmit,
1056 * before sending new DATA chunks, the sender MUST
1057 * first transmit any outstanding DATA chunks which
1058 * are marked for retransmission (limited by the
1061 if (!list_empty(&ctx->q->retransmit) &&
1062 !sctp_outq_flush_rtx(ctx, rtx_timeout))
1065 /* Apply Max.Burst limitation to the current transport in
1066 * case it will be used for new data. We are going to
1067 * rest it before we return, but we want to apply the limit
1068 * to the currently queued data.
1071 sctp_transport_burst_limited(ctx->transport);
1073 /* Finally, transmit new packets. */
1074 while ((chunk = sctp_outq_dequeue_data(ctx->q)) != NULL) {
1075 __u32 sid = ntohs(chunk->subh.data_hdr->stream);
1076 __u8 stream_state = SCTP_SO(&ctx->asoc->stream, sid)->state;
1078 /* Has this chunk expired? */
1079 if (sctp_chunk_abandoned(chunk)) {
1080 sctp_sched_dequeue_done(ctx->q, chunk);
1081 sctp_chunk_fail(chunk, 0);
1082 sctp_chunk_free(chunk);
1086 if (stream_state == SCTP_STREAM_CLOSED) {
1087 sctp_outq_head_data(ctx->q, chunk);
1091 sctp_outq_select_transport(ctx, chunk);
1093 pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p skb->users:%d\n",
1094 __func__, ctx->q, chunk, chunk && chunk->chunk_hdr ?
1095 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
1096 "illegal chunk", ntohl(chunk->subh.data_hdr->tsn),
1097 chunk->skb ? chunk->skb->head : NULL, chunk->skb ?
1098 refcount_read(&chunk->skb->users) : -1);
1100 /* Add the chunk to the packet. */
1101 status = sctp_packet_transmit_chunk(ctx->packet, chunk, 0,
1103 if (status != SCTP_XMIT_OK) {
1104 /* We could not append this chunk, so put
1105 * the chunk back on the output queue.
1107 pr_debug("%s: could not transmit tsn:0x%x, status:%d\n",
1108 __func__, ntohl(chunk->subh.data_hdr->tsn),
1111 sctp_outq_head_data(ctx->q, chunk);
1115 /* The sender is in the SHUTDOWN-PENDING state,
1116 * The sender MAY set the I-bit in the DATA
1119 if (ctx->asoc->state == SCTP_STATE_SHUTDOWN_PENDING)
1120 chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM;
1121 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
1122 ctx->asoc->stats.ouodchunks++;
1124 ctx->asoc->stats.oodchunks++;
1126 /* Only now it's safe to consider this
1127 * chunk as sent, sched-wise.
1129 sctp_sched_dequeue_done(ctx->q, chunk);
1131 list_add_tail(&chunk->transmitted_list,
1132 &ctx->transport->transmitted);
1134 sctp_transport_reset_t3_rtx(ctx->transport);
1135 ctx->transport->last_time_sent = jiffies;
1137 /* Only let one DATA chunk get bundled with a
1138 * COOKIE-ECHO chunk.
1140 if (ctx->packet->has_cookie_echo)
1145 static void sctp_outq_flush_transports(struct sctp_flush_ctx *ctx)
1147 struct sock *sk = ctx->asoc->base.sk;
1148 struct list_head *ltransport;
1149 struct sctp_packet *packet;
1150 struct sctp_transport *t;
1153 while ((ltransport = sctp_list_dequeue(&ctx->transport_list)) != NULL) {
1154 t = list_entry(ltransport, struct sctp_transport, send_ready);
1155 packet = &t->packet;
1156 if (!sctp_packet_empty(packet)) {
1158 if (t->dst && __sk_dst_get(sk) != t->dst) {
1160 sk_setup_caps(sk, t->dst);
1163 error = sctp_packet_transmit(packet, ctx->gfp);
1165 ctx->q->asoc->base.sk->sk_err = -error;
1168 /* Clear the burst limited state, if any */
1169 sctp_transport_burst_reset(t);
1173 /* Try to flush an outqueue.
1175 * Description: Send everything in q which we legally can, subject to
1176 * congestion limitations.
1177 * * Note: This function can be called from multiple contexts so appropriate
1178 * locking concerns must be made. Today we use the sock lock to protect
1182 static void sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp)
1184 struct sctp_flush_ctx ctx = {
1187 .transport_list = LIST_HEAD_INIT(ctx.transport_list),
1195 * When bundling control chunks with DATA chunks, an
1196 * endpoint MUST place control chunks first in the outbound
1197 * SCTP packet. The transmitter MUST transmit DATA chunks
1198 * within a SCTP packet in increasing order of TSN.
1202 sctp_outq_flush_ctrl(&ctx);
1204 if (q->asoc->src_out_of_asoc_ok)
1205 goto sctp_flush_out;
1207 sctp_outq_flush_data(&ctx, rtx_timeout);
1211 sctp_outq_flush_transports(&ctx);
1214 /* Update unack_data based on the incoming SACK chunk */
1215 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1216 struct sctp_sackhdr *sack)
1218 union sctp_sack_variable *frags;
1222 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1224 frags = sack->variable;
1225 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1226 unack_data -= ((ntohs(frags[i].gab.end) -
1227 ntohs(frags[i].gab.start) + 1));
1230 assoc->unack_data = unack_data;
1233 /* This is where we REALLY process a SACK.
1235 * Process the SACK against the outqueue. Mostly, this just frees
1236 * things off the transmitted queue.
1238 int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk)
1240 struct sctp_association *asoc = q->asoc;
1241 struct sctp_sackhdr *sack = chunk->subh.sack_hdr;
1242 struct sctp_transport *transport;
1243 struct sctp_chunk *tchunk = NULL;
1244 struct list_head *lchunk, *transport_list, *temp;
1245 union sctp_sack_variable *frags = sack->variable;
1246 __u32 sack_ctsn, ctsn, tsn;
1247 __u32 highest_tsn, highest_new_tsn;
1249 unsigned int outstanding;
1250 struct sctp_transport *primary = asoc->peer.primary_path;
1251 int count_of_newacks = 0;
1255 /* Grab the association's destination address list. */
1256 transport_list = &asoc->peer.transport_addr_list;
1258 /* SCTP path tracepoint for congestion control debugging. */
1259 if (trace_sctp_probe_path_enabled()) {
1260 list_for_each_entry(transport, transport_list, transports)
1261 trace_sctp_probe_path(transport, asoc);
1264 sack_ctsn = ntohl(sack->cum_tsn_ack);
1265 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1266 asoc->stats.gapcnt += gap_ack_blocks;
1268 * SFR-CACC algorithm:
1269 * On receipt of a SACK the sender SHOULD execute the
1270 * following statements.
1272 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1273 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1274 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1276 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1277 * is set the receiver of the SACK MUST take the following actions:
1279 * A) Initialize the cacc_saw_newack to 0 for all destination
1282 * Only bother if changeover_active is set. Otherwise, this is
1283 * totally suboptimal to do on every SACK.
1285 if (primary->cacc.changeover_active) {
1286 u8 clear_cycling = 0;
1288 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1289 primary->cacc.changeover_active = 0;
1293 if (clear_cycling || gap_ack_blocks) {
1294 list_for_each_entry(transport, transport_list,
1297 transport->cacc.cycling_changeover = 0;
1299 transport->cacc.cacc_saw_newack = 0;
1304 /* Get the highest TSN in the sack. */
1305 highest_tsn = sack_ctsn;
1307 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1309 if (TSN_lt(asoc->highest_sacked, highest_tsn))
1310 asoc->highest_sacked = highest_tsn;
1312 highest_new_tsn = sack_ctsn;
1314 /* Run through the retransmit queue. Credit bytes received
1315 * and free those chunks that we can.
1317 sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn);
1319 /* Run through the transmitted queue.
1320 * Credit bytes received and free those chunks which we can.
1322 * This is a MASSIVE candidate for optimization.
1324 list_for_each_entry(transport, transport_list, transports) {
1325 sctp_check_transmitted(q, &transport->transmitted,
1326 transport, &chunk->source, sack,
1329 * SFR-CACC algorithm:
1330 * C) Let count_of_newacks be the number of
1331 * destinations for which cacc_saw_newack is set.
1333 if (transport->cacc.cacc_saw_newack)
1337 /* Move the Cumulative TSN Ack Point if appropriate. */
1338 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) {
1339 asoc->ctsn_ack_point = sack_ctsn;
1343 if (gap_ack_blocks) {
1345 if (asoc->fast_recovery && accum_moved)
1346 highest_new_tsn = highest_tsn;
1348 list_for_each_entry(transport, transport_list, transports)
1349 sctp_mark_missing(q, &transport->transmitted, transport,
1350 highest_new_tsn, count_of_newacks);
1353 /* Update unack_data field in the assoc. */
1354 sctp_sack_update_unack_data(asoc, sack);
1356 ctsn = asoc->ctsn_ack_point;
1358 /* Throw away stuff rotting on the sack queue. */
1359 list_for_each_safe(lchunk, temp, &q->sacked) {
1360 tchunk = list_entry(lchunk, struct sctp_chunk,
1362 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1363 if (TSN_lte(tsn, ctsn)) {
1364 list_del_init(&tchunk->transmitted_list);
1365 if (asoc->peer.prsctp_capable &&
1366 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags))
1367 asoc->sent_cnt_removable--;
1368 sctp_chunk_free(tchunk);
1372 /* ii) Set rwnd equal to the newly received a_rwnd minus the
1373 * number of bytes still outstanding after processing the
1374 * Cumulative TSN Ack and the Gap Ack Blocks.
1377 sack_a_rwnd = ntohl(sack->a_rwnd);
1378 asoc->peer.zero_window_announced = !sack_a_rwnd;
1379 outstanding = q->outstanding_bytes;
1381 if (outstanding < sack_a_rwnd)
1382 sack_a_rwnd -= outstanding;
1386 asoc->peer.rwnd = sack_a_rwnd;
1388 asoc->stream.si->generate_ftsn(q, sack_ctsn);
1390 pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__, sack_ctsn);
1391 pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, "
1392 "advertised peer ack point:0x%x\n", __func__, asoc, ctsn,
1393 asoc->adv_peer_ack_point);
1395 return sctp_outq_is_empty(q);
1398 /* Is the outqueue empty?
1399 * The queue is empty when we have not pending data, no in-flight data
1400 * and nothing pending retransmissions.
1402 int sctp_outq_is_empty(const struct sctp_outq *q)
1404 return q->out_qlen == 0 && q->outstanding_bytes == 0 &&
1405 list_empty(&q->retransmit);
1408 /********************************************************************
1409 * 2nd Level Abstractions
1410 ********************************************************************/
1412 /* Go through a transport's transmitted list or the association's retransmit
1413 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1414 * The retransmit list will not have an associated transport.
1416 * I added coherent debug information output. --xguo
1418 * Instead of printing 'sacked' or 'kept' for each TSN on the
1419 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1420 * KEPT TSN6-TSN7, etc.
1422 static void sctp_check_transmitted(struct sctp_outq *q,
1423 struct list_head *transmitted_queue,
1424 struct sctp_transport *transport,
1425 union sctp_addr *saddr,
1426 struct sctp_sackhdr *sack,
1427 __u32 *highest_new_tsn_in_sack)
1429 struct list_head *lchunk;
1430 struct sctp_chunk *tchunk;
1431 struct list_head tlist;
1435 __u8 restart_timer = 0;
1436 int bytes_acked = 0;
1437 int migrate_bytes = 0;
1438 bool forward_progress = false;
1440 sack_ctsn = ntohl(sack->cum_tsn_ack);
1442 INIT_LIST_HEAD(&tlist);
1444 /* The while loop will skip empty transmitted queues. */
1445 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1446 tchunk = list_entry(lchunk, struct sctp_chunk,
1449 if (sctp_chunk_abandoned(tchunk)) {
1450 /* Move the chunk to abandoned list. */
1451 sctp_insert_list(&q->abandoned, lchunk);
1453 /* If this chunk has not been acked, stop
1454 * considering it as 'outstanding'.
1456 if (transmitted_queue != &q->retransmit &&
1457 !tchunk->tsn_gap_acked) {
1458 if (tchunk->transport)
1459 tchunk->transport->flight_size -=
1460 sctp_data_size(tchunk);
1461 q->outstanding_bytes -= sctp_data_size(tchunk);
1466 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1467 if (sctp_acked(sack, tsn)) {
1468 /* If this queue is the retransmit queue, the
1469 * retransmit timer has already reclaimed
1470 * the outstanding bytes for this chunk, so only
1471 * count bytes associated with a transport.
1473 if (transport && !tchunk->tsn_gap_acked) {
1474 /* If this chunk is being used for RTT
1475 * measurement, calculate the RTT and update
1476 * the RTO using this value.
1478 * 6.3.1 C5) Karn's algorithm: RTT measurements
1479 * MUST NOT be made using packets that were
1480 * retransmitted (and thus for which it is
1481 * ambiguous whether the reply was for the
1482 * first instance of the packet or a later
1485 if (!sctp_chunk_retransmitted(tchunk) &&
1486 tchunk->rtt_in_progress) {
1487 tchunk->rtt_in_progress = 0;
1488 rtt = jiffies - tchunk->sent_at;
1489 sctp_transport_update_rto(transport,
1493 if (TSN_lte(tsn, sack_ctsn)) {
1495 * SFR-CACC algorithm:
1496 * 2) If the SACK contains gap acks
1497 * and the flag CHANGEOVER_ACTIVE is
1498 * set the receiver of the SACK MUST
1499 * take the following action:
1501 * B) For each TSN t being acked that
1502 * has not been acked in any SACK so
1503 * far, set cacc_saw_newack to 1 for
1504 * the destination that the TSN was
1507 if (sack->num_gap_ack_blocks &&
1508 q->asoc->peer.primary_path->cacc.
1510 transport->cacc.cacc_saw_newack
1515 /* If the chunk hasn't been marked as ACKED,
1516 * mark it and account bytes_acked if the
1517 * chunk had a valid transport (it will not
1518 * have a transport if ASCONF had deleted it
1519 * while DATA was outstanding).
1521 if (!tchunk->tsn_gap_acked) {
1522 tchunk->tsn_gap_acked = 1;
1523 if (TSN_lt(*highest_new_tsn_in_sack, tsn))
1524 *highest_new_tsn_in_sack = tsn;
1525 bytes_acked += sctp_data_size(tchunk);
1526 if (!tchunk->transport)
1527 migrate_bytes += sctp_data_size(tchunk);
1528 forward_progress = true;
1531 if (TSN_lte(tsn, sack_ctsn)) {
1532 /* RFC 2960 6.3.2 Retransmission Timer Rules
1534 * R3) Whenever a SACK is received
1535 * that acknowledges the DATA chunk
1536 * with the earliest outstanding TSN
1537 * for that address, restart T3-rtx
1538 * timer for that address with its
1542 forward_progress = true;
1544 list_add_tail(&tchunk->transmitted_list,
1547 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1548 * M2) Each time a SACK arrives reporting
1549 * 'Stray DATA chunk(s)' record the highest TSN
1550 * reported as newly acknowledged, call this
1551 * value 'HighestTSNinSack'. A newly
1552 * acknowledged DATA chunk is one not
1553 * previously acknowledged in a SACK.
1555 * When the SCTP sender of data receives a SACK
1556 * chunk that acknowledges, for the first time,
1557 * the receipt of a DATA chunk, all the still
1558 * unacknowledged DATA chunks whose TSN is
1559 * older than that newly acknowledged DATA
1560 * chunk, are qualified as 'Stray DATA chunks'.
1562 list_add_tail(lchunk, &tlist);
1565 if (tchunk->tsn_gap_acked) {
1566 pr_debug("%s: receiver reneged on data TSN:0x%x\n",
1569 tchunk->tsn_gap_acked = 0;
1571 if (tchunk->transport)
1572 bytes_acked -= sctp_data_size(tchunk);
1574 /* RFC 2960 6.3.2 Retransmission Timer Rules
1576 * R4) Whenever a SACK is received missing a
1577 * TSN that was previously acknowledged via a
1578 * Gap Ack Block, start T3-rtx for the
1579 * destination address to which the DATA
1580 * chunk was originally
1581 * transmitted if it is not already running.
1586 list_add_tail(lchunk, &tlist);
1592 struct sctp_association *asoc = transport->asoc;
1594 /* We may have counted DATA that was migrated
1595 * to this transport due to DEL-IP operation.
1596 * Subtract those bytes, since the were never
1597 * send on this transport and shouldn't be
1598 * credited to this transport.
1600 bytes_acked -= migrate_bytes;
1602 /* 8.2. When an outstanding TSN is acknowledged,
1603 * the endpoint shall clear the error counter of
1604 * the destination transport address to which the
1605 * DATA chunk was last sent.
1606 * The association's overall error counter is
1609 transport->error_count = 0;
1610 transport->asoc->overall_error_count = 0;
1611 forward_progress = true;
1614 * While in SHUTDOWN PENDING, we may have started
1615 * the T5 shutdown guard timer after reaching the
1616 * retransmission limit. Stop that timer as soon
1617 * as the receiver acknowledged any data.
1619 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING &&
1620 del_timer(&asoc->timers
1621 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]))
1622 sctp_association_put(asoc);
1624 /* Mark the destination transport address as
1625 * active if it is not so marked.
1627 if ((transport->state == SCTP_INACTIVE ||
1628 transport->state == SCTP_UNCONFIRMED) &&
1629 sctp_cmp_addr_exact(&transport->ipaddr, saddr)) {
1630 sctp_assoc_control_transport(
1634 SCTP_RECEIVED_SACK);
1637 sctp_transport_raise_cwnd(transport, sack_ctsn,
1640 transport->flight_size -= bytes_acked;
1641 if (transport->flight_size == 0)
1642 transport->partial_bytes_acked = 0;
1643 q->outstanding_bytes -= bytes_acked + migrate_bytes;
1645 /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1646 * When a sender is doing zero window probing, it
1647 * should not timeout the association if it continues
1648 * to receive new packets from the receiver. The
1649 * reason is that the receiver MAY keep its window
1650 * closed for an indefinite time.
1651 * A sender is doing zero window probing when the
1652 * receiver's advertised window is zero, and there is
1653 * only one data chunk in flight to the receiver.
1655 * Allow the association to timeout while in SHUTDOWN
1656 * PENDING or SHUTDOWN RECEIVED in case the receiver
1657 * stays in zero window mode forever.
1659 if (!q->asoc->peer.rwnd &&
1660 !list_empty(&tlist) &&
1661 (sack_ctsn+2 == q->asoc->next_tsn) &&
1662 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) {
1663 pr_debug("%s: sack received for zero window "
1664 "probe:%u\n", __func__, sack_ctsn);
1666 q->asoc->overall_error_count = 0;
1667 transport->error_count = 0;
1671 /* RFC 2960 6.3.2 Retransmission Timer Rules
1673 * R2) Whenever all outstanding data sent to an address have
1674 * been acknowledged, turn off the T3-rtx timer of that
1677 if (!transport->flight_size) {
1678 if (del_timer(&transport->T3_rtx_timer))
1679 sctp_transport_put(transport);
1680 } else if (restart_timer) {
1681 if (!mod_timer(&transport->T3_rtx_timer,
1682 jiffies + transport->rto))
1683 sctp_transport_hold(transport);
1686 if (forward_progress) {
1688 sctp_transport_dst_confirm(transport);
1692 list_splice(&tlist, transmitted_queue);
1695 /* Mark chunks as missing and consequently may get retransmitted. */
1696 static void sctp_mark_missing(struct sctp_outq *q,
1697 struct list_head *transmitted_queue,
1698 struct sctp_transport *transport,
1699 __u32 highest_new_tsn_in_sack,
1700 int count_of_newacks)
1702 struct sctp_chunk *chunk;
1704 char do_fast_retransmit = 0;
1705 struct sctp_association *asoc = q->asoc;
1706 struct sctp_transport *primary = asoc->peer.primary_path;
1708 list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1710 tsn = ntohl(chunk->subh.data_hdr->tsn);
1712 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1713 * 'Unacknowledged TSN's', if the TSN number of an
1714 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1715 * value, increment the 'TSN.Missing.Report' count on that
1716 * chunk if it has NOT been fast retransmitted or marked for
1717 * fast retransmit already.
1719 if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
1720 !chunk->tsn_gap_acked &&
1721 TSN_lt(tsn, highest_new_tsn_in_sack)) {
1723 /* SFR-CACC may require us to skip marking
1724 * this chunk as missing.
1726 if (!transport || !sctp_cacc_skip(primary,
1728 count_of_newacks, tsn)) {
1729 chunk->tsn_missing_report++;
1731 pr_debug("%s: tsn:0x%x missing counter:%d\n",
1732 __func__, tsn, chunk->tsn_missing_report);
1736 * M4) If any DATA chunk is found to have a
1737 * 'TSN.Missing.Report'
1738 * value larger than or equal to 3, mark that chunk for
1739 * retransmission and start the fast retransmit procedure.
1742 if (chunk->tsn_missing_report >= 3) {
1743 chunk->fast_retransmit = SCTP_NEED_FRTX;
1744 do_fast_retransmit = 1;
1749 if (do_fast_retransmit)
1750 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1752 pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, "
1753 "flight_size:%d, pba:%d\n", __func__, transport,
1754 transport->cwnd, transport->ssthresh,
1755 transport->flight_size, transport->partial_bytes_acked);
1759 /* Is the given TSN acked by this packet? */
1760 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1762 __u32 ctsn = ntohl(sack->cum_tsn_ack);
1763 union sctp_sack_variable *frags;
1764 __u16 tsn_offset, blocks;
1767 if (TSN_lte(tsn, ctsn))
1770 /* 3.3.4 Selective Acknowledgment (SACK) (3):
1773 * These fields contain the Gap Ack Blocks. They are repeated
1774 * for each Gap Ack Block up to the number of Gap Ack Blocks
1775 * defined in the Number of Gap Ack Blocks field. All DATA
1776 * chunks with TSNs greater than or equal to (Cumulative TSN
1777 * Ack + Gap Ack Block Start) and less than or equal to
1778 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1779 * Block are assumed to have been received correctly.
1782 frags = sack->variable;
1783 blocks = ntohs(sack->num_gap_ack_blocks);
1784 tsn_offset = tsn - ctsn;
1785 for (i = 0; i < blocks; ++i) {
1786 if (tsn_offset >= ntohs(frags[i].gab.start) &&
1787 tsn_offset <= ntohs(frags[i].gab.end))
1796 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1797 int nskips, __be16 stream)
1801 for (i = 0; i < nskips; i++) {
1802 if (skiplist[i].stream == stream)
1808 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1809 void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1811 struct sctp_association *asoc = q->asoc;
1812 struct sctp_chunk *ftsn_chunk = NULL;
1813 struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1817 struct sctp_chunk *chunk;
1818 struct list_head *lchunk, *temp;
1820 if (!asoc->peer.prsctp_capable)
1823 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1826 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1827 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1829 if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1830 asoc->adv_peer_ack_point = ctsn;
1832 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1833 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1834 * the chunk next in the out-queue space is marked as "abandoned" as
1835 * shown in the following example:
1837 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1838 * and the Advanced.Peer.Ack.Point is updated to this value:
1840 * out-queue at the end of ==> out-queue after Adv.Ack.Point
1841 * normal SACK processing local advancement
1843 * Adv.Ack.Pt-> 102 acked 102 acked
1844 * 103 abandoned 103 abandoned
1845 * 104 abandoned Adv.Ack.P-> 104 abandoned
1847 * 106 acked 106 acked
1850 * In this example, the data sender successfully advanced the
1851 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1853 list_for_each_safe(lchunk, temp, &q->abandoned) {
1854 chunk = list_entry(lchunk, struct sctp_chunk,
1856 tsn = ntohl(chunk->subh.data_hdr->tsn);
1858 /* Remove any chunks in the abandoned queue that are acked by
1861 if (TSN_lte(tsn, ctsn)) {
1862 list_del_init(lchunk);
1863 sctp_chunk_free(chunk);
1865 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1866 asoc->adv_peer_ack_point = tsn;
1867 if (chunk->chunk_hdr->flags &
1868 SCTP_DATA_UNORDERED)
1870 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1872 chunk->subh.data_hdr->stream);
1873 ftsn_skip_arr[skip_pos].stream =
1874 chunk->subh.data_hdr->stream;
1875 ftsn_skip_arr[skip_pos].ssn =
1876 chunk->subh.data_hdr->ssn;
1877 if (skip_pos == nskips)
1886 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1887 * is greater than the Cumulative TSN ACK carried in the received
1888 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1889 * chunk containing the latest value of the
1890 * "Advanced.Peer.Ack.Point".
1892 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1893 * list each stream and sequence number in the forwarded TSN. This
1894 * information will enable the receiver to easily find any
1895 * stranded TSN's waiting on stream reorder queues. Each stream
1896 * SHOULD only be reported once; this means that if multiple
1897 * abandoned messages occur in the same stream then only the
1898 * highest abandoned stream sequence number is reported. If the
1899 * total size of the FORWARD TSN does NOT fit in a single MTU then
1900 * the sender of the FORWARD TSN SHOULD lower the
1901 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1904 if (asoc->adv_peer_ack_point > ctsn)
1905 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1906 nskips, &ftsn_skip_arr[0]);
1909 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1910 SCTP_INC_STATS(asoc->base.net, SCTP_MIB_OUTCTRLCHUNKS);