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 = SCTP_SO(&q->asoc->stream, 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 = SCTP_SO(&q->asoc->stream, 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_sk(asoc->base.sk)->default_ss);
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 = SCTP_SO(&asoc->stream, 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 -= chk->skb->truesize + sizeof(struct sctp_chunk);
396 static int sctp_prsctp_prune_unsent(struct sctp_association *asoc,
397 struct sctp_sndrcvinfo *sinfo, int msg_len)
399 struct sctp_outq *q = &asoc->outqueue;
400 struct sctp_chunk *chk, *temp;
402 q->sched->unsched_all(&asoc->stream);
404 list_for_each_entry_safe(chk, temp, &q->out_chunk_list, list) {
405 if (!chk->msg->abandoned &&
406 (!(chk->chunk_hdr->flags & SCTP_DATA_FIRST_FRAG) ||
407 !SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) ||
408 chk->sinfo.sinfo_timetolive <= sinfo->sinfo_timetolive))
411 chk->msg->abandoned = 1;
412 sctp_sched_dequeue_common(q, chk);
413 asoc->sent_cnt_removable--;
414 asoc->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++;
415 if (chk->sinfo.sinfo_stream < asoc->stream.outcnt) {
416 struct sctp_stream_out *streamout =
417 SCTP_SO(&asoc->stream, chk->sinfo.sinfo_stream);
419 streamout->ext->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++;
422 msg_len -= chk->skb->truesize + sizeof(struct sctp_chunk);
423 sctp_chunk_free(chk);
428 q->sched->sched_all(&asoc->stream);
433 /* Abandon the chunks according their priorities */
434 void sctp_prsctp_prune(struct sctp_association *asoc,
435 struct sctp_sndrcvinfo *sinfo, int msg_len)
437 struct sctp_transport *transport;
439 if (!asoc->peer.prsctp_capable || !asoc->sent_cnt_removable)
442 msg_len = sctp_prsctp_prune_sent(asoc, sinfo,
443 &asoc->outqueue.retransmit,
448 list_for_each_entry(transport, &asoc->peer.transport_addr_list,
450 msg_len = sctp_prsctp_prune_sent(asoc, sinfo,
451 &transport->transmitted,
457 sctp_prsctp_prune_unsent(asoc, sinfo, msg_len);
460 /* Mark all the eligible packets on a transport for retransmission. */
461 void sctp_retransmit_mark(struct sctp_outq *q,
462 struct sctp_transport *transport,
465 struct list_head *lchunk, *ltemp;
466 struct sctp_chunk *chunk;
468 /* Walk through the specified transmitted queue. */
469 list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
470 chunk = list_entry(lchunk, struct sctp_chunk,
473 /* If the chunk is abandoned, move it to abandoned list. */
474 if (sctp_chunk_abandoned(chunk)) {
475 list_del_init(lchunk);
476 sctp_insert_list(&q->abandoned, lchunk);
478 /* If this chunk has not been previousely acked,
479 * stop considering it 'outstanding'. Our peer
480 * will most likely never see it since it will
481 * not be retransmitted
483 if (!chunk->tsn_gap_acked) {
484 if (chunk->transport)
485 chunk->transport->flight_size -=
486 sctp_data_size(chunk);
487 q->outstanding_bytes -= sctp_data_size(chunk);
488 q->asoc->peer.rwnd += sctp_data_size(chunk);
493 /* If we are doing retransmission due to a timeout or pmtu
494 * discovery, only the chunks that are not yet acked should
495 * be added to the retransmit queue.
497 if ((reason == SCTP_RTXR_FAST_RTX &&
498 (chunk->fast_retransmit == SCTP_NEED_FRTX)) ||
499 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) {
500 /* RFC 2960 6.2.1 Processing a Received SACK
502 * C) Any time a DATA chunk is marked for
503 * retransmission (via either T3-rtx timer expiration
504 * (Section 6.3.3) or via fast retransmit
505 * (Section 7.2.4)), add the data size of those
506 * chunks to the rwnd.
508 q->asoc->peer.rwnd += sctp_data_size(chunk);
509 q->outstanding_bytes -= sctp_data_size(chunk);
510 if (chunk->transport)
511 transport->flight_size -= sctp_data_size(chunk);
513 /* sctpimpguide-05 Section 2.8.2
514 * M5) If a T3-rtx timer expires, the
515 * 'TSN.Missing.Report' of all affected TSNs is set
518 chunk->tsn_missing_report = 0;
520 /* If a chunk that is being used for RTT measurement
521 * has to be retransmitted, we cannot use this chunk
522 * anymore for RTT measurements. Reset rto_pending so
523 * that a new RTT measurement is started when a new
524 * data chunk is sent.
526 if (chunk->rtt_in_progress) {
527 chunk->rtt_in_progress = 0;
528 transport->rto_pending = 0;
531 /* Move the chunk to the retransmit queue. The chunks
532 * on the retransmit queue are always kept in order.
534 list_del_init(lchunk);
535 sctp_insert_list(&q->retransmit, lchunk);
539 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, "
540 "flight_size:%d, pba:%d\n", __func__, transport, reason,
541 transport->cwnd, transport->ssthresh, transport->flight_size,
542 transport->partial_bytes_acked);
545 /* Mark all the eligible packets on a transport for retransmission and force
548 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
549 enum sctp_retransmit_reason reason)
551 struct net *net = sock_net(q->asoc->base.sk);
554 case SCTP_RTXR_T3_RTX:
555 SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS);
556 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
557 /* Update the retran path if the T3-rtx timer has expired for
558 * the current retran path.
560 if (transport == transport->asoc->peer.retran_path)
561 sctp_assoc_update_retran_path(transport->asoc);
562 transport->asoc->rtx_data_chunks +=
563 transport->asoc->unack_data;
565 case SCTP_RTXR_FAST_RTX:
566 SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS);
567 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
570 case SCTP_RTXR_PMTUD:
571 SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS);
573 case SCTP_RTXR_T1_RTX:
574 SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS);
575 transport->asoc->init_retries++;
581 sctp_retransmit_mark(q, transport, reason);
583 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
584 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
585 * following the procedures outlined in C1 - C5.
587 if (reason == SCTP_RTXR_T3_RTX)
588 q->asoc->stream.si->generate_ftsn(q, q->asoc->ctsn_ack_point);
590 /* Flush the queues only on timeout, since fast_rtx is only
591 * triggered during sack processing and the queue
592 * will be flushed at the end.
594 if (reason != SCTP_RTXR_FAST_RTX)
595 sctp_outq_flush(q, /* rtx_timeout */ 1, GFP_ATOMIC);
599 * Transmit DATA chunks on the retransmit queue. Upon return from
600 * __sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
601 * need to be transmitted by the caller.
602 * We assume that pkt->transport has already been set.
604 * The return value is a normal kernel error return value.
606 static int __sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
607 int rtx_timeout, int *start_timer, gfp_t gfp)
609 struct sctp_transport *transport = pkt->transport;
610 struct sctp_chunk *chunk, *chunk1;
611 struct list_head *lqueue;
612 enum sctp_xmit status;
618 lqueue = &q->retransmit;
619 fast_rtx = q->fast_rtx;
621 /* This loop handles time-out retransmissions, fast retransmissions,
622 * and retransmissions due to opening of whindow.
624 * RFC 2960 6.3.3 Handle T3-rtx Expiration
626 * E3) Determine how many of the earliest (i.e., lowest TSN)
627 * outstanding DATA chunks for the address for which the
628 * T3-rtx has expired will fit into a single packet, subject
629 * to the MTU constraint for the path corresponding to the
630 * destination transport address to which the retransmission
631 * is being sent (this may be different from the address for
632 * which the timer expires [see Section 6.4]). Call this value
633 * K. Bundle and retransmit those K DATA chunks in a single
634 * packet to the destination endpoint.
636 * [Just to be painfully clear, if we are retransmitting
637 * because a timeout just happened, we should send only ONE
638 * packet of retransmitted data.]
640 * For fast retransmissions we also send only ONE packet. However,
641 * if we are just flushing the queue due to open window, we'll
642 * try to send as much as possible.
644 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
645 /* If the chunk is abandoned, move it to abandoned list. */
646 if (sctp_chunk_abandoned(chunk)) {
647 list_del_init(&chunk->transmitted_list);
648 sctp_insert_list(&q->abandoned,
649 &chunk->transmitted_list);
653 /* Make sure that Gap Acked TSNs are not retransmitted. A
654 * simple approach is just to move such TSNs out of the
655 * way and into a 'transmitted' queue and skip to the
658 if (chunk->tsn_gap_acked) {
659 list_move_tail(&chunk->transmitted_list,
660 &transport->transmitted);
664 /* If we are doing fast retransmit, ignore non-fast_rtransmit
667 if (fast_rtx && !chunk->fast_retransmit)
671 /* Attempt to append this chunk to the packet. */
672 status = sctp_packet_append_chunk(pkt, chunk);
675 case SCTP_XMIT_PMTU_FULL:
676 if (!pkt->has_data && !pkt->has_cookie_echo) {
677 /* If this packet did not contain DATA then
678 * retransmission did not happen, so do it
679 * again. We'll ignore the error here since
680 * control chunks are already freed so there
681 * is nothing we can do.
683 sctp_packet_transmit(pkt, gfp);
687 /* Send this packet. */
688 error = sctp_packet_transmit(pkt, gfp);
690 /* If we are retransmitting, we should only
691 * send a single packet.
692 * Otherwise, try appending this chunk again.
694 if (rtx_timeout || fast_rtx)
699 /* Bundle next chunk in the next round. */
702 case SCTP_XMIT_RWND_FULL:
703 /* Send this packet. */
704 error = sctp_packet_transmit(pkt, gfp);
706 /* Stop sending DATA as there is no more room
712 case SCTP_XMIT_DELAY:
713 /* Send this packet. */
714 error = sctp_packet_transmit(pkt, gfp);
716 /* Stop sending DATA because of nagle delay. */
721 /* The append was successful, so add this chunk to
722 * the transmitted list.
724 list_move_tail(&chunk->transmitted_list,
725 &transport->transmitted);
727 /* Mark the chunk as ineligible for fast retransmit
728 * after it is retransmitted.
730 if (chunk->fast_retransmit == SCTP_NEED_FRTX)
731 chunk->fast_retransmit = SCTP_DONT_FRTX;
733 q->asoc->stats.rtxchunks++;
737 /* Set the timer if there were no errors */
738 if (!error && !timer)
745 /* If we are here due to a retransmit timeout or a fast
746 * retransmit and if there are any chunks left in the retransmit
747 * queue that could not fit in the PMTU sized packet, they need
748 * to be marked as ineligible for a subsequent fast retransmit.
750 if (rtx_timeout || fast_rtx) {
751 list_for_each_entry(chunk1, lqueue, transmitted_list) {
752 if (chunk1->fast_retransmit == SCTP_NEED_FRTX)
753 chunk1->fast_retransmit = SCTP_DONT_FRTX;
757 *start_timer = timer;
759 /* Clear fast retransmit hint */
766 /* Cork the outqueue so queued chunks are really queued. */
767 void sctp_outq_uncork(struct sctp_outq *q, gfp_t gfp)
772 sctp_outq_flush(q, 0, gfp);
775 static int sctp_packet_singleton(struct sctp_transport *transport,
776 struct sctp_chunk *chunk, gfp_t gfp)
778 const struct sctp_association *asoc = transport->asoc;
779 const __u16 sport = asoc->base.bind_addr.port;
780 const __u16 dport = asoc->peer.port;
781 const __u32 vtag = asoc->peer.i.init_tag;
782 struct sctp_packet singleton;
784 sctp_packet_init(&singleton, transport, sport, dport);
785 sctp_packet_config(&singleton, vtag, 0);
786 sctp_packet_append_chunk(&singleton, chunk);
787 return sctp_packet_transmit(&singleton, gfp);
790 /* Struct to hold the context during sctp outq flush */
791 struct sctp_flush_ctx {
793 /* Current transport being used. It's NOT the same as curr active one */
794 struct sctp_transport *transport;
795 /* These transports have chunks to send. */
796 struct list_head transport_list;
797 struct sctp_association *asoc;
798 /* Packet on the current transport above */
799 struct sctp_packet *packet;
803 /* transport: current transport */
804 static void sctp_outq_select_transport(struct sctp_flush_ctx *ctx,
805 struct sctp_chunk *chunk)
807 struct sctp_transport *new_transport = chunk->transport;
809 if (!new_transport) {
810 if (!sctp_chunk_is_data(chunk)) {
811 /* If we have a prior transport pointer, see if
812 * the destination address of the chunk
813 * matches the destination address of the
814 * current transport. If not a match, then
815 * try to look up the transport with a given
816 * destination address. We do this because
817 * after processing ASCONFs, we may have new
818 * transports created.
820 if (ctx->transport && sctp_cmp_addr_exact(&chunk->dest,
821 &ctx->transport->ipaddr))
822 new_transport = ctx->transport;
824 new_transport = sctp_assoc_lookup_paddr(ctx->asoc,
828 /* if we still don't have a new transport, then
829 * use the current active path.
832 new_transport = ctx->asoc->peer.active_path;
836 switch (new_transport->state) {
838 case SCTP_UNCONFIRMED:
840 /* If the chunk is Heartbeat or Heartbeat Ack,
841 * send it to chunk->transport, even if it's
844 * 3.3.6 Heartbeat Acknowledgement:
846 * A HEARTBEAT ACK is always sent to the source IP
847 * address of the IP datagram containing the
848 * HEARTBEAT chunk to which this ack is responding.
851 * ASCONF_ACKs also must be sent to the source.
853 type = chunk->chunk_hdr->type;
854 if (type != SCTP_CID_HEARTBEAT &&
855 type != SCTP_CID_HEARTBEAT_ACK &&
856 type != SCTP_CID_ASCONF_ACK)
857 new_transport = ctx->asoc->peer.active_path;
864 /* Are we switching transports? Take care of transport locks. */
865 if (new_transport != ctx->transport) {
866 ctx->transport = new_transport;
867 ctx->packet = &ctx->transport->packet;
869 if (list_empty(&ctx->transport->send_ready))
870 list_add_tail(&ctx->transport->send_ready,
871 &ctx->transport_list);
873 sctp_packet_config(ctx->packet,
874 ctx->asoc->peer.i.init_tag,
875 ctx->asoc->peer.ecn_capable);
876 /* We've switched transports, so apply the
877 * Burst limit to the new transport.
879 sctp_transport_burst_limited(ctx->transport);
883 static void sctp_outq_flush_ctrl(struct sctp_flush_ctx *ctx)
885 struct sctp_chunk *chunk, *tmp;
886 enum sctp_xmit status;
887 int one_packet, error;
889 list_for_each_entry_safe(chunk, tmp, &ctx->q->control_chunk_list, list) {
893 * F1) This means that until such time as the ASCONF
894 * containing the add is acknowledged, the sender MUST
895 * NOT use the new IP address as a source for ANY SCTP
896 * packet except on carrying an ASCONF Chunk.
898 if (ctx->asoc->src_out_of_asoc_ok &&
899 chunk->chunk_hdr->type != SCTP_CID_ASCONF)
902 list_del_init(&chunk->list);
904 /* Pick the right transport to use. Should always be true for
905 * the first chunk as we don't have a transport by then.
907 sctp_outq_select_transport(ctx, chunk);
909 switch (chunk->chunk_hdr->type) {
912 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
913 * COMPLETE with any other chunks. [Send them immediately.]
916 case SCTP_CID_INIT_ACK:
917 case SCTP_CID_SHUTDOWN_COMPLETE:
918 error = sctp_packet_singleton(ctx->transport, chunk,
921 ctx->asoc->base.sk->sk_err = -error;
927 if (sctp_test_T_bit(chunk))
928 ctx->packet->vtag = ctx->asoc->c.my_vtag;
931 /* The following chunks are "response" chunks, i.e.
932 * they are generated in response to something we
933 * received. If we are sending these, then we can
934 * send only 1 packet containing these chunks.
936 case SCTP_CID_HEARTBEAT_ACK:
937 case SCTP_CID_SHUTDOWN_ACK:
938 case SCTP_CID_COOKIE_ACK:
939 case SCTP_CID_COOKIE_ECHO:
941 case SCTP_CID_ECN_CWR:
942 case SCTP_CID_ASCONF_ACK:
947 case SCTP_CID_HEARTBEAT:
948 case SCTP_CID_SHUTDOWN:
949 case SCTP_CID_ECN_ECNE:
950 case SCTP_CID_ASCONF:
951 case SCTP_CID_FWD_TSN:
952 case SCTP_CID_I_FWD_TSN:
953 case SCTP_CID_RECONF:
954 status = sctp_packet_transmit_chunk(ctx->packet, chunk,
955 one_packet, ctx->gfp);
956 if (status != SCTP_XMIT_OK) {
957 /* put the chunk back */
958 list_add(&chunk->list, &ctx->q->control_chunk_list);
962 ctx->asoc->stats.octrlchunks++;
963 /* PR-SCTP C5) If a FORWARD TSN is sent, the
964 * sender MUST assure that at least one T3-rtx
967 if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN ||
968 chunk->chunk_hdr->type == SCTP_CID_I_FWD_TSN) {
969 sctp_transport_reset_t3_rtx(ctx->transport);
970 ctx->transport->last_time_sent = jiffies;
973 if (chunk == ctx->asoc->strreset_chunk)
974 sctp_transport_reset_reconf_timer(ctx->transport);
979 /* We built a chunk with an illegal type! */
985 /* Returns false if new data shouldn't be sent */
986 static bool sctp_outq_flush_rtx(struct sctp_flush_ctx *ctx,
989 int error, start_timer = 0;
991 if (ctx->asoc->peer.retran_path->state == SCTP_UNCONFIRMED)
994 if (ctx->transport != ctx->asoc->peer.retran_path) {
995 /* Switch transports & prepare the packet. */
996 ctx->transport = ctx->asoc->peer.retran_path;
997 ctx->packet = &ctx->transport->packet;
999 if (list_empty(&ctx->transport->send_ready))
1000 list_add_tail(&ctx->transport->send_ready,
1001 &ctx->transport_list);
1003 sctp_packet_config(ctx->packet, ctx->asoc->peer.i.init_tag,
1004 ctx->asoc->peer.ecn_capable);
1007 error = __sctp_outq_flush_rtx(ctx->q, ctx->packet, rtx_timeout,
1008 &start_timer, ctx->gfp);
1010 ctx->asoc->base.sk->sk_err = -error;
1013 sctp_transport_reset_t3_rtx(ctx->transport);
1014 ctx->transport->last_time_sent = jiffies;
1017 /* This can happen on COOKIE-ECHO resend. Only
1018 * one chunk can get bundled with a COOKIE-ECHO.
1020 if (ctx->packet->has_cookie_echo)
1023 /* Don't send new data if there is still data
1024 * waiting to retransmit.
1026 if (!list_empty(&ctx->q->retransmit))
1032 static void sctp_outq_flush_data(struct sctp_flush_ctx *ctx,
1035 struct sctp_chunk *chunk;
1036 enum sctp_xmit status;
1038 /* Is it OK to send data chunks? */
1039 switch (ctx->asoc->state) {
1040 case SCTP_STATE_COOKIE_ECHOED:
1041 /* Only allow bundling when this packet has a COOKIE-ECHO
1044 if (!ctx->packet || !ctx->packet->has_cookie_echo)
1048 case SCTP_STATE_ESTABLISHED:
1049 case SCTP_STATE_SHUTDOWN_PENDING:
1050 case SCTP_STATE_SHUTDOWN_RECEIVED:
1058 /* RFC 2960 6.1 Transmission of DATA Chunks
1060 * C) When the time comes for the sender to transmit,
1061 * before sending new DATA chunks, the sender MUST
1062 * first transmit any outstanding DATA chunks which
1063 * are marked for retransmission (limited by the
1066 if (!list_empty(&ctx->q->retransmit) &&
1067 !sctp_outq_flush_rtx(ctx, rtx_timeout))
1070 /* Apply Max.Burst limitation to the current transport in
1071 * case it will be used for new data. We are going to
1072 * rest it before we return, but we want to apply the limit
1073 * to the currently queued data.
1076 sctp_transport_burst_limited(ctx->transport);
1078 /* Finally, transmit new packets. */
1079 while ((chunk = sctp_outq_dequeue_data(ctx->q)) != NULL) {
1080 __u32 sid = ntohs(chunk->subh.data_hdr->stream);
1081 __u8 stream_state = SCTP_SO(&ctx->asoc->stream, sid)->state;
1083 /* Has this chunk expired? */
1084 if (sctp_chunk_abandoned(chunk)) {
1085 sctp_sched_dequeue_done(ctx->q, chunk);
1086 sctp_chunk_fail(chunk, 0);
1087 sctp_chunk_free(chunk);
1091 if (stream_state == SCTP_STREAM_CLOSED) {
1092 sctp_outq_head_data(ctx->q, chunk);
1096 sctp_outq_select_transport(ctx, chunk);
1098 pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p skb->users:%d\n",
1099 __func__, ctx->q, chunk, chunk && chunk->chunk_hdr ?
1100 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
1101 "illegal chunk", ntohl(chunk->subh.data_hdr->tsn),
1102 chunk->skb ? chunk->skb->head : NULL, chunk->skb ?
1103 refcount_read(&chunk->skb->users) : -1);
1105 /* Add the chunk to the packet. */
1106 status = sctp_packet_transmit_chunk(ctx->packet, chunk, 0,
1108 if (status != SCTP_XMIT_OK) {
1109 /* We could not append this chunk, so put
1110 * the chunk back on the output queue.
1112 pr_debug("%s: could not transmit tsn:0x%x, status:%d\n",
1113 __func__, ntohl(chunk->subh.data_hdr->tsn),
1116 sctp_outq_head_data(ctx->q, chunk);
1120 /* The sender is in the SHUTDOWN-PENDING state,
1121 * The sender MAY set the I-bit in the DATA
1124 if (ctx->asoc->state == SCTP_STATE_SHUTDOWN_PENDING)
1125 chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM;
1126 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
1127 ctx->asoc->stats.ouodchunks++;
1129 ctx->asoc->stats.oodchunks++;
1131 /* Only now it's safe to consider this
1132 * chunk as sent, sched-wise.
1134 sctp_sched_dequeue_done(ctx->q, chunk);
1136 list_add_tail(&chunk->transmitted_list,
1137 &ctx->transport->transmitted);
1139 sctp_transport_reset_t3_rtx(ctx->transport);
1140 ctx->transport->last_time_sent = jiffies;
1142 /* Only let one DATA chunk get bundled with a
1143 * COOKIE-ECHO chunk.
1145 if (ctx->packet->has_cookie_echo)
1150 static void sctp_outq_flush_transports(struct sctp_flush_ctx *ctx)
1152 struct list_head *ltransport;
1153 struct sctp_packet *packet;
1154 struct sctp_transport *t;
1157 while ((ltransport = sctp_list_dequeue(&ctx->transport_list)) != NULL) {
1158 t = list_entry(ltransport, struct sctp_transport, send_ready);
1159 packet = &t->packet;
1160 if (!sctp_packet_empty(packet)) {
1161 error = sctp_packet_transmit(packet, ctx->gfp);
1163 ctx->q->asoc->base.sk->sk_err = -error;
1166 /* Clear the burst limited state, if any */
1167 sctp_transport_burst_reset(t);
1171 /* Try to flush an outqueue.
1173 * Description: Send everything in q which we legally can, subject to
1174 * congestion limitations.
1175 * * Note: This function can be called from multiple contexts so appropriate
1176 * locking concerns must be made. Today we use the sock lock to protect
1180 static void sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp)
1182 struct sctp_flush_ctx ctx = {
1185 .transport_list = LIST_HEAD_INIT(ctx.transport_list),
1193 * When bundling control chunks with DATA chunks, an
1194 * endpoint MUST place control chunks first in the outbound
1195 * SCTP packet. The transmitter MUST transmit DATA chunks
1196 * within a SCTP packet in increasing order of TSN.
1200 sctp_outq_flush_ctrl(&ctx);
1202 if (q->asoc->src_out_of_asoc_ok)
1203 goto sctp_flush_out;
1205 sctp_outq_flush_data(&ctx, rtx_timeout);
1209 sctp_outq_flush_transports(&ctx);
1212 /* Update unack_data based on the incoming SACK chunk */
1213 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1214 struct sctp_sackhdr *sack)
1216 union sctp_sack_variable *frags;
1220 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1222 frags = sack->variable;
1223 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1224 unack_data -= ((ntohs(frags[i].gab.end) -
1225 ntohs(frags[i].gab.start) + 1));
1228 assoc->unack_data = unack_data;
1231 /* This is where we REALLY process a SACK.
1233 * Process the SACK against the outqueue. Mostly, this just frees
1234 * things off the transmitted queue.
1236 int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk)
1238 struct sctp_association *asoc = q->asoc;
1239 struct sctp_sackhdr *sack = chunk->subh.sack_hdr;
1240 struct sctp_transport *transport;
1241 struct sctp_chunk *tchunk = NULL;
1242 struct list_head *lchunk, *transport_list, *temp;
1243 union sctp_sack_variable *frags = sack->variable;
1244 __u32 sack_ctsn, ctsn, tsn;
1245 __u32 highest_tsn, highest_new_tsn;
1247 unsigned int outstanding;
1248 struct sctp_transport *primary = asoc->peer.primary_path;
1249 int count_of_newacks = 0;
1253 /* Grab the association's destination address list. */
1254 transport_list = &asoc->peer.transport_addr_list;
1256 sack_ctsn = ntohl(sack->cum_tsn_ack);
1257 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1258 asoc->stats.gapcnt += gap_ack_blocks;
1260 * SFR-CACC algorithm:
1261 * On receipt of a SACK the sender SHOULD execute the
1262 * following statements.
1264 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1265 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1266 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1268 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1269 * is set the receiver of the SACK MUST take the following actions:
1271 * A) Initialize the cacc_saw_newack to 0 for all destination
1274 * Only bother if changeover_active is set. Otherwise, this is
1275 * totally suboptimal to do on every SACK.
1277 if (primary->cacc.changeover_active) {
1278 u8 clear_cycling = 0;
1280 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1281 primary->cacc.changeover_active = 0;
1285 if (clear_cycling || gap_ack_blocks) {
1286 list_for_each_entry(transport, transport_list,
1289 transport->cacc.cycling_changeover = 0;
1291 transport->cacc.cacc_saw_newack = 0;
1296 /* Get the highest TSN in the sack. */
1297 highest_tsn = sack_ctsn;
1299 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1301 if (TSN_lt(asoc->highest_sacked, highest_tsn))
1302 asoc->highest_sacked = highest_tsn;
1304 highest_new_tsn = sack_ctsn;
1306 /* Run through the retransmit queue. Credit bytes received
1307 * and free those chunks that we can.
1309 sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn);
1311 /* Run through the transmitted queue.
1312 * Credit bytes received and free those chunks which we can.
1314 * This is a MASSIVE candidate for optimization.
1316 list_for_each_entry(transport, transport_list, transports) {
1317 sctp_check_transmitted(q, &transport->transmitted,
1318 transport, &chunk->source, sack,
1321 * SFR-CACC algorithm:
1322 * C) Let count_of_newacks be the number of
1323 * destinations for which cacc_saw_newack is set.
1325 if (transport->cacc.cacc_saw_newack)
1329 /* Move the Cumulative TSN Ack Point if appropriate. */
1330 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) {
1331 asoc->ctsn_ack_point = sack_ctsn;
1335 if (gap_ack_blocks) {
1337 if (asoc->fast_recovery && accum_moved)
1338 highest_new_tsn = highest_tsn;
1340 list_for_each_entry(transport, transport_list, transports)
1341 sctp_mark_missing(q, &transport->transmitted, transport,
1342 highest_new_tsn, count_of_newacks);
1345 /* Update unack_data field in the assoc. */
1346 sctp_sack_update_unack_data(asoc, sack);
1348 ctsn = asoc->ctsn_ack_point;
1350 /* Throw away stuff rotting on the sack queue. */
1351 list_for_each_safe(lchunk, temp, &q->sacked) {
1352 tchunk = list_entry(lchunk, struct sctp_chunk,
1354 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1355 if (TSN_lte(tsn, ctsn)) {
1356 list_del_init(&tchunk->transmitted_list);
1357 if (asoc->peer.prsctp_capable &&
1358 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags))
1359 asoc->sent_cnt_removable--;
1360 sctp_chunk_free(tchunk);
1364 /* ii) Set rwnd equal to the newly received a_rwnd minus the
1365 * number of bytes still outstanding after processing the
1366 * Cumulative TSN Ack and the Gap Ack Blocks.
1369 sack_a_rwnd = ntohl(sack->a_rwnd);
1370 asoc->peer.zero_window_announced = !sack_a_rwnd;
1371 outstanding = q->outstanding_bytes;
1373 if (outstanding < sack_a_rwnd)
1374 sack_a_rwnd -= outstanding;
1378 asoc->peer.rwnd = sack_a_rwnd;
1380 asoc->stream.si->generate_ftsn(q, sack_ctsn);
1382 pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__, sack_ctsn);
1383 pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, "
1384 "advertised peer ack point:0x%x\n", __func__, asoc, ctsn,
1385 asoc->adv_peer_ack_point);
1387 return sctp_outq_is_empty(q);
1390 /* Is the outqueue empty?
1391 * The queue is empty when we have not pending data, no in-flight data
1392 * and nothing pending retransmissions.
1394 int sctp_outq_is_empty(const struct sctp_outq *q)
1396 return q->out_qlen == 0 && q->outstanding_bytes == 0 &&
1397 list_empty(&q->retransmit);
1400 /********************************************************************
1401 * 2nd Level Abstractions
1402 ********************************************************************/
1404 /* Go through a transport's transmitted list or the association's retransmit
1405 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1406 * The retransmit list will not have an associated transport.
1408 * I added coherent debug information output. --xguo
1410 * Instead of printing 'sacked' or 'kept' for each TSN on the
1411 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1412 * KEPT TSN6-TSN7, etc.
1414 static void sctp_check_transmitted(struct sctp_outq *q,
1415 struct list_head *transmitted_queue,
1416 struct sctp_transport *transport,
1417 union sctp_addr *saddr,
1418 struct sctp_sackhdr *sack,
1419 __u32 *highest_new_tsn_in_sack)
1421 struct list_head *lchunk;
1422 struct sctp_chunk *tchunk;
1423 struct list_head tlist;
1427 __u8 restart_timer = 0;
1428 int bytes_acked = 0;
1429 int migrate_bytes = 0;
1430 bool forward_progress = false;
1432 sack_ctsn = ntohl(sack->cum_tsn_ack);
1434 INIT_LIST_HEAD(&tlist);
1436 /* The while loop will skip empty transmitted queues. */
1437 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1438 tchunk = list_entry(lchunk, struct sctp_chunk,
1441 if (sctp_chunk_abandoned(tchunk)) {
1442 /* Move the chunk to abandoned list. */
1443 sctp_insert_list(&q->abandoned, lchunk);
1445 /* If this chunk has not been acked, stop
1446 * considering it as 'outstanding'.
1448 if (transmitted_queue != &q->retransmit &&
1449 !tchunk->tsn_gap_acked) {
1450 if (tchunk->transport)
1451 tchunk->transport->flight_size -=
1452 sctp_data_size(tchunk);
1453 q->outstanding_bytes -= sctp_data_size(tchunk);
1458 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1459 if (sctp_acked(sack, tsn)) {
1460 /* If this queue is the retransmit queue, the
1461 * retransmit timer has already reclaimed
1462 * the outstanding bytes for this chunk, so only
1463 * count bytes associated with a transport.
1465 if (transport && !tchunk->tsn_gap_acked) {
1466 /* If this chunk is being used for RTT
1467 * measurement, calculate the RTT and update
1468 * the RTO using this value.
1470 * 6.3.1 C5) Karn's algorithm: RTT measurements
1471 * MUST NOT be made using packets that were
1472 * retransmitted (and thus for which it is
1473 * ambiguous whether the reply was for the
1474 * first instance of the packet or a later
1477 if (!sctp_chunk_retransmitted(tchunk) &&
1478 tchunk->rtt_in_progress) {
1479 tchunk->rtt_in_progress = 0;
1480 rtt = jiffies - tchunk->sent_at;
1481 sctp_transport_update_rto(transport,
1485 if (TSN_lte(tsn, sack_ctsn)) {
1487 * SFR-CACC algorithm:
1488 * 2) If the SACK contains gap acks
1489 * and the flag CHANGEOVER_ACTIVE is
1490 * set the receiver of the SACK MUST
1491 * take the following action:
1493 * B) For each TSN t being acked that
1494 * has not been acked in any SACK so
1495 * far, set cacc_saw_newack to 1 for
1496 * the destination that the TSN was
1499 if (sack->num_gap_ack_blocks &&
1500 q->asoc->peer.primary_path->cacc.
1502 transport->cacc.cacc_saw_newack
1507 /* If the chunk hasn't been marked as ACKED,
1508 * mark it and account bytes_acked if the
1509 * chunk had a valid transport (it will not
1510 * have a transport if ASCONF had deleted it
1511 * while DATA was outstanding).
1513 if (!tchunk->tsn_gap_acked) {
1514 tchunk->tsn_gap_acked = 1;
1515 if (TSN_lt(*highest_new_tsn_in_sack, tsn))
1516 *highest_new_tsn_in_sack = tsn;
1517 bytes_acked += sctp_data_size(tchunk);
1518 if (!tchunk->transport)
1519 migrate_bytes += sctp_data_size(tchunk);
1520 forward_progress = true;
1523 if (TSN_lte(tsn, sack_ctsn)) {
1524 /* RFC 2960 6.3.2 Retransmission Timer Rules
1526 * R3) Whenever a SACK is received
1527 * that acknowledges the DATA chunk
1528 * with the earliest outstanding TSN
1529 * for that address, restart T3-rtx
1530 * timer for that address with its
1534 forward_progress = true;
1536 list_add_tail(&tchunk->transmitted_list,
1539 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1540 * M2) Each time a SACK arrives reporting
1541 * 'Stray DATA chunk(s)' record the highest TSN
1542 * reported as newly acknowledged, call this
1543 * value 'HighestTSNinSack'. A newly
1544 * acknowledged DATA chunk is one not
1545 * previously acknowledged in a SACK.
1547 * When the SCTP sender of data receives a SACK
1548 * chunk that acknowledges, for the first time,
1549 * the receipt of a DATA chunk, all the still
1550 * unacknowledged DATA chunks whose TSN is
1551 * older than that newly acknowledged DATA
1552 * chunk, are qualified as 'Stray DATA chunks'.
1554 list_add_tail(lchunk, &tlist);
1557 if (tchunk->tsn_gap_acked) {
1558 pr_debug("%s: receiver reneged on data TSN:0x%x\n",
1561 tchunk->tsn_gap_acked = 0;
1563 if (tchunk->transport)
1564 bytes_acked -= sctp_data_size(tchunk);
1566 /* RFC 2960 6.3.2 Retransmission Timer Rules
1568 * R4) Whenever a SACK is received missing a
1569 * TSN that was previously acknowledged via a
1570 * Gap Ack Block, start T3-rtx for the
1571 * destination address to which the DATA
1572 * chunk was originally
1573 * transmitted if it is not already running.
1578 list_add_tail(lchunk, &tlist);
1584 struct sctp_association *asoc = transport->asoc;
1586 /* We may have counted DATA that was migrated
1587 * to this transport due to DEL-IP operation.
1588 * Subtract those bytes, since the were never
1589 * send on this transport and shouldn't be
1590 * credited to this transport.
1592 bytes_acked -= migrate_bytes;
1594 /* 8.2. When an outstanding TSN is acknowledged,
1595 * the endpoint shall clear the error counter of
1596 * the destination transport address to which the
1597 * DATA chunk was last sent.
1598 * The association's overall error counter is
1601 transport->error_count = 0;
1602 transport->asoc->overall_error_count = 0;
1603 forward_progress = true;
1606 * While in SHUTDOWN PENDING, we may have started
1607 * the T5 shutdown guard timer after reaching the
1608 * retransmission limit. Stop that timer as soon
1609 * as the receiver acknowledged any data.
1611 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING &&
1612 del_timer(&asoc->timers
1613 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]))
1614 sctp_association_put(asoc);
1616 /* Mark the destination transport address as
1617 * active if it is not so marked.
1619 if ((transport->state == SCTP_INACTIVE ||
1620 transport->state == SCTP_UNCONFIRMED) &&
1621 sctp_cmp_addr_exact(&transport->ipaddr, saddr)) {
1622 sctp_assoc_control_transport(
1626 SCTP_RECEIVED_SACK);
1629 sctp_transport_raise_cwnd(transport, sack_ctsn,
1632 transport->flight_size -= bytes_acked;
1633 if (transport->flight_size == 0)
1634 transport->partial_bytes_acked = 0;
1635 q->outstanding_bytes -= bytes_acked + migrate_bytes;
1637 /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1638 * When a sender is doing zero window probing, it
1639 * should not timeout the association if it continues
1640 * to receive new packets from the receiver. The
1641 * reason is that the receiver MAY keep its window
1642 * closed for an indefinite time.
1643 * A sender is doing zero window probing when the
1644 * receiver's advertised window is zero, and there is
1645 * only one data chunk in flight to the receiver.
1647 * Allow the association to timeout while in SHUTDOWN
1648 * PENDING or SHUTDOWN RECEIVED in case the receiver
1649 * stays in zero window mode forever.
1651 if (!q->asoc->peer.rwnd &&
1652 !list_empty(&tlist) &&
1653 (sack_ctsn+2 == q->asoc->next_tsn) &&
1654 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) {
1655 pr_debug("%s: sack received for zero window "
1656 "probe:%u\n", __func__, sack_ctsn);
1658 q->asoc->overall_error_count = 0;
1659 transport->error_count = 0;
1663 /* RFC 2960 6.3.2 Retransmission Timer Rules
1665 * R2) Whenever all outstanding data sent to an address have
1666 * been acknowledged, turn off the T3-rtx timer of that
1669 if (!transport->flight_size) {
1670 if (del_timer(&transport->T3_rtx_timer))
1671 sctp_transport_put(transport);
1672 } else if (restart_timer) {
1673 if (!mod_timer(&transport->T3_rtx_timer,
1674 jiffies + transport->rto))
1675 sctp_transport_hold(transport);
1678 if (forward_progress) {
1680 sctp_transport_dst_confirm(transport);
1684 list_splice(&tlist, transmitted_queue);
1687 /* Mark chunks as missing and consequently may get retransmitted. */
1688 static void sctp_mark_missing(struct sctp_outq *q,
1689 struct list_head *transmitted_queue,
1690 struct sctp_transport *transport,
1691 __u32 highest_new_tsn_in_sack,
1692 int count_of_newacks)
1694 struct sctp_chunk *chunk;
1696 char do_fast_retransmit = 0;
1697 struct sctp_association *asoc = q->asoc;
1698 struct sctp_transport *primary = asoc->peer.primary_path;
1700 list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1702 tsn = ntohl(chunk->subh.data_hdr->tsn);
1704 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1705 * 'Unacknowledged TSN's', if the TSN number of an
1706 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1707 * value, increment the 'TSN.Missing.Report' count on that
1708 * chunk if it has NOT been fast retransmitted or marked for
1709 * fast retransmit already.
1711 if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
1712 !chunk->tsn_gap_acked &&
1713 TSN_lt(tsn, highest_new_tsn_in_sack)) {
1715 /* SFR-CACC may require us to skip marking
1716 * this chunk as missing.
1718 if (!transport || !sctp_cacc_skip(primary,
1720 count_of_newacks, tsn)) {
1721 chunk->tsn_missing_report++;
1723 pr_debug("%s: tsn:0x%x missing counter:%d\n",
1724 __func__, tsn, chunk->tsn_missing_report);
1728 * M4) If any DATA chunk is found to have a
1729 * 'TSN.Missing.Report'
1730 * value larger than or equal to 3, mark that chunk for
1731 * retransmission and start the fast retransmit procedure.
1734 if (chunk->tsn_missing_report >= 3) {
1735 chunk->fast_retransmit = SCTP_NEED_FRTX;
1736 do_fast_retransmit = 1;
1741 if (do_fast_retransmit)
1742 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1744 pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, "
1745 "flight_size:%d, pba:%d\n", __func__, transport,
1746 transport->cwnd, transport->ssthresh,
1747 transport->flight_size, transport->partial_bytes_acked);
1751 /* Is the given TSN acked by this packet? */
1752 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1754 __u32 ctsn = ntohl(sack->cum_tsn_ack);
1755 union sctp_sack_variable *frags;
1756 __u16 tsn_offset, blocks;
1759 if (TSN_lte(tsn, ctsn))
1762 /* 3.3.4 Selective Acknowledgment (SACK) (3):
1765 * These fields contain the Gap Ack Blocks. They are repeated
1766 * for each Gap Ack Block up to the number of Gap Ack Blocks
1767 * defined in the Number of Gap Ack Blocks field. All DATA
1768 * chunks with TSNs greater than or equal to (Cumulative TSN
1769 * Ack + Gap Ack Block Start) and less than or equal to
1770 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1771 * Block are assumed to have been received correctly.
1774 frags = sack->variable;
1775 blocks = ntohs(sack->num_gap_ack_blocks);
1776 tsn_offset = tsn - ctsn;
1777 for (i = 0; i < blocks; ++i) {
1778 if (tsn_offset >= ntohs(frags[i].gab.start) &&
1779 tsn_offset <= ntohs(frags[i].gab.end))
1788 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1789 int nskips, __be16 stream)
1793 for (i = 0; i < nskips; i++) {
1794 if (skiplist[i].stream == stream)
1800 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1801 void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1803 struct sctp_association *asoc = q->asoc;
1804 struct sctp_chunk *ftsn_chunk = NULL;
1805 struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1809 struct sctp_chunk *chunk;
1810 struct list_head *lchunk, *temp;
1812 if (!asoc->peer.prsctp_capable)
1815 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1818 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1819 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1821 if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1822 asoc->adv_peer_ack_point = ctsn;
1824 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1825 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1826 * the chunk next in the out-queue space is marked as "abandoned" as
1827 * shown in the following example:
1829 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1830 * and the Advanced.Peer.Ack.Point is updated to this value:
1832 * out-queue at the end of ==> out-queue after Adv.Ack.Point
1833 * normal SACK processing local advancement
1835 * Adv.Ack.Pt-> 102 acked 102 acked
1836 * 103 abandoned 103 abandoned
1837 * 104 abandoned Adv.Ack.P-> 104 abandoned
1839 * 106 acked 106 acked
1842 * In this example, the data sender successfully advanced the
1843 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1845 list_for_each_safe(lchunk, temp, &q->abandoned) {
1846 chunk = list_entry(lchunk, struct sctp_chunk,
1848 tsn = ntohl(chunk->subh.data_hdr->tsn);
1850 /* Remove any chunks in the abandoned queue that are acked by
1853 if (TSN_lte(tsn, ctsn)) {
1854 list_del_init(lchunk);
1855 sctp_chunk_free(chunk);
1857 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1858 asoc->adv_peer_ack_point = tsn;
1859 if (chunk->chunk_hdr->flags &
1860 SCTP_DATA_UNORDERED)
1862 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1864 chunk->subh.data_hdr->stream);
1865 ftsn_skip_arr[skip_pos].stream =
1866 chunk->subh.data_hdr->stream;
1867 ftsn_skip_arr[skip_pos].ssn =
1868 chunk->subh.data_hdr->ssn;
1869 if (skip_pos == nskips)
1878 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1879 * is greater than the Cumulative TSN ACK carried in the received
1880 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1881 * chunk containing the latest value of the
1882 * "Advanced.Peer.Ack.Point".
1884 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1885 * list each stream and sequence number in the forwarded TSN. This
1886 * information will enable the receiver to easily find any
1887 * stranded TSN's waiting on stream reorder queues. Each stream
1888 * SHOULD only be reported once; this means that if multiple
1889 * abandoned messages occur in the same stream then only the
1890 * highest abandoned stream sequence number is reported. If the
1891 * total size of the FORWARD TSN does NOT fit in a single MTU then
1892 * the sender of the FORWARD TSN SHOULD lower the
1893 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1896 if (asoc->adv_peer_ack_point > ctsn)
1897 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1898 nskips, &ftsn_skip_arr[0]);
1901 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1902 SCTP_INC_STATS(sock_net(asoc->base.sk), SCTP_MIB_OUTCTRLCHUNKS);