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>
54 /* Declare internal functions here. */
55 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn);
56 static void sctp_check_transmitted(struct sctp_outq *q,
57 struct list_head *transmitted_queue,
58 struct sctp_transport *transport,
59 union sctp_addr *saddr,
60 struct sctp_sackhdr *sack,
61 __u32 *highest_new_tsn);
63 static void sctp_mark_missing(struct sctp_outq *q,
64 struct list_head *transmitted_queue,
65 struct sctp_transport *transport,
66 __u32 highest_new_tsn,
67 int count_of_newacks);
69 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn);
71 static void sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp);
73 /* Add data to the front of the queue. */
74 static inline void sctp_outq_head_data(struct sctp_outq *q,
75 struct sctp_chunk *ch)
77 list_add(&ch->list, &q->out_chunk_list);
78 q->out_qlen += ch->skb->len;
81 /* Take data from the front of the queue. */
82 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
84 struct sctp_chunk *ch = NULL;
86 if (!list_empty(&q->out_chunk_list)) {
87 struct list_head *entry = q->out_chunk_list.next;
89 ch = list_entry(entry, struct sctp_chunk, list);
91 q->out_qlen -= ch->skb->len;
95 /* Add data chunk to the end of the queue. */
96 static inline void sctp_outq_tail_data(struct sctp_outq *q,
97 struct sctp_chunk *ch)
99 list_add_tail(&ch->list, &q->out_chunk_list);
100 q->out_qlen += ch->skb->len;
104 * SFR-CACC algorithm:
105 * D) If count_of_newacks is greater than or equal to 2
106 * and t was not sent to the current primary then the
107 * sender MUST NOT increment missing report count for t.
109 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary,
110 struct sctp_transport *transport,
111 int count_of_newacks)
113 if (count_of_newacks >= 2 && transport != primary)
119 * SFR-CACC algorithm:
120 * F) If count_of_newacks is less than 2, let d be the
121 * destination to which t was sent. If cacc_saw_newack
122 * is 0 for destination d, then the sender MUST NOT
123 * increment missing report count for t.
125 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport,
126 int count_of_newacks)
128 if (count_of_newacks < 2 &&
129 (transport && !transport->cacc.cacc_saw_newack))
135 * SFR-CACC algorithm:
136 * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
137 * execute steps C, D, F.
139 * C has been implemented in sctp_outq_sack
141 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
142 struct sctp_transport *transport,
143 int count_of_newacks)
145 if (!primary->cacc.cycling_changeover) {
146 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
148 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
156 * SFR-CACC algorithm:
157 * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
158 * than next_tsn_at_change of the current primary, then
159 * the sender MUST NOT increment missing report count
162 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
164 if (primary->cacc.cycling_changeover &&
165 TSN_lt(tsn, primary->cacc.next_tsn_at_change))
171 * SFR-CACC algorithm:
172 * 3) If the missing report count for TSN t is to be
173 * incremented according to [RFC2960] and
174 * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
175 * then the sender MUST further execute steps 3.1 and
176 * 3.2 to determine if the missing report count for
177 * TSN t SHOULD NOT be incremented.
179 * 3.3) If 3.1 and 3.2 do not dictate that the missing
180 * report count for t should not be incremented, then
181 * the sender SHOULD increment missing report count for
182 * t (according to [RFC2960] and [SCTP_STEWART_2002]).
184 static inline int sctp_cacc_skip(struct sctp_transport *primary,
185 struct sctp_transport *transport,
186 int count_of_newacks,
189 if (primary->cacc.changeover_active &&
190 (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) ||
191 sctp_cacc_skip_3_2(primary, tsn)))
196 /* Initialize an existing sctp_outq. This does the boring stuff.
197 * You still need to define handlers if you really want to DO
198 * something with this structure...
200 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
202 memset(q, 0, sizeof(struct sctp_outq));
205 INIT_LIST_HEAD(&q->out_chunk_list);
206 INIT_LIST_HEAD(&q->control_chunk_list);
207 INIT_LIST_HEAD(&q->retransmit);
208 INIT_LIST_HEAD(&q->sacked);
209 INIT_LIST_HEAD(&q->abandoned);
212 /* Free the outqueue structure and any related pending chunks.
214 static void __sctp_outq_teardown(struct sctp_outq *q)
216 struct sctp_transport *transport;
217 struct list_head *lchunk, *temp;
218 struct sctp_chunk *chunk, *tmp;
220 /* Throw away unacknowledged chunks. */
221 list_for_each_entry(transport, &q->asoc->peer.transport_addr_list,
223 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
224 chunk = list_entry(lchunk, struct sctp_chunk,
226 /* Mark as part of a failed message. */
227 sctp_chunk_fail(chunk, q->error);
228 sctp_chunk_free(chunk);
232 /* Throw away chunks that have been gap ACKed. */
233 list_for_each_safe(lchunk, temp, &q->sacked) {
234 list_del_init(lchunk);
235 chunk = list_entry(lchunk, struct sctp_chunk,
237 sctp_chunk_fail(chunk, q->error);
238 sctp_chunk_free(chunk);
241 /* Throw away any chunks in the retransmit queue. */
242 list_for_each_safe(lchunk, temp, &q->retransmit) {
243 list_del_init(lchunk);
244 chunk = list_entry(lchunk, struct sctp_chunk,
246 sctp_chunk_fail(chunk, q->error);
247 sctp_chunk_free(chunk);
250 /* Throw away any chunks that are in the abandoned queue. */
251 list_for_each_safe(lchunk, temp, &q->abandoned) {
252 list_del_init(lchunk);
253 chunk = list_entry(lchunk, struct sctp_chunk,
255 sctp_chunk_fail(chunk, q->error);
256 sctp_chunk_free(chunk);
259 /* Throw away any leftover data chunks. */
260 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
262 /* Mark as send failure. */
263 sctp_chunk_fail(chunk, q->error);
264 sctp_chunk_free(chunk);
267 /* Throw away any leftover control chunks. */
268 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
269 list_del_init(&chunk->list);
270 sctp_chunk_free(chunk);
274 void sctp_outq_teardown(struct sctp_outq *q)
276 __sctp_outq_teardown(q);
277 sctp_outq_init(q->asoc, q);
280 /* Free the outqueue structure and any related pending chunks. */
281 void sctp_outq_free(struct sctp_outq *q)
283 /* Throw away leftover chunks. */
284 __sctp_outq_teardown(q);
287 /* Put a new chunk in an sctp_outq. */
288 void sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk, gfp_t gfp)
290 struct net *net = sock_net(q->asoc->base.sk);
292 pr_debug("%s: outq:%p, chunk:%p[%s]\n", __func__, q, chunk,
293 chunk && chunk->chunk_hdr ?
294 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
297 /* If it is data, queue it up, otherwise, send it
300 if (sctp_chunk_is_data(chunk)) {
301 pr_debug("%s: outqueueing: outq:%p, chunk:%p[%s])\n",
302 __func__, q, chunk, chunk && chunk->chunk_hdr ?
303 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
306 sctp_outq_tail_data(q, chunk);
307 if (chunk->asoc->peer.prsctp_capable &&
308 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags))
309 chunk->asoc->sent_cnt_removable++;
310 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
311 SCTP_INC_STATS(net, SCTP_MIB_OUTUNORDERCHUNKS);
313 SCTP_INC_STATS(net, SCTP_MIB_OUTORDERCHUNKS);
315 list_add_tail(&chunk->list, &q->control_chunk_list);
316 SCTP_INC_STATS(net, SCTP_MIB_OUTCTRLCHUNKS);
320 sctp_outq_flush(q, 0, gfp);
323 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list
324 * and the abandoned list are in ascending order.
326 static void sctp_insert_list(struct list_head *head, struct list_head *new)
328 struct list_head *pos;
329 struct sctp_chunk *nchunk, *lchunk;
333 nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
334 ntsn = ntohl(nchunk->subh.data_hdr->tsn);
336 list_for_each(pos, head) {
337 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
338 ltsn = ntohl(lchunk->subh.data_hdr->tsn);
339 if (TSN_lt(ntsn, ltsn)) {
340 list_add(new, pos->prev);
346 list_add_tail(new, head);
349 static int sctp_prsctp_prune_sent(struct sctp_association *asoc,
350 struct sctp_sndrcvinfo *sinfo,
351 struct list_head *queue, int msg_len)
353 struct sctp_chunk *chk, *temp;
355 list_for_each_entry_safe(chk, temp, queue, transmitted_list) {
356 if (!SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) ||
357 chk->sinfo.sinfo_timetolive <= sinfo->sinfo_timetolive)
360 list_del_init(&chk->transmitted_list);
361 sctp_insert_list(&asoc->outqueue.abandoned,
362 &chk->transmitted_list);
364 asoc->sent_cnt_removable--;
365 asoc->abandoned_sent[SCTP_PR_INDEX(PRIO)]++;
367 if (!chk->tsn_gap_acked) {
369 chk->transport->flight_size -=
371 asoc->outqueue.outstanding_bytes -= sctp_data_size(chk);
374 msg_len -= SCTP_DATA_SNDSIZE(chk) +
375 sizeof(struct sk_buff) +
376 sizeof(struct sctp_chunk);
384 static int sctp_prsctp_prune_unsent(struct sctp_association *asoc,
385 struct sctp_sndrcvinfo *sinfo, int msg_len)
387 struct sctp_outq *q = &asoc->outqueue;
388 struct sctp_chunk *chk, *temp;
390 list_for_each_entry_safe(chk, temp, &q->out_chunk_list, list) {
391 if (!SCTP_PR_PRIO_ENABLED(chk->sinfo.sinfo_flags) ||
392 chk->sinfo.sinfo_timetolive <= sinfo->sinfo_timetolive)
395 list_del_init(&chk->list);
396 q->out_qlen -= chk->skb->len;
397 asoc->sent_cnt_removable--;
398 asoc->abandoned_unsent[SCTP_PR_INDEX(PRIO)]++;
400 msg_len -= SCTP_DATA_SNDSIZE(chk) +
401 sizeof(struct sk_buff) +
402 sizeof(struct sctp_chunk);
403 sctp_chunk_free(chk);
411 /* Abandon the chunks according their priorities */
412 void sctp_prsctp_prune(struct sctp_association *asoc,
413 struct sctp_sndrcvinfo *sinfo, int msg_len)
415 struct sctp_transport *transport;
417 if (!asoc->peer.prsctp_capable || !asoc->sent_cnt_removable)
420 msg_len = sctp_prsctp_prune_sent(asoc, sinfo,
421 &asoc->outqueue.retransmit,
426 list_for_each_entry(transport, &asoc->peer.transport_addr_list,
428 msg_len = sctp_prsctp_prune_sent(asoc, sinfo,
429 &transport->transmitted,
435 sctp_prsctp_prune_unsent(asoc, sinfo, msg_len);
438 /* Mark all the eligible packets on a transport for retransmission. */
439 void sctp_retransmit_mark(struct sctp_outq *q,
440 struct sctp_transport *transport,
443 struct list_head *lchunk, *ltemp;
444 struct sctp_chunk *chunk;
446 /* Walk through the specified transmitted queue. */
447 list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
448 chunk = list_entry(lchunk, struct sctp_chunk,
451 /* If the chunk is abandoned, move it to abandoned list. */
452 if (sctp_chunk_abandoned(chunk)) {
453 list_del_init(lchunk);
454 sctp_insert_list(&q->abandoned, lchunk);
456 /* If this chunk has not been previousely acked,
457 * stop considering it 'outstanding'. Our peer
458 * will most likely never see it since it will
459 * not be retransmitted
461 if (!chunk->tsn_gap_acked) {
462 if (chunk->transport)
463 chunk->transport->flight_size -=
464 sctp_data_size(chunk);
465 q->outstanding_bytes -= sctp_data_size(chunk);
466 q->asoc->peer.rwnd += sctp_data_size(chunk);
471 /* If we are doing retransmission due to a timeout or pmtu
472 * discovery, only the chunks that are not yet acked should
473 * be added to the retransmit queue.
475 if ((reason == SCTP_RTXR_FAST_RTX &&
476 (chunk->fast_retransmit == SCTP_NEED_FRTX)) ||
477 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) {
478 /* RFC 2960 6.2.1 Processing a Received SACK
480 * C) Any time a DATA chunk is marked for
481 * retransmission (via either T3-rtx timer expiration
482 * (Section 6.3.3) or via fast retransmit
483 * (Section 7.2.4)), add the data size of those
484 * chunks to the rwnd.
486 q->asoc->peer.rwnd += sctp_data_size(chunk);
487 q->outstanding_bytes -= sctp_data_size(chunk);
488 if (chunk->transport)
489 transport->flight_size -= sctp_data_size(chunk);
491 /* sctpimpguide-05 Section 2.8.2
492 * M5) If a T3-rtx timer expires, the
493 * 'TSN.Missing.Report' of all affected TSNs is set
496 chunk->tsn_missing_report = 0;
498 /* If a chunk that is being used for RTT measurement
499 * has to be retransmitted, we cannot use this chunk
500 * anymore for RTT measurements. Reset rto_pending so
501 * that a new RTT measurement is started when a new
502 * data chunk is sent.
504 if (chunk->rtt_in_progress) {
505 chunk->rtt_in_progress = 0;
506 transport->rto_pending = 0;
509 /* Move the chunk to the retransmit queue. The chunks
510 * on the retransmit queue are always kept in order.
512 list_del_init(lchunk);
513 sctp_insert_list(&q->retransmit, lchunk);
517 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, "
518 "flight_size:%d, pba:%d\n", __func__, transport, reason,
519 transport->cwnd, transport->ssthresh, transport->flight_size,
520 transport->partial_bytes_acked);
523 /* Mark all the eligible packets on a transport for retransmission and force
526 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
527 sctp_retransmit_reason_t reason)
529 struct net *net = sock_net(q->asoc->base.sk);
532 case SCTP_RTXR_T3_RTX:
533 SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS);
534 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
535 /* Update the retran path if the T3-rtx timer has expired for
536 * the current retran path.
538 if (transport == transport->asoc->peer.retran_path)
539 sctp_assoc_update_retran_path(transport->asoc);
540 transport->asoc->rtx_data_chunks +=
541 transport->asoc->unack_data;
543 case SCTP_RTXR_FAST_RTX:
544 SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS);
545 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
548 case SCTP_RTXR_PMTUD:
549 SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS);
551 case SCTP_RTXR_T1_RTX:
552 SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS);
553 transport->asoc->init_retries++;
559 sctp_retransmit_mark(q, transport, reason);
561 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
562 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
563 * following the procedures outlined in C1 - C5.
565 if (reason == SCTP_RTXR_T3_RTX)
566 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
568 /* Flush the queues only on timeout, since fast_rtx is only
569 * triggered during sack processing and the queue
570 * will be flushed at the end.
572 if (reason != SCTP_RTXR_FAST_RTX)
573 sctp_outq_flush(q, /* rtx_timeout */ 1, GFP_ATOMIC);
577 * Transmit DATA chunks on the retransmit queue. Upon return from
578 * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
579 * need to be transmitted by the caller.
580 * We assume that pkt->transport has already been set.
582 * The return value is a normal kernel error return value.
584 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
585 int rtx_timeout, int *start_timer)
587 struct list_head *lqueue;
588 struct sctp_transport *transport = pkt->transport;
590 struct sctp_chunk *chunk, *chunk1;
596 lqueue = &q->retransmit;
597 fast_rtx = q->fast_rtx;
599 /* This loop handles time-out retransmissions, fast retransmissions,
600 * and retransmissions due to opening of whindow.
602 * RFC 2960 6.3.3 Handle T3-rtx Expiration
604 * E3) Determine how many of the earliest (i.e., lowest TSN)
605 * outstanding DATA chunks for the address for which the
606 * T3-rtx has expired will fit into a single packet, subject
607 * to the MTU constraint for the path corresponding to the
608 * destination transport address to which the retransmission
609 * is being sent (this may be different from the address for
610 * which the timer expires [see Section 6.4]). Call this value
611 * K. Bundle and retransmit those K DATA chunks in a single
612 * packet to the destination endpoint.
614 * [Just to be painfully clear, if we are retransmitting
615 * because a timeout just happened, we should send only ONE
616 * packet of retransmitted data.]
618 * For fast retransmissions we also send only ONE packet. However,
619 * if we are just flushing the queue due to open window, we'll
620 * try to send as much as possible.
622 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
623 /* If the chunk is abandoned, move it to abandoned list. */
624 if (sctp_chunk_abandoned(chunk)) {
625 list_del_init(&chunk->transmitted_list);
626 sctp_insert_list(&q->abandoned,
627 &chunk->transmitted_list);
631 /* Make sure that Gap Acked TSNs are not retransmitted. A
632 * simple approach is just to move such TSNs out of the
633 * way and into a 'transmitted' queue and skip to the
636 if (chunk->tsn_gap_acked) {
637 list_move_tail(&chunk->transmitted_list,
638 &transport->transmitted);
642 /* If we are doing fast retransmit, ignore non-fast_rtransmit
645 if (fast_rtx && !chunk->fast_retransmit)
649 /* Attempt to append this chunk to the packet. */
650 status = sctp_packet_append_chunk(pkt, chunk);
653 case SCTP_XMIT_PMTU_FULL:
654 if (!pkt->has_data && !pkt->has_cookie_echo) {
655 /* If this packet did not contain DATA then
656 * retransmission did not happen, so do it
657 * again. We'll ignore the error here since
658 * control chunks are already freed so there
659 * is nothing we can do.
661 sctp_packet_transmit(pkt, GFP_ATOMIC);
665 /* Send this packet. */
666 error = sctp_packet_transmit(pkt, GFP_ATOMIC);
668 /* If we are retransmitting, we should only
669 * send a single packet.
670 * Otherwise, try appending this chunk again.
672 if (rtx_timeout || fast_rtx)
677 /* Bundle next chunk in the next round. */
680 case SCTP_XMIT_RWND_FULL:
681 /* Send this packet. */
682 error = sctp_packet_transmit(pkt, GFP_ATOMIC);
684 /* Stop sending DATA as there is no more room
690 case SCTP_XMIT_DELAY:
691 /* Send this packet. */
692 error = sctp_packet_transmit(pkt, GFP_ATOMIC);
694 /* Stop sending DATA because of nagle delay. */
699 /* The append was successful, so add this chunk to
700 * the transmitted list.
702 list_move_tail(&chunk->transmitted_list,
703 &transport->transmitted);
705 /* Mark the chunk as ineligible for fast retransmit
706 * after it is retransmitted.
708 if (chunk->fast_retransmit == SCTP_NEED_FRTX)
709 chunk->fast_retransmit = SCTP_DONT_FRTX;
711 q->asoc->stats.rtxchunks++;
715 /* Set the timer if there were no errors */
716 if (!error && !timer)
723 /* If we are here due to a retransmit timeout or a fast
724 * retransmit and if there are any chunks left in the retransmit
725 * queue that could not fit in the PMTU sized packet, they need
726 * to be marked as ineligible for a subsequent fast retransmit.
728 if (rtx_timeout || fast_rtx) {
729 list_for_each_entry(chunk1, lqueue, transmitted_list) {
730 if (chunk1->fast_retransmit == SCTP_NEED_FRTX)
731 chunk1->fast_retransmit = SCTP_DONT_FRTX;
735 *start_timer = timer;
737 /* Clear fast retransmit hint */
744 /* Cork the outqueue so queued chunks are really queued. */
745 void sctp_outq_uncork(struct sctp_outq *q, gfp_t gfp)
750 sctp_outq_flush(q, 0, gfp);
755 * Try to flush an outqueue.
757 * Description: Send everything in q which we legally can, subject to
758 * congestion limitations.
759 * * Note: This function can be called from multiple contexts so appropriate
760 * locking concerns must be made. Today we use the sock lock to protect
763 static void sctp_outq_flush(struct sctp_outq *q, int rtx_timeout, gfp_t gfp)
765 struct sctp_packet *packet;
766 struct sctp_packet singleton;
767 struct sctp_association *asoc = q->asoc;
768 __u16 sport = asoc->base.bind_addr.port;
769 __u16 dport = asoc->peer.port;
770 __u32 vtag = asoc->peer.i.init_tag;
771 struct sctp_transport *transport = NULL;
772 struct sctp_transport *new_transport;
773 struct sctp_chunk *chunk, *tmp;
779 /* These transports have chunks to send. */
780 struct list_head transport_list;
781 struct list_head *ltransport;
783 INIT_LIST_HEAD(&transport_list);
789 * When bundling control chunks with DATA chunks, an
790 * endpoint MUST place control chunks first in the outbound
791 * SCTP packet. The transmitter MUST transmit DATA chunks
792 * within a SCTP packet in increasing order of TSN.
796 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
798 * F1) This means that until such time as the ASCONF
799 * containing the add is acknowledged, the sender MUST
800 * NOT use the new IP address as a source for ANY SCTP
801 * packet except on carrying an ASCONF Chunk.
803 if (asoc->src_out_of_asoc_ok &&
804 chunk->chunk_hdr->type != SCTP_CID_ASCONF)
807 list_del_init(&chunk->list);
809 /* Pick the right transport to use. */
810 new_transport = chunk->transport;
812 if (!new_transport) {
814 * If we have a prior transport pointer, see if
815 * the destination address of the chunk
816 * matches the destination address of the
817 * current transport. If not a match, then
818 * try to look up the transport with a given
819 * destination address. We do this because
820 * after processing ASCONFs, we may have new
821 * transports created.
824 sctp_cmp_addr_exact(&chunk->dest,
826 new_transport = transport;
828 new_transport = sctp_assoc_lookup_paddr(asoc,
831 /* if we still don't have a new transport, then
832 * use the current active path.
835 new_transport = asoc->peer.active_path;
836 } else if ((new_transport->state == SCTP_INACTIVE) ||
837 (new_transport->state == SCTP_UNCONFIRMED) ||
838 (new_transport->state == SCTP_PF)) {
839 /* If the chunk is Heartbeat or Heartbeat Ack,
840 * send it to chunk->transport, even if it's
843 * 3.3.6 Heartbeat Acknowledgement:
845 * A HEARTBEAT ACK is always sent to the source IP
846 * address of the IP datagram containing the
847 * HEARTBEAT chunk to which this ack is responding.
850 * ASCONF_ACKs also must be sent to the source.
852 if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
853 chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK &&
854 chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK)
855 new_transport = asoc->peer.active_path;
858 /* Are we switching transports?
859 * Take care of transport locks.
861 if (new_transport != transport) {
862 transport = new_transport;
863 if (list_empty(&transport->send_ready)) {
864 list_add_tail(&transport->send_ready,
867 packet = &transport->packet;
868 sctp_packet_config(packet, vtag,
869 asoc->peer.ecn_capable);
872 switch (chunk->chunk_hdr->type) {
876 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
877 * COMPLETE with any other chunks. [Send them immediately.]
880 case SCTP_CID_INIT_ACK:
881 case SCTP_CID_SHUTDOWN_COMPLETE:
882 sctp_packet_init(&singleton, transport, sport, dport);
883 sctp_packet_config(&singleton, vtag, 0);
884 sctp_packet_append_chunk(&singleton, chunk);
885 error = sctp_packet_transmit(&singleton, gfp);
887 asoc->base.sk->sk_err = -error;
893 if (sctp_test_T_bit(chunk)) {
894 packet->vtag = asoc->c.my_vtag;
896 /* The following chunks are "response" chunks, i.e.
897 * they are generated in response to something we
898 * received. If we are sending these, then we can
899 * send only 1 packet containing these chunks.
901 case SCTP_CID_HEARTBEAT_ACK:
902 case SCTP_CID_SHUTDOWN_ACK:
903 case SCTP_CID_COOKIE_ACK:
904 case SCTP_CID_COOKIE_ECHO:
906 case SCTP_CID_ECN_CWR:
907 case SCTP_CID_ASCONF_ACK:
912 case SCTP_CID_HEARTBEAT:
913 case SCTP_CID_SHUTDOWN:
914 case SCTP_CID_ECN_ECNE:
915 case SCTP_CID_ASCONF:
916 case SCTP_CID_FWD_TSN:
917 case SCTP_CID_RECONF:
918 status = sctp_packet_transmit_chunk(packet, chunk,
920 if (status != SCTP_XMIT_OK) {
921 /* put the chunk back */
922 list_add(&chunk->list, &q->control_chunk_list);
926 asoc->stats.octrlchunks++;
927 /* PR-SCTP C5) If a FORWARD TSN is sent, the
928 * sender MUST assure that at least one T3-rtx
931 if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN) {
932 sctp_transport_reset_t3_rtx(transport);
933 transport->last_time_sent = jiffies;
936 if (chunk == asoc->strreset_chunk)
937 sctp_transport_reset_reconf_timer(transport);
942 /* We built a chunk with an illegal type! */
947 if (q->asoc->src_out_of_asoc_ok)
950 /* Is it OK to send data chunks? */
951 switch (asoc->state) {
952 case SCTP_STATE_COOKIE_ECHOED:
953 /* Only allow bundling when this packet has a COOKIE-ECHO
956 if (!packet || !packet->has_cookie_echo)
960 case SCTP_STATE_ESTABLISHED:
961 case SCTP_STATE_SHUTDOWN_PENDING:
962 case SCTP_STATE_SHUTDOWN_RECEIVED:
964 * RFC 2960 6.1 Transmission of DATA Chunks
966 * C) When the time comes for the sender to transmit,
967 * before sending new DATA chunks, the sender MUST
968 * first transmit any outstanding DATA chunks which
969 * are marked for retransmission (limited by the
972 if (!list_empty(&q->retransmit)) {
973 if (asoc->peer.retran_path->state == SCTP_UNCONFIRMED)
975 if (transport == asoc->peer.retran_path)
978 /* Switch transports & prepare the packet. */
980 transport = asoc->peer.retran_path;
982 if (list_empty(&transport->send_ready)) {
983 list_add_tail(&transport->send_ready,
987 packet = &transport->packet;
988 sctp_packet_config(packet, vtag,
989 asoc->peer.ecn_capable);
991 error = sctp_outq_flush_rtx(q, packet,
992 rtx_timeout, &start_timer);
994 asoc->base.sk->sk_err = -error;
997 sctp_transport_reset_t3_rtx(transport);
998 transport->last_time_sent = jiffies;
1001 /* This can happen on COOKIE-ECHO resend. Only
1002 * one chunk can get bundled with a COOKIE-ECHO.
1004 if (packet->has_cookie_echo)
1005 goto sctp_flush_out;
1007 /* Don't send new data if there is still data
1008 * waiting to retransmit.
1010 if (!list_empty(&q->retransmit))
1011 goto sctp_flush_out;
1014 /* Apply Max.Burst limitation to the current transport in
1015 * case it will be used for new data. We are going to
1016 * rest it before we return, but we want to apply the limit
1017 * to the currently queued data.
1020 sctp_transport_burst_limited(transport);
1022 /* Finally, transmit new packets. */
1023 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
1024 __u32 sid = ntohs(chunk->subh.data_hdr->stream);
1026 /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
1027 * stream identifier.
1029 if (chunk->sinfo.sinfo_stream >= asoc->stream->outcnt) {
1031 /* Mark as failed send. */
1032 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
1033 if (asoc->peer.prsctp_capable &&
1034 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags))
1035 asoc->sent_cnt_removable--;
1036 sctp_chunk_free(chunk);
1040 /* Has this chunk expired? */
1041 if (sctp_chunk_abandoned(chunk)) {
1042 sctp_chunk_fail(chunk, 0);
1043 sctp_chunk_free(chunk);
1047 if (asoc->stream->out[sid].state == SCTP_STREAM_CLOSED) {
1048 sctp_outq_head_data(q, chunk);
1049 goto sctp_flush_out;
1052 /* If there is a specified transport, use it.
1053 * Otherwise, we want to use the active path.
1055 new_transport = chunk->transport;
1056 if (!new_transport ||
1057 ((new_transport->state == SCTP_INACTIVE) ||
1058 (new_transport->state == SCTP_UNCONFIRMED) ||
1059 (new_transport->state == SCTP_PF)))
1060 new_transport = asoc->peer.active_path;
1061 if (new_transport->state == SCTP_UNCONFIRMED) {
1062 WARN_ONCE(1, "Attempt to send packet on unconfirmed path.");
1063 sctp_chunk_fail(chunk, 0);
1064 sctp_chunk_free(chunk);
1068 /* Change packets if necessary. */
1069 if (new_transport != transport) {
1070 transport = new_transport;
1072 /* Schedule to have this transport's
1075 if (list_empty(&transport->send_ready)) {
1076 list_add_tail(&transport->send_ready,
1080 packet = &transport->packet;
1081 sctp_packet_config(packet, vtag,
1082 asoc->peer.ecn_capable);
1083 /* We've switched transports, so apply the
1084 * Burst limit to the new transport.
1086 sctp_transport_burst_limited(transport);
1089 pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p "
1091 __func__, q, chunk, chunk && chunk->chunk_hdr ?
1092 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
1093 "illegal chunk", ntohl(chunk->subh.data_hdr->tsn),
1094 chunk->skb ? chunk->skb->head : NULL, chunk->skb ?
1095 atomic_read(&chunk->skb->users) : -1);
1097 /* Add the chunk to the packet. */
1098 status = sctp_packet_transmit_chunk(packet, chunk, 0, gfp);
1101 case SCTP_XMIT_PMTU_FULL:
1102 case SCTP_XMIT_RWND_FULL:
1103 case SCTP_XMIT_DELAY:
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(q, chunk);
1112 goto sctp_flush_out;
1115 /* The sender is in the SHUTDOWN-PENDING state,
1116 * The sender MAY set the I-bit in the DATA
1119 if (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 asoc->stats.ouodchunks++;
1124 asoc->stats.oodchunks++;
1132 /* BUG: We assume that the sctp_packet_transmit()
1133 * call below will succeed all the time and add the
1134 * chunk to the transmitted list and restart the
1136 * It is possible that the call can fail under OOM
1139 * Is this really a problem? Won't this behave
1142 list_add_tail(&chunk->transmitted_list,
1143 &transport->transmitted);
1145 sctp_transport_reset_t3_rtx(transport);
1146 transport->last_time_sent = jiffies;
1148 /* Only let one DATA chunk get bundled with a
1149 * COOKIE-ECHO chunk.
1151 if (packet->has_cookie_echo)
1152 goto sctp_flush_out;
1163 /* Before returning, examine all the transports touched in
1164 * this call. Right now, we bluntly force clear all the
1165 * transports. Things might change after we implement Nagle.
1166 * But such an examination is still required.
1170 while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL) {
1171 struct sctp_transport *t = list_entry(ltransport,
1172 struct sctp_transport,
1174 packet = &t->packet;
1175 if (!sctp_packet_empty(packet)) {
1176 error = sctp_packet_transmit(packet, gfp);
1178 asoc->base.sk->sk_err = -error;
1181 /* Clear the burst limited state, if any */
1182 sctp_transport_burst_reset(t);
1186 /* Update unack_data based on the incoming SACK chunk */
1187 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1188 struct sctp_sackhdr *sack)
1190 sctp_sack_variable_t *frags;
1194 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1196 frags = sack->variable;
1197 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1198 unack_data -= ((ntohs(frags[i].gab.end) -
1199 ntohs(frags[i].gab.start) + 1));
1202 assoc->unack_data = unack_data;
1205 /* This is where we REALLY process a SACK.
1207 * Process the SACK against the outqueue. Mostly, this just frees
1208 * things off the transmitted queue.
1210 int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk)
1212 struct sctp_association *asoc = q->asoc;
1213 struct sctp_sackhdr *sack = chunk->subh.sack_hdr;
1214 struct sctp_transport *transport;
1215 struct sctp_chunk *tchunk = NULL;
1216 struct list_head *lchunk, *transport_list, *temp;
1217 sctp_sack_variable_t *frags = sack->variable;
1218 __u32 sack_ctsn, ctsn, tsn;
1219 __u32 highest_tsn, highest_new_tsn;
1221 unsigned int outstanding;
1222 struct sctp_transport *primary = asoc->peer.primary_path;
1223 int count_of_newacks = 0;
1227 /* Grab the association's destination address list. */
1228 transport_list = &asoc->peer.transport_addr_list;
1230 sack_ctsn = ntohl(sack->cum_tsn_ack);
1231 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1232 asoc->stats.gapcnt += gap_ack_blocks;
1234 * SFR-CACC algorithm:
1235 * On receipt of a SACK the sender SHOULD execute the
1236 * following statements.
1238 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1239 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1240 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1242 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1243 * is set the receiver of the SACK MUST take the following actions:
1245 * A) Initialize the cacc_saw_newack to 0 for all destination
1248 * Only bother if changeover_active is set. Otherwise, this is
1249 * totally suboptimal to do on every SACK.
1251 if (primary->cacc.changeover_active) {
1252 u8 clear_cycling = 0;
1254 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1255 primary->cacc.changeover_active = 0;
1259 if (clear_cycling || gap_ack_blocks) {
1260 list_for_each_entry(transport, transport_list,
1263 transport->cacc.cycling_changeover = 0;
1265 transport->cacc.cacc_saw_newack = 0;
1270 /* Get the highest TSN in the sack. */
1271 highest_tsn = sack_ctsn;
1273 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1275 if (TSN_lt(asoc->highest_sacked, highest_tsn))
1276 asoc->highest_sacked = highest_tsn;
1278 highest_new_tsn = sack_ctsn;
1280 /* Run through the retransmit queue. Credit bytes received
1281 * and free those chunks that we can.
1283 sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn);
1285 /* Run through the transmitted queue.
1286 * Credit bytes received and free those chunks which we can.
1288 * This is a MASSIVE candidate for optimization.
1290 list_for_each_entry(transport, transport_list, transports) {
1291 sctp_check_transmitted(q, &transport->transmitted,
1292 transport, &chunk->source, sack,
1295 * SFR-CACC algorithm:
1296 * C) Let count_of_newacks be the number of
1297 * destinations for which cacc_saw_newack is set.
1299 if (transport->cacc.cacc_saw_newack)
1303 /* Move the Cumulative TSN Ack Point if appropriate. */
1304 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) {
1305 asoc->ctsn_ack_point = sack_ctsn;
1309 if (gap_ack_blocks) {
1311 if (asoc->fast_recovery && accum_moved)
1312 highest_new_tsn = highest_tsn;
1314 list_for_each_entry(transport, transport_list, transports)
1315 sctp_mark_missing(q, &transport->transmitted, transport,
1316 highest_new_tsn, count_of_newacks);
1319 /* Update unack_data field in the assoc. */
1320 sctp_sack_update_unack_data(asoc, sack);
1322 ctsn = asoc->ctsn_ack_point;
1324 /* Throw away stuff rotting on the sack queue. */
1325 list_for_each_safe(lchunk, temp, &q->sacked) {
1326 tchunk = list_entry(lchunk, struct sctp_chunk,
1328 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1329 if (TSN_lte(tsn, ctsn)) {
1330 list_del_init(&tchunk->transmitted_list);
1331 if (asoc->peer.prsctp_capable &&
1332 SCTP_PR_PRIO_ENABLED(chunk->sinfo.sinfo_flags))
1333 asoc->sent_cnt_removable--;
1334 sctp_chunk_free(tchunk);
1338 /* ii) Set rwnd equal to the newly received a_rwnd minus the
1339 * number of bytes still outstanding after processing the
1340 * Cumulative TSN Ack and the Gap Ack Blocks.
1343 sack_a_rwnd = ntohl(sack->a_rwnd);
1344 asoc->peer.zero_window_announced = !sack_a_rwnd;
1345 outstanding = q->outstanding_bytes;
1347 if (outstanding < sack_a_rwnd)
1348 sack_a_rwnd -= outstanding;
1352 asoc->peer.rwnd = sack_a_rwnd;
1354 sctp_generate_fwdtsn(q, sack_ctsn);
1356 pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__, sack_ctsn);
1357 pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, "
1358 "advertised peer ack point:0x%x\n", __func__, asoc, ctsn,
1359 asoc->adv_peer_ack_point);
1361 return sctp_outq_is_empty(q);
1364 /* Is the outqueue empty?
1365 * The queue is empty when we have not pending data, no in-flight data
1366 * and nothing pending retransmissions.
1368 int sctp_outq_is_empty(const struct sctp_outq *q)
1370 return q->out_qlen == 0 && q->outstanding_bytes == 0 &&
1371 list_empty(&q->retransmit);
1374 /********************************************************************
1375 * 2nd Level Abstractions
1376 ********************************************************************/
1378 /* Go through a transport's transmitted list or the association's retransmit
1379 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1380 * The retransmit list will not have an associated transport.
1382 * I added coherent debug information output. --xguo
1384 * Instead of printing 'sacked' or 'kept' for each TSN on the
1385 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1386 * KEPT TSN6-TSN7, etc.
1388 static void sctp_check_transmitted(struct sctp_outq *q,
1389 struct list_head *transmitted_queue,
1390 struct sctp_transport *transport,
1391 union sctp_addr *saddr,
1392 struct sctp_sackhdr *sack,
1393 __u32 *highest_new_tsn_in_sack)
1395 struct list_head *lchunk;
1396 struct sctp_chunk *tchunk;
1397 struct list_head tlist;
1401 __u8 restart_timer = 0;
1402 int bytes_acked = 0;
1403 int migrate_bytes = 0;
1404 bool forward_progress = false;
1406 sack_ctsn = ntohl(sack->cum_tsn_ack);
1408 INIT_LIST_HEAD(&tlist);
1410 /* The while loop will skip empty transmitted queues. */
1411 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1412 tchunk = list_entry(lchunk, struct sctp_chunk,
1415 if (sctp_chunk_abandoned(tchunk)) {
1416 /* Move the chunk to abandoned list. */
1417 sctp_insert_list(&q->abandoned, lchunk);
1419 /* If this chunk has not been acked, stop
1420 * considering it as 'outstanding'.
1422 if (!tchunk->tsn_gap_acked) {
1423 if (tchunk->transport)
1424 tchunk->transport->flight_size -=
1425 sctp_data_size(tchunk);
1426 q->outstanding_bytes -= sctp_data_size(tchunk);
1431 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1432 if (sctp_acked(sack, tsn)) {
1433 /* If this queue is the retransmit queue, the
1434 * retransmit timer has already reclaimed
1435 * the outstanding bytes for this chunk, so only
1436 * count bytes associated with a transport.
1439 /* If this chunk is being used for RTT
1440 * measurement, calculate the RTT and update
1441 * the RTO using this value.
1443 * 6.3.1 C5) Karn's algorithm: RTT measurements
1444 * MUST NOT be made using packets that were
1445 * retransmitted (and thus for which it is
1446 * ambiguous whether the reply was for the
1447 * first instance of the packet or a later
1450 if (!tchunk->tsn_gap_acked &&
1451 !sctp_chunk_retransmitted(tchunk) &&
1452 tchunk->rtt_in_progress) {
1453 tchunk->rtt_in_progress = 0;
1454 rtt = jiffies - tchunk->sent_at;
1455 sctp_transport_update_rto(transport,
1460 /* If the chunk hasn't been marked as ACKED,
1461 * mark it and account bytes_acked if the
1462 * chunk had a valid transport (it will not
1463 * have a transport if ASCONF had deleted it
1464 * while DATA was outstanding).
1466 if (!tchunk->tsn_gap_acked) {
1467 tchunk->tsn_gap_acked = 1;
1468 if (TSN_lt(*highest_new_tsn_in_sack, tsn))
1469 *highest_new_tsn_in_sack = tsn;
1470 bytes_acked += sctp_data_size(tchunk);
1471 if (!tchunk->transport)
1472 migrate_bytes += sctp_data_size(tchunk);
1473 forward_progress = true;
1476 if (TSN_lte(tsn, sack_ctsn)) {
1477 /* RFC 2960 6.3.2 Retransmission Timer Rules
1479 * R3) Whenever a SACK is received
1480 * that acknowledges the DATA chunk
1481 * with the earliest outstanding TSN
1482 * for that address, restart T3-rtx
1483 * timer for that address with its
1487 forward_progress = true;
1489 if (!tchunk->tsn_gap_acked) {
1491 * SFR-CACC algorithm:
1492 * 2) If the SACK contains gap acks
1493 * and the flag CHANGEOVER_ACTIVE is
1494 * set the receiver of the SACK MUST
1495 * take the following action:
1497 * B) For each TSN t being acked that
1498 * has not been acked in any SACK so
1499 * far, set cacc_saw_newack to 1 for
1500 * the destination that the TSN was
1504 sack->num_gap_ack_blocks &&
1505 q->asoc->peer.primary_path->cacc.
1507 transport->cacc.cacc_saw_newack
1511 list_add_tail(&tchunk->transmitted_list,
1514 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1515 * M2) Each time a SACK arrives reporting
1516 * 'Stray DATA chunk(s)' record the highest TSN
1517 * reported as newly acknowledged, call this
1518 * value 'HighestTSNinSack'. A newly
1519 * acknowledged DATA chunk is one not
1520 * previously acknowledged in a SACK.
1522 * When the SCTP sender of data receives a SACK
1523 * chunk that acknowledges, for the first time,
1524 * the receipt of a DATA chunk, all the still
1525 * unacknowledged DATA chunks whose TSN is
1526 * older than that newly acknowledged DATA
1527 * chunk, are qualified as 'Stray DATA chunks'.
1529 list_add_tail(lchunk, &tlist);
1532 if (tchunk->tsn_gap_acked) {
1533 pr_debug("%s: receiver reneged on data TSN:0x%x\n",
1536 tchunk->tsn_gap_acked = 0;
1538 if (tchunk->transport)
1539 bytes_acked -= sctp_data_size(tchunk);
1541 /* RFC 2960 6.3.2 Retransmission Timer Rules
1543 * R4) Whenever a SACK is received missing a
1544 * TSN that was previously acknowledged via a
1545 * Gap Ack Block, start T3-rtx for the
1546 * destination address to which the DATA
1547 * chunk was originally
1548 * transmitted if it is not already running.
1553 list_add_tail(lchunk, &tlist);
1559 struct sctp_association *asoc = transport->asoc;
1561 /* We may have counted DATA that was migrated
1562 * to this transport due to DEL-IP operation.
1563 * Subtract those bytes, since the were never
1564 * send on this transport and shouldn't be
1565 * credited to this transport.
1567 bytes_acked -= migrate_bytes;
1569 /* 8.2. When an outstanding TSN is acknowledged,
1570 * the endpoint shall clear the error counter of
1571 * the destination transport address to which the
1572 * DATA chunk was last sent.
1573 * The association's overall error counter is
1576 transport->error_count = 0;
1577 transport->asoc->overall_error_count = 0;
1578 forward_progress = true;
1581 * While in SHUTDOWN PENDING, we may have started
1582 * the T5 shutdown guard timer after reaching the
1583 * retransmission limit. Stop that timer as soon
1584 * as the receiver acknowledged any data.
1586 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING &&
1587 del_timer(&asoc->timers
1588 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]))
1589 sctp_association_put(asoc);
1591 /* Mark the destination transport address as
1592 * active if it is not so marked.
1594 if ((transport->state == SCTP_INACTIVE ||
1595 transport->state == SCTP_UNCONFIRMED) &&
1596 sctp_cmp_addr_exact(&transport->ipaddr, saddr)) {
1597 sctp_assoc_control_transport(
1601 SCTP_RECEIVED_SACK);
1604 sctp_transport_raise_cwnd(transport, sack_ctsn,
1607 transport->flight_size -= bytes_acked;
1608 if (transport->flight_size == 0)
1609 transport->partial_bytes_acked = 0;
1610 q->outstanding_bytes -= bytes_acked + migrate_bytes;
1612 /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1613 * When a sender is doing zero window probing, it
1614 * should not timeout the association if it continues
1615 * to receive new packets from the receiver. The
1616 * reason is that the receiver MAY keep its window
1617 * closed for an indefinite time.
1618 * A sender is doing zero window probing when the
1619 * receiver's advertised window is zero, and there is
1620 * only one data chunk in flight to the receiver.
1622 * Allow the association to timeout while in SHUTDOWN
1623 * PENDING or SHUTDOWN RECEIVED in case the receiver
1624 * stays in zero window mode forever.
1626 if (!q->asoc->peer.rwnd &&
1627 !list_empty(&tlist) &&
1628 (sack_ctsn+2 == q->asoc->next_tsn) &&
1629 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) {
1630 pr_debug("%s: sack received for zero window "
1631 "probe:%u\n", __func__, sack_ctsn);
1633 q->asoc->overall_error_count = 0;
1634 transport->error_count = 0;
1638 /* RFC 2960 6.3.2 Retransmission Timer Rules
1640 * R2) Whenever all outstanding data sent to an address have
1641 * been acknowledged, turn off the T3-rtx timer of that
1644 if (!transport->flight_size) {
1645 if (del_timer(&transport->T3_rtx_timer))
1646 sctp_transport_put(transport);
1647 } else if (restart_timer) {
1648 if (!mod_timer(&transport->T3_rtx_timer,
1649 jiffies + transport->rto))
1650 sctp_transport_hold(transport);
1653 if (forward_progress) {
1655 sctp_transport_dst_confirm(transport);
1659 list_splice(&tlist, transmitted_queue);
1662 /* Mark chunks as missing and consequently may get retransmitted. */
1663 static void sctp_mark_missing(struct sctp_outq *q,
1664 struct list_head *transmitted_queue,
1665 struct sctp_transport *transport,
1666 __u32 highest_new_tsn_in_sack,
1667 int count_of_newacks)
1669 struct sctp_chunk *chunk;
1671 char do_fast_retransmit = 0;
1672 struct sctp_association *asoc = q->asoc;
1673 struct sctp_transport *primary = asoc->peer.primary_path;
1675 list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1677 tsn = ntohl(chunk->subh.data_hdr->tsn);
1679 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1680 * 'Unacknowledged TSN's', if the TSN number of an
1681 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1682 * value, increment the 'TSN.Missing.Report' count on that
1683 * chunk if it has NOT been fast retransmitted or marked for
1684 * fast retransmit already.
1686 if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
1687 !chunk->tsn_gap_acked &&
1688 TSN_lt(tsn, highest_new_tsn_in_sack)) {
1690 /* SFR-CACC may require us to skip marking
1691 * this chunk as missing.
1693 if (!transport || !sctp_cacc_skip(primary,
1695 count_of_newacks, tsn)) {
1696 chunk->tsn_missing_report++;
1698 pr_debug("%s: tsn:0x%x missing counter:%d\n",
1699 __func__, tsn, chunk->tsn_missing_report);
1703 * M4) If any DATA chunk is found to have a
1704 * 'TSN.Missing.Report'
1705 * value larger than or equal to 3, mark that chunk for
1706 * retransmission and start the fast retransmit procedure.
1709 if (chunk->tsn_missing_report >= 3) {
1710 chunk->fast_retransmit = SCTP_NEED_FRTX;
1711 do_fast_retransmit = 1;
1716 if (do_fast_retransmit)
1717 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1719 pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, "
1720 "flight_size:%d, pba:%d\n", __func__, transport,
1721 transport->cwnd, transport->ssthresh,
1722 transport->flight_size, transport->partial_bytes_acked);
1726 /* Is the given TSN acked by this packet? */
1727 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1730 sctp_sack_variable_t *frags;
1731 __u16 tsn_offset, blocks;
1732 __u32 ctsn = ntohl(sack->cum_tsn_ack);
1734 if (TSN_lte(tsn, ctsn))
1737 /* 3.3.4 Selective Acknowledgement (SACK) (3):
1740 * These fields contain the Gap Ack Blocks. They are repeated
1741 * for each Gap Ack Block up to the number of Gap Ack Blocks
1742 * defined in the Number of Gap Ack Blocks field. All DATA
1743 * chunks with TSNs greater than or equal to (Cumulative TSN
1744 * Ack + Gap Ack Block Start) and less than or equal to
1745 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1746 * Block are assumed to have been received correctly.
1749 frags = sack->variable;
1750 blocks = ntohs(sack->num_gap_ack_blocks);
1751 tsn_offset = tsn - ctsn;
1752 for (i = 0; i < blocks; ++i) {
1753 if (tsn_offset >= ntohs(frags[i].gab.start) &&
1754 tsn_offset <= ntohs(frags[i].gab.end))
1763 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1764 int nskips, __be16 stream)
1768 for (i = 0; i < nskips; i++) {
1769 if (skiplist[i].stream == stream)
1775 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1776 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1778 struct sctp_association *asoc = q->asoc;
1779 struct sctp_chunk *ftsn_chunk = NULL;
1780 struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1784 struct sctp_chunk *chunk;
1785 struct list_head *lchunk, *temp;
1787 if (!asoc->peer.prsctp_capable)
1790 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1793 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1794 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1796 if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1797 asoc->adv_peer_ack_point = ctsn;
1799 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1800 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1801 * the chunk next in the out-queue space is marked as "abandoned" as
1802 * shown in the following example:
1804 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1805 * and the Advanced.Peer.Ack.Point is updated to this value:
1807 * out-queue at the end of ==> out-queue after Adv.Ack.Point
1808 * normal SACK processing local advancement
1810 * Adv.Ack.Pt-> 102 acked 102 acked
1811 * 103 abandoned 103 abandoned
1812 * 104 abandoned Adv.Ack.P-> 104 abandoned
1814 * 106 acked 106 acked
1817 * In this example, the data sender successfully advanced the
1818 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1820 list_for_each_safe(lchunk, temp, &q->abandoned) {
1821 chunk = list_entry(lchunk, struct sctp_chunk,
1823 tsn = ntohl(chunk->subh.data_hdr->tsn);
1825 /* Remove any chunks in the abandoned queue that are acked by
1828 if (TSN_lte(tsn, ctsn)) {
1829 list_del_init(lchunk);
1830 sctp_chunk_free(chunk);
1832 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1833 asoc->adv_peer_ack_point = tsn;
1834 if (chunk->chunk_hdr->flags &
1835 SCTP_DATA_UNORDERED)
1837 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1839 chunk->subh.data_hdr->stream);
1840 ftsn_skip_arr[skip_pos].stream =
1841 chunk->subh.data_hdr->stream;
1842 ftsn_skip_arr[skip_pos].ssn =
1843 chunk->subh.data_hdr->ssn;
1844 if (skip_pos == nskips)
1853 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1854 * is greater than the Cumulative TSN ACK carried in the received
1855 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1856 * chunk containing the latest value of the
1857 * "Advanced.Peer.Ack.Point".
1859 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1860 * list each stream and sequence number in the forwarded TSN. This
1861 * information will enable the receiver to easily find any
1862 * stranded TSN's waiting on stream reorder queues. Each stream
1863 * SHOULD only be reported once; this means that if multiple
1864 * abandoned messages occur in the same stream then only the
1865 * highest abandoned stream sequence number is reported. If the
1866 * total size of the FORWARD TSN does NOT fit in a single MTU then
1867 * the sender of the FORWARD TSN SHOULD lower the
1868 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1871 if (asoc->adv_peer_ack_point > ctsn)
1872 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1873 nskips, &ftsn_skip_arr[0]);
1876 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1877 SCTP_INC_STATS(sock_net(asoc->base.sk), SCTP_MIB_OUTCTRLCHUNKS);