xen: fix xen_flush_tlb_others
[linux/fpc-iii.git] / net / sctp / outqueue.c
blobbc411c89621635159713716a1747253f6d138a27
1 /* SCTP kernel implementation
2 * (C) Copyright IBM Corp. 2001, 2004
3 * Copyright (c) 1999-2000 Cisco, Inc.
4 * Copyright (c) 1999-2001 Motorola, Inc.
5 * Copyright (c) 2001-2003 Intel Corp.
7 * This file is part of the SCTP kernel implementation
9 * These functions implement the sctp_outq class. The outqueue handles
10 * bundling and queueing of outgoing SCTP chunks.
12 * This SCTP implementation is free software;
13 * you can redistribute it and/or modify it under the terms of
14 * the GNU General Public License as published by
15 * the Free Software Foundation; either version 2, or (at your option)
16 * any later version.
18 * This SCTP implementation is distributed in the hope that it
19 * will be useful, but WITHOUT ANY WARRANTY; without even the implied
20 * ************************
21 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
22 * See the GNU General Public License for more details.
24 * You should have received a copy of the GNU General Public License
25 * along with GNU CC; see the file COPYING. If not, write to
26 * the Free Software Foundation, 59 Temple Place - Suite 330,
27 * Boston, MA 02111-1307, USA.
29 * Please send any bug reports or fixes you make to the
30 * email address(es):
31 * lksctp developers <lksctp-developers@lists.sourceforge.net>
33 * Or submit a bug report through the following website:
34 * http://www.sf.net/projects/lksctp
36 * Written or modified by:
37 * La Monte H.P. Yarroll <piggy@acm.org>
38 * Karl Knutson <karl@athena.chicago.il.us>
39 * Perry Melange <pmelange@null.cc.uic.edu>
40 * Xingang Guo <xingang.guo@intel.com>
41 * Hui Huang <hui.huang@nokia.com>
42 * Sridhar Samudrala <sri@us.ibm.com>
43 * Jon Grimm <jgrimm@us.ibm.com>
45 * Any bugs reported given to us we will try to fix... any fixes shared will
46 * be incorporated into the next SCTP release.
49 #include <linux/types.h>
50 #include <linux/list.h> /* For struct list_head */
51 #include <linux/socket.h>
52 #include <linux/ip.h>
53 #include <net/sock.h> /* For skb_set_owner_w */
55 #include <net/sctp/sctp.h>
56 #include <net/sctp/sm.h>
58 /* Declare internal functions here. */
59 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn);
60 static void sctp_check_transmitted(struct sctp_outq *q,
61 struct list_head *transmitted_queue,
62 struct sctp_transport *transport,
63 struct sctp_sackhdr *sack,
64 __u32 highest_new_tsn);
66 static void sctp_mark_missing(struct sctp_outq *q,
67 struct list_head *transmitted_queue,
68 struct sctp_transport *transport,
69 __u32 highest_new_tsn,
70 int count_of_newacks);
72 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn);
74 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout);
76 /* Add data to the front of the queue. */
77 static inline void sctp_outq_head_data(struct sctp_outq *q,
78 struct sctp_chunk *ch)
80 list_add(&ch->list, &q->out_chunk_list);
81 q->out_qlen += ch->skb->len;
82 return;
85 /* Take data from the front of the queue. */
86 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
88 struct sctp_chunk *ch = NULL;
90 if (!list_empty(&q->out_chunk_list)) {
91 struct list_head *entry = q->out_chunk_list.next;
93 ch = list_entry(entry, struct sctp_chunk, list);
94 list_del_init(entry);
95 q->out_qlen -= ch->skb->len;
97 return ch;
99 /* Add data chunk to the end of the queue. */
100 static inline void sctp_outq_tail_data(struct sctp_outq *q,
101 struct sctp_chunk *ch)
103 list_add_tail(&ch->list, &q->out_chunk_list);
104 q->out_qlen += ch->skb->len;
105 return;
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)
119 return 1;
120 return 0;
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 && !transport->cacc.cacc_saw_newack)
134 return 1;
135 return 0;
139 * SFR-CACC algorithm:
140 * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
141 * execute steps C, D, F.
143 * C has been implemented in sctp_outq_sack
145 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
146 struct sctp_transport *transport,
147 int count_of_newacks)
149 if (!primary->cacc.cycling_changeover) {
150 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
151 return 1;
152 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
153 return 1;
154 return 0;
156 return 0;
160 * SFR-CACC algorithm:
161 * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
162 * than next_tsn_at_change of the current primary, then
163 * the sender MUST NOT increment missing report count
164 * for t.
166 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
168 if (primary->cacc.cycling_changeover &&
169 TSN_lt(tsn, primary->cacc.next_tsn_at_change))
170 return 1;
171 return 0;
175 * SFR-CACC algorithm:
176 * 3) If the missing report count for TSN t is to be
177 * incremented according to [RFC2960] and
178 * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
179 * then the sender MUST futher execute steps 3.1 and
180 * 3.2 to determine if the missing report count for
181 * TSN t SHOULD NOT be incremented.
183 * 3.3) If 3.1 and 3.2 do not dictate that the missing
184 * report count for t should not be incremented, then
185 * the sender SOULD increment missing report count for
186 * t (according to [RFC2960] and [SCTP_STEWART_2002]).
188 static inline int sctp_cacc_skip(struct sctp_transport *primary,
189 struct sctp_transport *transport,
190 int count_of_newacks,
191 __u32 tsn)
193 if (primary->cacc.changeover_active &&
194 (sctp_cacc_skip_3_1(primary, transport, count_of_newacks)
195 || sctp_cacc_skip_3_2(primary, tsn)))
196 return 1;
197 return 0;
200 /* Initialize an existing sctp_outq. This does the boring stuff.
201 * You still need to define handlers if you really want to DO
202 * something with this structure...
204 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
206 q->asoc = asoc;
207 INIT_LIST_HEAD(&q->out_chunk_list);
208 INIT_LIST_HEAD(&q->control_chunk_list);
209 INIT_LIST_HEAD(&q->retransmit);
210 INIT_LIST_HEAD(&q->sacked);
211 INIT_LIST_HEAD(&q->abandoned);
213 q->fast_rtx = 0;
214 q->outstanding_bytes = 0;
215 q->empty = 1;
216 q->cork = 0;
218 q->malloced = 0;
219 q->out_qlen = 0;
222 /* Free the outqueue structure and any related pending chunks.
224 void sctp_outq_teardown(struct sctp_outq *q)
226 struct sctp_transport *transport;
227 struct list_head *lchunk, *temp;
228 struct sctp_chunk *chunk, *tmp;
230 /* Throw away unacknowledged chunks. */
231 list_for_each_entry(transport, &q->asoc->peer.transport_addr_list,
232 transports) {
233 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
234 chunk = list_entry(lchunk, struct sctp_chunk,
235 transmitted_list);
236 /* Mark as part of a failed message. */
237 sctp_chunk_fail(chunk, q->error);
238 sctp_chunk_free(chunk);
242 /* Throw away chunks that have been gap ACKed. */
243 list_for_each_safe(lchunk, temp, &q->sacked) {
244 list_del_init(lchunk);
245 chunk = list_entry(lchunk, struct sctp_chunk,
246 transmitted_list);
247 sctp_chunk_fail(chunk, q->error);
248 sctp_chunk_free(chunk);
251 /* Throw away any chunks in the retransmit queue. */
252 list_for_each_safe(lchunk, temp, &q->retransmit) {
253 list_del_init(lchunk);
254 chunk = list_entry(lchunk, struct sctp_chunk,
255 transmitted_list);
256 sctp_chunk_fail(chunk, q->error);
257 sctp_chunk_free(chunk);
260 /* Throw away any chunks that are in the abandoned queue. */
261 list_for_each_safe(lchunk, temp, &q->abandoned) {
262 list_del_init(lchunk);
263 chunk = list_entry(lchunk, struct sctp_chunk,
264 transmitted_list);
265 sctp_chunk_fail(chunk, q->error);
266 sctp_chunk_free(chunk);
269 /* Throw away any leftover data chunks. */
270 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
272 /* Mark as send failure. */
273 sctp_chunk_fail(chunk, q->error);
274 sctp_chunk_free(chunk);
277 q->error = 0;
279 /* Throw away any leftover control chunks. */
280 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
281 list_del_init(&chunk->list);
282 sctp_chunk_free(chunk);
286 /* Free the outqueue structure and any related pending chunks. */
287 void sctp_outq_free(struct sctp_outq *q)
289 /* Throw away leftover chunks. */
290 sctp_outq_teardown(q);
292 /* If we were kmalloc()'d, free the memory. */
293 if (q->malloced)
294 kfree(q);
297 /* Put a new chunk in an sctp_outq. */
298 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk)
300 int error = 0;
302 SCTP_DEBUG_PRINTK("sctp_outq_tail(%p, %p[%s])\n",
303 q, chunk, chunk && chunk->chunk_hdr ?
304 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
305 : "Illegal Chunk");
307 /* If it is data, queue it up, otherwise, send it
308 * immediately.
310 if (SCTP_CID_DATA == chunk->chunk_hdr->type) {
311 /* Is it OK to queue data chunks? */
312 /* From 9. Termination of Association
314 * When either endpoint performs a shutdown, the
315 * association on each peer will stop accepting new
316 * data from its user and only deliver data in queue
317 * at the time of sending or receiving the SHUTDOWN
318 * chunk.
320 switch (q->asoc->state) {
321 case SCTP_STATE_EMPTY:
322 case SCTP_STATE_CLOSED:
323 case SCTP_STATE_SHUTDOWN_PENDING:
324 case SCTP_STATE_SHUTDOWN_SENT:
325 case SCTP_STATE_SHUTDOWN_RECEIVED:
326 case SCTP_STATE_SHUTDOWN_ACK_SENT:
327 /* Cannot send after transport endpoint shutdown */
328 error = -ESHUTDOWN;
329 break;
331 default:
332 SCTP_DEBUG_PRINTK("outqueueing (%p, %p[%s])\n",
333 q, chunk, chunk && chunk->chunk_hdr ?
334 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
335 : "Illegal Chunk");
337 sctp_outq_tail_data(q, chunk);
338 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
339 SCTP_INC_STATS(SCTP_MIB_OUTUNORDERCHUNKS);
340 else
341 SCTP_INC_STATS(SCTP_MIB_OUTORDERCHUNKS);
342 q->empty = 0;
343 break;
345 } else {
346 list_add_tail(&chunk->list, &q->control_chunk_list);
347 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
350 if (error < 0)
351 return error;
353 if (!q->cork)
354 error = sctp_outq_flush(q, 0);
356 return error;
359 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list
360 * and the abandoned list are in ascending order.
362 static void sctp_insert_list(struct list_head *head, struct list_head *new)
364 struct list_head *pos;
365 struct sctp_chunk *nchunk, *lchunk;
366 __u32 ntsn, ltsn;
367 int done = 0;
369 nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
370 ntsn = ntohl(nchunk->subh.data_hdr->tsn);
372 list_for_each(pos, head) {
373 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
374 ltsn = ntohl(lchunk->subh.data_hdr->tsn);
375 if (TSN_lt(ntsn, ltsn)) {
376 list_add(new, pos->prev);
377 done = 1;
378 break;
381 if (!done)
382 list_add_tail(new, head);
385 /* Mark all the eligible packets on a transport for retransmission. */
386 void sctp_retransmit_mark(struct sctp_outq *q,
387 struct sctp_transport *transport,
388 __u8 reason)
390 struct list_head *lchunk, *ltemp;
391 struct sctp_chunk *chunk;
393 /* Walk through the specified transmitted queue. */
394 list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
395 chunk = list_entry(lchunk, struct sctp_chunk,
396 transmitted_list);
398 /* If the chunk is abandoned, move it to abandoned list. */
399 if (sctp_chunk_abandoned(chunk)) {
400 list_del_init(lchunk);
401 sctp_insert_list(&q->abandoned, lchunk);
403 /* If this chunk has not been previousely acked,
404 * stop considering it 'outstanding'. Our peer
405 * will most likely never see it since it will
406 * not be retransmitted
408 if (!chunk->tsn_gap_acked) {
409 chunk->transport->flight_size -=
410 sctp_data_size(chunk);
411 q->outstanding_bytes -= sctp_data_size(chunk);
412 q->asoc->peer.rwnd += (sctp_data_size(chunk) +
413 sizeof(struct sk_buff));
415 continue;
418 /* If we are doing retransmission due to a timeout or pmtu
419 * discovery, only the chunks that are not yet acked should
420 * be added to the retransmit queue.
422 if ((reason == SCTP_RTXR_FAST_RTX &&
423 (chunk->fast_retransmit == SCTP_NEED_FRTX)) ||
424 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) {
425 /* If this chunk was sent less then 1 rto ago, do not
426 * retransmit this chunk, but give the peer time
427 * to acknowlege it. Do this only when
428 * retransmitting due to T3 timeout.
430 if (reason == SCTP_RTXR_T3_RTX &&
431 (jiffies - chunk->sent_at) < transport->last_rto)
432 continue;
434 /* RFC 2960 6.2.1 Processing a Received SACK
436 * C) Any time a DATA chunk is marked for
437 * retransmission (via either T3-rtx timer expiration
438 * (Section 6.3.3) or via fast retransmit
439 * (Section 7.2.4)), add the data size of those
440 * chunks to the rwnd.
442 q->asoc->peer.rwnd += (sctp_data_size(chunk) +
443 sizeof(struct sk_buff));
444 q->outstanding_bytes -= sctp_data_size(chunk);
445 transport->flight_size -= sctp_data_size(chunk);
447 /* sctpimpguide-05 Section 2.8.2
448 * M5) If a T3-rtx timer expires, the
449 * 'TSN.Missing.Report' of all affected TSNs is set
450 * to 0.
452 chunk->tsn_missing_report = 0;
454 /* If a chunk that is being used for RTT measurement
455 * has to be retransmitted, we cannot use this chunk
456 * anymore for RTT measurements. Reset rto_pending so
457 * that a new RTT measurement is started when a new
458 * data chunk is sent.
460 if (chunk->rtt_in_progress) {
461 chunk->rtt_in_progress = 0;
462 transport->rto_pending = 0;
465 /* Move the chunk to the retransmit queue. The chunks
466 * on the retransmit queue are always kept in order.
468 list_del_init(lchunk);
469 sctp_insert_list(&q->retransmit, lchunk);
473 SCTP_DEBUG_PRINTK("%s: transport: %p, reason: %d, "
474 "cwnd: %d, ssthresh: %d, flight_size: %d, "
475 "pba: %d\n", __func__,
476 transport, reason,
477 transport->cwnd, transport->ssthresh,
478 transport->flight_size,
479 transport->partial_bytes_acked);
483 /* Mark all the eligible packets on a transport for retransmission and force
484 * one packet out.
486 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
487 sctp_retransmit_reason_t reason)
489 int error = 0;
491 switch(reason) {
492 case SCTP_RTXR_T3_RTX:
493 SCTP_INC_STATS(SCTP_MIB_T3_RETRANSMITS);
494 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
495 /* Update the retran path if the T3-rtx timer has expired for
496 * the current retran path.
498 if (transport == transport->asoc->peer.retran_path)
499 sctp_assoc_update_retran_path(transport->asoc);
500 transport->asoc->rtx_data_chunks +=
501 transport->asoc->unack_data;
502 break;
503 case SCTP_RTXR_FAST_RTX:
504 SCTP_INC_STATS(SCTP_MIB_FAST_RETRANSMITS);
505 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
506 q->fast_rtx = 1;
507 break;
508 case SCTP_RTXR_PMTUD:
509 SCTP_INC_STATS(SCTP_MIB_PMTUD_RETRANSMITS);
510 break;
511 case SCTP_RTXR_T1_RTX:
512 SCTP_INC_STATS(SCTP_MIB_T1_RETRANSMITS);
513 transport->asoc->init_retries++;
514 break;
515 default:
516 BUG();
519 sctp_retransmit_mark(q, transport, reason);
521 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
522 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
523 * following the procedures outlined in C1 - C5.
525 if (reason == SCTP_RTXR_T3_RTX)
526 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
528 /* Flush the queues only on timeout, since fast_rtx is only
529 * triggered during sack processing and the queue
530 * will be flushed at the end.
532 if (reason != SCTP_RTXR_FAST_RTX)
533 error = sctp_outq_flush(q, /* rtx_timeout */ 1);
535 if (error)
536 q->asoc->base.sk->sk_err = -error;
540 * Transmit DATA chunks on the retransmit queue. Upon return from
541 * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
542 * need to be transmitted by the caller.
543 * We assume that pkt->transport has already been set.
545 * The return value is a normal kernel error return value.
547 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
548 int rtx_timeout, int *start_timer)
550 struct list_head *lqueue;
551 struct sctp_transport *transport = pkt->transport;
552 sctp_xmit_t status;
553 struct sctp_chunk *chunk, *chunk1;
554 struct sctp_association *asoc;
555 int fast_rtx;
556 int error = 0;
557 int timer = 0;
558 int done = 0;
560 asoc = q->asoc;
561 lqueue = &q->retransmit;
562 fast_rtx = q->fast_rtx;
564 /* This loop handles time-out retransmissions, fast retransmissions,
565 * and retransmissions due to opening of whindow.
567 * RFC 2960 6.3.3 Handle T3-rtx Expiration
569 * E3) Determine how many of the earliest (i.e., lowest TSN)
570 * outstanding DATA chunks for the address for which the
571 * T3-rtx has expired will fit into a single packet, subject
572 * to the MTU constraint for the path corresponding to the
573 * destination transport address to which the retransmission
574 * is being sent (this may be different from the address for
575 * which the timer expires [see Section 6.4]). Call this value
576 * K. Bundle and retransmit those K DATA chunks in a single
577 * packet to the destination endpoint.
579 * [Just to be painfully clear, if we are retransmitting
580 * because a timeout just happened, we should send only ONE
581 * packet of retransmitted data.]
583 * For fast retransmissions we also send only ONE packet. However,
584 * if we are just flushing the queue due to open window, we'll
585 * try to send as much as possible.
587 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
589 /* Make sure that Gap Acked TSNs are not retransmitted. A
590 * simple approach is just to move such TSNs out of the
591 * way and into a 'transmitted' queue and skip to the
592 * next chunk.
594 if (chunk->tsn_gap_acked) {
595 list_del(&chunk->transmitted_list);
596 list_add_tail(&chunk->transmitted_list,
597 &transport->transmitted);
598 continue;
601 /* If we are doing fast retransmit, ignore non-fast_rtransmit
602 * chunks
604 if (fast_rtx && !chunk->fast_retransmit)
605 continue;
607 /* Attempt to append this chunk to the packet. */
608 status = sctp_packet_append_chunk(pkt, chunk);
610 switch (status) {
611 case SCTP_XMIT_PMTU_FULL:
612 /* Send this packet. */
613 error = sctp_packet_transmit(pkt);
615 /* If we are retransmitting, we should only
616 * send a single packet.
618 if (rtx_timeout || fast_rtx)
619 done = 1;
621 /* Bundle next chunk in the next round. */
622 break;
624 case SCTP_XMIT_RWND_FULL:
625 /* Send this packet. */
626 error = sctp_packet_transmit(pkt);
628 /* Stop sending DATA as there is no more room
629 * at the receiver.
631 done = 1;
632 break;
634 case SCTP_XMIT_NAGLE_DELAY:
635 /* Send this packet. */
636 error = sctp_packet_transmit(pkt);
638 /* Stop sending DATA because of nagle delay. */
639 done = 1;
640 break;
642 default:
643 /* The append was successful, so add this chunk to
644 * the transmitted list.
646 list_del(&chunk->transmitted_list);
647 list_add_tail(&chunk->transmitted_list,
648 &transport->transmitted);
650 /* Mark the chunk as ineligible for fast retransmit
651 * after it is retransmitted.
653 if (chunk->fast_retransmit == SCTP_NEED_FRTX)
654 chunk->fast_retransmit = SCTP_DONT_FRTX;
656 /* Force start T3-rtx timer when fast retransmitting
657 * the earliest outstanding TSN
659 if (!timer && fast_rtx &&
660 ntohl(chunk->subh.data_hdr->tsn) ==
661 asoc->ctsn_ack_point + 1)
662 timer = 2;
664 q->empty = 0;
665 break;
668 /* Set the timer if there were no errors */
669 if (!error && !timer)
670 timer = 1;
672 if (done)
673 break;
676 /* If we are here due to a retransmit timeout or a fast
677 * retransmit and if there are any chunks left in the retransmit
678 * queue that could not fit in the PMTU sized packet, they need
679 * to be marked as ineligible for a subsequent fast retransmit.
681 if (rtx_timeout || fast_rtx) {
682 list_for_each_entry(chunk1, lqueue, transmitted_list) {
683 if (chunk1->fast_retransmit == SCTP_NEED_FRTX)
684 chunk1->fast_retransmit = SCTP_DONT_FRTX;
688 *start_timer = timer;
690 /* Clear fast retransmit hint */
691 if (fast_rtx)
692 q->fast_rtx = 0;
694 return error;
697 /* Cork the outqueue so queued chunks are really queued. */
698 int sctp_outq_uncork(struct sctp_outq *q)
700 int error = 0;
701 if (q->cork)
702 q->cork = 0;
703 error = sctp_outq_flush(q, 0);
704 return error;
709 * Try to flush an outqueue.
711 * Description: Send everything in q which we legally can, subject to
712 * congestion limitations.
713 * * Note: This function can be called from multiple contexts so appropriate
714 * locking concerns must be made. Today we use the sock lock to protect
715 * this function.
717 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
719 struct sctp_packet *packet;
720 struct sctp_packet singleton;
721 struct sctp_association *asoc = q->asoc;
722 __u16 sport = asoc->base.bind_addr.port;
723 __u16 dport = asoc->peer.port;
724 __u32 vtag = asoc->peer.i.init_tag;
725 struct sctp_transport *transport = NULL;
726 struct sctp_transport *new_transport;
727 struct sctp_chunk *chunk, *tmp;
728 sctp_xmit_t status;
729 int error = 0;
730 int start_timer = 0;
731 int one_packet = 0;
733 /* These transports have chunks to send. */
734 struct list_head transport_list;
735 struct list_head *ltransport;
737 INIT_LIST_HEAD(&transport_list);
738 packet = NULL;
741 * 6.10 Bundling
742 * ...
743 * When bundling control chunks with DATA chunks, an
744 * endpoint MUST place control chunks first in the outbound
745 * SCTP packet. The transmitter MUST transmit DATA chunks
746 * within a SCTP packet in increasing order of TSN.
747 * ...
750 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
751 list_del_init(&chunk->list);
753 /* Pick the right transport to use. */
754 new_transport = chunk->transport;
756 if (!new_transport) {
758 * If we have a prior transport pointer, see if
759 * the destination address of the chunk
760 * matches the destination address of the
761 * current transport. If not a match, then
762 * try to look up the transport with a given
763 * destination address. We do this because
764 * after processing ASCONFs, we may have new
765 * transports created.
767 if (transport &&
768 sctp_cmp_addr_exact(&chunk->dest,
769 &transport->ipaddr))
770 new_transport = transport;
771 else
772 new_transport = sctp_assoc_lookup_paddr(asoc,
773 &chunk->dest);
775 /* if we still don't have a new transport, then
776 * use the current active path.
778 if (!new_transport)
779 new_transport = asoc->peer.active_path;
780 } else if ((new_transport->state == SCTP_INACTIVE) ||
781 (new_transport->state == SCTP_UNCONFIRMED)) {
782 /* If the chunk is Heartbeat or Heartbeat Ack,
783 * send it to chunk->transport, even if it's
784 * inactive.
786 * 3.3.6 Heartbeat Acknowledgement:
787 * ...
788 * A HEARTBEAT ACK is always sent to the source IP
789 * address of the IP datagram containing the
790 * HEARTBEAT chunk to which this ack is responding.
791 * ...
793 * ASCONF_ACKs also must be sent to the source.
795 if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
796 chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK &&
797 chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK)
798 new_transport = asoc->peer.active_path;
801 /* Are we switching transports?
802 * Take care of transport locks.
804 if (new_transport != transport) {
805 transport = new_transport;
806 if (list_empty(&transport->send_ready)) {
807 list_add_tail(&transport->send_ready,
808 &transport_list);
810 packet = &transport->packet;
811 sctp_packet_config(packet, vtag,
812 asoc->peer.ecn_capable);
815 switch (chunk->chunk_hdr->type) {
817 * 6.10 Bundling
818 * ...
819 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
820 * COMPLETE with any other chunks. [Send them immediately.]
822 case SCTP_CID_INIT:
823 case SCTP_CID_INIT_ACK:
824 case SCTP_CID_SHUTDOWN_COMPLETE:
825 sctp_packet_init(&singleton, transport, sport, dport);
826 sctp_packet_config(&singleton, vtag, 0);
827 sctp_packet_append_chunk(&singleton, chunk);
828 error = sctp_packet_transmit(&singleton);
829 if (error < 0)
830 return error;
831 break;
833 case SCTP_CID_ABORT:
834 if (sctp_test_T_bit(chunk)) {
835 packet->vtag = asoc->c.my_vtag;
837 /* The following chunks are "response" chunks, i.e.
838 * they are generated in response to something we
839 * received. If we are sending these, then we can
840 * send only 1 packet containing these chunks.
842 case SCTP_CID_HEARTBEAT_ACK:
843 case SCTP_CID_SHUTDOWN_ACK:
844 case SCTP_CID_COOKIE_ACK:
845 case SCTP_CID_COOKIE_ECHO:
846 case SCTP_CID_ERROR:
847 case SCTP_CID_ECN_CWR:
848 case SCTP_CID_ASCONF_ACK:
849 one_packet = 1;
850 /* Fall throught */
852 case SCTP_CID_SACK:
853 case SCTP_CID_HEARTBEAT:
854 case SCTP_CID_SHUTDOWN:
855 case SCTP_CID_ECN_ECNE:
856 case SCTP_CID_ASCONF:
857 case SCTP_CID_FWD_TSN:
858 status = sctp_packet_transmit_chunk(packet, chunk,
859 one_packet);
860 if (status != SCTP_XMIT_OK) {
861 /* put the chunk back */
862 list_add(&chunk->list, &q->control_chunk_list);
864 break;
866 default:
867 /* We built a chunk with an illegal type! */
868 BUG();
872 /* Is it OK to send data chunks? */
873 switch (asoc->state) {
874 case SCTP_STATE_COOKIE_ECHOED:
875 /* Only allow bundling when this packet has a COOKIE-ECHO
876 * chunk.
878 if (!packet || !packet->has_cookie_echo)
879 break;
881 /* fallthru */
882 case SCTP_STATE_ESTABLISHED:
883 case SCTP_STATE_SHUTDOWN_PENDING:
884 case SCTP_STATE_SHUTDOWN_RECEIVED:
886 * RFC 2960 6.1 Transmission of DATA Chunks
888 * C) When the time comes for the sender to transmit,
889 * before sending new DATA chunks, the sender MUST
890 * first transmit any outstanding DATA chunks which
891 * are marked for retransmission (limited by the
892 * current cwnd).
894 if (!list_empty(&q->retransmit)) {
895 if (transport == asoc->peer.retran_path)
896 goto retran;
898 /* Switch transports & prepare the packet. */
900 transport = asoc->peer.retran_path;
902 if (list_empty(&transport->send_ready)) {
903 list_add_tail(&transport->send_ready,
904 &transport_list);
907 packet = &transport->packet;
908 sctp_packet_config(packet, vtag,
909 asoc->peer.ecn_capable);
910 retran:
911 error = sctp_outq_flush_rtx(q, packet,
912 rtx_timeout, &start_timer);
914 if (start_timer)
915 sctp_transport_reset_timers(transport,
916 start_timer-1);
918 /* This can happen on COOKIE-ECHO resend. Only
919 * one chunk can get bundled with a COOKIE-ECHO.
921 if (packet->has_cookie_echo)
922 goto sctp_flush_out;
924 /* Don't send new data if there is still data
925 * waiting to retransmit.
927 if (!list_empty(&q->retransmit))
928 goto sctp_flush_out;
931 /* Finally, transmit new packets. */
932 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
933 /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
934 * stream identifier.
936 if (chunk->sinfo.sinfo_stream >=
937 asoc->c.sinit_num_ostreams) {
939 /* Mark as failed send. */
940 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
941 sctp_chunk_free(chunk);
942 continue;
945 /* Has this chunk expired? */
946 if (sctp_chunk_abandoned(chunk)) {
947 sctp_chunk_fail(chunk, 0);
948 sctp_chunk_free(chunk);
949 continue;
952 /* If there is a specified transport, use it.
953 * Otherwise, we want to use the active path.
955 new_transport = chunk->transport;
956 if (!new_transport ||
957 ((new_transport->state == SCTP_INACTIVE) ||
958 (new_transport->state == SCTP_UNCONFIRMED)))
959 new_transport = asoc->peer.active_path;
961 /* Change packets if necessary. */
962 if (new_transport != transport) {
963 transport = new_transport;
965 /* Schedule to have this transport's
966 * packet flushed.
968 if (list_empty(&transport->send_ready)) {
969 list_add_tail(&transport->send_ready,
970 &transport_list);
973 packet = &transport->packet;
974 sctp_packet_config(packet, vtag,
975 asoc->peer.ecn_capable);
978 SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
979 q, chunk,
980 chunk && chunk->chunk_hdr ?
981 sctp_cname(SCTP_ST_CHUNK(
982 chunk->chunk_hdr->type))
983 : "Illegal Chunk");
985 SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
986 "%p skb->users %d.\n",
987 ntohl(chunk->subh.data_hdr->tsn),
988 chunk->skb ?chunk->skb->head : NULL,
989 chunk->skb ?
990 atomic_read(&chunk->skb->users) : -1);
992 /* Add the chunk to the packet. */
993 status = sctp_packet_transmit_chunk(packet, chunk, 0);
995 switch (status) {
996 case SCTP_XMIT_PMTU_FULL:
997 case SCTP_XMIT_RWND_FULL:
998 case SCTP_XMIT_NAGLE_DELAY:
999 /* We could not append this chunk, so put
1000 * the chunk back on the output queue.
1002 SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
1003 "not transmit TSN: 0x%x, status: %d\n",
1004 ntohl(chunk->subh.data_hdr->tsn),
1005 status);
1006 sctp_outq_head_data(q, chunk);
1007 goto sctp_flush_out;
1008 break;
1010 case SCTP_XMIT_OK:
1011 break;
1013 default:
1014 BUG();
1017 /* BUG: We assume that the sctp_packet_transmit()
1018 * call below will succeed all the time and add the
1019 * chunk to the transmitted list and restart the
1020 * timers.
1021 * It is possible that the call can fail under OOM
1022 * conditions.
1024 * Is this really a problem? Won't this behave
1025 * like a lost TSN?
1027 list_add_tail(&chunk->transmitted_list,
1028 &transport->transmitted);
1030 sctp_transport_reset_timers(transport, 0);
1032 q->empty = 0;
1034 /* Only let one DATA chunk get bundled with a
1035 * COOKIE-ECHO chunk.
1037 if (packet->has_cookie_echo)
1038 goto sctp_flush_out;
1040 break;
1042 default:
1043 /* Do nothing. */
1044 break;
1047 sctp_flush_out:
1049 /* Before returning, examine all the transports touched in
1050 * this call. Right now, we bluntly force clear all the
1051 * transports. Things might change after we implement Nagle.
1052 * But such an examination is still required.
1054 * --xguo
1056 while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
1057 struct sctp_transport *t = list_entry(ltransport,
1058 struct sctp_transport,
1059 send_ready);
1060 packet = &t->packet;
1061 if (!sctp_packet_empty(packet))
1062 error = sctp_packet_transmit(packet);
1065 return error;
1068 /* Update unack_data based on the incoming SACK chunk */
1069 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1070 struct sctp_sackhdr *sack)
1072 sctp_sack_variable_t *frags;
1073 __u16 unack_data;
1074 int i;
1076 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1078 frags = sack->variable;
1079 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1080 unack_data -= ((ntohs(frags[i].gab.end) -
1081 ntohs(frags[i].gab.start) + 1));
1084 assoc->unack_data = unack_data;
1087 /* Return the highest new tsn that is acknowledged by the given SACK chunk. */
1088 static __u32 sctp_highest_new_tsn(struct sctp_sackhdr *sack,
1089 struct sctp_association *asoc)
1091 struct sctp_transport *transport;
1092 struct sctp_chunk *chunk;
1093 __u32 highest_new_tsn, tsn;
1094 struct list_head *transport_list = &asoc->peer.transport_addr_list;
1096 highest_new_tsn = ntohl(sack->cum_tsn_ack);
1098 list_for_each_entry(transport, transport_list, transports) {
1099 list_for_each_entry(chunk, &transport->transmitted,
1100 transmitted_list) {
1101 tsn = ntohl(chunk->subh.data_hdr->tsn);
1103 if (!chunk->tsn_gap_acked &&
1104 TSN_lt(highest_new_tsn, tsn) &&
1105 sctp_acked(sack, tsn))
1106 highest_new_tsn = tsn;
1110 return highest_new_tsn;
1113 /* This is where we REALLY process a SACK.
1115 * Process the SACK against the outqueue. Mostly, this just frees
1116 * things off the transmitted queue.
1118 int sctp_outq_sack(struct sctp_outq *q, struct sctp_sackhdr *sack)
1120 struct sctp_association *asoc = q->asoc;
1121 struct sctp_transport *transport;
1122 struct sctp_chunk *tchunk = NULL;
1123 struct list_head *lchunk, *transport_list, *temp;
1124 sctp_sack_variable_t *frags = sack->variable;
1125 __u32 sack_ctsn, ctsn, tsn;
1126 __u32 highest_tsn, highest_new_tsn;
1127 __u32 sack_a_rwnd;
1128 unsigned outstanding;
1129 struct sctp_transport *primary = asoc->peer.primary_path;
1130 int count_of_newacks = 0;
1131 int gap_ack_blocks;
1133 /* Grab the association's destination address list. */
1134 transport_list = &asoc->peer.transport_addr_list;
1136 sack_ctsn = ntohl(sack->cum_tsn_ack);
1137 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1139 * SFR-CACC algorithm:
1140 * On receipt of a SACK the sender SHOULD execute the
1141 * following statements.
1143 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1144 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1145 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1146 * all destinations.
1147 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1148 * is set the receiver of the SACK MUST take the following actions:
1150 * A) Initialize the cacc_saw_newack to 0 for all destination
1151 * addresses.
1153 * Only bother if changeover_active is set. Otherwise, this is
1154 * totally suboptimal to do on every SACK.
1156 if (primary->cacc.changeover_active) {
1157 u8 clear_cycling = 0;
1159 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1160 primary->cacc.changeover_active = 0;
1161 clear_cycling = 1;
1164 if (clear_cycling || gap_ack_blocks) {
1165 list_for_each_entry(transport, transport_list,
1166 transports) {
1167 if (clear_cycling)
1168 transport->cacc.cycling_changeover = 0;
1169 if (gap_ack_blocks)
1170 transport->cacc.cacc_saw_newack = 0;
1175 /* Get the highest TSN in the sack. */
1176 highest_tsn = sack_ctsn;
1177 if (gap_ack_blocks)
1178 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1180 if (TSN_lt(asoc->highest_sacked, highest_tsn)) {
1181 highest_new_tsn = highest_tsn;
1182 asoc->highest_sacked = highest_tsn;
1183 } else {
1184 highest_new_tsn = sctp_highest_new_tsn(sack, asoc);
1188 /* Run through the retransmit queue. Credit bytes received
1189 * and free those chunks that we can.
1191 sctp_check_transmitted(q, &q->retransmit, NULL, sack, highest_new_tsn);
1193 /* Run through the transmitted queue.
1194 * Credit bytes received and free those chunks which we can.
1196 * This is a MASSIVE candidate for optimization.
1198 list_for_each_entry(transport, transport_list, transports) {
1199 sctp_check_transmitted(q, &transport->transmitted,
1200 transport, sack, highest_new_tsn);
1202 * SFR-CACC algorithm:
1203 * C) Let count_of_newacks be the number of
1204 * destinations for which cacc_saw_newack is set.
1206 if (transport->cacc.cacc_saw_newack)
1207 count_of_newacks ++;
1210 if (gap_ack_blocks) {
1211 list_for_each_entry(transport, transport_list, transports)
1212 sctp_mark_missing(q, &transport->transmitted, transport,
1213 highest_new_tsn, count_of_newacks);
1216 /* Move the Cumulative TSN Ack Point if appropriate. */
1217 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn))
1218 asoc->ctsn_ack_point = sack_ctsn;
1220 /* Update unack_data field in the assoc. */
1221 sctp_sack_update_unack_data(asoc, sack);
1223 ctsn = asoc->ctsn_ack_point;
1225 /* Throw away stuff rotting on the sack queue. */
1226 list_for_each_safe(lchunk, temp, &q->sacked) {
1227 tchunk = list_entry(lchunk, struct sctp_chunk,
1228 transmitted_list);
1229 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1230 if (TSN_lte(tsn, ctsn)) {
1231 list_del_init(&tchunk->transmitted_list);
1232 sctp_chunk_free(tchunk);
1236 /* ii) Set rwnd equal to the newly received a_rwnd minus the
1237 * number of bytes still outstanding after processing the
1238 * Cumulative TSN Ack and the Gap Ack Blocks.
1241 sack_a_rwnd = ntohl(sack->a_rwnd);
1242 outstanding = q->outstanding_bytes;
1244 if (outstanding < sack_a_rwnd)
1245 sack_a_rwnd -= outstanding;
1246 else
1247 sack_a_rwnd = 0;
1249 asoc->peer.rwnd = sack_a_rwnd;
1251 sctp_generate_fwdtsn(q, sack_ctsn);
1253 SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n",
1254 __func__, sack_ctsn);
1255 SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, "
1256 "%p is 0x%x. Adv peer ack point: 0x%x\n",
1257 __func__, asoc, ctsn, asoc->adv_peer_ack_point);
1259 /* See if all chunks are acked.
1260 * Make sure the empty queue handler will get run later.
1262 q->empty = (list_empty(&q->out_chunk_list) &&
1263 list_empty(&q->retransmit));
1264 if (!q->empty)
1265 goto finish;
1267 list_for_each_entry(transport, transport_list, transports) {
1268 q->empty = q->empty && list_empty(&transport->transmitted);
1269 if (!q->empty)
1270 goto finish;
1273 SCTP_DEBUG_PRINTK("sack queue is empty.\n");
1274 finish:
1275 return q->empty;
1278 /* Is the outqueue empty? */
1279 int sctp_outq_is_empty(const struct sctp_outq *q)
1281 return q->empty;
1284 /********************************************************************
1285 * 2nd Level Abstractions
1286 ********************************************************************/
1288 /* Go through a transport's transmitted list or the association's retransmit
1289 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1290 * The retransmit list will not have an associated transport.
1292 * I added coherent debug information output. --xguo
1294 * Instead of printing 'sacked' or 'kept' for each TSN on the
1295 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1296 * KEPT TSN6-TSN7, etc.
1298 static void sctp_check_transmitted(struct sctp_outq *q,
1299 struct list_head *transmitted_queue,
1300 struct sctp_transport *transport,
1301 struct sctp_sackhdr *sack,
1302 __u32 highest_new_tsn_in_sack)
1304 struct list_head *lchunk;
1305 struct sctp_chunk *tchunk;
1306 struct list_head tlist;
1307 __u32 tsn;
1308 __u32 sack_ctsn;
1309 __u32 rtt;
1310 __u8 restart_timer = 0;
1311 int bytes_acked = 0;
1313 /* These state variables are for coherent debug output. --xguo */
1315 #if SCTP_DEBUG
1316 __u32 dbg_ack_tsn = 0; /* An ACKed TSN range starts here... */
1317 __u32 dbg_last_ack_tsn = 0; /* ...and finishes here. */
1318 __u32 dbg_kept_tsn = 0; /* An un-ACKed range starts here... */
1319 __u32 dbg_last_kept_tsn = 0; /* ...and finishes here. */
1321 /* 0 : The last TSN was ACKed.
1322 * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1323 * -1: We need to initialize.
1325 int dbg_prt_state = -1;
1326 #endif /* SCTP_DEBUG */
1328 sack_ctsn = ntohl(sack->cum_tsn_ack);
1330 INIT_LIST_HEAD(&tlist);
1332 /* The while loop will skip empty transmitted queues. */
1333 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1334 tchunk = list_entry(lchunk, struct sctp_chunk,
1335 transmitted_list);
1337 if (sctp_chunk_abandoned(tchunk)) {
1338 /* Move the chunk to abandoned list. */
1339 sctp_insert_list(&q->abandoned, lchunk);
1341 /* If this chunk has not been acked, stop
1342 * considering it as 'outstanding'.
1344 if (!tchunk->tsn_gap_acked) {
1345 tchunk->transport->flight_size -=
1346 sctp_data_size(tchunk);
1347 q->outstanding_bytes -= sctp_data_size(tchunk);
1349 continue;
1352 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1353 if (sctp_acked(sack, tsn)) {
1354 /* If this queue is the retransmit queue, the
1355 * retransmit timer has already reclaimed
1356 * the outstanding bytes for this chunk, so only
1357 * count bytes associated with a transport.
1359 if (transport) {
1360 /* If this chunk is being used for RTT
1361 * measurement, calculate the RTT and update
1362 * the RTO using this value.
1364 * 6.3.1 C5) Karn's algorithm: RTT measurements
1365 * MUST NOT be made using packets that were
1366 * retransmitted (and thus for which it is
1367 * ambiguous whether the reply was for the
1368 * first instance of the packet or a later
1369 * instance).
1371 if (!tchunk->tsn_gap_acked &&
1372 !tchunk->resent &&
1373 tchunk->rtt_in_progress) {
1374 tchunk->rtt_in_progress = 0;
1375 rtt = jiffies - tchunk->sent_at;
1376 sctp_transport_update_rto(transport,
1377 rtt);
1380 if (TSN_lte(tsn, sack_ctsn)) {
1381 /* RFC 2960 6.3.2 Retransmission Timer Rules
1383 * R3) Whenever a SACK is received
1384 * that acknowledges the DATA chunk
1385 * with the earliest outstanding TSN
1386 * for that address, restart T3-rtx
1387 * timer for that address with its
1388 * current RTO.
1390 restart_timer = 1;
1392 if (!tchunk->tsn_gap_acked) {
1393 tchunk->tsn_gap_acked = 1;
1394 bytes_acked += sctp_data_size(tchunk);
1396 * SFR-CACC algorithm:
1397 * 2) If the SACK contains gap acks
1398 * and the flag CHANGEOVER_ACTIVE is
1399 * set the receiver of the SACK MUST
1400 * take the following action:
1402 * B) For each TSN t being acked that
1403 * has not been acked in any SACK so
1404 * far, set cacc_saw_newack to 1 for
1405 * the destination that the TSN was
1406 * sent to.
1408 if (transport &&
1409 sack->num_gap_ack_blocks &&
1410 q->asoc->peer.primary_path->cacc.
1411 changeover_active)
1412 transport->cacc.cacc_saw_newack
1413 = 1;
1416 list_add_tail(&tchunk->transmitted_list,
1417 &q->sacked);
1418 } else {
1419 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1420 * M2) Each time a SACK arrives reporting
1421 * 'Stray DATA chunk(s)' record the highest TSN
1422 * reported as newly acknowledged, call this
1423 * value 'HighestTSNinSack'. A newly
1424 * acknowledged DATA chunk is one not
1425 * previously acknowledged in a SACK.
1427 * When the SCTP sender of data receives a SACK
1428 * chunk that acknowledges, for the first time,
1429 * the receipt of a DATA chunk, all the still
1430 * unacknowledged DATA chunks whose TSN is
1431 * older than that newly acknowledged DATA
1432 * chunk, are qualified as 'Stray DATA chunks'.
1434 if (!tchunk->tsn_gap_acked) {
1435 tchunk->tsn_gap_acked = 1;
1436 bytes_acked += sctp_data_size(tchunk);
1438 list_add_tail(lchunk, &tlist);
1441 #if SCTP_DEBUG
1442 switch (dbg_prt_state) {
1443 case 0: /* last TSN was ACKed */
1444 if (dbg_last_ack_tsn + 1 == tsn) {
1445 /* This TSN belongs to the
1446 * current ACK range.
1448 break;
1451 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1452 /* Display the end of the
1453 * current range.
1455 SCTP_DEBUG_PRINTK("-%08x",
1456 dbg_last_ack_tsn);
1459 /* Start a new range. */
1460 SCTP_DEBUG_PRINTK(",%08x", tsn);
1461 dbg_ack_tsn = tsn;
1462 break;
1464 case 1: /* The last TSN was NOT ACKed. */
1465 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1466 /* Display the end of current range. */
1467 SCTP_DEBUG_PRINTK("-%08x",
1468 dbg_last_kept_tsn);
1471 SCTP_DEBUG_PRINTK("\n");
1473 /* FALL THROUGH... */
1474 default:
1475 /* This is the first-ever TSN we examined. */
1476 /* Start a new range of ACK-ed TSNs. */
1477 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn);
1478 dbg_prt_state = 0;
1479 dbg_ack_tsn = tsn;
1482 dbg_last_ack_tsn = tsn;
1483 #endif /* SCTP_DEBUG */
1485 } else {
1486 if (tchunk->tsn_gap_acked) {
1487 SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1488 "data TSN: 0x%x\n",
1489 __func__,
1490 tsn);
1491 tchunk->tsn_gap_acked = 0;
1493 bytes_acked -= sctp_data_size(tchunk);
1495 /* RFC 2960 6.3.2 Retransmission Timer Rules
1497 * R4) Whenever a SACK is received missing a
1498 * TSN that was previously acknowledged via a
1499 * Gap Ack Block, start T3-rtx for the
1500 * destination address to which the DATA
1501 * chunk was originally
1502 * transmitted if it is not already running.
1504 restart_timer = 1;
1507 list_add_tail(lchunk, &tlist);
1509 #if SCTP_DEBUG
1510 /* See the above comments on ACK-ed TSNs. */
1511 switch (dbg_prt_state) {
1512 case 1:
1513 if (dbg_last_kept_tsn + 1 == tsn)
1514 break;
1516 if (dbg_last_kept_tsn != dbg_kept_tsn)
1517 SCTP_DEBUG_PRINTK("-%08x",
1518 dbg_last_kept_tsn);
1520 SCTP_DEBUG_PRINTK(",%08x", tsn);
1521 dbg_kept_tsn = tsn;
1522 break;
1524 case 0:
1525 if (dbg_last_ack_tsn != dbg_ack_tsn)
1526 SCTP_DEBUG_PRINTK("-%08x",
1527 dbg_last_ack_tsn);
1528 SCTP_DEBUG_PRINTK("\n");
1530 /* FALL THROUGH... */
1531 default:
1532 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn);
1533 dbg_prt_state = 1;
1534 dbg_kept_tsn = tsn;
1537 dbg_last_kept_tsn = tsn;
1538 #endif /* SCTP_DEBUG */
1542 #if SCTP_DEBUG
1543 /* Finish off the last range, displaying its ending TSN. */
1544 switch (dbg_prt_state) {
1545 case 0:
1546 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1547 SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_ack_tsn);
1548 } else {
1549 SCTP_DEBUG_PRINTK("\n");
1551 break;
1553 case 1:
1554 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1555 SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_kept_tsn);
1556 } else {
1557 SCTP_DEBUG_PRINTK("\n");
1560 #endif /* SCTP_DEBUG */
1561 if (transport) {
1562 if (bytes_acked) {
1563 /* 8.2. When an outstanding TSN is acknowledged,
1564 * the endpoint shall clear the error counter of
1565 * the destination transport address to which the
1566 * DATA chunk was last sent.
1567 * The association's overall error counter is
1568 * also cleared.
1570 transport->error_count = 0;
1571 transport->asoc->overall_error_count = 0;
1573 /* Mark the destination transport address as
1574 * active if it is not so marked.
1576 if ((transport->state == SCTP_INACTIVE) ||
1577 (transport->state == SCTP_UNCONFIRMED)) {
1578 sctp_assoc_control_transport(
1579 transport->asoc,
1580 transport,
1581 SCTP_TRANSPORT_UP,
1582 SCTP_RECEIVED_SACK);
1585 sctp_transport_raise_cwnd(transport, sack_ctsn,
1586 bytes_acked);
1588 transport->flight_size -= bytes_acked;
1589 if (transport->flight_size == 0)
1590 transport->partial_bytes_acked = 0;
1591 q->outstanding_bytes -= bytes_acked;
1592 } else {
1593 /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1594 * When a sender is doing zero window probing, it
1595 * should not timeout the association if it continues
1596 * to receive new packets from the receiver. The
1597 * reason is that the receiver MAY keep its window
1598 * closed for an indefinite time.
1599 * A sender is doing zero window probing when the
1600 * receiver's advertised window is zero, and there is
1601 * only one data chunk in flight to the receiver.
1603 if (!q->asoc->peer.rwnd &&
1604 !list_empty(&tlist) &&
1605 (sack_ctsn+2 == q->asoc->next_tsn)) {
1606 SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1607 "window probe: %u\n",
1608 __func__, sack_ctsn);
1609 q->asoc->overall_error_count = 0;
1610 transport->error_count = 0;
1614 /* RFC 2960 6.3.2 Retransmission Timer Rules
1616 * R2) Whenever all outstanding data sent to an address have
1617 * been acknowledged, turn off the T3-rtx timer of that
1618 * address.
1620 if (!transport->flight_size) {
1621 if (timer_pending(&transport->T3_rtx_timer) &&
1622 del_timer(&transport->T3_rtx_timer)) {
1623 sctp_transport_put(transport);
1625 } else if (restart_timer) {
1626 if (!mod_timer(&transport->T3_rtx_timer,
1627 jiffies + transport->rto))
1628 sctp_transport_hold(transport);
1632 list_splice(&tlist, transmitted_queue);
1635 /* Mark chunks as missing and consequently may get retransmitted. */
1636 static void sctp_mark_missing(struct sctp_outq *q,
1637 struct list_head *transmitted_queue,
1638 struct sctp_transport *transport,
1639 __u32 highest_new_tsn_in_sack,
1640 int count_of_newacks)
1642 struct sctp_chunk *chunk;
1643 __u32 tsn;
1644 char do_fast_retransmit = 0;
1645 struct sctp_transport *primary = q->asoc->peer.primary_path;
1647 list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1649 tsn = ntohl(chunk->subh.data_hdr->tsn);
1651 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1652 * 'Unacknowledged TSN's', if the TSN number of an
1653 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1654 * value, increment the 'TSN.Missing.Report' count on that
1655 * chunk if it has NOT been fast retransmitted or marked for
1656 * fast retransmit already.
1658 if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
1659 !chunk->tsn_gap_acked &&
1660 TSN_lt(tsn, highest_new_tsn_in_sack)) {
1662 /* SFR-CACC may require us to skip marking
1663 * this chunk as missing.
1665 if (!transport || !sctp_cacc_skip(primary, transport,
1666 count_of_newacks, tsn)) {
1667 chunk->tsn_missing_report++;
1669 SCTP_DEBUG_PRINTK(
1670 "%s: TSN 0x%x missing counter: %d\n",
1671 __func__, tsn,
1672 chunk->tsn_missing_report);
1676 * M4) If any DATA chunk is found to have a
1677 * 'TSN.Missing.Report'
1678 * value larger than or equal to 3, mark that chunk for
1679 * retransmission and start the fast retransmit procedure.
1682 if (chunk->tsn_missing_report >= 3) {
1683 chunk->fast_retransmit = SCTP_NEED_FRTX;
1684 do_fast_retransmit = 1;
1688 if (transport) {
1689 if (do_fast_retransmit)
1690 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1692 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1693 "ssthresh: %d, flight_size: %d, pba: %d\n",
1694 __func__, transport, transport->cwnd,
1695 transport->ssthresh, transport->flight_size,
1696 transport->partial_bytes_acked);
1700 /* Is the given TSN acked by this packet? */
1701 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1703 int i;
1704 sctp_sack_variable_t *frags;
1705 __u16 gap;
1706 __u32 ctsn = ntohl(sack->cum_tsn_ack);
1708 if (TSN_lte(tsn, ctsn))
1709 goto pass;
1711 /* 3.3.4 Selective Acknowledgement (SACK) (3):
1713 * Gap Ack Blocks:
1714 * These fields contain the Gap Ack Blocks. They are repeated
1715 * for each Gap Ack Block up to the number of Gap Ack Blocks
1716 * defined in the Number of Gap Ack Blocks field. All DATA
1717 * chunks with TSNs greater than or equal to (Cumulative TSN
1718 * Ack + Gap Ack Block Start) and less than or equal to
1719 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1720 * Block are assumed to have been received correctly.
1723 frags = sack->variable;
1724 gap = tsn - ctsn;
1725 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1726 if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1727 TSN_lte(gap, ntohs(frags[i].gab.end)))
1728 goto pass;
1731 return 0;
1732 pass:
1733 return 1;
1736 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1737 int nskips, __be16 stream)
1739 int i;
1741 for (i = 0; i < nskips; i++) {
1742 if (skiplist[i].stream == stream)
1743 return i;
1745 return i;
1748 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1749 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1751 struct sctp_association *asoc = q->asoc;
1752 struct sctp_chunk *ftsn_chunk = NULL;
1753 struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1754 int nskips = 0;
1755 int skip_pos = 0;
1756 __u32 tsn;
1757 struct sctp_chunk *chunk;
1758 struct list_head *lchunk, *temp;
1760 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1761 * received SACK.
1763 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1764 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1766 if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1767 asoc->adv_peer_ack_point = ctsn;
1769 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1770 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1771 * the chunk next in the out-queue space is marked as "abandoned" as
1772 * shown in the following example:
1774 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1775 * and the Advanced.Peer.Ack.Point is updated to this value:
1777 * out-queue at the end of ==> out-queue after Adv.Ack.Point
1778 * normal SACK processing local advancement
1779 * ... ...
1780 * Adv.Ack.Pt-> 102 acked 102 acked
1781 * 103 abandoned 103 abandoned
1782 * 104 abandoned Adv.Ack.P-> 104 abandoned
1783 * 105 105
1784 * 106 acked 106 acked
1785 * ... ...
1787 * In this example, the data sender successfully advanced the
1788 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1790 list_for_each_safe(lchunk, temp, &q->abandoned) {
1791 chunk = list_entry(lchunk, struct sctp_chunk,
1792 transmitted_list);
1793 tsn = ntohl(chunk->subh.data_hdr->tsn);
1795 /* Remove any chunks in the abandoned queue that are acked by
1796 * the ctsn.
1798 if (TSN_lte(tsn, ctsn)) {
1799 list_del_init(lchunk);
1800 sctp_chunk_free(chunk);
1801 } else {
1802 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1803 asoc->adv_peer_ack_point = tsn;
1804 if (chunk->chunk_hdr->flags &
1805 SCTP_DATA_UNORDERED)
1806 continue;
1807 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1808 nskips,
1809 chunk->subh.data_hdr->stream);
1810 ftsn_skip_arr[skip_pos].stream =
1811 chunk->subh.data_hdr->stream;
1812 ftsn_skip_arr[skip_pos].ssn =
1813 chunk->subh.data_hdr->ssn;
1814 if (skip_pos == nskips)
1815 nskips++;
1816 if (nskips == 10)
1817 break;
1818 } else
1819 break;
1823 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1824 * is greater than the Cumulative TSN ACK carried in the received
1825 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1826 * chunk containing the latest value of the
1827 * "Advanced.Peer.Ack.Point".
1829 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1830 * list each stream and sequence number in the forwarded TSN. This
1831 * information will enable the receiver to easily find any
1832 * stranded TSN's waiting on stream reorder queues. Each stream
1833 * SHOULD only be reported once; this means that if multiple
1834 * abandoned messages occur in the same stream then only the
1835 * highest abandoned stream sequence number is reported. If the
1836 * total size of the FORWARD TSN does NOT fit in a single MTU then
1837 * the sender of the FORWARD TSN SHOULD lower the
1838 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1839 * single MTU.
1841 if (asoc->adv_peer_ack_point > ctsn)
1842 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1843 nskips, &ftsn_skip_arr[0]);
1845 if (ftsn_chunk) {
1846 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1847 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);