spi-topcliff-pch: Fix issue for transmitting over 4KByte
[zen-stable.git] / net / sctp / outqueue.c
blobcfeb1d4a1ee6ca730595959946ced911b7442baa
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
51 #include <linux/types.h>
52 #include <linux/list.h> /* For struct list_head */
53 #include <linux/socket.h>
54 #include <linux/ip.h>
55 #include <linux/slab.h>
56 #include <net/sock.h> /* For skb_set_owner_w */
58 #include <net/sctp/sctp.h>
59 #include <net/sctp/sm.h>
61 /* Declare internal functions here. */
62 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn);
63 static void sctp_check_transmitted(struct sctp_outq *q,
64 struct list_head *transmitted_queue,
65 struct sctp_transport *transport,
66 struct sctp_sackhdr *sack,
67 __u32 *highest_new_tsn);
69 static void sctp_mark_missing(struct sctp_outq *q,
70 struct list_head *transmitted_queue,
71 struct sctp_transport *transport,
72 __u32 highest_new_tsn,
73 int count_of_newacks);
75 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn);
77 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout);
79 /* Add data to the front of the queue. */
80 static inline void sctp_outq_head_data(struct sctp_outq *q,
81 struct sctp_chunk *ch)
83 list_add(&ch->list, &q->out_chunk_list);
84 q->out_qlen += ch->skb->len;
87 /* Take data from the front of the queue. */
88 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
90 struct sctp_chunk *ch = NULL;
92 if (!list_empty(&q->out_chunk_list)) {
93 struct list_head *entry = q->out_chunk_list.next;
95 ch = list_entry(entry, struct sctp_chunk, list);
96 list_del_init(entry);
97 q->out_qlen -= ch->skb->len;
99 return ch;
101 /* Add data chunk to the end of the queue. */
102 static inline void sctp_outq_tail_data(struct sctp_outq *q,
103 struct sctp_chunk *ch)
105 list_add_tail(&ch->list, &q->out_chunk_list);
106 q->out_qlen += ch->skb->len;
110 * SFR-CACC algorithm:
111 * D) If count_of_newacks is greater than or equal to 2
112 * and t was not sent to the current primary then the
113 * sender MUST NOT increment missing report count for t.
115 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary,
116 struct sctp_transport *transport,
117 int count_of_newacks)
119 if (count_of_newacks >=2 && transport != primary)
120 return 1;
121 return 0;
125 * SFR-CACC algorithm:
126 * F) If count_of_newacks is less than 2, let d be the
127 * destination to which t was sent. If cacc_saw_newack
128 * is 0 for destination d, then the sender MUST NOT
129 * increment missing report count for t.
131 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport,
132 int count_of_newacks)
134 if (count_of_newacks < 2 &&
135 (transport && !transport->cacc.cacc_saw_newack))
136 return 1;
137 return 0;
141 * SFR-CACC algorithm:
142 * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
143 * execute steps C, D, F.
145 * C has been implemented in sctp_outq_sack
147 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
148 struct sctp_transport *transport,
149 int count_of_newacks)
151 if (!primary->cacc.cycling_changeover) {
152 if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
153 return 1;
154 if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
155 return 1;
156 return 0;
158 return 0;
162 * SFR-CACC algorithm:
163 * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
164 * than next_tsn_at_change of the current primary, then
165 * the sender MUST NOT increment missing report count
166 * for t.
168 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
170 if (primary->cacc.cycling_changeover &&
171 TSN_lt(tsn, primary->cacc.next_tsn_at_change))
172 return 1;
173 return 0;
177 * SFR-CACC algorithm:
178 * 3) If the missing report count for TSN t is to be
179 * incremented according to [RFC2960] and
180 * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
181 * then the sender MUST further execute steps 3.1 and
182 * 3.2 to determine if the missing report count for
183 * TSN t SHOULD NOT be incremented.
185 * 3.3) If 3.1 and 3.2 do not dictate that the missing
186 * report count for t should not be incremented, then
187 * the sender SHOULD increment missing report count for
188 * t (according to [RFC2960] and [SCTP_STEWART_2002]).
190 static inline int sctp_cacc_skip(struct sctp_transport *primary,
191 struct sctp_transport *transport,
192 int count_of_newacks,
193 __u32 tsn)
195 if (primary->cacc.changeover_active &&
196 (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) ||
197 sctp_cacc_skip_3_2(primary, tsn)))
198 return 1;
199 return 0;
202 /* Initialize an existing sctp_outq. This does the boring stuff.
203 * You still need to define handlers if you really want to DO
204 * something with this structure...
206 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
208 q->asoc = asoc;
209 INIT_LIST_HEAD(&q->out_chunk_list);
210 INIT_LIST_HEAD(&q->control_chunk_list);
211 INIT_LIST_HEAD(&q->retransmit);
212 INIT_LIST_HEAD(&q->sacked);
213 INIT_LIST_HEAD(&q->abandoned);
215 q->fast_rtx = 0;
216 q->outstanding_bytes = 0;
217 q->empty = 1;
218 q->cork = 0;
220 q->malloced = 0;
221 q->out_qlen = 0;
224 /* Free the outqueue structure and any related pending chunks.
226 void sctp_outq_teardown(struct sctp_outq *q)
228 struct sctp_transport *transport;
229 struct list_head *lchunk, *temp;
230 struct sctp_chunk *chunk, *tmp;
232 /* Throw away unacknowledged chunks. */
233 list_for_each_entry(transport, &q->asoc->peer.transport_addr_list,
234 transports) {
235 while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
236 chunk = list_entry(lchunk, struct sctp_chunk,
237 transmitted_list);
238 /* Mark as part of a failed message. */
239 sctp_chunk_fail(chunk, q->error);
240 sctp_chunk_free(chunk);
244 /* Throw away chunks that have been gap ACKed. */
245 list_for_each_safe(lchunk, temp, &q->sacked) {
246 list_del_init(lchunk);
247 chunk = list_entry(lchunk, struct sctp_chunk,
248 transmitted_list);
249 sctp_chunk_fail(chunk, q->error);
250 sctp_chunk_free(chunk);
253 /* Throw away any chunks in the retransmit queue. */
254 list_for_each_safe(lchunk, temp, &q->retransmit) {
255 list_del_init(lchunk);
256 chunk = list_entry(lchunk, struct sctp_chunk,
257 transmitted_list);
258 sctp_chunk_fail(chunk, q->error);
259 sctp_chunk_free(chunk);
262 /* Throw away any chunks that are in the abandoned queue. */
263 list_for_each_safe(lchunk, temp, &q->abandoned) {
264 list_del_init(lchunk);
265 chunk = list_entry(lchunk, struct sctp_chunk,
266 transmitted_list);
267 sctp_chunk_fail(chunk, q->error);
268 sctp_chunk_free(chunk);
271 /* Throw away any leftover data chunks. */
272 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
274 /* Mark as send failure. */
275 sctp_chunk_fail(chunk, q->error);
276 sctp_chunk_free(chunk);
279 q->error = 0;
281 /* Throw away any leftover control chunks. */
282 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
283 list_del_init(&chunk->list);
284 sctp_chunk_free(chunk);
288 /* Free the outqueue structure and any related pending chunks. */
289 void sctp_outq_free(struct sctp_outq *q)
291 /* Throw away leftover chunks. */
292 sctp_outq_teardown(q);
294 /* If we were kmalloc()'d, free the memory. */
295 if (q->malloced)
296 kfree(q);
299 /* Put a new chunk in an sctp_outq. */
300 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk)
302 int error = 0;
304 SCTP_DEBUG_PRINTK("sctp_outq_tail(%p, %p[%s])\n",
305 q, chunk, chunk && chunk->chunk_hdr ?
306 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
307 : "Illegal Chunk");
309 /* If it is data, queue it up, otherwise, send it
310 * immediately.
312 if (sctp_chunk_is_data(chunk)) {
313 /* Is it OK to queue data chunks? */
314 /* From 9. Termination of Association
316 * When either endpoint performs a shutdown, the
317 * association on each peer will stop accepting new
318 * data from its user and only deliver data in queue
319 * at the time of sending or receiving the SHUTDOWN
320 * chunk.
322 switch (q->asoc->state) {
323 case SCTP_STATE_CLOSED:
324 case SCTP_STATE_SHUTDOWN_PENDING:
325 case SCTP_STATE_SHUTDOWN_SENT:
326 case SCTP_STATE_SHUTDOWN_RECEIVED:
327 case SCTP_STATE_SHUTDOWN_ACK_SENT:
328 /* Cannot send after transport endpoint shutdown */
329 error = -ESHUTDOWN;
330 break;
332 default:
333 SCTP_DEBUG_PRINTK("outqueueing (%p, %p[%s])\n",
334 q, chunk, chunk && chunk->chunk_hdr ?
335 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
336 : "Illegal Chunk");
338 sctp_outq_tail_data(q, chunk);
339 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
340 SCTP_INC_STATS(SCTP_MIB_OUTUNORDERCHUNKS);
341 else
342 SCTP_INC_STATS(SCTP_MIB_OUTORDERCHUNKS);
343 q->empty = 0;
344 break;
346 } else {
347 list_add_tail(&chunk->list, &q->control_chunk_list);
348 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
351 if (error < 0)
352 return error;
354 if (!q->cork)
355 error = sctp_outq_flush(q, 0);
357 return error;
360 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list
361 * and the abandoned list are in ascending order.
363 static void sctp_insert_list(struct list_head *head, struct list_head *new)
365 struct list_head *pos;
366 struct sctp_chunk *nchunk, *lchunk;
367 __u32 ntsn, ltsn;
368 int done = 0;
370 nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
371 ntsn = ntohl(nchunk->subh.data_hdr->tsn);
373 list_for_each(pos, head) {
374 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
375 ltsn = ntohl(lchunk->subh.data_hdr->tsn);
376 if (TSN_lt(ntsn, ltsn)) {
377 list_add(new, pos->prev);
378 done = 1;
379 break;
382 if (!done)
383 list_add_tail(new, head);
386 /* Mark all the eligible packets on a transport for retransmission. */
387 void sctp_retransmit_mark(struct sctp_outq *q,
388 struct sctp_transport *transport,
389 __u8 reason)
391 struct list_head *lchunk, *ltemp;
392 struct sctp_chunk *chunk;
394 /* Walk through the specified transmitted queue. */
395 list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
396 chunk = list_entry(lchunk, struct sctp_chunk,
397 transmitted_list);
399 /* If the chunk is abandoned, move it to abandoned list. */
400 if (sctp_chunk_abandoned(chunk)) {
401 list_del_init(lchunk);
402 sctp_insert_list(&q->abandoned, lchunk);
404 /* If this chunk has not been previousely acked,
405 * stop considering it 'outstanding'. Our peer
406 * will most likely never see it since it will
407 * not be retransmitted
409 if (!chunk->tsn_gap_acked) {
410 if (chunk->transport)
411 chunk->transport->flight_size -=
412 sctp_data_size(chunk);
413 q->outstanding_bytes -= sctp_data_size(chunk);
414 q->asoc->peer.rwnd += sctp_data_size(chunk);
416 continue;
419 /* If we are doing retransmission due to a timeout or pmtu
420 * discovery, only the chunks that are not yet acked should
421 * be added to the retransmit queue.
423 if ((reason == SCTP_RTXR_FAST_RTX &&
424 (chunk->fast_retransmit == SCTP_NEED_FRTX)) ||
425 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) {
426 /* RFC 2960 6.2.1 Processing a Received SACK
428 * C) Any time a DATA chunk is marked for
429 * retransmission (via either T3-rtx timer expiration
430 * (Section 6.3.3) or via fast retransmit
431 * (Section 7.2.4)), add the data size of those
432 * chunks to the rwnd.
434 q->asoc->peer.rwnd += sctp_data_size(chunk);
435 q->outstanding_bytes -= sctp_data_size(chunk);
436 if (chunk->transport)
437 transport->flight_size -= sctp_data_size(chunk);
439 /* sctpimpguide-05 Section 2.8.2
440 * M5) If a T3-rtx timer expires, the
441 * 'TSN.Missing.Report' of all affected TSNs is set
442 * to 0.
444 chunk->tsn_missing_report = 0;
446 /* If a chunk that is being used for RTT measurement
447 * has to be retransmitted, we cannot use this chunk
448 * anymore for RTT measurements. Reset rto_pending so
449 * that a new RTT measurement is started when a new
450 * data chunk is sent.
452 if (chunk->rtt_in_progress) {
453 chunk->rtt_in_progress = 0;
454 transport->rto_pending = 0;
457 /* Move the chunk to the retransmit queue. The chunks
458 * on the retransmit queue are always kept in order.
460 list_del_init(lchunk);
461 sctp_insert_list(&q->retransmit, lchunk);
465 SCTP_DEBUG_PRINTK("%s: transport: %p, reason: %d, "
466 "cwnd: %d, ssthresh: %d, flight_size: %d, "
467 "pba: %d\n", __func__,
468 transport, reason,
469 transport->cwnd, transport->ssthresh,
470 transport->flight_size,
471 transport->partial_bytes_acked);
475 /* Mark all the eligible packets on a transport for retransmission and force
476 * one packet out.
478 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
479 sctp_retransmit_reason_t reason)
481 int error = 0;
483 switch(reason) {
484 case SCTP_RTXR_T3_RTX:
485 SCTP_INC_STATS(SCTP_MIB_T3_RETRANSMITS);
486 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
487 /* Update the retran path if the T3-rtx timer has expired for
488 * the current retran path.
490 if (transport == transport->asoc->peer.retran_path)
491 sctp_assoc_update_retran_path(transport->asoc);
492 transport->asoc->rtx_data_chunks +=
493 transport->asoc->unack_data;
494 break;
495 case SCTP_RTXR_FAST_RTX:
496 SCTP_INC_STATS(SCTP_MIB_FAST_RETRANSMITS);
497 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
498 q->fast_rtx = 1;
499 break;
500 case SCTP_RTXR_PMTUD:
501 SCTP_INC_STATS(SCTP_MIB_PMTUD_RETRANSMITS);
502 break;
503 case SCTP_RTXR_T1_RTX:
504 SCTP_INC_STATS(SCTP_MIB_T1_RETRANSMITS);
505 transport->asoc->init_retries++;
506 break;
507 default:
508 BUG();
511 sctp_retransmit_mark(q, transport, reason);
513 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
514 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
515 * following the procedures outlined in C1 - C5.
517 if (reason == SCTP_RTXR_T3_RTX)
518 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
520 /* Flush the queues only on timeout, since fast_rtx is only
521 * triggered during sack processing and the queue
522 * will be flushed at the end.
524 if (reason != SCTP_RTXR_FAST_RTX)
525 error = sctp_outq_flush(q, /* rtx_timeout */ 1);
527 if (error)
528 q->asoc->base.sk->sk_err = -error;
532 * Transmit DATA chunks on the retransmit queue. Upon return from
533 * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
534 * need to be transmitted by the caller.
535 * We assume that pkt->transport has already been set.
537 * The return value is a normal kernel error return value.
539 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
540 int rtx_timeout, int *start_timer)
542 struct list_head *lqueue;
543 struct sctp_transport *transport = pkt->transport;
544 sctp_xmit_t status;
545 struct sctp_chunk *chunk, *chunk1;
546 int fast_rtx;
547 int error = 0;
548 int timer = 0;
549 int done = 0;
551 lqueue = &q->retransmit;
552 fast_rtx = q->fast_rtx;
554 /* This loop handles time-out retransmissions, fast retransmissions,
555 * and retransmissions due to opening of whindow.
557 * RFC 2960 6.3.3 Handle T3-rtx Expiration
559 * E3) Determine how many of the earliest (i.e., lowest TSN)
560 * outstanding DATA chunks for the address for which the
561 * T3-rtx has expired will fit into a single packet, subject
562 * to the MTU constraint for the path corresponding to the
563 * destination transport address to which the retransmission
564 * is being sent (this may be different from the address for
565 * which the timer expires [see Section 6.4]). Call this value
566 * K. Bundle and retransmit those K DATA chunks in a single
567 * packet to the destination endpoint.
569 * [Just to be painfully clear, if we are retransmitting
570 * because a timeout just happened, we should send only ONE
571 * packet of retransmitted data.]
573 * For fast retransmissions we also send only ONE packet. However,
574 * if we are just flushing the queue due to open window, we'll
575 * try to send as much as possible.
577 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
578 /* If the chunk is abandoned, move it to abandoned list. */
579 if (sctp_chunk_abandoned(chunk)) {
580 list_del_init(&chunk->transmitted_list);
581 sctp_insert_list(&q->abandoned,
582 &chunk->transmitted_list);
583 continue;
586 /* Make sure that Gap Acked TSNs are not retransmitted. A
587 * simple approach is just to move such TSNs out of the
588 * way and into a 'transmitted' queue and skip to the
589 * next chunk.
591 if (chunk->tsn_gap_acked) {
592 list_del(&chunk->transmitted_list);
593 list_add_tail(&chunk->transmitted_list,
594 &transport->transmitted);
595 continue;
598 /* If we are doing fast retransmit, ignore non-fast_rtransmit
599 * chunks
601 if (fast_rtx && !chunk->fast_retransmit)
602 continue;
604 redo:
605 /* Attempt to append this chunk to the packet. */
606 status = sctp_packet_append_chunk(pkt, chunk);
608 switch (status) {
609 case SCTP_XMIT_PMTU_FULL:
610 if (!pkt->has_data && !pkt->has_cookie_echo) {
611 /* If this packet did not contain DATA then
612 * retransmission did not happen, so do it
613 * again. We'll ignore the error here since
614 * control chunks are already freed so there
615 * is nothing we can do.
617 sctp_packet_transmit(pkt);
618 goto redo;
621 /* Send this packet. */
622 error = sctp_packet_transmit(pkt);
624 /* If we are retransmitting, we should only
625 * send a single packet.
626 * Otherwise, try appending this chunk again.
628 if (rtx_timeout || fast_rtx)
629 done = 1;
630 else
631 goto redo;
633 /* Bundle next chunk in the next round. */
634 break;
636 case SCTP_XMIT_RWND_FULL:
637 /* Send this packet. */
638 error = sctp_packet_transmit(pkt);
640 /* Stop sending DATA as there is no more room
641 * at the receiver.
643 done = 1;
644 break;
646 case SCTP_XMIT_NAGLE_DELAY:
647 /* Send this packet. */
648 error = sctp_packet_transmit(pkt);
650 /* Stop sending DATA because of nagle delay. */
651 done = 1;
652 break;
654 default:
655 /* The append was successful, so add this chunk to
656 * the transmitted list.
658 list_del(&chunk->transmitted_list);
659 list_add_tail(&chunk->transmitted_list,
660 &transport->transmitted);
662 /* Mark the chunk as ineligible for fast retransmit
663 * after it is retransmitted.
665 if (chunk->fast_retransmit == SCTP_NEED_FRTX)
666 chunk->fast_retransmit = SCTP_DONT_FRTX;
668 q->empty = 0;
669 break;
672 /* Set the timer if there were no errors */
673 if (!error && !timer)
674 timer = 1;
676 if (done)
677 break;
680 /* If we are here due to a retransmit timeout or a fast
681 * retransmit and if there are any chunks left in the retransmit
682 * queue that could not fit in the PMTU sized packet, they need
683 * to be marked as ineligible for a subsequent fast retransmit.
685 if (rtx_timeout || fast_rtx) {
686 list_for_each_entry(chunk1, lqueue, transmitted_list) {
687 if (chunk1->fast_retransmit == SCTP_NEED_FRTX)
688 chunk1->fast_retransmit = SCTP_DONT_FRTX;
692 *start_timer = timer;
694 /* Clear fast retransmit hint */
695 if (fast_rtx)
696 q->fast_rtx = 0;
698 return error;
701 /* Cork the outqueue so queued chunks are really queued. */
702 int sctp_outq_uncork(struct sctp_outq *q)
704 int error = 0;
705 if (q->cork)
706 q->cork = 0;
707 error = sctp_outq_flush(q, 0);
708 return error;
713 * Try to flush an outqueue.
715 * Description: Send everything in q which we legally can, subject to
716 * congestion limitations.
717 * * Note: This function can be called from multiple contexts so appropriate
718 * locking concerns must be made. Today we use the sock lock to protect
719 * this function.
721 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
723 struct sctp_packet *packet;
724 struct sctp_packet singleton;
725 struct sctp_association *asoc = q->asoc;
726 __u16 sport = asoc->base.bind_addr.port;
727 __u16 dport = asoc->peer.port;
728 __u32 vtag = asoc->peer.i.init_tag;
729 struct sctp_transport *transport = NULL;
730 struct sctp_transport *new_transport;
731 struct sctp_chunk *chunk, *tmp;
732 sctp_xmit_t status;
733 int error = 0;
734 int start_timer = 0;
735 int one_packet = 0;
737 /* These transports have chunks to send. */
738 struct list_head transport_list;
739 struct list_head *ltransport;
741 INIT_LIST_HEAD(&transport_list);
742 packet = NULL;
745 * 6.10 Bundling
746 * ...
747 * When bundling control chunks with DATA chunks, an
748 * endpoint MUST place control chunks first in the outbound
749 * SCTP packet. The transmitter MUST transmit DATA chunks
750 * within a SCTP packet in increasing order of TSN.
751 * ...
754 list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
755 /* RFC 5061, 5.3
756 * F1) This means that until such time as the ASCONF
757 * containing the add is acknowledged, the sender MUST
758 * NOT use the new IP address as a source for ANY SCTP
759 * packet except on carrying an ASCONF Chunk.
761 if (asoc->src_out_of_asoc_ok &&
762 chunk->chunk_hdr->type != SCTP_CID_ASCONF)
763 continue;
765 list_del_init(&chunk->list);
767 /* Pick the right transport to use. */
768 new_transport = chunk->transport;
770 if (!new_transport) {
772 * If we have a prior transport pointer, see if
773 * the destination address of the chunk
774 * matches the destination address of the
775 * current transport. If not a match, then
776 * try to look up the transport with a given
777 * destination address. We do this because
778 * after processing ASCONFs, we may have new
779 * transports created.
781 if (transport &&
782 sctp_cmp_addr_exact(&chunk->dest,
783 &transport->ipaddr))
784 new_transport = transport;
785 else
786 new_transport = sctp_assoc_lookup_paddr(asoc,
787 &chunk->dest);
789 /* if we still don't have a new transport, then
790 * use the current active path.
792 if (!new_transport)
793 new_transport = asoc->peer.active_path;
794 } else if ((new_transport->state == SCTP_INACTIVE) ||
795 (new_transport->state == SCTP_UNCONFIRMED)) {
796 /* If the chunk is Heartbeat or Heartbeat Ack,
797 * send it to chunk->transport, even if it's
798 * inactive.
800 * 3.3.6 Heartbeat Acknowledgement:
801 * ...
802 * A HEARTBEAT ACK is always sent to the source IP
803 * address of the IP datagram containing the
804 * HEARTBEAT chunk to which this ack is responding.
805 * ...
807 * ASCONF_ACKs also must be sent to the source.
809 if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
810 chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK &&
811 chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK)
812 new_transport = asoc->peer.active_path;
815 /* Are we switching transports?
816 * Take care of transport locks.
818 if (new_transport != transport) {
819 transport = new_transport;
820 if (list_empty(&transport->send_ready)) {
821 list_add_tail(&transport->send_ready,
822 &transport_list);
824 packet = &transport->packet;
825 sctp_packet_config(packet, vtag,
826 asoc->peer.ecn_capable);
829 switch (chunk->chunk_hdr->type) {
831 * 6.10 Bundling
832 * ...
833 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
834 * COMPLETE with any other chunks. [Send them immediately.]
836 case SCTP_CID_INIT:
837 case SCTP_CID_INIT_ACK:
838 case SCTP_CID_SHUTDOWN_COMPLETE:
839 sctp_packet_init(&singleton, transport, sport, dport);
840 sctp_packet_config(&singleton, vtag, 0);
841 sctp_packet_append_chunk(&singleton, chunk);
842 error = sctp_packet_transmit(&singleton);
843 if (error < 0)
844 return error;
845 break;
847 case SCTP_CID_ABORT:
848 if (sctp_test_T_bit(chunk)) {
849 packet->vtag = asoc->c.my_vtag;
851 /* The following chunks are "response" chunks, i.e.
852 * they are generated in response to something we
853 * received. If we are sending these, then we can
854 * send only 1 packet containing these chunks.
856 case SCTP_CID_HEARTBEAT_ACK:
857 case SCTP_CID_SHUTDOWN_ACK:
858 case SCTP_CID_COOKIE_ACK:
859 case SCTP_CID_COOKIE_ECHO:
860 case SCTP_CID_ERROR:
861 case SCTP_CID_ECN_CWR:
862 case SCTP_CID_ASCONF_ACK:
863 one_packet = 1;
864 /* Fall through */
866 case SCTP_CID_SACK:
867 case SCTP_CID_HEARTBEAT:
868 case SCTP_CID_SHUTDOWN:
869 case SCTP_CID_ECN_ECNE:
870 case SCTP_CID_ASCONF:
871 case SCTP_CID_FWD_TSN:
872 status = sctp_packet_transmit_chunk(packet, chunk,
873 one_packet);
874 if (status != SCTP_XMIT_OK) {
875 /* put the chunk back */
876 list_add(&chunk->list, &q->control_chunk_list);
877 } else if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN) {
878 /* PR-SCTP C5) If a FORWARD TSN is sent, the
879 * sender MUST assure that at least one T3-rtx
880 * timer is running.
882 sctp_transport_reset_timers(transport);
884 break;
886 default:
887 /* We built a chunk with an illegal type! */
888 BUG();
892 if (q->asoc->src_out_of_asoc_ok)
893 goto sctp_flush_out;
895 /* Is it OK to send data chunks? */
896 switch (asoc->state) {
897 case SCTP_STATE_COOKIE_ECHOED:
898 /* Only allow bundling when this packet has a COOKIE-ECHO
899 * chunk.
901 if (!packet || !packet->has_cookie_echo)
902 break;
904 /* fallthru */
905 case SCTP_STATE_ESTABLISHED:
906 case SCTP_STATE_SHUTDOWN_PENDING:
907 case SCTP_STATE_SHUTDOWN_RECEIVED:
909 * RFC 2960 6.1 Transmission of DATA Chunks
911 * C) When the time comes for the sender to transmit,
912 * before sending new DATA chunks, the sender MUST
913 * first transmit any outstanding DATA chunks which
914 * are marked for retransmission (limited by the
915 * current cwnd).
917 if (!list_empty(&q->retransmit)) {
918 if (asoc->peer.retran_path->state == SCTP_UNCONFIRMED)
919 goto sctp_flush_out;
920 if (transport == asoc->peer.retran_path)
921 goto retran;
923 /* Switch transports & prepare the packet. */
925 transport = asoc->peer.retran_path;
927 if (list_empty(&transport->send_ready)) {
928 list_add_tail(&transport->send_ready,
929 &transport_list);
932 packet = &transport->packet;
933 sctp_packet_config(packet, vtag,
934 asoc->peer.ecn_capable);
935 retran:
936 error = sctp_outq_flush_rtx(q, packet,
937 rtx_timeout, &start_timer);
939 if (start_timer)
940 sctp_transport_reset_timers(transport);
942 /* This can happen on COOKIE-ECHO resend. Only
943 * one chunk can get bundled with a COOKIE-ECHO.
945 if (packet->has_cookie_echo)
946 goto sctp_flush_out;
948 /* Don't send new data if there is still data
949 * waiting to retransmit.
951 if (!list_empty(&q->retransmit))
952 goto sctp_flush_out;
955 /* Apply Max.Burst limitation to the current transport in
956 * case it will be used for new data. We are going to
957 * rest it before we return, but we want to apply the limit
958 * to the currently queued data.
960 if (transport)
961 sctp_transport_burst_limited(transport);
963 /* Finally, transmit new packets. */
964 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
965 /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
966 * stream identifier.
968 if (chunk->sinfo.sinfo_stream >=
969 asoc->c.sinit_num_ostreams) {
971 /* Mark as failed send. */
972 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
973 sctp_chunk_free(chunk);
974 continue;
977 /* Has this chunk expired? */
978 if (sctp_chunk_abandoned(chunk)) {
979 sctp_chunk_fail(chunk, 0);
980 sctp_chunk_free(chunk);
981 continue;
984 /* If there is a specified transport, use it.
985 * Otherwise, we want to use the active path.
987 new_transport = chunk->transport;
988 if (!new_transport ||
989 ((new_transport->state == SCTP_INACTIVE) ||
990 (new_transport->state == SCTP_UNCONFIRMED)))
991 new_transport = asoc->peer.active_path;
992 if (new_transport->state == SCTP_UNCONFIRMED)
993 continue;
995 /* Change packets if necessary. */
996 if (new_transport != transport) {
997 transport = new_transport;
999 /* Schedule to have this transport's
1000 * packet flushed.
1002 if (list_empty(&transport->send_ready)) {
1003 list_add_tail(&transport->send_ready,
1004 &transport_list);
1007 packet = &transport->packet;
1008 sctp_packet_config(packet, vtag,
1009 asoc->peer.ecn_capable);
1010 /* We've switched transports, so apply the
1011 * Burst limit to the new transport.
1013 sctp_transport_burst_limited(transport);
1016 SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
1017 q, chunk,
1018 chunk && chunk->chunk_hdr ?
1019 sctp_cname(SCTP_ST_CHUNK(
1020 chunk->chunk_hdr->type))
1021 : "Illegal Chunk");
1023 SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
1024 "%p skb->users %d.\n",
1025 ntohl(chunk->subh.data_hdr->tsn),
1026 chunk->skb ?chunk->skb->head : NULL,
1027 chunk->skb ?
1028 atomic_read(&chunk->skb->users) : -1);
1030 /* Add the chunk to the packet. */
1031 status = sctp_packet_transmit_chunk(packet, chunk, 0);
1033 switch (status) {
1034 case SCTP_XMIT_PMTU_FULL:
1035 case SCTP_XMIT_RWND_FULL:
1036 case SCTP_XMIT_NAGLE_DELAY:
1037 /* We could not append this chunk, so put
1038 * the chunk back on the output queue.
1040 SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
1041 "not transmit TSN: 0x%x, status: %d\n",
1042 ntohl(chunk->subh.data_hdr->tsn),
1043 status);
1044 sctp_outq_head_data(q, chunk);
1045 goto sctp_flush_out;
1046 break;
1048 case SCTP_XMIT_OK:
1049 /* The sender is in the SHUTDOWN-PENDING state,
1050 * The sender MAY set the I-bit in the DATA
1051 * chunk header.
1053 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING)
1054 chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM;
1056 break;
1058 default:
1059 BUG();
1062 /* BUG: We assume that the sctp_packet_transmit()
1063 * call below will succeed all the time and add the
1064 * chunk to the transmitted list and restart the
1065 * timers.
1066 * It is possible that the call can fail under OOM
1067 * conditions.
1069 * Is this really a problem? Won't this behave
1070 * like a lost TSN?
1072 list_add_tail(&chunk->transmitted_list,
1073 &transport->transmitted);
1075 sctp_transport_reset_timers(transport);
1077 q->empty = 0;
1079 /* Only let one DATA chunk get bundled with a
1080 * COOKIE-ECHO chunk.
1082 if (packet->has_cookie_echo)
1083 goto sctp_flush_out;
1085 break;
1087 default:
1088 /* Do nothing. */
1089 break;
1092 sctp_flush_out:
1094 /* Before returning, examine all the transports touched in
1095 * this call. Right now, we bluntly force clear all the
1096 * transports. Things might change after we implement Nagle.
1097 * But such an examination is still required.
1099 * --xguo
1101 while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
1102 struct sctp_transport *t = list_entry(ltransport,
1103 struct sctp_transport,
1104 send_ready);
1105 packet = &t->packet;
1106 if (!sctp_packet_empty(packet))
1107 error = sctp_packet_transmit(packet);
1109 /* Clear the burst limited state, if any */
1110 sctp_transport_burst_reset(t);
1113 return error;
1116 /* Update unack_data based on the incoming SACK chunk */
1117 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1118 struct sctp_sackhdr *sack)
1120 sctp_sack_variable_t *frags;
1121 __u16 unack_data;
1122 int i;
1124 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1126 frags = sack->variable;
1127 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1128 unack_data -= ((ntohs(frags[i].gab.end) -
1129 ntohs(frags[i].gab.start) + 1));
1132 assoc->unack_data = unack_data;
1135 /* This is where we REALLY process a SACK.
1137 * Process the SACK against the outqueue. Mostly, this just frees
1138 * things off the transmitted queue.
1140 int sctp_outq_sack(struct sctp_outq *q, struct sctp_sackhdr *sack)
1142 struct sctp_association *asoc = q->asoc;
1143 struct sctp_transport *transport;
1144 struct sctp_chunk *tchunk = NULL;
1145 struct list_head *lchunk, *transport_list, *temp;
1146 sctp_sack_variable_t *frags = sack->variable;
1147 __u32 sack_ctsn, ctsn, tsn;
1148 __u32 highest_tsn, highest_new_tsn;
1149 __u32 sack_a_rwnd;
1150 unsigned outstanding;
1151 struct sctp_transport *primary = asoc->peer.primary_path;
1152 int count_of_newacks = 0;
1153 int gap_ack_blocks;
1154 u8 accum_moved = 0;
1156 /* Grab the association's destination address list. */
1157 transport_list = &asoc->peer.transport_addr_list;
1159 sack_ctsn = ntohl(sack->cum_tsn_ack);
1160 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1162 * SFR-CACC algorithm:
1163 * On receipt of a SACK the sender SHOULD execute the
1164 * following statements.
1166 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1167 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1168 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1169 * all destinations.
1170 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1171 * is set the receiver of the SACK MUST take the following actions:
1173 * A) Initialize the cacc_saw_newack to 0 for all destination
1174 * addresses.
1176 * Only bother if changeover_active is set. Otherwise, this is
1177 * totally suboptimal to do on every SACK.
1179 if (primary->cacc.changeover_active) {
1180 u8 clear_cycling = 0;
1182 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1183 primary->cacc.changeover_active = 0;
1184 clear_cycling = 1;
1187 if (clear_cycling || gap_ack_blocks) {
1188 list_for_each_entry(transport, transport_list,
1189 transports) {
1190 if (clear_cycling)
1191 transport->cacc.cycling_changeover = 0;
1192 if (gap_ack_blocks)
1193 transport->cacc.cacc_saw_newack = 0;
1198 /* Get the highest TSN in the sack. */
1199 highest_tsn = sack_ctsn;
1200 if (gap_ack_blocks)
1201 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1203 if (TSN_lt(asoc->highest_sacked, highest_tsn))
1204 asoc->highest_sacked = highest_tsn;
1206 highest_new_tsn = sack_ctsn;
1208 /* Run through the retransmit queue. Credit bytes received
1209 * and free those chunks that we can.
1211 sctp_check_transmitted(q, &q->retransmit, NULL, sack, &highest_new_tsn);
1213 /* Run through the transmitted queue.
1214 * Credit bytes received and free those chunks which we can.
1216 * This is a MASSIVE candidate for optimization.
1218 list_for_each_entry(transport, transport_list, transports) {
1219 sctp_check_transmitted(q, &transport->transmitted,
1220 transport, sack, &highest_new_tsn);
1222 * SFR-CACC algorithm:
1223 * C) Let count_of_newacks be the number of
1224 * destinations for which cacc_saw_newack is set.
1226 if (transport->cacc.cacc_saw_newack)
1227 count_of_newacks ++;
1230 /* Move the Cumulative TSN Ack Point if appropriate. */
1231 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) {
1232 asoc->ctsn_ack_point = sack_ctsn;
1233 accum_moved = 1;
1236 if (gap_ack_blocks) {
1238 if (asoc->fast_recovery && accum_moved)
1239 highest_new_tsn = highest_tsn;
1241 list_for_each_entry(transport, transport_list, transports)
1242 sctp_mark_missing(q, &transport->transmitted, transport,
1243 highest_new_tsn, count_of_newacks);
1246 /* Update unack_data field in the assoc. */
1247 sctp_sack_update_unack_data(asoc, sack);
1249 ctsn = asoc->ctsn_ack_point;
1251 /* Throw away stuff rotting on the sack queue. */
1252 list_for_each_safe(lchunk, temp, &q->sacked) {
1253 tchunk = list_entry(lchunk, struct sctp_chunk,
1254 transmitted_list);
1255 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1256 if (TSN_lte(tsn, ctsn)) {
1257 list_del_init(&tchunk->transmitted_list);
1258 sctp_chunk_free(tchunk);
1262 /* ii) Set rwnd equal to the newly received a_rwnd minus the
1263 * number of bytes still outstanding after processing the
1264 * Cumulative TSN Ack and the Gap Ack Blocks.
1267 sack_a_rwnd = ntohl(sack->a_rwnd);
1268 outstanding = q->outstanding_bytes;
1270 if (outstanding < sack_a_rwnd)
1271 sack_a_rwnd -= outstanding;
1272 else
1273 sack_a_rwnd = 0;
1275 asoc->peer.rwnd = sack_a_rwnd;
1277 sctp_generate_fwdtsn(q, sack_ctsn);
1279 SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n",
1280 __func__, sack_ctsn);
1281 SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, "
1282 "%p is 0x%x. Adv peer ack point: 0x%x\n",
1283 __func__, asoc, ctsn, asoc->adv_peer_ack_point);
1285 /* See if all chunks are acked.
1286 * Make sure the empty queue handler will get run later.
1288 q->empty = (list_empty(&q->out_chunk_list) &&
1289 list_empty(&q->retransmit));
1290 if (!q->empty)
1291 goto finish;
1293 list_for_each_entry(transport, transport_list, transports) {
1294 q->empty = q->empty && list_empty(&transport->transmitted);
1295 if (!q->empty)
1296 goto finish;
1299 SCTP_DEBUG_PRINTK("sack queue is empty.\n");
1300 finish:
1301 return q->empty;
1304 /* Is the outqueue empty? */
1305 int sctp_outq_is_empty(const struct sctp_outq *q)
1307 return q->empty;
1310 /********************************************************************
1311 * 2nd Level Abstractions
1312 ********************************************************************/
1314 /* Go through a transport's transmitted list or the association's retransmit
1315 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1316 * The retransmit list will not have an associated transport.
1318 * I added coherent debug information output. --xguo
1320 * Instead of printing 'sacked' or 'kept' for each TSN on the
1321 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1322 * KEPT TSN6-TSN7, etc.
1324 static void sctp_check_transmitted(struct sctp_outq *q,
1325 struct list_head *transmitted_queue,
1326 struct sctp_transport *transport,
1327 struct sctp_sackhdr *sack,
1328 __u32 *highest_new_tsn_in_sack)
1330 struct list_head *lchunk;
1331 struct sctp_chunk *tchunk;
1332 struct list_head tlist;
1333 __u32 tsn;
1334 __u32 sack_ctsn;
1335 __u32 rtt;
1336 __u8 restart_timer = 0;
1337 int bytes_acked = 0;
1338 int migrate_bytes = 0;
1340 /* These state variables are for coherent debug output. --xguo */
1342 #if SCTP_DEBUG
1343 __u32 dbg_ack_tsn = 0; /* An ACKed TSN range starts here... */
1344 __u32 dbg_last_ack_tsn = 0; /* ...and finishes here. */
1345 __u32 dbg_kept_tsn = 0; /* An un-ACKed range starts here... */
1346 __u32 dbg_last_kept_tsn = 0; /* ...and finishes here. */
1348 /* 0 : The last TSN was ACKed.
1349 * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1350 * -1: We need to initialize.
1352 int dbg_prt_state = -1;
1353 #endif /* SCTP_DEBUG */
1355 sack_ctsn = ntohl(sack->cum_tsn_ack);
1357 INIT_LIST_HEAD(&tlist);
1359 /* The while loop will skip empty transmitted queues. */
1360 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1361 tchunk = list_entry(lchunk, struct sctp_chunk,
1362 transmitted_list);
1364 if (sctp_chunk_abandoned(tchunk)) {
1365 /* Move the chunk to abandoned list. */
1366 sctp_insert_list(&q->abandoned, lchunk);
1368 /* If this chunk has not been acked, stop
1369 * considering it as 'outstanding'.
1371 if (!tchunk->tsn_gap_acked) {
1372 if (tchunk->transport)
1373 tchunk->transport->flight_size -=
1374 sctp_data_size(tchunk);
1375 q->outstanding_bytes -= sctp_data_size(tchunk);
1377 continue;
1380 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1381 if (sctp_acked(sack, tsn)) {
1382 /* If this queue is the retransmit queue, the
1383 * retransmit timer has already reclaimed
1384 * the outstanding bytes for this chunk, so only
1385 * count bytes associated with a transport.
1387 if (transport) {
1388 /* If this chunk is being used for RTT
1389 * measurement, calculate the RTT and update
1390 * the RTO using this value.
1392 * 6.3.1 C5) Karn's algorithm: RTT measurements
1393 * MUST NOT be made using packets that were
1394 * retransmitted (and thus for which it is
1395 * ambiguous whether the reply was for the
1396 * first instance of the packet or a later
1397 * instance).
1399 if (!tchunk->tsn_gap_acked &&
1400 tchunk->rtt_in_progress) {
1401 tchunk->rtt_in_progress = 0;
1402 rtt = jiffies - tchunk->sent_at;
1403 sctp_transport_update_rto(transport,
1404 rtt);
1408 /* If the chunk hasn't been marked as ACKED,
1409 * mark it and account bytes_acked if the
1410 * chunk had a valid transport (it will not
1411 * have a transport if ASCONF had deleted it
1412 * while DATA was outstanding).
1414 if (!tchunk->tsn_gap_acked) {
1415 tchunk->tsn_gap_acked = 1;
1416 *highest_new_tsn_in_sack = tsn;
1417 bytes_acked += sctp_data_size(tchunk);
1418 if (!tchunk->transport)
1419 migrate_bytes += sctp_data_size(tchunk);
1422 if (TSN_lte(tsn, sack_ctsn)) {
1423 /* RFC 2960 6.3.2 Retransmission Timer Rules
1425 * R3) Whenever a SACK is received
1426 * that acknowledges the DATA chunk
1427 * with the earliest outstanding TSN
1428 * for that address, restart T3-rtx
1429 * timer for that address with its
1430 * current RTO.
1432 restart_timer = 1;
1434 if (!tchunk->tsn_gap_acked) {
1436 * SFR-CACC algorithm:
1437 * 2) If the SACK contains gap acks
1438 * and the flag CHANGEOVER_ACTIVE is
1439 * set the receiver of the SACK MUST
1440 * take the following action:
1442 * B) For each TSN t being acked that
1443 * has not been acked in any SACK so
1444 * far, set cacc_saw_newack to 1 for
1445 * the destination that the TSN was
1446 * sent to.
1448 if (transport &&
1449 sack->num_gap_ack_blocks &&
1450 q->asoc->peer.primary_path->cacc.
1451 changeover_active)
1452 transport->cacc.cacc_saw_newack
1453 = 1;
1456 list_add_tail(&tchunk->transmitted_list,
1457 &q->sacked);
1458 } else {
1459 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1460 * M2) Each time a SACK arrives reporting
1461 * 'Stray DATA chunk(s)' record the highest TSN
1462 * reported as newly acknowledged, call this
1463 * value 'HighestTSNinSack'. A newly
1464 * acknowledged DATA chunk is one not
1465 * previously acknowledged in a SACK.
1467 * When the SCTP sender of data receives a SACK
1468 * chunk that acknowledges, for the first time,
1469 * the receipt of a DATA chunk, all the still
1470 * unacknowledged DATA chunks whose TSN is
1471 * older than that newly acknowledged DATA
1472 * chunk, are qualified as 'Stray DATA chunks'.
1474 list_add_tail(lchunk, &tlist);
1477 #if SCTP_DEBUG
1478 switch (dbg_prt_state) {
1479 case 0: /* last TSN was ACKed */
1480 if (dbg_last_ack_tsn + 1 == tsn) {
1481 /* This TSN belongs to the
1482 * current ACK range.
1484 break;
1487 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1488 /* Display the end of the
1489 * current range.
1491 SCTP_DEBUG_PRINTK_CONT("-%08x",
1492 dbg_last_ack_tsn);
1495 /* Start a new range. */
1496 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn);
1497 dbg_ack_tsn = tsn;
1498 break;
1500 case 1: /* The last TSN was NOT ACKed. */
1501 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1502 /* Display the end of current range. */
1503 SCTP_DEBUG_PRINTK_CONT("-%08x",
1504 dbg_last_kept_tsn);
1507 SCTP_DEBUG_PRINTK_CONT("\n");
1509 /* FALL THROUGH... */
1510 default:
1511 /* This is the first-ever TSN we examined. */
1512 /* Start a new range of ACK-ed TSNs. */
1513 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn);
1514 dbg_prt_state = 0;
1515 dbg_ack_tsn = tsn;
1518 dbg_last_ack_tsn = tsn;
1519 #endif /* SCTP_DEBUG */
1521 } else {
1522 if (tchunk->tsn_gap_acked) {
1523 SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1524 "data TSN: 0x%x\n",
1525 __func__,
1526 tsn);
1527 tchunk->tsn_gap_acked = 0;
1529 if (tchunk->transport)
1530 bytes_acked -= sctp_data_size(tchunk);
1532 /* RFC 2960 6.3.2 Retransmission Timer Rules
1534 * R4) Whenever a SACK is received missing a
1535 * TSN that was previously acknowledged via a
1536 * Gap Ack Block, start T3-rtx for the
1537 * destination address to which the DATA
1538 * chunk was originally
1539 * transmitted if it is not already running.
1541 restart_timer = 1;
1544 list_add_tail(lchunk, &tlist);
1546 #if SCTP_DEBUG
1547 /* See the above comments on ACK-ed TSNs. */
1548 switch (dbg_prt_state) {
1549 case 1:
1550 if (dbg_last_kept_tsn + 1 == tsn)
1551 break;
1553 if (dbg_last_kept_tsn != dbg_kept_tsn)
1554 SCTP_DEBUG_PRINTK_CONT("-%08x",
1555 dbg_last_kept_tsn);
1557 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn);
1558 dbg_kept_tsn = tsn;
1559 break;
1561 case 0:
1562 if (dbg_last_ack_tsn != dbg_ack_tsn)
1563 SCTP_DEBUG_PRINTK_CONT("-%08x",
1564 dbg_last_ack_tsn);
1565 SCTP_DEBUG_PRINTK_CONT("\n");
1567 /* FALL THROUGH... */
1568 default:
1569 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn);
1570 dbg_prt_state = 1;
1571 dbg_kept_tsn = tsn;
1574 dbg_last_kept_tsn = tsn;
1575 #endif /* SCTP_DEBUG */
1579 #if SCTP_DEBUG
1580 /* Finish off the last range, displaying its ending TSN. */
1581 switch (dbg_prt_state) {
1582 case 0:
1583 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1584 SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_ack_tsn);
1585 } else {
1586 SCTP_DEBUG_PRINTK_CONT("\n");
1588 break;
1590 case 1:
1591 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1592 SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_kept_tsn);
1593 } else {
1594 SCTP_DEBUG_PRINTK_CONT("\n");
1597 #endif /* SCTP_DEBUG */
1598 if (transport) {
1599 if (bytes_acked) {
1600 struct sctp_association *asoc = transport->asoc;
1602 /* We may have counted DATA that was migrated
1603 * to this transport due to DEL-IP operation.
1604 * Subtract those bytes, since the were never
1605 * send on this transport and shouldn't be
1606 * credited to this transport.
1608 bytes_acked -= migrate_bytes;
1610 /* 8.2. When an outstanding TSN is acknowledged,
1611 * the endpoint shall clear the error counter of
1612 * the destination transport address to which the
1613 * DATA chunk was last sent.
1614 * The association's overall error counter is
1615 * also cleared.
1617 transport->error_count = 0;
1618 transport->asoc->overall_error_count = 0;
1621 * While in SHUTDOWN PENDING, we may have started
1622 * the T5 shutdown guard timer after reaching the
1623 * retransmission limit. Stop that timer as soon
1624 * as the receiver acknowledged any data.
1626 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING &&
1627 del_timer(&asoc->timers
1628 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]))
1629 sctp_association_put(asoc);
1631 /* Mark the destination transport address as
1632 * active if it is not so marked.
1634 if ((transport->state == SCTP_INACTIVE) ||
1635 (transport->state == SCTP_UNCONFIRMED)) {
1636 sctp_assoc_control_transport(
1637 transport->asoc,
1638 transport,
1639 SCTP_TRANSPORT_UP,
1640 SCTP_RECEIVED_SACK);
1643 sctp_transport_raise_cwnd(transport, sack_ctsn,
1644 bytes_acked);
1646 transport->flight_size -= bytes_acked;
1647 if (transport->flight_size == 0)
1648 transport->partial_bytes_acked = 0;
1649 q->outstanding_bytes -= bytes_acked + migrate_bytes;
1650 } else {
1651 /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1652 * When a sender is doing zero window probing, it
1653 * should not timeout the association if it continues
1654 * to receive new packets from the receiver. The
1655 * reason is that the receiver MAY keep its window
1656 * closed for an indefinite time.
1657 * A sender is doing zero window probing when the
1658 * receiver's advertised window is zero, and there is
1659 * only one data chunk in flight to the receiver.
1661 * Allow the association to timeout while in SHUTDOWN
1662 * PENDING or SHUTDOWN RECEIVED in case the receiver
1663 * stays in zero window mode forever.
1665 if (!q->asoc->peer.rwnd &&
1666 !list_empty(&tlist) &&
1667 (sack_ctsn+2 == q->asoc->next_tsn) &&
1668 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) {
1669 SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1670 "window probe: %u\n",
1671 __func__, sack_ctsn);
1672 q->asoc->overall_error_count = 0;
1673 transport->error_count = 0;
1677 /* RFC 2960 6.3.2 Retransmission Timer Rules
1679 * R2) Whenever all outstanding data sent to an address have
1680 * been acknowledged, turn off the T3-rtx timer of that
1681 * address.
1683 if (!transport->flight_size) {
1684 if (timer_pending(&transport->T3_rtx_timer) &&
1685 del_timer(&transport->T3_rtx_timer)) {
1686 sctp_transport_put(transport);
1688 } else if (restart_timer) {
1689 if (!mod_timer(&transport->T3_rtx_timer,
1690 jiffies + transport->rto))
1691 sctp_transport_hold(transport);
1695 list_splice(&tlist, transmitted_queue);
1698 /* Mark chunks as missing and consequently may get retransmitted. */
1699 static void sctp_mark_missing(struct sctp_outq *q,
1700 struct list_head *transmitted_queue,
1701 struct sctp_transport *transport,
1702 __u32 highest_new_tsn_in_sack,
1703 int count_of_newacks)
1705 struct sctp_chunk *chunk;
1706 __u32 tsn;
1707 char do_fast_retransmit = 0;
1708 struct sctp_association *asoc = q->asoc;
1709 struct sctp_transport *primary = asoc->peer.primary_path;
1711 list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1713 tsn = ntohl(chunk->subh.data_hdr->tsn);
1715 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1716 * 'Unacknowledged TSN's', if the TSN number of an
1717 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1718 * value, increment the 'TSN.Missing.Report' count on that
1719 * chunk if it has NOT been fast retransmitted or marked for
1720 * fast retransmit already.
1722 if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
1723 !chunk->tsn_gap_acked &&
1724 TSN_lt(tsn, highest_new_tsn_in_sack)) {
1726 /* SFR-CACC may require us to skip marking
1727 * this chunk as missing.
1729 if (!transport || !sctp_cacc_skip(primary,
1730 chunk->transport,
1731 count_of_newacks, tsn)) {
1732 chunk->tsn_missing_report++;
1734 SCTP_DEBUG_PRINTK(
1735 "%s: TSN 0x%x missing counter: %d\n",
1736 __func__, tsn,
1737 chunk->tsn_missing_report);
1741 * M4) If any DATA chunk is found to have a
1742 * 'TSN.Missing.Report'
1743 * value larger than or equal to 3, mark that chunk for
1744 * retransmission and start the fast retransmit procedure.
1747 if (chunk->tsn_missing_report >= 3) {
1748 chunk->fast_retransmit = SCTP_NEED_FRTX;
1749 do_fast_retransmit = 1;
1753 if (transport) {
1754 if (do_fast_retransmit)
1755 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1757 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1758 "ssthresh: %d, flight_size: %d, pba: %d\n",
1759 __func__, transport, transport->cwnd,
1760 transport->ssthresh, transport->flight_size,
1761 transport->partial_bytes_acked);
1765 /* Is the given TSN acked by this packet? */
1766 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1768 int i;
1769 sctp_sack_variable_t *frags;
1770 __u16 gap;
1771 __u32 ctsn = ntohl(sack->cum_tsn_ack);
1773 if (TSN_lte(tsn, ctsn))
1774 goto pass;
1776 /* 3.3.4 Selective Acknowledgement (SACK) (3):
1778 * Gap Ack Blocks:
1779 * These fields contain the Gap Ack Blocks. They are repeated
1780 * for each Gap Ack Block up to the number of Gap Ack Blocks
1781 * defined in the Number of Gap Ack Blocks field. All DATA
1782 * chunks with TSNs greater than or equal to (Cumulative TSN
1783 * Ack + Gap Ack Block Start) and less than or equal to
1784 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1785 * Block are assumed to have been received correctly.
1788 frags = sack->variable;
1789 gap = tsn - ctsn;
1790 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1791 if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1792 TSN_lte(gap, ntohs(frags[i].gab.end)))
1793 goto pass;
1796 return 0;
1797 pass:
1798 return 1;
1801 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1802 int nskips, __be16 stream)
1804 int i;
1806 for (i = 0; i < nskips; i++) {
1807 if (skiplist[i].stream == stream)
1808 return i;
1810 return i;
1813 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1814 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1816 struct sctp_association *asoc = q->asoc;
1817 struct sctp_chunk *ftsn_chunk = NULL;
1818 struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1819 int nskips = 0;
1820 int skip_pos = 0;
1821 __u32 tsn;
1822 struct sctp_chunk *chunk;
1823 struct list_head *lchunk, *temp;
1825 if (!asoc->peer.prsctp_capable)
1826 return;
1828 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1829 * received SACK.
1831 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1832 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1834 if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1835 asoc->adv_peer_ack_point = ctsn;
1837 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1838 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1839 * the chunk next in the out-queue space is marked as "abandoned" as
1840 * shown in the following example:
1842 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1843 * and the Advanced.Peer.Ack.Point is updated to this value:
1845 * out-queue at the end of ==> out-queue after Adv.Ack.Point
1846 * normal SACK processing local advancement
1847 * ... ...
1848 * Adv.Ack.Pt-> 102 acked 102 acked
1849 * 103 abandoned 103 abandoned
1850 * 104 abandoned Adv.Ack.P-> 104 abandoned
1851 * 105 105
1852 * 106 acked 106 acked
1853 * ... ...
1855 * In this example, the data sender successfully advanced the
1856 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1858 list_for_each_safe(lchunk, temp, &q->abandoned) {
1859 chunk = list_entry(lchunk, struct sctp_chunk,
1860 transmitted_list);
1861 tsn = ntohl(chunk->subh.data_hdr->tsn);
1863 /* Remove any chunks in the abandoned queue that are acked by
1864 * the ctsn.
1866 if (TSN_lte(tsn, ctsn)) {
1867 list_del_init(lchunk);
1868 sctp_chunk_free(chunk);
1869 } else {
1870 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1871 asoc->adv_peer_ack_point = tsn;
1872 if (chunk->chunk_hdr->flags &
1873 SCTP_DATA_UNORDERED)
1874 continue;
1875 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1876 nskips,
1877 chunk->subh.data_hdr->stream);
1878 ftsn_skip_arr[skip_pos].stream =
1879 chunk->subh.data_hdr->stream;
1880 ftsn_skip_arr[skip_pos].ssn =
1881 chunk->subh.data_hdr->ssn;
1882 if (skip_pos == nskips)
1883 nskips++;
1884 if (nskips == 10)
1885 break;
1886 } else
1887 break;
1891 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1892 * is greater than the Cumulative TSN ACK carried in the received
1893 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1894 * chunk containing the latest value of the
1895 * "Advanced.Peer.Ack.Point".
1897 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1898 * list each stream and sequence number in the forwarded TSN. This
1899 * information will enable the receiver to easily find any
1900 * stranded TSN's waiting on stream reorder queues. Each stream
1901 * SHOULD only be reported once; this means that if multiple
1902 * abandoned messages occur in the same stream then only the
1903 * highest abandoned stream sequence number is reported. If the
1904 * total size of the FORWARD TSN does NOT fit in a single MTU then
1905 * the sender of the FORWARD TSN SHOULD lower the
1906 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1907 * single MTU.
1909 if (asoc->adv_peer_ack_point > ctsn)
1910 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1911 nskips, &ftsn_skip_arr[0]);
1913 if (ftsn_chunk) {
1914 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1915 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);