igb: Simplify how we populate the RSS key
[linux-2.6/cjktty.git] / net / sctp / outqueue.c
blobd16632e1503a56a4c592157936ab568a2be3a3a4
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 struct net *net = sock_net(q->asoc->base.sk);
303 int error = 0;
305 SCTP_DEBUG_PRINTK("sctp_outq_tail(%p, %p[%s])\n",
306 q, chunk, chunk && chunk->chunk_hdr ?
307 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
308 : "Illegal Chunk");
310 /* If it is data, queue it up, otherwise, send it
311 * immediately.
313 if (sctp_chunk_is_data(chunk)) {
314 /* Is it OK to queue data chunks? */
315 /* From 9. Termination of Association
317 * When either endpoint performs a shutdown, the
318 * association on each peer will stop accepting new
319 * data from its user and only deliver data in queue
320 * at the time of sending or receiving the SHUTDOWN
321 * chunk.
323 switch (q->asoc->state) {
324 case SCTP_STATE_CLOSED:
325 case SCTP_STATE_SHUTDOWN_PENDING:
326 case SCTP_STATE_SHUTDOWN_SENT:
327 case SCTP_STATE_SHUTDOWN_RECEIVED:
328 case SCTP_STATE_SHUTDOWN_ACK_SENT:
329 /* Cannot send after transport endpoint shutdown */
330 error = -ESHUTDOWN;
331 break;
333 default:
334 SCTP_DEBUG_PRINTK("outqueueing (%p, %p[%s])\n",
335 q, chunk, chunk && chunk->chunk_hdr ?
336 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
337 : "Illegal Chunk");
339 sctp_outq_tail_data(q, chunk);
340 if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
341 SCTP_INC_STATS(net, SCTP_MIB_OUTUNORDERCHUNKS);
342 else
343 SCTP_INC_STATS(net, SCTP_MIB_OUTORDERCHUNKS);
344 q->empty = 0;
345 break;
347 } else {
348 list_add_tail(&chunk->list, &q->control_chunk_list);
349 SCTP_INC_STATS(net, SCTP_MIB_OUTCTRLCHUNKS);
352 if (error < 0)
353 return error;
355 if (!q->cork)
356 error = sctp_outq_flush(q, 0);
358 return error;
361 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list
362 * and the abandoned list are in ascending order.
364 static void sctp_insert_list(struct list_head *head, struct list_head *new)
366 struct list_head *pos;
367 struct sctp_chunk *nchunk, *lchunk;
368 __u32 ntsn, ltsn;
369 int done = 0;
371 nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
372 ntsn = ntohl(nchunk->subh.data_hdr->tsn);
374 list_for_each(pos, head) {
375 lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
376 ltsn = ntohl(lchunk->subh.data_hdr->tsn);
377 if (TSN_lt(ntsn, ltsn)) {
378 list_add(new, pos->prev);
379 done = 1;
380 break;
383 if (!done)
384 list_add_tail(new, head);
387 /* Mark all the eligible packets on a transport for retransmission. */
388 void sctp_retransmit_mark(struct sctp_outq *q,
389 struct sctp_transport *transport,
390 __u8 reason)
392 struct list_head *lchunk, *ltemp;
393 struct sctp_chunk *chunk;
395 /* Walk through the specified transmitted queue. */
396 list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
397 chunk = list_entry(lchunk, struct sctp_chunk,
398 transmitted_list);
400 /* If the chunk is abandoned, move it to abandoned list. */
401 if (sctp_chunk_abandoned(chunk)) {
402 list_del_init(lchunk);
403 sctp_insert_list(&q->abandoned, lchunk);
405 /* If this chunk has not been previousely acked,
406 * stop considering it 'outstanding'. Our peer
407 * will most likely never see it since it will
408 * not be retransmitted
410 if (!chunk->tsn_gap_acked) {
411 if (chunk->transport)
412 chunk->transport->flight_size -=
413 sctp_data_size(chunk);
414 q->outstanding_bytes -= sctp_data_size(chunk);
415 q->asoc->peer.rwnd += sctp_data_size(chunk);
417 continue;
420 /* If we are doing retransmission due to a timeout or pmtu
421 * discovery, only the chunks that are not yet acked should
422 * be added to the retransmit queue.
424 if ((reason == SCTP_RTXR_FAST_RTX &&
425 (chunk->fast_retransmit == SCTP_NEED_FRTX)) ||
426 (reason != SCTP_RTXR_FAST_RTX && !chunk->tsn_gap_acked)) {
427 /* RFC 2960 6.2.1 Processing a Received SACK
429 * C) Any time a DATA chunk is marked for
430 * retransmission (via either T3-rtx timer expiration
431 * (Section 6.3.3) or via fast retransmit
432 * (Section 7.2.4)), add the data size of those
433 * chunks to the rwnd.
435 q->asoc->peer.rwnd += sctp_data_size(chunk);
436 q->outstanding_bytes -= sctp_data_size(chunk);
437 if (chunk->transport)
438 transport->flight_size -= sctp_data_size(chunk);
440 /* sctpimpguide-05 Section 2.8.2
441 * M5) If a T3-rtx timer expires, the
442 * 'TSN.Missing.Report' of all affected TSNs is set
443 * to 0.
445 chunk->tsn_missing_report = 0;
447 /* If a chunk that is being used for RTT measurement
448 * has to be retransmitted, we cannot use this chunk
449 * anymore for RTT measurements. Reset rto_pending so
450 * that a new RTT measurement is started when a new
451 * data chunk is sent.
453 if (chunk->rtt_in_progress) {
454 chunk->rtt_in_progress = 0;
455 transport->rto_pending = 0;
458 /* Move the chunk to the retransmit queue. The chunks
459 * on the retransmit queue are always kept in order.
461 list_del_init(lchunk);
462 sctp_insert_list(&q->retransmit, lchunk);
466 SCTP_DEBUG_PRINTK("%s: transport: %p, reason: %d, "
467 "cwnd: %d, ssthresh: %d, flight_size: %d, "
468 "pba: %d\n", __func__,
469 transport, reason,
470 transport->cwnd, transport->ssthresh,
471 transport->flight_size,
472 transport->partial_bytes_acked);
476 /* Mark all the eligible packets on a transport for retransmission and force
477 * one packet out.
479 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
480 sctp_retransmit_reason_t reason)
482 struct net *net = sock_net(q->asoc->base.sk);
483 int error = 0;
485 switch(reason) {
486 case SCTP_RTXR_T3_RTX:
487 SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS);
488 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
489 /* Update the retran path if the T3-rtx timer has expired for
490 * the current retran path.
492 if (transport == transport->asoc->peer.retran_path)
493 sctp_assoc_update_retran_path(transport->asoc);
494 transport->asoc->rtx_data_chunks +=
495 transport->asoc->unack_data;
496 break;
497 case SCTP_RTXR_FAST_RTX:
498 SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS);
499 sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
500 q->fast_rtx = 1;
501 break;
502 case SCTP_RTXR_PMTUD:
503 SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS);
504 break;
505 case SCTP_RTXR_T1_RTX:
506 SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS);
507 transport->asoc->init_retries++;
508 break;
509 default:
510 BUG();
513 sctp_retransmit_mark(q, transport, reason);
515 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
516 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
517 * following the procedures outlined in C1 - C5.
519 if (reason == SCTP_RTXR_T3_RTX)
520 sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
522 /* Flush the queues only on timeout, since fast_rtx is only
523 * triggered during sack processing and the queue
524 * will be flushed at the end.
526 if (reason != SCTP_RTXR_FAST_RTX)
527 error = sctp_outq_flush(q, /* rtx_timeout */ 1);
529 if (error)
530 q->asoc->base.sk->sk_err = -error;
534 * Transmit DATA chunks on the retransmit queue. Upon return from
535 * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
536 * need to be transmitted by the caller.
537 * We assume that pkt->transport has already been set.
539 * The return value is a normal kernel error return value.
541 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
542 int rtx_timeout, int *start_timer)
544 struct list_head *lqueue;
545 struct sctp_transport *transport = pkt->transport;
546 sctp_xmit_t status;
547 struct sctp_chunk *chunk, *chunk1;
548 int fast_rtx;
549 int error = 0;
550 int timer = 0;
551 int done = 0;
553 lqueue = &q->retransmit;
554 fast_rtx = q->fast_rtx;
556 /* This loop handles time-out retransmissions, fast retransmissions,
557 * and retransmissions due to opening of whindow.
559 * RFC 2960 6.3.3 Handle T3-rtx Expiration
561 * E3) Determine how many of the earliest (i.e., lowest TSN)
562 * outstanding DATA chunks for the address for which the
563 * T3-rtx has expired will fit into a single packet, subject
564 * to the MTU constraint for the path corresponding to the
565 * destination transport address to which the retransmission
566 * is being sent (this may be different from the address for
567 * which the timer expires [see Section 6.4]). Call this value
568 * K. Bundle and retransmit those K DATA chunks in a single
569 * packet to the destination endpoint.
571 * [Just to be painfully clear, if we are retransmitting
572 * because a timeout just happened, we should send only ONE
573 * packet of retransmitted data.]
575 * For fast retransmissions we also send only ONE packet. However,
576 * if we are just flushing the queue due to open window, we'll
577 * try to send as much as possible.
579 list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
580 /* If the chunk is abandoned, move it to abandoned list. */
581 if (sctp_chunk_abandoned(chunk)) {
582 list_del_init(&chunk->transmitted_list);
583 sctp_insert_list(&q->abandoned,
584 &chunk->transmitted_list);
585 continue;
588 /* Make sure that Gap Acked TSNs are not retransmitted. A
589 * simple approach is just to move such TSNs out of the
590 * way and into a 'transmitted' queue and skip to the
591 * next chunk.
593 if (chunk->tsn_gap_acked) {
594 list_move_tail(&chunk->transmitted_list,
595 &transport->transmitted);
596 continue;
599 /* If we are doing fast retransmit, ignore non-fast_rtransmit
600 * chunks
602 if (fast_rtx && !chunk->fast_retransmit)
603 continue;
605 redo:
606 /* Attempt to append this chunk to the packet. */
607 status = sctp_packet_append_chunk(pkt, chunk);
609 switch (status) {
610 case SCTP_XMIT_PMTU_FULL:
611 if (!pkt->has_data && !pkt->has_cookie_echo) {
612 /* If this packet did not contain DATA then
613 * retransmission did not happen, so do it
614 * again. We'll ignore the error here since
615 * control chunks are already freed so there
616 * is nothing we can do.
618 sctp_packet_transmit(pkt);
619 goto redo;
622 /* Send this packet. */
623 error = sctp_packet_transmit(pkt);
625 /* If we are retransmitting, we should only
626 * send a single packet.
627 * Otherwise, try appending this chunk again.
629 if (rtx_timeout || fast_rtx)
630 done = 1;
631 else
632 goto redo;
634 /* Bundle next chunk in the next round. */
635 break;
637 case SCTP_XMIT_RWND_FULL:
638 /* Send this packet. */
639 error = sctp_packet_transmit(pkt);
641 /* Stop sending DATA as there is no more room
642 * at the receiver.
644 done = 1;
645 break;
647 case SCTP_XMIT_NAGLE_DELAY:
648 /* Send this packet. */
649 error = sctp_packet_transmit(pkt);
651 /* Stop sending DATA because of nagle delay. */
652 done = 1;
653 break;
655 default:
656 /* The append was successful, so add this chunk to
657 * the transmitted list.
659 list_move_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 (new_transport->state == SCTP_PF)) {
797 /* If the chunk is Heartbeat or Heartbeat Ack,
798 * send it to chunk->transport, even if it's
799 * inactive.
801 * 3.3.6 Heartbeat Acknowledgement:
802 * ...
803 * A HEARTBEAT ACK is always sent to the source IP
804 * address of the IP datagram containing the
805 * HEARTBEAT chunk to which this ack is responding.
806 * ...
808 * ASCONF_ACKs also must be sent to the source.
810 if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
811 chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK &&
812 chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK)
813 new_transport = asoc->peer.active_path;
816 /* Are we switching transports?
817 * Take care of transport locks.
819 if (new_transport != transport) {
820 transport = new_transport;
821 if (list_empty(&transport->send_ready)) {
822 list_add_tail(&transport->send_ready,
823 &transport_list);
825 packet = &transport->packet;
826 sctp_packet_config(packet, vtag,
827 asoc->peer.ecn_capable);
830 switch (chunk->chunk_hdr->type) {
832 * 6.10 Bundling
833 * ...
834 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
835 * COMPLETE with any other chunks. [Send them immediately.]
837 case SCTP_CID_INIT:
838 case SCTP_CID_INIT_ACK:
839 case SCTP_CID_SHUTDOWN_COMPLETE:
840 sctp_packet_init(&singleton, transport, sport, dport);
841 sctp_packet_config(&singleton, vtag, 0);
842 sctp_packet_append_chunk(&singleton, chunk);
843 error = sctp_packet_transmit(&singleton);
844 if (error < 0)
845 return error;
846 break;
848 case SCTP_CID_ABORT:
849 if (sctp_test_T_bit(chunk)) {
850 packet->vtag = asoc->c.my_vtag;
852 /* The following chunks are "response" chunks, i.e.
853 * they are generated in response to something we
854 * received. If we are sending these, then we can
855 * send only 1 packet containing these chunks.
857 case SCTP_CID_HEARTBEAT_ACK:
858 case SCTP_CID_SHUTDOWN_ACK:
859 case SCTP_CID_COOKIE_ACK:
860 case SCTP_CID_COOKIE_ECHO:
861 case SCTP_CID_ERROR:
862 case SCTP_CID_ECN_CWR:
863 case SCTP_CID_ASCONF_ACK:
864 one_packet = 1;
865 /* Fall through */
867 case SCTP_CID_SACK:
868 case SCTP_CID_HEARTBEAT:
869 case SCTP_CID_SHUTDOWN:
870 case SCTP_CID_ECN_ECNE:
871 case SCTP_CID_ASCONF:
872 case SCTP_CID_FWD_TSN:
873 status = sctp_packet_transmit_chunk(packet, chunk,
874 one_packet);
875 if (status != SCTP_XMIT_OK) {
876 /* put the chunk back */
877 list_add(&chunk->list, &q->control_chunk_list);
878 } else if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN) {
879 /* PR-SCTP C5) If a FORWARD TSN is sent, the
880 * sender MUST assure that at least one T3-rtx
881 * timer is running.
883 sctp_transport_reset_timers(transport);
885 break;
887 default:
888 /* We built a chunk with an illegal type! */
889 BUG();
893 if (q->asoc->src_out_of_asoc_ok)
894 goto sctp_flush_out;
896 /* Is it OK to send data chunks? */
897 switch (asoc->state) {
898 case SCTP_STATE_COOKIE_ECHOED:
899 /* Only allow bundling when this packet has a COOKIE-ECHO
900 * chunk.
902 if (!packet || !packet->has_cookie_echo)
903 break;
905 /* fallthru */
906 case SCTP_STATE_ESTABLISHED:
907 case SCTP_STATE_SHUTDOWN_PENDING:
908 case SCTP_STATE_SHUTDOWN_RECEIVED:
910 * RFC 2960 6.1 Transmission of DATA Chunks
912 * C) When the time comes for the sender to transmit,
913 * before sending new DATA chunks, the sender MUST
914 * first transmit any outstanding DATA chunks which
915 * are marked for retransmission (limited by the
916 * current cwnd).
918 if (!list_empty(&q->retransmit)) {
919 if (asoc->peer.retran_path->state == SCTP_UNCONFIRMED)
920 goto sctp_flush_out;
921 if (transport == asoc->peer.retran_path)
922 goto retran;
924 /* Switch transports & prepare the packet. */
926 transport = asoc->peer.retran_path;
928 if (list_empty(&transport->send_ready)) {
929 list_add_tail(&transport->send_ready,
930 &transport_list);
933 packet = &transport->packet;
934 sctp_packet_config(packet, vtag,
935 asoc->peer.ecn_capable);
936 retran:
937 error = sctp_outq_flush_rtx(q, packet,
938 rtx_timeout, &start_timer);
940 if (start_timer)
941 sctp_transport_reset_timers(transport);
943 /* This can happen on COOKIE-ECHO resend. Only
944 * one chunk can get bundled with a COOKIE-ECHO.
946 if (packet->has_cookie_echo)
947 goto sctp_flush_out;
949 /* Don't send new data if there is still data
950 * waiting to retransmit.
952 if (!list_empty(&q->retransmit))
953 goto sctp_flush_out;
956 /* Apply Max.Burst limitation to the current transport in
957 * case it will be used for new data. We are going to
958 * rest it before we return, but we want to apply the limit
959 * to the currently queued data.
961 if (transport)
962 sctp_transport_burst_limited(transport);
964 /* Finally, transmit new packets. */
965 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
966 /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
967 * stream identifier.
969 if (chunk->sinfo.sinfo_stream >=
970 asoc->c.sinit_num_ostreams) {
972 /* Mark as failed send. */
973 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
974 sctp_chunk_free(chunk);
975 continue;
978 /* Has this chunk expired? */
979 if (sctp_chunk_abandoned(chunk)) {
980 sctp_chunk_fail(chunk, 0);
981 sctp_chunk_free(chunk);
982 continue;
985 /* If there is a specified transport, use it.
986 * Otherwise, we want to use the active path.
988 new_transport = chunk->transport;
989 if (!new_transport ||
990 ((new_transport->state == SCTP_INACTIVE) ||
991 (new_transport->state == SCTP_UNCONFIRMED) ||
992 (new_transport->state == SCTP_PF)))
993 new_transport = asoc->peer.active_path;
994 if (new_transport->state == SCTP_UNCONFIRMED)
995 continue;
997 /* Change packets if necessary. */
998 if (new_transport != transport) {
999 transport = new_transport;
1001 /* Schedule to have this transport's
1002 * packet flushed.
1004 if (list_empty(&transport->send_ready)) {
1005 list_add_tail(&transport->send_ready,
1006 &transport_list);
1009 packet = &transport->packet;
1010 sctp_packet_config(packet, vtag,
1011 asoc->peer.ecn_capable);
1012 /* We've switched transports, so apply the
1013 * Burst limit to the new transport.
1015 sctp_transport_burst_limited(transport);
1018 SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
1019 q, chunk,
1020 chunk && chunk->chunk_hdr ?
1021 sctp_cname(SCTP_ST_CHUNK(
1022 chunk->chunk_hdr->type))
1023 : "Illegal Chunk");
1025 SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
1026 "%p skb->users %d.\n",
1027 ntohl(chunk->subh.data_hdr->tsn),
1028 chunk->skb ?chunk->skb->head : NULL,
1029 chunk->skb ?
1030 atomic_read(&chunk->skb->users) : -1);
1032 /* Add the chunk to the packet. */
1033 status = sctp_packet_transmit_chunk(packet, chunk, 0);
1035 switch (status) {
1036 case SCTP_XMIT_PMTU_FULL:
1037 case SCTP_XMIT_RWND_FULL:
1038 case SCTP_XMIT_NAGLE_DELAY:
1039 /* We could not append this chunk, so put
1040 * the chunk back on the output queue.
1042 SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
1043 "not transmit TSN: 0x%x, status: %d\n",
1044 ntohl(chunk->subh.data_hdr->tsn),
1045 status);
1046 sctp_outq_head_data(q, chunk);
1047 goto sctp_flush_out;
1048 break;
1050 case SCTP_XMIT_OK:
1051 /* The sender is in the SHUTDOWN-PENDING state,
1052 * The sender MAY set the I-bit in the DATA
1053 * chunk header.
1055 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING)
1056 chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM;
1058 break;
1060 default:
1061 BUG();
1064 /* BUG: We assume that the sctp_packet_transmit()
1065 * call below will succeed all the time and add the
1066 * chunk to the transmitted list and restart the
1067 * timers.
1068 * It is possible that the call can fail under OOM
1069 * conditions.
1071 * Is this really a problem? Won't this behave
1072 * like a lost TSN?
1074 list_add_tail(&chunk->transmitted_list,
1075 &transport->transmitted);
1077 sctp_transport_reset_timers(transport);
1079 q->empty = 0;
1081 /* Only let one DATA chunk get bundled with a
1082 * COOKIE-ECHO chunk.
1084 if (packet->has_cookie_echo)
1085 goto sctp_flush_out;
1087 break;
1089 default:
1090 /* Do nothing. */
1091 break;
1094 sctp_flush_out:
1096 /* Before returning, examine all the transports touched in
1097 * this call. Right now, we bluntly force clear all the
1098 * transports. Things might change after we implement Nagle.
1099 * But such an examination is still required.
1101 * --xguo
1103 while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
1104 struct sctp_transport *t = list_entry(ltransport,
1105 struct sctp_transport,
1106 send_ready);
1107 packet = &t->packet;
1108 if (!sctp_packet_empty(packet))
1109 error = sctp_packet_transmit(packet);
1111 /* Clear the burst limited state, if any */
1112 sctp_transport_burst_reset(t);
1115 return error;
1118 /* Update unack_data based on the incoming SACK chunk */
1119 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1120 struct sctp_sackhdr *sack)
1122 sctp_sack_variable_t *frags;
1123 __u16 unack_data;
1124 int i;
1126 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1128 frags = sack->variable;
1129 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1130 unack_data -= ((ntohs(frags[i].gab.end) -
1131 ntohs(frags[i].gab.start) + 1));
1134 assoc->unack_data = unack_data;
1137 /* This is where we REALLY process a SACK.
1139 * Process the SACK against the outqueue. Mostly, this just frees
1140 * things off the transmitted queue.
1142 int sctp_outq_sack(struct sctp_outq *q, struct sctp_sackhdr *sack)
1144 struct sctp_association *asoc = q->asoc;
1145 struct sctp_transport *transport;
1146 struct sctp_chunk *tchunk = NULL;
1147 struct list_head *lchunk, *transport_list, *temp;
1148 sctp_sack_variable_t *frags = sack->variable;
1149 __u32 sack_ctsn, ctsn, tsn;
1150 __u32 highest_tsn, highest_new_tsn;
1151 __u32 sack_a_rwnd;
1152 unsigned int outstanding;
1153 struct sctp_transport *primary = asoc->peer.primary_path;
1154 int count_of_newacks = 0;
1155 int gap_ack_blocks;
1156 u8 accum_moved = 0;
1158 /* Grab the association's destination address list. */
1159 transport_list = &asoc->peer.transport_addr_list;
1161 sack_ctsn = ntohl(sack->cum_tsn_ack);
1162 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1164 * SFR-CACC algorithm:
1165 * On receipt of a SACK the sender SHOULD execute the
1166 * following statements.
1168 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1169 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1170 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1171 * all destinations.
1172 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1173 * is set the receiver of the SACK MUST take the following actions:
1175 * A) Initialize the cacc_saw_newack to 0 for all destination
1176 * addresses.
1178 * Only bother if changeover_active is set. Otherwise, this is
1179 * totally suboptimal to do on every SACK.
1181 if (primary->cacc.changeover_active) {
1182 u8 clear_cycling = 0;
1184 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1185 primary->cacc.changeover_active = 0;
1186 clear_cycling = 1;
1189 if (clear_cycling || gap_ack_blocks) {
1190 list_for_each_entry(transport, transport_list,
1191 transports) {
1192 if (clear_cycling)
1193 transport->cacc.cycling_changeover = 0;
1194 if (gap_ack_blocks)
1195 transport->cacc.cacc_saw_newack = 0;
1200 /* Get the highest TSN in the sack. */
1201 highest_tsn = sack_ctsn;
1202 if (gap_ack_blocks)
1203 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1205 if (TSN_lt(asoc->highest_sacked, highest_tsn))
1206 asoc->highest_sacked = highest_tsn;
1208 highest_new_tsn = sack_ctsn;
1210 /* Run through the retransmit queue. Credit bytes received
1211 * and free those chunks that we can.
1213 sctp_check_transmitted(q, &q->retransmit, NULL, sack, &highest_new_tsn);
1215 /* Run through the transmitted queue.
1216 * Credit bytes received and free those chunks which we can.
1218 * This is a MASSIVE candidate for optimization.
1220 list_for_each_entry(transport, transport_list, transports) {
1221 sctp_check_transmitted(q, &transport->transmitted,
1222 transport, sack, &highest_new_tsn);
1224 * SFR-CACC algorithm:
1225 * C) Let count_of_newacks be the number of
1226 * destinations for which cacc_saw_newack is set.
1228 if (transport->cacc.cacc_saw_newack)
1229 count_of_newacks ++;
1232 /* Move the Cumulative TSN Ack Point if appropriate. */
1233 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) {
1234 asoc->ctsn_ack_point = sack_ctsn;
1235 accum_moved = 1;
1238 if (gap_ack_blocks) {
1240 if (asoc->fast_recovery && accum_moved)
1241 highest_new_tsn = highest_tsn;
1243 list_for_each_entry(transport, transport_list, transports)
1244 sctp_mark_missing(q, &transport->transmitted, transport,
1245 highest_new_tsn, count_of_newacks);
1248 /* Update unack_data field in the assoc. */
1249 sctp_sack_update_unack_data(asoc, sack);
1251 ctsn = asoc->ctsn_ack_point;
1253 /* Throw away stuff rotting on the sack queue. */
1254 list_for_each_safe(lchunk, temp, &q->sacked) {
1255 tchunk = list_entry(lchunk, struct sctp_chunk,
1256 transmitted_list);
1257 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1258 if (TSN_lte(tsn, ctsn)) {
1259 list_del_init(&tchunk->transmitted_list);
1260 sctp_chunk_free(tchunk);
1264 /* ii) Set rwnd equal to the newly received a_rwnd minus the
1265 * number of bytes still outstanding after processing the
1266 * Cumulative TSN Ack and the Gap Ack Blocks.
1269 sack_a_rwnd = ntohl(sack->a_rwnd);
1270 outstanding = q->outstanding_bytes;
1272 if (outstanding < sack_a_rwnd)
1273 sack_a_rwnd -= outstanding;
1274 else
1275 sack_a_rwnd = 0;
1277 asoc->peer.rwnd = sack_a_rwnd;
1279 sctp_generate_fwdtsn(q, sack_ctsn);
1281 SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n",
1282 __func__, sack_ctsn);
1283 SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, "
1284 "%p is 0x%x. Adv peer ack point: 0x%x\n",
1285 __func__, asoc, ctsn, asoc->adv_peer_ack_point);
1287 /* See if all chunks are acked.
1288 * Make sure the empty queue handler will get run later.
1290 q->empty = (list_empty(&q->out_chunk_list) &&
1291 list_empty(&q->retransmit));
1292 if (!q->empty)
1293 goto finish;
1295 list_for_each_entry(transport, transport_list, transports) {
1296 q->empty = q->empty && list_empty(&transport->transmitted);
1297 if (!q->empty)
1298 goto finish;
1301 SCTP_DEBUG_PRINTK("sack queue is empty.\n");
1302 finish:
1303 return q->empty;
1306 /* Is the outqueue empty? */
1307 int sctp_outq_is_empty(const struct sctp_outq *q)
1309 return q->empty;
1312 /********************************************************************
1313 * 2nd Level Abstractions
1314 ********************************************************************/
1316 /* Go through a transport's transmitted list or the association's retransmit
1317 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1318 * The retransmit list will not have an associated transport.
1320 * I added coherent debug information output. --xguo
1322 * Instead of printing 'sacked' or 'kept' for each TSN on the
1323 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1324 * KEPT TSN6-TSN7, etc.
1326 static void sctp_check_transmitted(struct sctp_outq *q,
1327 struct list_head *transmitted_queue,
1328 struct sctp_transport *transport,
1329 struct sctp_sackhdr *sack,
1330 __u32 *highest_new_tsn_in_sack)
1332 struct list_head *lchunk;
1333 struct sctp_chunk *tchunk;
1334 struct list_head tlist;
1335 __u32 tsn;
1336 __u32 sack_ctsn;
1337 __u32 rtt;
1338 __u8 restart_timer = 0;
1339 int bytes_acked = 0;
1340 int migrate_bytes = 0;
1342 /* These state variables are for coherent debug output. --xguo */
1344 #if SCTP_DEBUG
1345 __u32 dbg_ack_tsn = 0; /* An ACKed TSN range starts here... */
1346 __u32 dbg_last_ack_tsn = 0; /* ...and finishes here. */
1347 __u32 dbg_kept_tsn = 0; /* An un-ACKed range starts here... */
1348 __u32 dbg_last_kept_tsn = 0; /* ...and finishes here. */
1350 /* 0 : The last TSN was ACKed.
1351 * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1352 * -1: We need to initialize.
1354 int dbg_prt_state = -1;
1355 #endif /* SCTP_DEBUG */
1357 sack_ctsn = ntohl(sack->cum_tsn_ack);
1359 INIT_LIST_HEAD(&tlist);
1361 /* The while loop will skip empty transmitted queues. */
1362 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1363 tchunk = list_entry(lchunk, struct sctp_chunk,
1364 transmitted_list);
1366 if (sctp_chunk_abandoned(tchunk)) {
1367 /* Move the chunk to abandoned list. */
1368 sctp_insert_list(&q->abandoned, lchunk);
1370 /* If this chunk has not been acked, stop
1371 * considering it as 'outstanding'.
1373 if (!tchunk->tsn_gap_acked) {
1374 if (tchunk->transport)
1375 tchunk->transport->flight_size -=
1376 sctp_data_size(tchunk);
1377 q->outstanding_bytes -= sctp_data_size(tchunk);
1379 continue;
1382 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1383 if (sctp_acked(sack, tsn)) {
1384 /* If this queue is the retransmit queue, the
1385 * retransmit timer has already reclaimed
1386 * the outstanding bytes for this chunk, so only
1387 * count bytes associated with a transport.
1389 if (transport) {
1390 /* If this chunk is being used for RTT
1391 * measurement, calculate the RTT and update
1392 * the RTO using this value.
1394 * 6.3.1 C5) Karn's algorithm: RTT measurements
1395 * MUST NOT be made using packets that were
1396 * retransmitted (and thus for which it is
1397 * ambiguous whether the reply was for the
1398 * first instance of the packet or a later
1399 * instance).
1401 if (!tchunk->tsn_gap_acked &&
1402 tchunk->rtt_in_progress) {
1403 tchunk->rtt_in_progress = 0;
1404 rtt = jiffies - tchunk->sent_at;
1405 sctp_transport_update_rto(transport,
1406 rtt);
1410 /* If the chunk hasn't been marked as ACKED,
1411 * mark it and account bytes_acked if the
1412 * chunk had a valid transport (it will not
1413 * have a transport if ASCONF had deleted it
1414 * while DATA was outstanding).
1416 if (!tchunk->tsn_gap_acked) {
1417 tchunk->tsn_gap_acked = 1;
1418 *highest_new_tsn_in_sack = tsn;
1419 bytes_acked += sctp_data_size(tchunk);
1420 if (!tchunk->transport)
1421 migrate_bytes += sctp_data_size(tchunk);
1424 if (TSN_lte(tsn, sack_ctsn)) {
1425 /* RFC 2960 6.3.2 Retransmission Timer Rules
1427 * R3) Whenever a SACK is received
1428 * that acknowledges the DATA chunk
1429 * with the earliest outstanding TSN
1430 * for that address, restart T3-rtx
1431 * timer for that address with its
1432 * current RTO.
1434 restart_timer = 1;
1436 if (!tchunk->tsn_gap_acked) {
1438 * SFR-CACC algorithm:
1439 * 2) If the SACK contains gap acks
1440 * and the flag CHANGEOVER_ACTIVE is
1441 * set the receiver of the SACK MUST
1442 * take the following action:
1444 * B) For each TSN t being acked that
1445 * has not been acked in any SACK so
1446 * far, set cacc_saw_newack to 1 for
1447 * the destination that the TSN was
1448 * sent to.
1450 if (transport &&
1451 sack->num_gap_ack_blocks &&
1452 q->asoc->peer.primary_path->cacc.
1453 changeover_active)
1454 transport->cacc.cacc_saw_newack
1455 = 1;
1458 list_add_tail(&tchunk->transmitted_list,
1459 &q->sacked);
1460 } else {
1461 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1462 * M2) Each time a SACK arrives reporting
1463 * 'Stray DATA chunk(s)' record the highest TSN
1464 * reported as newly acknowledged, call this
1465 * value 'HighestTSNinSack'. A newly
1466 * acknowledged DATA chunk is one not
1467 * previously acknowledged in a SACK.
1469 * When the SCTP sender of data receives a SACK
1470 * chunk that acknowledges, for the first time,
1471 * the receipt of a DATA chunk, all the still
1472 * unacknowledged DATA chunks whose TSN is
1473 * older than that newly acknowledged DATA
1474 * chunk, are qualified as 'Stray DATA chunks'.
1476 list_add_tail(lchunk, &tlist);
1479 #if SCTP_DEBUG
1480 switch (dbg_prt_state) {
1481 case 0: /* last TSN was ACKed */
1482 if (dbg_last_ack_tsn + 1 == tsn) {
1483 /* This TSN belongs to the
1484 * current ACK range.
1486 break;
1489 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1490 /* Display the end of the
1491 * current range.
1493 SCTP_DEBUG_PRINTK_CONT("-%08x",
1494 dbg_last_ack_tsn);
1497 /* Start a new range. */
1498 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn);
1499 dbg_ack_tsn = tsn;
1500 break;
1502 case 1: /* The last TSN was NOT ACKed. */
1503 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1504 /* Display the end of current range. */
1505 SCTP_DEBUG_PRINTK_CONT("-%08x",
1506 dbg_last_kept_tsn);
1509 SCTP_DEBUG_PRINTK_CONT("\n");
1511 /* FALL THROUGH... */
1512 default:
1513 /* This is the first-ever TSN we examined. */
1514 /* Start a new range of ACK-ed TSNs. */
1515 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn);
1516 dbg_prt_state = 0;
1517 dbg_ack_tsn = tsn;
1520 dbg_last_ack_tsn = tsn;
1521 #endif /* SCTP_DEBUG */
1523 } else {
1524 if (tchunk->tsn_gap_acked) {
1525 SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1526 "data TSN: 0x%x\n",
1527 __func__,
1528 tsn);
1529 tchunk->tsn_gap_acked = 0;
1531 if (tchunk->transport)
1532 bytes_acked -= sctp_data_size(tchunk);
1534 /* RFC 2960 6.3.2 Retransmission Timer Rules
1536 * R4) Whenever a SACK is received missing a
1537 * TSN that was previously acknowledged via a
1538 * Gap Ack Block, start T3-rtx for the
1539 * destination address to which the DATA
1540 * chunk was originally
1541 * transmitted if it is not already running.
1543 restart_timer = 1;
1546 list_add_tail(lchunk, &tlist);
1548 #if SCTP_DEBUG
1549 /* See the above comments on ACK-ed TSNs. */
1550 switch (dbg_prt_state) {
1551 case 1:
1552 if (dbg_last_kept_tsn + 1 == tsn)
1553 break;
1555 if (dbg_last_kept_tsn != dbg_kept_tsn)
1556 SCTP_DEBUG_PRINTK_CONT("-%08x",
1557 dbg_last_kept_tsn);
1559 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn);
1560 dbg_kept_tsn = tsn;
1561 break;
1563 case 0:
1564 if (dbg_last_ack_tsn != dbg_ack_tsn)
1565 SCTP_DEBUG_PRINTK_CONT("-%08x",
1566 dbg_last_ack_tsn);
1567 SCTP_DEBUG_PRINTK_CONT("\n");
1569 /* FALL THROUGH... */
1570 default:
1571 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn);
1572 dbg_prt_state = 1;
1573 dbg_kept_tsn = tsn;
1576 dbg_last_kept_tsn = tsn;
1577 #endif /* SCTP_DEBUG */
1581 #if SCTP_DEBUG
1582 /* Finish off the last range, displaying its ending TSN. */
1583 switch (dbg_prt_state) {
1584 case 0:
1585 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1586 SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_ack_tsn);
1587 } else {
1588 SCTP_DEBUG_PRINTK_CONT("\n");
1590 break;
1592 case 1:
1593 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1594 SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_kept_tsn);
1595 } else {
1596 SCTP_DEBUG_PRINTK_CONT("\n");
1599 #endif /* SCTP_DEBUG */
1600 if (transport) {
1601 if (bytes_acked) {
1602 struct sctp_association *asoc = transport->asoc;
1604 /* We may have counted DATA that was migrated
1605 * to this transport due to DEL-IP operation.
1606 * Subtract those bytes, since the were never
1607 * send on this transport and shouldn't be
1608 * credited to this transport.
1610 bytes_acked -= migrate_bytes;
1612 /* 8.2. When an outstanding TSN is acknowledged,
1613 * the endpoint shall clear the error counter of
1614 * the destination transport address to which the
1615 * DATA chunk was last sent.
1616 * The association's overall error counter is
1617 * also cleared.
1619 transport->error_count = 0;
1620 transport->asoc->overall_error_count = 0;
1623 * While in SHUTDOWN PENDING, we may have started
1624 * the T5 shutdown guard timer after reaching the
1625 * retransmission limit. Stop that timer as soon
1626 * as the receiver acknowledged any data.
1628 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING &&
1629 del_timer(&asoc->timers
1630 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]))
1631 sctp_association_put(asoc);
1633 /* Mark the destination transport address as
1634 * active if it is not so marked.
1636 if ((transport->state == SCTP_INACTIVE) ||
1637 (transport->state == SCTP_UNCONFIRMED)) {
1638 sctp_assoc_control_transport(
1639 transport->asoc,
1640 transport,
1641 SCTP_TRANSPORT_UP,
1642 SCTP_RECEIVED_SACK);
1645 sctp_transport_raise_cwnd(transport, sack_ctsn,
1646 bytes_acked);
1648 transport->flight_size -= bytes_acked;
1649 if (transport->flight_size == 0)
1650 transport->partial_bytes_acked = 0;
1651 q->outstanding_bytes -= bytes_acked + migrate_bytes;
1652 } else {
1653 /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1654 * When a sender is doing zero window probing, it
1655 * should not timeout the association if it continues
1656 * to receive new packets from the receiver. The
1657 * reason is that the receiver MAY keep its window
1658 * closed for an indefinite time.
1659 * A sender is doing zero window probing when the
1660 * receiver's advertised window is zero, and there is
1661 * only one data chunk in flight to the receiver.
1663 * Allow the association to timeout while in SHUTDOWN
1664 * PENDING or SHUTDOWN RECEIVED in case the receiver
1665 * stays in zero window mode forever.
1667 if (!q->asoc->peer.rwnd &&
1668 !list_empty(&tlist) &&
1669 (sack_ctsn+2 == q->asoc->next_tsn) &&
1670 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) {
1671 SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1672 "window probe: %u\n",
1673 __func__, sack_ctsn);
1674 q->asoc->overall_error_count = 0;
1675 transport->error_count = 0;
1679 /* RFC 2960 6.3.2 Retransmission Timer Rules
1681 * R2) Whenever all outstanding data sent to an address have
1682 * been acknowledged, turn off the T3-rtx timer of that
1683 * address.
1685 if (!transport->flight_size) {
1686 if (timer_pending(&transport->T3_rtx_timer) &&
1687 del_timer(&transport->T3_rtx_timer)) {
1688 sctp_transport_put(transport);
1690 } else if (restart_timer) {
1691 if (!mod_timer(&transport->T3_rtx_timer,
1692 jiffies + transport->rto))
1693 sctp_transport_hold(transport);
1697 list_splice(&tlist, transmitted_queue);
1700 /* Mark chunks as missing and consequently may get retransmitted. */
1701 static void sctp_mark_missing(struct sctp_outq *q,
1702 struct list_head *transmitted_queue,
1703 struct sctp_transport *transport,
1704 __u32 highest_new_tsn_in_sack,
1705 int count_of_newacks)
1707 struct sctp_chunk *chunk;
1708 __u32 tsn;
1709 char do_fast_retransmit = 0;
1710 struct sctp_association *asoc = q->asoc;
1711 struct sctp_transport *primary = asoc->peer.primary_path;
1713 list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1715 tsn = ntohl(chunk->subh.data_hdr->tsn);
1717 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1718 * 'Unacknowledged TSN's', if the TSN number of an
1719 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1720 * value, increment the 'TSN.Missing.Report' count on that
1721 * chunk if it has NOT been fast retransmitted or marked for
1722 * fast retransmit already.
1724 if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
1725 !chunk->tsn_gap_acked &&
1726 TSN_lt(tsn, highest_new_tsn_in_sack)) {
1728 /* SFR-CACC may require us to skip marking
1729 * this chunk as missing.
1731 if (!transport || !sctp_cacc_skip(primary,
1732 chunk->transport,
1733 count_of_newacks, tsn)) {
1734 chunk->tsn_missing_report++;
1736 SCTP_DEBUG_PRINTK(
1737 "%s: TSN 0x%x missing counter: %d\n",
1738 __func__, tsn,
1739 chunk->tsn_missing_report);
1743 * M4) If any DATA chunk is found to have a
1744 * 'TSN.Missing.Report'
1745 * value larger than or equal to 3, mark that chunk for
1746 * retransmission and start the fast retransmit procedure.
1749 if (chunk->tsn_missing_report >= 3) {
1750 chunk->fast_retransmit = SCTP_NEED_FRTX;
1751 do_fast_retransmit = 1;
1755 if (transport) {
1756 if (do_fast_retransmit)
1757 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1759 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1760 "ssthresh: %d, flight_size: %d, pba: %d\n",
1761 __func__, transport, transport->cwnd,
1762 transport->ssthresh, transport->flight_size,
1763 transport->partial_bytes_acked);
1767 /* Is the given TSN acked by this packet? */
1768 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1770 int i;
1771 sctp_sack_variable_t *frags;
1772 __u16 gap;
1773 __u32 ctsn = ntohl(sack->cum_tsn_ack);
1775 if (TSN_lte(tsn, ctsn))
1776 goto pass;
1778 /* 3.3.4 Selective Acknowledgement (SACK) (3):
1780 * Gap Ack Blocks:
1781 * These fields contain the Gap Ack Blocks. They are repeated
1782 * for each Gap Ack Block up to the number of Gap Ack Blocks
1783 * defined in the Number of Gap Ack Blocks field. All DATA
1784 * chunks with TSNs greater than or equal to (Cumulative TSN
1785 * Ack + Gap Ack Block Start) and less than or equal to
1786 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1787 * Block are assumed to have been received correctly.
1790 frags = sack->variable;
1791 gap = tsn - ctsn;
1792 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1793 if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1794 TSN_lte(gap, ntohs(frags[i].gab.end)))
1795 goto pass;
1798 return 0;
1799 pass:
1800 return 1;
1803 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1804 int nskips, __be16 stream)
1806 int i;
1808 for (i = 0; i < nskips; i++) {
1809 if (skiplist[i].stream == stream)
1810 return i;
1812 return i;
1815 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1816 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1818 struct sctp_association *asoc = q->asoc;
1819 struct sctp_chunk *ftsn_chunk = NULL;
1820 struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1821 int nskips = 0;
1822 int skip_pos = 0;
1823 __u32 tsn;
1824 struct sctp_chunk *chunk;
1825 struct list_head *lchunk, *temp;
1827 if (!asoc->peer.prsctp_capable)
1828 return;
1830 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1831 * received SACK.
1833 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1834 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1836 if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1837 asoc->adv_peer_ack_point = ctsn;
1839 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1840 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1841 * the chunk next in the out-queue space is marked as "abandoned" as
1842 * shown in the following example:
1844 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1845 * and the Advanced.Peer.Ack.Point is updated to this value:
1847 * out-queue at the end of ==> out-queue after Adv.Ack.Point
1848 * normal SACK processing local advancement
1849 * ... ...
1850 * Adv.Ack.Pt-> 102 acked 102 acked
1851 * 103 abandoned 103 abandoned
1852 * 104 abandoned Adv.Ack.P-> 104 abandoned
1853 * 105 105
1854 * 106 acked 106 acked
1855 * ... ...
1857 * In this example, the data sender successfully advanced the
1858 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1860 list_for_each_safe(lchunk, temp, &q->abandoned) {
1861 chunk = list_entry(lchunk, struct sctp_chunk,
1862 transmitted_list);
1863 tsn = ntohl(chunk->subh.data_hdr->tsn);
1865 /* Remove any chunks in the abandoned queue that are acked by
1866 * the ctsn.
1868 if (TSN_lte(tsn, ctsn)) {
1869 list_del_init(lchunk);
1870 sctp_chunk_free(chunk);
1871 } else {
1872 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1873 asoc->adv_peer_ack_point = tsn;
1874 if (chunk->chunk_hdr->flags &
1875 SCTP_DATA_UNORDERED)
1876 continue;
1877 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1878 nskips,
1879 chunk->subh.data_hdr->stream);
1880 ftsn_skip_arr[skip_pos].stream =
1881 chunk->subh.data_hdr->stream;
1882 ftsn_skip_arr[skip_pos].ssn =
1883 chunk->subh.data_hdr->ssn;
1884 if (skip_pos == nskips)
1885 nskips++;
1886 if (nskips == 10)
1887 break;
1888 } else
1889 break;
1893 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1894 * is greater than the Cumulative TSN ACK carried in the received
1895 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1896 * chunk containing the latest value of the
1897 * "Advanced.Peer.Ack.Point".
1899 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1900 * list each stream and sequence number in the forwarded TSN. This
1901 * information will enable the receiver to easily find any
1902 * stranded TSN's waiting on stream reorder queues. Each stream
1903 * SHOULD only be reported once; this means that if multiple
1904 * abandoned messages occur in the same stream then only the
1905 * highest abandoned stream sequence number is reported. If the
1906 * total size of the FORWARD TSN does NOT fit in a single MTU then
1907 * the sender of the FORWARD TSN SHOULD lower the
1908 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1909 * single MTU.
1911 if (asoc->adv_peer_ack_point > ctsn)
1912 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1913 nskips, &ftsn_skip_arr[0]);
1915 if (ftsn_chunk) {
1916 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1917 SCTP_INC_STATS(sock_net(asoc->base.sk), SCTP_MIB_OUTCTRLCHUNKS);