ARM: 7409/1: Do not call flush_cache_user_range with mmap_sem held
[linux/fpc-iii.git] / net / sctp / outqueue.c
blob1f2938fbf9b73c14c9e990a445af4470500444be
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 list_del_init(&chunk->list);
757 /* Pick the right transport to use. */
758 new_transport = chunk->transport;
760 if (!new_transport) {
762 * If we have a prior transport pointer, see if
763 * the destination address of the chunk
764 * matches the destination address of the
765 * current transport. If not a match, then
766 * try to look up the transport with a given
767 * destination address. We do this because
768 * after processing ASCONFs, we may have new
769 * transports created.
771 if (transport &&
772 sctp_cmp_addr_exact(&chunk->dest,
773 &transport->ipaddr))
774 new_transport = transport;
775 else
776 new_transport = sctp_assoc_lookup_paddr(asoc,
777 &chunk->dest);
779 /* if we still don't have a new transport, then
780 * use the current active path.
782 if (!new_transport)
783 new_transport = asoc->peer.active_path;
784 } else if ((new_transport->state == SCTP_INACTIVE) ||
785 (new_transport->state == SCTP_UNCONFIRMED)) {
786 /* If the chunk is Heartbeat or Heartbeat Ack,
787 * send it to chunk->transport, even if it's
788 * inactive.
790 * 3.3.6 Heartbeat Acknowledgement:
791 * ...
792 * A HEARTBEAT ACK is always sent to the source IP
793 * address of the IP datagram containing the
794 * HEARTBEAT chunk to which this ack is responding.
795 * ...
797 * ASCONF_ACKs also must be sent to the source.
799 if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
800 chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK &&
801 chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK)
802 new_transport = asoc->peer.active_path;
805 /* Are we switching transports?
806 * Take care of transport locks.
808 if (new_transport != transport) {
809 transport = new_transport;
810 if (list_empty(&transport->send_ready)) {
811 list_add_tail(&transport->send_ready,
812 &transport_list);
814 packet = &transport->packet;
815 sctp_packet_config(packet, vtag,
816 asoc->peer.ecn_capable);
819 switch (chunk->chunk_hdr->type) {
821 * 6.10 Bundling
822 * ...
823 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
824 * COMPLETE with any other chunks. [Send them immediately.]
826 case SCTP_CID_INIT:
827 case SCTP_CID_INIT_ACK:
828 case SCTP_CID_SHUTDOWN_COMPLETE:
829 sctp_packet_init(&singleton, transport, sport, dport);
830 sctp_packet_config(&singleton, vtag, 0);
831 sctp_packet_append_chunk(&singleton, chunk);
832 error = sctp_packet_transmit(&singleton);
833 if (error < 0)
834 return error;
835 break;
837 case SCTP_CID_ABORT:
838 if (sctp_test_T_bit(chunk)) {
839 packet->vtag = asoc->c.my_vtag;
841 /* The following chunks are "response" chunks, i.e.
842 * they are generated in response to something we
843 * received. If we are sending these, then we can
844 * send only 1 packet containing these chunks.
846 case SCTP_CID_HEARTBEAT_ACK:
847 case SCTP_CID_SHUTDOWN_ACK:
848 case SCTP_CID_COOKIE_ACK:
849 case SCTP_CID_COOKIE_ECHO:
850 case SCTP_CID_ERROR:
851 case SCTP_CID_ECN_CWR:
852 case SCTP_CID_ASCONF_ACK:
853 one_packet = 1;
854 /* Fall through */
856 case SCTP_CID_SACK:
857 case SCTP_CID_HEARTBEAT:
858 case SCTP_CID_SHUTDOWN:
859 case SCTP_CID_ECN_ECNE:
860 case SCTP_CID_ASCONF:
861 case SCTP_CID_FWD_TSN:
862 status = sctp_packet_transmit_chunk(packet, chunk,
863 one_packet);
864 if (status != SCTP_XMIT_OK) {
865 /* put the chunk back */
866 list_add(&chunk->list, &q->control_chunk_list);
867 } else if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN) {
868 /* PR-SCTP C5) If a FORWARD TSN is sent, the
869 * sender MUST assure that at least one T3-rtx
870 * timer is running.
872 sctp_transport_reset_timers(transport);
874 break;
876 default:
877 /* We built a chunk with an illegal type! */
878 BUG();
882 /* Is it OK to send data chunks? */
883 switch (asoc->state) {
884 case SCTP_STATE_COOKIE_ECHOED:
885 /* Only allow bundling when this packet has a COOKIE-ECHO
886 * chunk.
888 if (!packet || !packet->has_cookie_echo)
889 break;
891 /* fallthru */
892 case SCTP_STATE_ESTABLISHED:
893 case SCTP_STATE_SHUTDOWN_PENDING:
894 case SCTP_STATE_SHUTDOWN_RECEIVED:
896 * RFC 2960 6.1 Transmission of DATA Chunks
898 * C) When the time comes for the sender to transmit,
899 * before sending new DATA chunks, the sender MUST
900 * first transmit any outstanding DATA chunks which
901 * are marked for retransmission (limited by the
902 * current cwnd).
904 if (!list_empty(&q->retransmit)) {
905 if (transport == asoc->peer.retran_path)
906 goto retran;
908 /* Switch transports & prepare the packet. */
910 transport = asoc->peer.retran_path;
912 if (list_empty(&transport->send_ready)) {
913 list_add_tail(&transport->send_ready,
914 &transport_list);
917 packet = &transport->packet;
918 sctp_packet_config(packet, vtag,
919 asoc->peer.ecn_capable);
920 retran:
921 error = sctp_outq_flush_rtx(q, packet,
922 rtx_timeout, &start_timer);
924 if (start_timer)
925 sctp_transport_reset_timers(transport);
927 /* This can happen on COOKIE-ECHO resend. Only
928 * one chunk can get bundled with a COOKIE-ECHO.
930 if (packet->has_cookie_echo)
931 goto sctp_flush_out;
933 /* Don't send new data if there is still data
934 * waiting to retransmit.
936 if (!list_empty(&q->retransmit))
937 goto sctp_flush_out;
940 /* Apply Max.Burst limitation to the current transport in
941 * case it will be used for new data. We are going to
942 * rest it before we return, but we want to apply the limit
943 * to the currently queued data.
945 if (transport)
946 sctp_transport_burst_limited(transport);
948 /* Finally, transmit new packets. */
949 while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
950 /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
951 * stream identifier.
953 if (chunk->sinfo.sinfo_stream >=
954 asoc->c.sinit_num_ostreams) {
956 /* Mark as failed send. */
957 sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
958 sctp_chunk_free(chunk);
959 continue;
962 /* Has this chunk expired? */
963 if (sctp_chunk_abandoned(chunk)) {
964 sctp_chunk_fail(chunk, 0);
965 sctp_chunk_free(chunk);
966 continue;
969 /* If there is a specified transport, use it.
970 * Otherwise, we want to use the active path.
972 new_transport = chunk->transport;
973 if (!new_transport ||
974 ((new_transport->state == SCTP_INACTIVE) ||
975 (new_transport->state == SCTP_UNCONFIRMED)))
976 new_transport = asoc->peer.active_path;
978 /* Change packets if necessary. */
979 if (new_transport != transport) {
980 transport = new_transport;
982 /* Schedule to have this transport's
983 * packet flushed.
985 if (list_empty(&transport->send_ready)) {
986 list_add_tail(&transport->send_ready,
987 &transport_list);
990 packet = &transport->packet;
991 sctp_packet_config(packet, vtag,
992 asoc->peer.ecn_capable);
993 /* We've switched transports, so apply the
994 * Burst limit to the new transport.
996 sctp_transport_burst_limited(transport);
999 SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
1000 q, chunk,
1001 chunk && chunk->chunk_hdr ?
1002 sctp_cname(SCTP_ST_CHUNK(
1003 chunk->chunk_hdr->type))
1004 : "Illegal Chunk");
1006 SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
1007 "%p skb->users %d.\n",
1008 ntohl(chunk->subh.data_hdr->tsn),
1009 chunk->skb ?chunk->skb->head : NULL,
1010 chunk->skb ?
1011 atomic_read(&chunk->skb->users) : -1);
1013 /* Add the chunk to the packet. */
1014 status = sctp_packet_transmit_chunk(packet, chunk, 0);
1016 switch (status) {
1017 case SCTP_XMIT_PMTU_FULL:
1018 case SCTP_XMIT_RWND_FULL:
1019 case SCTP_XMIT_NAGLE_DELAY:
1020 /* We could not append this chunk, so put
1021 * the chunk back on the output queue.
1023 SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
1024 "not transmit TSN: 0x%x, status: %d\n",
1025 ntohl(chunk->subh.data_hdr->tsn),
1026 status);
1027 sctp_outq_head_data(q, chunk);
1028 goto sctp_flush_out;
1029 break;
1031 case SCTP_XMIT_OK:
1032 /* The sender is in the SHUTDOWN-PENDING state,
1033 * The sender MAY set the I-bit in the DATA
1034 * chunk header.
1036 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING)
1037 chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM;
1039 break;
1041 default:
1042 BUG();
1045 /* BUG: We assume that the sctp_packet_transmit()
1046 * call below will succeed all the time and add the
1047 * chunk to the transmitted list and restart the
1048 * timers.
1049 * It is possible that the call can fail under OOM
1050 * conditions.
1052 * Is this really a problem? Won't this behave
1053 * like a lost TSN?
1055 list_add_tail(&chunk->transmitted_list,
1056 &transport->transmitted);
1058 sctp_transport_reset_timers(transport);
1060 q->empty = 0;
1062 /* Only let one DATA chunk get bundled with a
1063 * COOKIE-ECHO chunk.
1065 if (packet->has_cookie_echo)
1066 goto sctp_flush_out;
1068 break;
1070 default:
1071 /* Do nothing. */
1072 break;
1075 sctp_flush_out:
1077 /* Before returning, examine all the transports touched in
1078 * this call. Right now, we bluntly force clear all the
1079 * transports. Things might change after we implement Nagle.
1080 * But such an examination is still required.
1082 * --xguo
1084 while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
1085 struct sctp_transport *t = list_entry(ltransport,
1086 struct sctp_transport,
1087 send_ready);
1088 packet = &t->packet;
1089 if (!sctp_packet_empty(packet))
1090 error = sctp_packet_transmit(packet);
1092 /* Clear the burst limited state, if any */
1093 sctp_transport_burst_reset(t);
1096 return error;
1099 /* Update unack_data based on the incoming SACK chunk */
1100 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1101 struct sctp_sackhdr *sack)
1103 sctp_sack_variable_t *frags;
1104 __u16 unack_data;
1105 int i;
1107 unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1109 frags = sack->variable;
1110 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1111 unack_data -= ((ntohs(frags[i].gab.end) -
1112 ntohs(frags[i].gab.start) + 1));
1115 assoc->unack_data = unack_data;
1118 /* This is where we REALLY process a SACK.
1120 * Process the SACK against the outqueue. Mostly, this just frees
1121 * things off the transmitted queue.
1123 int sctp_outq_sack(struct sctp_outq *q, struct sctp_sackhdr *sack)
1125 struct sctp_association *asoc = q->asoc;
1126 struct sctp_transport *transport;
1127 struct sctp_chunk *tchunk = NULL;
1128 struct list_head *lchunk, *transport_list, *temp;
1129 sctp_sack_variable_t *frags = sack->variable;
1130 __u32 sack_ctsn, ctsn, tsn;
1131 __u32 highest_tsn, highest_new_tsn;
1132 __u32 sack_a_rwnd;
1133 unsigned outstanding;
1134 struct sctp_transport *primary = asoc->peer.primary_path;
1135 int count_of_newacks = 0;
1136 int gap_ack_blocks;
1137 u8 accum_moved = 0;
1139 /* Grab the association's destination address list. */
1140 transport_list = &asoc->peer.transport_addr_list;
1142 sack_ctsn = ntohl(sack->cum_tsn_ack);
1143 gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1145 * SFR-CACC algorithm:
1146 * On receipt of a SACK the sender SHOULD execute the
1147 * following statements.
1149 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1150 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1151 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1152 * all destinations.
1153 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1154 * is set the receiver of the SACK MUST take the following actions:
1156 * A) Initialize the cacc_saw_newack to 0 for all destination
1157 * addresses.
1159 * Only bother if changeover_active is set. Otherwise, this is
1160 * totally suboptimal to do on every SACK.
1162 if (primary->cacc.changeover_active) {
1163 u8 clear_cycling = 0;
1165 if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1166 primary->cacc.changeover_active = 0;
1167 clear_cycling = 1;
1170 if (clear_cycling || gap_ack_blocks) {
1171 list_for_each_entry(transport, transport_list,
1172 transports) {
1173 if (clear_cycling)
1174 transport->cacc.cycling_changeover = 0;
1175 if (gap_ack_blocks)
1176 transport->cacc.cacc_saw_newack = 0;
1181 /* Get the highest TSN in the sack. */
1182 highest_tsn = sack_ctsn;
1183 if (gap_ack_blocks)
1184 highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1186 if (TSN_lt(asoc->highest_sacked, highest_tsn))
1187 asoc->highest_sacked = highest_tsn;
1189 highest_new_tsn = sack_ctsn;
1191 /* Run through the retransmit queue. Credit bytes received
1192 * and free those chunks that we can.
1194 sctp_check_transmitted(q, &q->retransmit, NULL, sack, &highest_new_tsn);
1196 /* Run through the transmitted queue.
1197 * Credit bytes received and free those chunks which we can.
1199 * This is a MASSIVE candidate for optimization.
1201 list_for_each_entry(transport, transport_list, transports) {
1202 sctp_check_transmitted(q, &transport->transmitted,
1203 transport, sack, &highest_new_tsn);
1205 * SFR-CACC algorithm:
1206 * C) Let count_of_newacks be the number of
1207 * destinations for which cacc_saw_newack is set.
1209 if (transport->cacc.cacc_saw_newack)
1210 count_of_newacks ++;
1213 /* Move the Cumulative TSN Ack Point if appropriate. */
1214 if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) {
1215 asoc->ctsn_ack_point = sack_ctsn;
1216 accum_moved = 1;
1219 if (gap_ack_blocks) {
1221 if (asoc->fast_recovery && accum_moved)
1222 highest_new_tsn = highest_tsn;
1224 list_for_each_entry(transport, transport_list, transports)
1225 sctp_mark_missing(q, &transport->transmitted, transport,
1226 highest_new_tsn, count_of_newacks);
1229 /* Update unack_data field in the assoc. */
1230 sctp_sack_update_unack_data(asoc, sack);
1232 ctsn = asoc->ctsn_ack_point;
1234 /* Throw away stuff rotting on the sack queue. */
1235 list_for_each_safe(lchunk, temp, &q->sacked) {
1236 tchunk = list_entry(lchunk, struct sctp_chunk,
1237 transmitted_list);
1238 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1239 if (TSN_lte(tsn, ctsn)) {
1240 list_del_init(&tchunk->transmitted_list);
1241 sctp_chunk_free(tchunk);
1245 /* ii) Set rwnd equal to the newly received a_rwnd minus the
1246 * number of bytes still outstanding after processing the
1247 * Cumulative TSN Ack and the Gap Ack Blocks.
1250 sack_a_rwnd = ntohl(sack->a_rwnd);
1251 outstanding = q->outstanding_bytes;
1253 if (outstanding < sack_a_rwnd)
1254 sack_a_rwnd -= outstanding;
1255 else
1256 sack_a_rwnd = 0;
1258 asoc->peer.rwnd = sack_a_rwnd;
1260 sctp_generate_fwdtsn(q, sack_ctsn);
1262 SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n",
1263 __func__, sack_ctsn);
1264 SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, "
1265 "%p is 0x%x. Adv peer ack point: 0x%x\n",
1266 __func__, asoc, ctsn, asoc->adv_peer_ack_point);
1268 /* See if all chunks are acked.
1269 * Make sure the empty queue handler will get run later.
1271 q->empty = (list_empty(&q->out_chunk_list) &&
1272 list_empty(&q->retransmit));
1273 if (!q->empty)
1274 goto finish;
1276 list_for_each_entry(transport, transport_list, transports) {
1277 q->empty = q->empty && list_empty(&transport->transmitted);
1278 if (!q->empty)
1279 goto finish;
1282 SCTP_DEBUG_PRINTK("sack queue is empty.\n");
1283 finish:
1284 return q->empty;
1287 /* Is the outqueue empty? */
1288 int sctp_outq_is_empty(const struct sctp_outq *q)
1290 return q->empty;
1293 /********************************************************************
1294 * 2nd Level Abstractions
1295 ********************************************************************/
1297 /* Go through a transport's transmitted list or the association's retransmit
1298 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1299 * The retransmit list will not have an associated transport.
1301 * I added coherent debug information output. --xguo
1303 * Instead of printing 'sacked' or 'kept' for each TSN on the
1304 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1305 * KEPT TSN6-TSN7, etc.
1307 static void sctp_check_transmitted(struct sctp_outq *q,
1308 struct list_head *transmitted_queue,
1309 struct sctp_transport *transport,
1310 struct sctp_sackhdr *sack,
1311 __u32 *highest_new_tsn_in_sack)
1313 struct list_head *lchunk;
1314 struct sctp_chunk *tchunk;
1315 struct list_head tlist;
1316 __u32 tsn;
1317 __u32 sack_ctsn;
1318 __u32 rtt;
1319 __u8 restart_timer = 0;
1320 int bytes_acked = 0;
1321 int migrate_bytes = 0;
1323 /* These state variables are for coherent debug output. --xguo */
1325 #if SCTP_DEBUG
1326 __u32 dbg_ack_tsn = 0; /* An ACKed TSN range starts here... */
1327 __u32 dbg_last_ack_tsn = 0; /* ...and finishes here. */
1328 __u32 dbg_kept_tsn = 0; /* An un-ACKed range starts here... */
1329 __u32 dbg_last_kept_tsn = 0; /* ...and finishes here. */
1331 /* 0 : The last TSN was ACKed.
1332 * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1333 * -1: We need to initialize.
1335 int dbg_prt_state = -1;
1336 #endif /* SCTP_DEBUG */
1338 sack_ctsn = ntohl(sack->cum_tsn_ack);
1340 INIT_LIST_HEAD(&tlist);
1342 /* The while loop will skip empty transmitted queues. */
1343 while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1344 tchunk = list_entry(lchunk, struct sctp_chunk,
1345 transmitted_list);
1347 if (sctp_chunk_abandoned(tchunk)) {
1348 /* Move the chunk to abandoned list. */
1349 sctp_insert_list(&q->abandoned, lchunk);
1351 /* If this chunk has not been acked, stop
1352 * considering it as 'outstanding'.
1354 if (!tchunk->tsn_gap_acked) {
1355 if (tchunk->transport)
1356 tchunk->transport->flight_size -=
1357 sctp_data_size(tchunk);
1358 q->outstanding_bytes -= sctp_data_size(tchunk);
1360 continue;
1363 tsn = ntohl(tchunk->subh.data_hdr->tsn);
1364 if (sctp_acked(sack, tsn)) {
1365 /* If this queue is the retransmit queue, the
1366 * retransmit timer has already reclaimed
1367 * the outstanding bytes for this chunk, so only
1368 * count bytes associated with a transport.
1370 if (transport) {
1371 /* If this chunk is being used for RTT
1372 * measurement, calculate the RTT and update
1373 * the RTO using this value.
1375 * 6.3.1 C5) Karn's algorithm: RTT measurements
1376 * MUST NOT be made using packets that were
1377 * retransmitted (and thus for which it is
1378 * ambiguous whether the reply was for the
1379 * first instance of the packet or a later
1380 * instance).
1382 if (!tchunk->tsn_gap_acked &&
1383 tchunk->rtt_in_progress) {
1384 tchunk->rtt_in_progress = 0;
1385 rtt = jiffies - tchunk->sent_at;
1386 sctp_transport_update_rto(transport,
1387 rtt);
1391 /* If the chunk hasn't been marked as ACKED,
1392 * mark it and account bytes_acked if the
1393 * chunk had a valid transport (it will not
1394 * have a transport if ASCONF had deleted it
1395 * while DATA was outstanding).
1397 if (!tchunk->tsn_gap_acked) {
1398 tchunk->tsn_gap_acked = 1;
1399 *highest_new_tsn_in_sack = tsn;
1400 bytes_acked += sctp_data_size(tchunk);
1401 if (!tchunk->transport)
1402 migrate_bytes += sctp_data_size(tchunk);
1405 if (TSN_lte(tsn, sack_ctsn)) {
1406 /* RFC 2960 6.3.2 Retransmission Timer Rules
1408 * R3) Whenever a SACK is received
1409 * that acknowledges the DATA chunk
1410 * with the earliest outstanding TSN
1411 * for that address, restart T3-rtx
1412 * timer for that address with its
1413 * current RTO.
1415 restart_timer = 1;
1417 if (!tchunk->tsn_gap_acked) {
1419 * SFR-CACC algorithm:
1420 * 2) If the SACK contains gap acks
1421 * and the flag CHANGEOVER_ACTIVE is
1422 * set the receiver of the SACK MUST
1423 * take the following action:
1425 * B) For each TSN t being acked that
1426 * has not been acked in any SACK so
1427 * far, set cacc_saw_newack to 1 for
1428 * the destination that the TSN was
1429 * sent to.
1431 if (transport &&
1432 sack->num_gap_ack_blocks &&
1433 q->asoc->peer.primary_path->cacc.
1434 changeover_active)
1435 transport->cacc.cacc_saw_newack
1436 = 1;
1439 list_add_tail(&tchunk->transmitted_list,
1440 &q->sacked);
1441 } else {
1442 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1443 * M2) Each time a SACK arrives reporting
1444 * 'Stray DATA chunk(s)' record the highest TSN
1445 * reported as newly acknowledged, call this
1446 * value 'HighestTSNinSack'. A newly
1447 * acknowledged DATA chunk is one not
1448 * previously acknowledged in a SACK.
1450 * When the SCTP sender of data receives a SACK
1451 * chunk that acknowledges, for the first time,
1452 * the receipt of a DATA chunk, all the still
1453 * unacknowledged DATA chunks whose TSN is
1454 * older than that newly acknowledged DATA
1455 * chunk, are qualified as 'Stray DATA chunks'.
1457 list_add_tail(lchunk, &tlist);
1460 #if SCTP_DEBUG
1461 switch (dbg_prt_state) {
1462 case 0: /* last TSN was ACKed */
1463 if (dbg_last_ack_tsn + 1 == tsn) {
1464 /* This TSN belongs to the
1465 * current ACK range.
1467 break;
1470 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1471 /* Display the end of the
1472 * current range.
1474 SCTP_DEBUG_PRINTK_CONT("-%08x",
1475 dbg_last_ack_tsn);
1478 /* Start a new range. */
1479 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn);
1480 dbg_ack_tsn = tsn;
1481 break;
1483 case 1: /* The last TSN was NOT ACKed. */
1484 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1485 /* Display the end of current range. */
1486 SCTP_DEBUG_PRINTK_CONT("-%08x",
1487 dbg_last_kept_tsn);
1490 SCTP_DEBUG_PRINTK_CONT("\n");
1492 /* FALL THROUGH... */
1493 default:
1494 /* This is the first-ever TSN we examined. */
1495 /* Start a new range of ACK-ed TSNs. */
1496 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn);
1497 dbg_prt_state = 0;
1498 dbg_ack_tsn = tsn;
1501 dbg_last_ack_tsn = tsn;
1502 #endif /* SCTP_DEBUG */
1504 } else {
1505 if (tchunk->tsn_gap_acked) {
1506 SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1507 "data TSN: 0x%x\n",
1508 __func__,
1509 tsn);
1510 tchunk->tsn_gap_acked = 0;
1512 if (tchunk->transport)
1513 bytes_acked -= sctp_data_size(tchunk);
1515 /* RFC 2960 6.3.2 Retransmission Timer Rules
1517 * R4) Whenever a SACK is received missing a
1518 * TSN that was previously acknowledged via a
1519 * Gap Ack Block, start T3-rtx for the
1520 * destination address to which the DATA
1521 * chunk was originally
1522 * transmitted if it is not already running.
1524 restart_timer = 1;
1527 list_add_tail(lchunk, &tlist);
1529 #if SCTP_DEBUG
1530 /* See the above comments on ACK-ed TSNs. */
1531 switch (dbg_prt_state) {
1532 case 1:
1533 if (dbg_last_kept_tsn + 1 == tsn)
1534 break;
1536 if (dbg_last_kept_tsn != dbg_kept_tsn)
1537 SCTP_DEBUG_PRINTK_CONT("-%08x",
1538 dbg_last_kept_tsn);
1540 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn);
1541 dbg_kept_tsn = tsn;
1542 break;
1544 case 0:
1545 if (dbg_last_ack_tsn != dbg_ack_tsn)
1546 SCTP_DEBUG_PRINTK_CONT("-%08x",
1547 dbg_last_ack_tsn);
1548 SCTP_DEBUG_PRINTK_CONT("\n");
1550 /* FALL THROUGH... */
1551 default:
1552 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn);
1553 dbg_prt_state = 1;
1554 dbg_kept_tsn = tsn;
1557 dbg_last_kept_tsn = tsn;
1558 #endif /* SCTP_DEBUG */
1562 #if SCTP_DEBUG
1563 /* Finish off the last range, displaying its ending TSN. */
1564 switch (dbg_prt_state) {
1565 case 0:
1566 if (dbg_last_ack_tsn != dbg_ack_tsn) {
1567 SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_ack_tsn);
1568 } else {
1569 SCTP_DEBUG_PRINTK_CONT("\n");
1571 break;
1573 case 1:
1574 if (dbg_last_kept_tsn != dbg_kept_tsn) {
1575 SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_kept_tsn);
1576 } else {
1577 SCTP_DEBUG_PRINTK_CONT("\n");
1580 #endif /* SCTP_DEBUG */
1581 if (transport) {
1582 if (bytes_acked) {
1583 struct sctp_association *asoc = transport->asoc;
1585 /* We may have counted DATA that was migrated
1586 * to this transport due to DEL-IP operation.
1587 * Subtract those bytes, since the were never
1588 * send on this transport and shouldn't be
1589 * credited to this transport.
1591 bytes_acked -= migrate_bytes;
1593 /* 8.2. When an outstanding TSN is acknowledged,
1594 * the endpoint shall clear the error counter of
1595 * the destination transport address to which the
1596 * DATA chunk was last sent.
1597 * The association's overall error counter is
1598 * also cleared.
1600 transport->error_count = 0;
1601 transport->asoc->overall_error_count = 0;
1604 * While in SHUTDOWN PENDING, we may have started
1605 * the T5 shutdown guard timer after reaching the
1606 * retransmission limit. Stop that timer as soon
1607 * as the receiver acknowledged any data.
1609 if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING &&
1610 del_timer(&asoc->timers
1611 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]))
1612 sctp_association_put(asoc);
1614 /* Mark the destination transport address as
1615 * active if it is not so marked.
1617 if ((transport->state == SCTP_INACTIVE) ||
1618 (transport->state == SCTP_UNCONFIRMED)) {
1619 sctp_assoc_control_transport(
1620 transport->asoc,
1621 transport,
1622 SCTP_TRANSPORT_UP,
1623 SCTP_RECEIVED_SACK);
1626 sctp_transport_raise_cwnd(transport, sack_ctsn,
1627 bytes_acked);
1629 transport->flight_size -= bytes_acked;
1630 if (transport->flight_size == 0)
1631 transport->partial_bytes_acked = 0;
1632 q->outstanding_bytes -= bytes_acked + migrate_bytes;
1633 } else {
1634 /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1635 * When a sender is doing zero window probing, it
1636 * should not timeout the association if it continues
1637 * to receive new packets from the receiver. The
1638 * reason is that the receiver MAY keep its window
1639 * closed for an indefinite time.
1640 * A sender is doing zero window probing when the
1641 * receiver's advertised window is zero, and there is
1642 * only one data chunk in flight to the receiver.
1644 * Allow the association to timeout while in SHUTDOWN
1645 * PENDING or SHUTDOWN RECEIVED in case the receiver
1646 * stays in zero window mode forever.
1648 if (!q->asoc->peer.rwnd &&
1649 !list_empty(&tlist) &&
1650 (sack_ctsn+2 == q->asoc->next_tsn) &&
1651 q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) {
1652 SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1653 "window probe: %u\n",
1654 __func__, sack_ctsn);
1655 q->asoc->overall_error_count = 0;
1656 transport->error_count = 0;
1660 /* RFC 2960 6.3.2 Retransmission Timer Rules
1662 * R2) Whenever all outstanding data sent to an address have
1663 * been acknowledged, turn off the T3-rtx timer of that
1664 * address.
1666 if (!transport->flight_size) {
1667 if (timer_pending(&transport->T3_rtx_timer) &&
1668 del_timer(&transport->T3_rtx_timer)) {
1669 sctp_transport_put(transport);
1671 } else if (restart_timer) {
1672 if (!mod_timer(&transport->T3_rtx_timer,
1673 jiffies + transport->rto))
1674 sctp_transport_hold(transport);
1678 list_splice(&tlist, transmitted_queue);
1681 /* Mark chunks as missing and consequently may get retransmitted. */
1682 static void sctp_mark_missing(struct sctp_outq *q,
1683 struct list_head *transmitted_queue,
1684 struct sctp_transport *transport,
1685 __u32 highest_new_tsn_in_sack,
1686 int count_of_newacks)
1688 struct sctp_chunk *chunk;
1689 __u32 tsn;
1690 char do_fast_retransmit = 0;
1691 struct sctp_association *asoc = q->asoc;
1692 struct sctp_transport *primary = asoc->peer.primary_path;
1694 list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1696 tsn = ntohl(chunk->subh.data_hdr->tsn);
1698 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1699 * 'Unacknowledged TSN's', if the TSN number of an
1700 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1701 * value, increment the 'TSN.Missing.Report' count on that
1702 * chunk if it has NOT been fast retransmitted or marked for
1703 * fast retransmit already.
1705 if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
1706 !chunk->tsn_gap_acked &&
1707 TSN_lt(tsn, highest_new_tsn_in_sack)) {
1709 /* SFR-CACC may require us to skip marking
1710 * this chunk as missing.
1712 if (!transport || !sctp_cacc_skip(primary,
1713 chunk->transport,
1714 count_of_newacks, tsn)) {
1715 chunk->tsn_missing_report++;
1717 SCTP_DEBUG_PRINTK(
1718 "%s: TSN 0x%x missing counter: %d\n",
1719 __func__, tsn,
1720 chunk->tsn_missing_report);
1724 * M4) If any DATA chunk is found to have a
1725 * 'TSN.Missing.Report'
1726 * value larger than or equal to 3, mark that chunk for
1727 * retransmission and start the fast retransmit procedure.
1730 if (chunk->tsn_missing_report >= 3) {
1731 chunk->fast_retransmit = SCTP_NEED_FRTX;
1732 do_fast_retransmit = 1;
1736 if (transport) {
1737 if (do_fast_retransmit)
1738 sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1740 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1741 "ssthresh: %d, flight_size: %d, pba: %d\n",
1742 __func__, transport, transport->cwnd,
1743 transport->ssthresh, transport->flight_size,
1744 transport->partial_bytes_acked);
1748 /* Is the given TSN acked by this packet? */
1749 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1751 int i;
1752 sctp_sack_variable_t *frags;
1753 __u16 gap;
1754 __u32 ctsn = ntohl(sack->cum_tsn_ack);
1756 if (TSN_lte(tsn, ctsn))
1757 goto pass;
1759 /* 3.3.4 Selective Acknowledgement (SACK) (3):
1761 * Gap Ack Blocks:
1762 * These fields contain the Gap Ack Blocks. They are repeated
1763 * for each Gap Ack Block up to the number of Gap Ack Blocks
1764 * defined in the Number of Gap Ack Blocks field. All DATA
1765 * chunks with TSNs greater than or equal to (Cumulative TSN
1766 * Ack + Gap Ack Block Start) and less than or equal to
1767 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1768 * Block are assumed to have been received correctly.
1771 frags = sack->variable;
1772 gap = tsn - ctsn;
1773 for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1774 if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1775 TSN_lte(gap, ntohs(frags[i].gab.end)))
1776 goto pass;
1779 return 0;
1780 pass:
1781 return 1;
1784 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1785 int nskips, __be16 stream)
1787 int i;
1789 for (i = 0; i < nskips; i++) {
1790 if (skiplist[i].stream == stream)
1791 return i;
1793 return i;
1796 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1797 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1799 struct sctp_association *asoc = q->asoc;
1800 struct sctp_chunk *ftsn_chunk = NULL;
1801 struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1802 int nskips = 0;
1803 int skip_pos = 0;
1804 __u32 tsn;
1805 struct sctp_chunk *chunk;
1806 struct list_head *lchunk, *temp;
1808 if (!asoc->peer.prsctp_capable)
1809 return;
1811 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1812 * received SACK.
1814 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1815 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1817 if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1818 asoc->adv_peer_ack_point = ctsn;
1820 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1821 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1822 * the chunk next in the out-queue space is marked as "abandoned" as
1823 * shown in the following example:
1825 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1826 * and the Advanced.Peer.Ack.Point is updated to this value:
1828 * out-queue at the end of ==> out-queue after Adv.Ack.Point
1829 * normal SACK processing local advancement
1830 * ... ...
1831 * Adv.Ack.Pt-> 102 acked 102 acked
1832 * 103 abandoned 103 abandoned
1833 * 104 abandoned Adv.Ack.P-> 104 abandoned
1834 * 105 105
1835 * 106 acked 106 acked
1836 * ... ...
1838 * In this example, the data sender successfully advanced the
1839 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1841 list_for_each_safe(lchunk, temp, &q->abandoned) {
1842 chunk = list_entry(lchunk, struct sctp_chunk,
1843 transmitted_list);
1844 tsn = ntohl(chunk->subh.data_hdr->tsn);
1846 /* Remove any chunks in the abandoned queue that are acked by
1847 * the ctsn.
1849 if (TSN_lte(tsn, ctsn)) {
1850 list_del_init(lchunk);
1851 sctp_chunk_free(chunk);
1852 } else {
1853 if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1854 asoc->adv_peer_ack_point = tsn;
1855 if (chunk->chunk_hdr->flags &
1856 SCTP_DATA_UNORDERED)
1857 continue;
1858 skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1859 nskips,
1860 chunk->subh.data_hdr->stream);
1861 ftsn_skip_arr[skip_pos].stream =
1862 chunk->subh.data_hdr->stream;
1863 ftsn_skip_arr[skip_pos].ssn =
1864 chunk->subh.data_hdr->ssn;
1865 if (skip_pos == nskips)
1866 nskips++;
1867 if (nskips == 10)
1868 break;
1869 } else
1870 break;
1874 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1875 * is greater than the Cumulative TSN ACK carried in the received
1876 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1877 * chunk containing the latest value of the
1878 * "Advanced.Peer.Ack.Point".
1880 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1881 * list each stream and sequence number in the forwarded TSN. This
1882 * information will enable the receiver to easily find any
1883 * stranded TSN's waiting on stream reorder queues. Each stream
1884 * SHOULD only be reported once; this means that if multiple
1885 * abandoned messages occur in the same stream then only the
1886 * highest abandoned stream sequence number is reported. If the
1887 * total size of the FORWARD TSN does NOT fit in a single MTU then
1888 * the sender of the FORWARD TSN SHOULD lower the
1889 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1890 * single MTU.
1892 if (asoc->adv_peer_ack_point > ctsn)
1893 ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1894 nskips, &ftsn_skip_arr[0]);
1896 if (ftsn_chunk) {
1897 list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1898 SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);