blob: be4d63d5a5cc050d782a2355a3eb1b9ae61bda07 [file] [log] [blame]
Vlad Yasevich60c778b2008-01-11 09:57:09 -05001/* SCTP kernel implementation
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Copyright (c) 1999-2000 Cisco, Inc.
3 * Copyright (c) 1999-2001 Motorola, Inc.
4 * Copyright (c) 2001-2003 International Business Machines Corp.
5 * Copyright (c) 2001 Intel Corp.
6 * Copyright (c) 2001 La Monte H.P. Yarroll
7 *
Vlad Yasevich60c778b2008-01-11 09:57:09 -05008 * This file is part of the SCTP kernel implementation
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 *
10 * This module provides the abstraction for an SCTP tranport representing
11 * a remote transport address. For local transport addresses, we just use
12 * union sctp_addr.
13 *
Vlad Yasevich60c778b2008-01-11 09:57:09 -050014 * This SCTP implementation is free software;
Linus Torvalds1da177e2005-04-16 15:20:36 -070015 * you can redistribute it and/or modify it under the terms of
16 * the GNU General Public License as published by
17 * the Free Software Foundation; either version 2, or (at your option)
18 * any later version.
19 *
Vlad Yasevich60c778b2008-01-11 09:57:09 -050020 * This SCTP implementation is distributed in the hope that it
Linus Torvalds1da177e2005-04-16 15:20:36 -070021 * will be useful, but WITHOUT ANY WARRANTY; without even the implied
22 * ************************
23 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
24 * See the GNU General Public License for more details.
25 *
26 * You should have received a copy of the GNU General Public License
27 * along with GNU CC; see the file COPYING. If not, write to
28 * the Free Software Foundation, 59 Temple Place - Suite 330,
29 * Boston, MA 02111-1307, USA.
30 *
31 * Please send any bug reports or fixes you make to the
32 * email address(es):
33 * lksctp developers <lksctp-developers@lists.sourceforge.net>
34 *
35 * Or submit a bug report through the following website:
36 * http://www.sf.net/projects/lksctp
37 *
38 * Written or modified by:
39 * La Monte H.P. Yarroll <piggy@acm.org>
40 * Karl Knutson <karl@athena.chicago.il.us>
41 * Jon Grimm <jgrimm@us.ibm.com>
42 * Xingang Guo <xingang.guo@intel.com>
43 * Hui Huang <hui.huang@nokia.com>
44 * Sridhar Samudrala <sri@us.ibm.com>
45 * Ardelle Fan <ardelle.fan@intel.com>
46 *
47 * Any bugs reported given to us we will try to fix... any fixes shared will
48 * be incorporated into the next SCTP release.
49 */
50
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090051#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/types.h>
Sridhar Samudralaad8fec12006-07-21 14:48:50 -070053#include <linux/random.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#include <net/sctp/sctp.h>
55#include <net/sctp/sm.h>
56
57/* 1st Level Abstractions. */
58
59/* Initialize a new transport from provided memory. */
60static struct sctp_transport *sctp_transport_init(struct sctp_transport *peer,
61 const union sctp_addr *addr,
Al Virodd0fc662005-10-07 07:46:04 +010062 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -070063{
64 /* Copy in the address. */
65 peer->ipaddr = *addr;
66 peer->af_specific = sctp_get_af_specific(addr->sa.sa_family);
67 peer->asoc = NULL;
68
69 peer->dst = NULL;
70 memset(&peer->saddr, 0, sizeof(union sctp_addr));
71
72 /* From 6.3.1 RTO Calculation:
73 *
74 * C1) Until an RTT measurement has been made for a packet sent to the
75 * given destination transport address, set RTO to the protocol
76 * parameter 'RTO.Initial'.
77 */
Andrei Pelinescu-Onciul5fdd4ba2009-11-29 00:14:02 -080078 peer->rto = msecs_to_jiffies(sctp_rto_initial);
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 peer->rtt = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 peer->rttvar = 0;
81 peer->srtt = 0;
82 peer->rto_pending = 0;
Vlad Yasevichfaee47c2009-02-13 08:33:43 +000083 peer->hb_sent = 0;
Vlad Yasevicha6465232008-06-04 12:38:43 -070084 peer->fast_recovery = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
86 peer->last_time_heard = jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 peer->last_time_ecne_reduced = jiffies;
88
Frank Filz3f7a87d2005-06-20 13:14:57 -070089 peer->init_sent_count = 0;
90
Frank Filz52ccb8e2005-12-22 11:36:46 -080091 peer->param_flags = SPP_HB_DISABLE |
92 SPP_PMTUD_ENABLE |
93 SPP_SACKDELAY_ENABLE;
94 peer->hbinterval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
96 /* Initialize the default path max_retrans. */
Frank Filz52ccb8e2005-12-22 11:36:46 -080097 peer->pathmaxrxt = sctp_max_retrans_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 peer->error_count = 0;
99
100 INIT_LIST_HEAD(&peer->transmitted);
101 INIT_LIST_HEAD(&peer->send_ready);
102 INIT_LIST_HEAD(&peer->transports);
103
Florian Westphal6d0ccba2008-07-18 23:04:39 -0700104 peer->T3_rtx_timer.expires = 0;
105 peer->hb_timer.expires = 0;
106
Pavel Emelyanovb24b8a22008-01-23 21:20:07 -0800107 setup_timer(&peer->T3_rtx_timer, sctp_generate_t3_rtx_event,
108 (unsigned long)peer);
109 setup_timer(&peer->hb_timer, sctp_generate_heartbeat_event,
110 (unsigned long)peer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
Sridhar Samudralaad8fec12006-07-21 14:48:50 -0700112 /* Initialize the 64-bit random nonce sent with heartbeat. */
113 get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce));
114
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 atomic_set(&peer->refcnt, 1);
116 peer->dead = 0;
117
118 peer->malloced = 0;
119
120 /* Initialize the state information for SFR-CACC */
121 peer->cacc.changeover_active = 0;
122 peer->cacc.cycling_changeover = 0;
123 peer->cacc.next_tsn_at_change = 0;
124 peer->cacc.cacc_saw_newack = 0;
125
126 return peer;
127}
128
129/* Allocate and initialize a new transport. */
Alexey Dobriyan3182cd82005-07-11 20:57:47 -0700130struct sctp_transport *sctp_transport_new(const union sctp_addr *addr,
Al Virodd0fc662005-10-07 07:46:04 +0100131 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132{
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900133 struct sctp_transport *transport;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900135 transport = t_new(struct sctp_transport, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 if (!transport)
137 goto fail;
138
139 if (!sctp_transport_init(transport, addr, gfp))
140 goto fail_init;
141
142 transport->malloced = 1;
143 SCTP_DBG_OBJCNT_INC(transport);
144
145 return transport;
146
147fail_init:
148 kfree(transport);
149
150fail:
151 return NULL;
152}
153
154/* This transport is no longer needed. Free up if possible, or
155 * delay until it last reference count.
156 */
157void sctp_transport_free(struct sctp_transport *transport)
158{
159 transport->dead = 1;
160
161 /* Try to delete the heartbeat timer. */
162 if (del_timer(&transport->hb_timer))
163 sctp_transport_put(transport);
164
165 /* Delete the T3_rtx timer if it's active.
166 * There is no point in not doing this now and letting
167 * structure hang around in memory since we know
168 * the tranport is going away.
169 */
170 if (timer_pending(&transport->T3_rtx_timer) &&
171 del_timer(&transport->T3_rtx_timer))
172 sctp_transport_put(transport);
173
174
175 sctp_transport_put(transport);
176}
177
178/* Destroy the transport data structure.
179 * Assumes there are no more users of this structure.
180 */
181static void sctp_transport_destroy(struct sctp_transport *transport)
182{
183 SCTP_ASSERT(transport->dead, "Transport is not dead", return);
184
185 if (transport->asoc)
186 sctp_association_put(transport->asoc);
187
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900188 sctp_packet_free(&transport->packet);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
190 dst_release(transport->dst);
191 kfree(transport);
192 SCTP_DBG_OBJCNT_DEC(transport);
193}
194
195/* Start T3_rtx timer if it is not already running and update the heartbeat
196 * timer. This routine is called every time a DATA chunk is sent.
197 */
Vlad Yasevich62aeaff2008-06-04 12:39:11 -0700198void sctp_transport_reset_timers(struct sctp_transport *transport, int force)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199{
200 /* RFC 2960 6.3.2 Retransmission Timer Rules
201 *
202 * R1) Every time a DATA chunk is sent to any address(including a
203 * retransmission), if the T3-rtx timer of that address is not running
204 * start it running so that it will expire after the RTO of that
205 * address.
206 */
207
Vlad Yasevich62aeaff2008-06-04 12:39:11 -0700208 if (force || !timer_pending(&transport->T3_rtx_timer))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 if (!mod_timer(&transport->T3_rtx_timer,
210 jiffies + transport->rto))
211 sctp_transport_hold(transport);
212
213 /* When a data chunk is sent, reset the heartbeat interval. */
214 if (!mod_timer(&transport->hb_timer,
215 sctp_transport_timeout(transport)))
216 sctp_transport_hold(transport);
217}
218
219/* This transport has been assigned to an association.
220 * Initialize fields from the association or from the sock itself.
221 * Register the reference count in the association.
222 */
223void sctp_transport_set_owner(struct sctp_transport *transport,
224 struct sctp_association *asoc)
225{
226 transport->asoc = asoc;
227 sctp_association_hold(asoc);
228}
229
230/* Initialize the pmtu of a transport. */
231void sctp_transport_pmtu(struct sctp_transport *transport)
232{
233 struct dst_entry *dst;
234
235 dst = transport->af_specific->get_dst(NULL, &transport->ipaddr, NULL);
236
237 if (dst) {
Frank Filz52ccb8e2005-12-22 11:36:46 -0800238 transport->pathmtu = dst_mtu(dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 dst_release(dst);
240 } else
Frank Filz52ccb8e2005-12-22 11:36:46 -0800241 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242}
243
Vlad Yasevichc910b472007-06-07 13:47:03 -0400244/* this is a complete rip-off from __sk_dst_check
245 * the cookie is always 0 since this is how it's used in the
246 * pmtu code
247 */
248static struct dst_entry *sctp_transport_dst_check(struct sctp_transport *t)
249{
250 struct dst_entry *dst = t->dst;
251
252 if (dst && dst->obsolete && dst->ops->check(dst, 0) == NULL) {
253 dst_release(t->dst);
254 t->dst = NULL;
255 return NULL;
256 }
257
258 return dst;
259}
260
261void sctp_transport_update_pmtu(struct sctp_transport *t, u32 pmtu)
262{
263 struct dst_entry *dst;
264
265 if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) {
266 printk(KERN_WARNING "%s: Reported pmtu %d too low, "
267 "using default minimum of %d\n",
Harvey Harrison0dc47872008-03-05 20:47:47 -0800268 __func__, pmtu,
Vlad Yasevichc910b472007-06-07 13:47:03 -0400269 SCTP_DEFAULT_MINSEGMENT);
270 /* Use default minimum segment size and disable
271 * pmtu discovery on this transport.
272 */
273 t->pathmtu = SCTP_DEFAULT_MINSEGMENT;
Vlad Yasevichc910b472007-06-07 13:47:03 -0400274 } else {
275 t->pathmtu = pmtu;
276 }
277
278 dst = sctp_transport_dst_check(t);
279 if (dst)
280 dst->ops->update_pmtu(dst, pmtu);
281}
282
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283/* Caches the dst entry and source address for a transport's destination
284 * address.
285 */
286void sctp_transport_route(struct sctp_transport *transport,
287 union sctp_addr *saddr, struct sctp_sock *opt)
288{
289 struct sctp_association *asoc = transport->asoc;
290 struct sctp_af *af = transport->af_specific;
291 union sctp_addr *daddr = &transport->ipaddr;
292 struct dst_entry *dst;
293
294 dst = af->get_dst(asoc, daddr, saddr);
295
296 if (saddr)
297 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
298 else
YOSHIFUJI Hideakie5117102008-05-29 19:55:05 +0900299 af->get_saddr(opt, asoc, dst, daddr, &transport->saddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300
301 transport->dst = dst;
Frank Filz52ccb8e2005-12-22 11:36:46 -0800302 if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) {
303 return;
304 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 if (dst) {
Frank Filz52ccb8e2005-12-22 11:36:46 -0800306 transport->pathmtu = dst_mtu(dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307
308 /* Initialize sk->sk_rcv_saddr, if the transport is the
309 * association's active path for getsockname().
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900310 */
Vlad Yasevicha78102e2009-11-11 11:54:37 +0000311 if (asoc && (!asoc->peer.primary_path ||
312 (transport == asoc->peer.active_path)))
Neil Hormanbf031ff2005-12-02 20:32:29 -0800313 opt->pf->af->to_sk_saddr(&transport->saddr,
314 asoc->base.sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 } else
Frank Filz52ccb8e2005-12-22 11:36:46 -0800316 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317}
318
319/* Hold a reference to a transport. */
320void sctp_transport_hold(struct sctp_transport *transport)
321{
322 atomic_inc(&transport->refcnt);
323}
324
325/* Release a reference to a transport and clean up
326 * if there are no more references.
327 */
328void sctp_transport_put(struct sctp_transport *transport)
329{
330 if (atomic_dec_and_test(&transport->refcnt))
331 sctp_transport_destroy(transport);
332}
333
334/* Update transport's RTO based on the newly calculated RTT. */
335void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
336{
337 /* Check for valid transport. */
338 SCTP_ASSERT(tp, "NULL transport", return);
339
340 /* We should not be doing any RTO updates unless rto_pending is set. */
341 SCTP_ASSERT(tp->rto_pending, "rto_pending not set", return);
342
343 if (tp->rttvar || tp->srtt) {
344 /* 6.3.1 C3) When a new RTT measurement R' is made, set
345 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
346 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
347 */
348
349 /* Note: The above algorithm has been rewritten to
350 * express rto_beta and rto_alpha as inverse powers
351 * of two.
352 * For example, assuming the default value of RTO.Alpha of
353 * 1/8, rto_alpha would be expressed as 3.
354 */
355 tp->rttvar = tp->rttvar - (tp->rttvar >> sctp_rto_beta)
356 + ((abs(tp->srtt - rtt)) >> sctp_rto_beta);
357 tp->srtt = tp->srtt - (tp->srtt >> sctp_rto_alpha)
358 + (rtt >> sctp_rto_alpha);
359 } else {
360 /* 6.3.1 C2) When the first RTT measurement R is made, set
361 * SRTT <- R, RTTVAR <- R/2.
362 */
363 tp->srtt = rtt;
364 tp->rttvar = rtt >> 1;
365 }
366
367 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
368 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
369 */
370 if (tp->rttvar == 0)
371 tp->rttvar = SCTP_CLOCK_GRANULARITY;
372
373 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
374 tp->rto = tp->srtt + (tp->rttvar << 2);
375
376 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
377 * seconds then it is rounded up to RTO.Min seconds.
378 */
379 if (tp->rto < tp->asoc->rto_min)
380 tp->rto = tp->asoc->rto_min;
381
382 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
383 * at least RTO.max seconds.
384 */
385 if (tp->rto > tp->asoc->rto_max)
386 tp->rto = tp->asoc->rto_max;
387
388 tp->rtt = rtt;
389
390 /* Reset rto_pending so that a new RTT measurement is started when a
391 * new data chunk is sent.
392 */
393 tp->rto_pending = 0;
394
395 SCTP_DEBUG_PRINTK("%s: transport: %p, rtt: %d, srtt: %d "
Harvey Harrison0dc47872008-03-05 20:47:47 -0800396 "rttvar: %d, rto: %ld\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 tp, rtt, tp->srtt, tp->rttvar, tp->rto);
398}
399
400/* This routine updates the transport's cwnd and partial_bytes_acked
401 * parameters based on the bytes acked in the received SACK.
402 */
403void sctp_transport_raise_cwnd(struct sctp_transport *transport,
404 __u32 sack_ctsn, __u32 bytes_acked)
405{
406 __u32 cwnd, ssthresh, flight_size, pba, pmtu;
407
408 cwnd = transport->cwnd;
409 flight_size = transport->flight_size;
410
Vlad Yasevicha6465232008-06-04 12:38:43 -0700411 /* See if we need to exit Fast Recovery first */
412 if (transport->fast_recovery &&
413 TSN_lte(transport->fast_recovery_exit, sack_ctsn))
414 transport->fast_recovery = 0;
415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 /* The appropriate cwnd increase algorithm is performed if, and only
Vlad Yasevicha6465232008-06-04 12:38:43 -0700417 * if the cumulative TSN whould advanced and the congestion window is
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 * being fully utilized.
419 */
Vlad Yasevicha6465232008-06-04 12:38:43 -0700420 if (TSN_lte(sack_ctsn, transport->asoc->ctsn_ack_point) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 (flight_size < cwnd))
422 return;
423
424 ssthresh = transport->ssthresh;
425 pba = transport->partial_bytes_acked;
Frank Filz52ccb8e2005-12-22 11:36:46 -0800426 pmtu = transport->asoc->pathmtu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427
428 if (cwnd <= ssthresh) {
Vlad Yasevicha6465232008-06-04 12:38:43 -0700429 /* RFC 4960 7.2.1
430 * o When cwnd is less than or equal to ssthresh, an SCTP
431 * endpoint MUST use the slow-start algorithm to increase
432 * cwnd only if the current congestion window is being fully
433 * utilized, an incoming SACK advances the Cumulative TSN
434 * Ack Point, and the data sender is not in Fast Recovery.
435 * Only when these three conditions are met can the cwnd be
436 * increased; otherwise, the cwnd MUST not be increased.
437 * If these conditions are met, then cwnd MUST be increased
438 * by, at most, the lesser of 1) the total size of the
439 * previously outstanding DATA chunk(s) acknowledged, and
440 * 2) the destination's path MTU. This upper bound protects
441 * against the ACK-Splitting attack outlined in [SAVAGE99].
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 */
Vlad Yasevicha6465232008-06-04 12:38:43 -0700443 if (transport->fast_recovery)
444 return;
445
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 if (bytes_acked > pmtu)
447 cwnd += pmtu;
448 else
449 cwnd += bytes_acked;
450 SCTP_DEBUG_PRINTK("%s: SLOW START: transport: %p, "
451 "bytes_acked: %d, cwnd: %d, ssthresh: %d, "
452 "flight_size: %d, pba: %d\n",
Harvey Harrison0dc47872008-03-05 20:47:47 -0800453 __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 transport, bytes_acked, cwnd,
455 ssthresh, flight_size, pba);
456 } else {
457 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
458 * upon each SACK arrival that advances the Cumulative TSN Ack
459 * Point, increase partial_bytes_acked by the total number of
460 * bytes of all new chunks acknowledged in that SACK including
461 * chunks acknowledged by the new Cumulative TSN Ack and by
462 * Gap Ack Blocks.
463 *
464 * When partial_bytes_acked is equal to or greater than cwnd
465 * and before the arrival of the SACK the sender had cwnd or
466 * more bytes of data outstanding (i.e., before arrival of the
467 * SACK, flightsize was greater than or equal to cwnd),
468 * increase cwnd by MTU, and reset partial_bytes_acked to
469 * (partial_bytes_acked - cwnd).
470 */
471 pba += bytes_acked;
472 if (pba >= cwnd) {
473 cwnd += pmtu;
474 pba = ((cwnd < pba) ? (pba - cwnd) : 0);
475 }
476 SCTP_DEBUG_PRINTK("%s: CONGESTION AVOIDANCE: "
477 "transport: %p, bytes_acked: %d, cwnd: %d, "
478 "ssthresh: %d, flight_size: %d, pba: %d\n",
Harvey Harrison0dc47872008-03-05 20:47:47 -0800479 __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 transport, bytes_acked, cwnd,
481 ssthresh, flight_size, pba);
482 }
483
484 transport->cwnd = cwnd;
485 transport->partial_bytes_acked = pba;
486}
487
488/* This routine is used to lower the transport's cwnd when congestion is
489 * detected.
490 */
491void sctp_transport_lower_cwnd(struct sctp_transport *transport,
492 sctp_lower_cwnd_t reason)
493{
494 switch (reason) {
495 case SCTP_LOWER_CWND_T3_RTX:
496 /* RFC 2960 Section 7.2.3, sctpimpguide
497 * When the T3-rtx timer expires on an address, SCTP should
498 * perform slow start by:
499 * ssthresh = max(cwnd/2, 4*MTU)
500 * cwnd = 1*MTU
501 * partial_bytes_acked = 0
502 */
503 transport->ssthresh = max(transport->cwnd/2,
Frank Filz52ccb8e2005-12-22 11:36:46 -0800504 4*transport->asoc->pathmtu);
505 transport->cwnd = transport->asoc->pathmtu;
Vlad Yasevich33ce8282009-09-04 18:20:58 -0400506
507 /* T3-rtx also clears fast recovery on the transport */
508 transport->fast_recovery = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 break;
510
511 case SCTP_LOWER_CWND_FAST_RTX:
512 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
513 * destination address(es) to which the missing DATA chunks
514 * were last sent, according to the formula described in
515 * Section 7.2.3.
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900516 *
517 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 * losses from SACK (see Section 7.2.4), An endpoint
519 * should do the following:
520 * ssthresh = max(cwnd/2, 4*MTU)
521 * cwnd = ssthresh
522 * partial_bytes_acked = 0
523 */
Vlad Yasevicha6465232008-06-04 12:38:43 -0700524 if (transport->fast_recovery)
525 return;
526
527 /* Mark Fast recovery */
528 transport->fast_recovery = 1;
529 transport->fast_recovery_exit = transport->asoc->next_tsn - 1;
530
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 transport->ssthresh = max(transport->cwnd/2,
Frank Filz52ccb8e2005-12-22 11:36:46 -0800532 4*transport->asoc->pathmtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 transport->cwnd = transport->ssthresh;
534 break;
535
536 case SCTP_LOWER_CWND_ECNE:
537 /* RFC 2481 Section 6.1.2.
538 * If the sender receives an ECN-Echo ACK packet
539 * then the sender knows that congestion was encountered in the
540 * network on the path from the sender to the receiver. The
541 * indication of congestion should be treated just as a
542 * congestion loss in non-ECN Capable TCP. That is, the TCP
543 * source halves the congestion window "cwnd" and reduces the
544 * slow start threshold "ssthresh".
545 * A critical condition is that TCP does not react to
546 * congestion indications more than once every window of
547 * data (or more loosely more than once every round-trip time).
548 */
Wei Yongjunf61f6f82009-03-02 09:46:13 +0000549 if (time_after(jiffies, transport->last_time_ecne_reduced +
550 transport->rtt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 transport->ssthresh = max(transport->cwnd/2,
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900552 4*transport->asoc->pathmtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 transport->cwnd = transport->ssthresh;
554 transport->last_time_ecne_reduced = jiffies;
555 }
556 break;
557
558 case SCTP_LOWER_CWND_INACTIVE:
559 /* RFC 2960 Section 7.2.1, sctpimpguide
560 * When the endpoint does not transmit data on a given
561 * transport address, the cwnd of the transport address
562 * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
563 * NOTE: Although the draft recommends that this check needs
564 * to be done every RTO interval, we do it every hearbeat
565 * interval.
566 */
Vlad Yasevich245cba72009-11-23 15:53:58 -0500567 transport->cwnd = max(transport->cwnd/2,
568 4*transport->asoc->pathmtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700570 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
572 transport->partial_bytes_acked = 0;
573 SCTP_DEBUG_PRINTK("%s: transport: %p reason: %d cwnd: "
Harvey Harrison0dc47872008-03-05 20:47:47 -0800574 "%d ssthresh: %d\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 transport, reason,
576 transport->cwnd, transport->ssthresh);
577}
578
Vlad Yasevich46d5a802009-11-23 15:54:00 -0500579/* Apply Max.Burst limit to the congestion window:
580 * sctpimpguide-05 2.14.2
581 * D) When the time comes for the sender to
582 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
583 * first be applied to limit how many new DATA chunks may be sent.
584 * The limit is applied by adjusting cwnd as follows:
585 * if ((flightsize+ Max.Burst * MTU) < cwnd)
586 * cwnd = flightsize + Max.Burst * MTU
587 */
588
589void sctp_transport_burst_limited(struct sctp_transport *t)
590{
591 struct sctp_association *asoc = t->asoc;
592 u32 old_cwnd = t->cwnd;
593 u32 max_burst_bytes;
594
595 if (t->burst_limited)
596 return;
597
598 max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu);
599 if (max_burst_bytes < old_cwnd) {
600 t->cwnd = max_burst_bytes;
601 t->burst_limited = old_cwnd;
602 }
603}
604
605/* Restore the old cwnd congestion window, after the burst had it's
606 * desired effect.
607 */
608void sctp_transport_burst_reset(struct sctp_transport *t)
609{
610 if (t->burst_limited) {
611 t->cwnd = t->burst_limited;
612 t->burst_limited = 0;
613 }
614}
615
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616/* What is the next timeout value for this transport? */
617unsigned long sctp_transport_timeout(struct sctp_transport *t)
618{
619 unsigned long timeout;
Sridhar Samudralaad8fec12006-07-21 14:48:50 -0700620 timeout = t->rto + sctp_jitter(t->rto);
621 if (t->state != SCTP_UNCONFIRMED)
622 timeout += t->hbinterval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 timeout += jiffies;
624 return timeout;
625}
Vlad Yasevich749bf922007-03-19 17:02:30 -0700626
627/* Reset transport variables to their initial values */
628void sctp_transport_reset(struct sctp_transport *t)
629{
630 struct sctp_association *asoc = t->asoc;
631
632 /* RFC 2960 (bis), Section 5.2.4
633 * All the congestion control parameters (e.g., cwnd, ssthresh)
634 * related to this peer MUST be reset to their initial values
635 * (see Section 6.2.1)
636 */
637 t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
Vlad Yasevich46d5a802009-11-23 15:54:00 -0500638 t->burst_limited = 0;
Vlad Yasevich289f4242007-03-22 12:26:25 -0700639 t->ssthresh = asoc->peer.i.a_rwnd;
Andrei Pelinescu-Onciul5fdd4ba2009-11-29 00:14:02 -0800640 t->rto = asoc->rto_initial;
Vlad Yasevich749bf922007-03-19 17:02:30 -0700641 t->rtt = 0;
642 t->srtt = 0;
643 t->rttvar = 0;
644
645 /* Reset these additional varibles so that we have a clean
646 * slate.
647 */
648 t->partial_bytes_acked = 0;
649 t->flight_size = 0;
650 t->error_count = 0;
651 t->rto_pending = 0;
Vlad Yasevichfaee47c2009-02-13 08:33:43 +0000652 t->hb_sent = 0;
Vlad Yasevicha6465232008-06-04 12:38:43 -0700653 t->fast_recovery = 0;
Vlad Yasevich749bf922007-03-19 17:02:30 -0700654
655 /* Initialize the state information for SFR-CACC */
656 t->cacc.changeover_active = 0;
657 t->cacc.cycling_changeover = 0;
658 t->cacc.next_tsn_at_change = 0;
659 t->cacc.cacc_saw_newack = 0;
660}