blob: 6bc27200e6cafc8251ebc50b3bac6521a943abf6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/* SCTP kernel reference Implementation
2 * 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 *
8 * This file is part of the SCTP kernel reference Implementation
9 *
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 *
14 * The SCTP reference implementation is free software;
15 * 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 *
20 * The SCTP reference implementation is distributed in the hope that it
21 * 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
51#include <linux/types.h>
52#include <net/sctp/sctp.h>
53#include <net/sctp/sm.h>
54
55/* 1st Level Abstractions. */
56
57/* Initialize a new transport from provided memory. */
58static struct sctp_transport *sctp_transport_init(struct sctp_transport *peer,
59 const union sctp_addr *addr,
Al Virodd0fc662005-10-07 07:46:04 +010060 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -070061{
62 /* Copy in the address. */
63 peer->ipaddr = *addr;
64 peer->af_specific = sctp_get_af_specific(addr->sa.sa_family);
65 peer->asoc = NULL;
66
67 peer->dst = NULL;
68 memset(&peer->saddr, 0, sizeof(union sctp_addr));
69
70 /* From 6.3.1 RTO Calculation:
71 *
72 * C1) Until an RTT measurement has been made for a packet sent to the
73 * given destination transport address, set RTO to the protocol
74 * parameter 'RTO.Initial'.
75 */
76 peer->rtt = 0;
77 peer->rto = sctp_rto_initial;
78 peer->rttvar = 0;
79 peer->srtt = 0;
80 peer->rto_pending = 0;
81
82 peer->last_time_heard = jiffies;
83 peer->last_time_used = jiffies;
84 peer->last_time_ecne_reduced = jiffies;
85
Frank Filz3f7a87d2005-06-20 13:14:57 -070086 peer->init_sent_count = 0;
87
88 peer->state = SCTP_ACTIVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 peer->hb_allowed = 0;
90
91 /* Initialize the default path max_retrans. */
92 peer->max_retrans = sctp_max_retrans_path;
93 peer->error_count = 0;
94
95 INIT_LIST_HEAD(&peer->transmitted);
96 INIT_LIST_HEAD(&peer->send_ready);
97 INIT_LIST_HEAD(&peer->transports);
98
99 /* Set up the retransmission timer. */
100 init_timer(&peer->T3_rtx_timer);
101 peer->T3_rtx_timer.function = sctp_generate_t3_rtx_event;
102 peer->T3_rtx_timer.data = (unsigned long)peer;
103
104 /* Set up the heartbeat timer. */
105 init_timer(&peer->hb_timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 peer->hb_timer.function = sctp_generate_heartbeat_event;
107 peer->hb_timer.data = (unsigned long)peer;
108
109 atomic_set(&peer->refcnt, 1);
110 peer->dead = 0;
111
112 peer->malloced = 0;
113
114 /* Initialize the state information for SFR-CACC */
115 peer->cacc.changeover_active = 0;
116 peer->cacc.cycling_changeover = 0;
117 peer->cacc.next_tsn_at_change = 0;
118 peer->cacc.cacc_saw_newack = 0;
119
120 return peer;
121}
122
123/* Allocate and initialize a new transport. */
Alexey Dobriyan3182cd842005-07-11 20:57:47 -0700124struct sctp_transport *sctp_transport_new(const union sctp_addr *addr,
Al Virodd0fc662005-10-07 07:46:04 +0100125 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126{
127 struct sctp_transport *transport;
128
129 transport = t_new(struct sctp_transport, gfp);
130 if (!transport)
131 goto fail;
132
133 if (!sctp_transport_init(transport, addr, gfp))
134 goto fail_init;
135
136 transport->malloced = 1;
137 SCTP_DBG_OBJCNT_INC(transport);
138
139 return transport;
140
141fail_init:
142 kfree(transport);
143
144fail:
145 return NULL;
146}
147
148/* This transport is no longer needed. Free up if possible, or
149 * delay until it last reference count.
150 */
151void sctp_transport_free(struct sctp_transport *transport)
152{
153 transport->dead = 1;
154
155 /* Try to delete the heartbeat timer. */
156 if (del_timer(&transport->hb_timer))
157 sctp_transport_put(transport);
158
159 /* Delete the T3_rtx timer if it's active.
160 * There is no point in not doing this now and letting
161 * structure hang around in memory since we know
162 * the tranport is going away.
163 */
164 if (timer_pending(&transport->T3_rtx_timer) &&
165 del_timer(&transport->T3_rtx_timer))
166 sctp_transport_put(transport);
167
168
169 sctp_transport_put(transport);
170}
171
172/* Destroy the transport data structure.
173 * Assumes there are no more users of this structure.
174 */
175static void sctp_transport_destroy(struct sctp_transport *transport)
176{
177 SCTP_ASSERT(transport->dead, "Transport is not dead", return);
178
179 if (transport->asoc)
180 sctp_association_put(transport->asoc);
181
182 sctp_packet_free(&transport->packet);
183
184 dst_release(transport->dst);
185 kfree(transport);
186 SCTP_DBG_OBJCNT_DEC(transport);
187}
188
189/* Start T3_rtx timer if it is not already running and update the heartbeat
190 * timer. This routine is called every time a DATA chunk is sent.
191 */
192void sctp_transport_reset_timers(struct sctp_transport *transport)
193{
194 /* RFC 2960 6.3.2 Retransmission Timer Rules
195 *
196 * R1) Every time a DATA chunk is sent to any address(including a
197 * retransmission), if the T3-rtx timer of that address is not running
198 * start it running so that it will expire after the RTO of that
199 * address.
200 */
201
202 if (!timer_pending(&transport->T3_rtx_timer))
203 if (!mod_timer(&transport->T3_rtx_timer,
204 jiffies + transport->rto))
205 sctp_transport_hold(transport);
206
207 /* When a data chunk is sent, reset the heartbeat interval. */
208 if (!mod_timer(&transport->hb_timer,
209 sctp_transport_timeout(transport)))
210 sctp_transport_hold(transport);
211}
212
213/* This transport has been assigned to an association.
214 * Initialize fields from the association or from the sock itself.
215 * Register the reference count in the association.
216 */
217void sctp_transport_set_owner(struct sctp_transport *transport,
218 struct sctp_association *asoc)
219{
220 transport->asoc = asoc;
221 sctp_association_hold(asoc);
222}
223
224/* Initialize the pmtu of a transport. */
225void sctp_transport_pmtu(struct sctp_transport *transport)
226{
227 struct dst_entry *dst;
228
229 dst = transport->af_specific->get_dst(NULL, &transport->ipaddr, NULL);
230
231 if (dst) {
232 transport->pmtu = dst_mtu(dst);
233 dst_release(dst);
234 } else
235 transport->pmtu = SCTP_DEFAULT_MAXSEGMENT;
236}
237
238/* Caches the dst entry and source address for a transport's destination
239 * address.
240 */
241void sctp_transport_route(struct sctp_transport *transport,
242 union sctp_addr *saddr, struct sctp_sock *opt)
243{
244 struct sctp_association *asoc = transport->asoc;
245 struct sctp_af *af = transport->af_specific;
246 union sctp_addr *daddr = &transport->ipaddr;
247 struct dst_entry *dst;
248
249 dst = af->get_dst(asoc, daddr, saddr);
250
251 if (saddr)
252 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
253 else
254 af->get_saddr(asoc, dst, daddr, &transport->saddr);
255
256 transport->dst = dst;
257 if (dst) {
258 transport->pmtu = dst_mtu(dst);
259
260 /* Initialize sk->sk_rcv_saddr, if the transport is the
261 * association's active path for getsockname().
262 */
263 if (asoc && (transport == asoc->peer.active_path))
264 af->to_sk_saddr(&transport->saddr, asoc->base.sk);
265 } else
266 transport->pmtu = SCTP_DEFAULT_MAXSEGMENT;
267}
268
269/* Hold a reference to a transport. */
270void sctp_transport_hold(struct sctp_transport *transport)
271{
272 atomic_inc(&transport->refcnt);
273}
274
275/* Release a reference to a transport and clean up
276 * if there are no more references.
277 */
278void sctp_transport_put(struct sctp_transport *transport)
279{
280 if (atomic_dec_and_test(&transport->refcnt))
281 sctp_transport_destroy(transport);
282}
283
284/* Update transport's RTO based on the newly calculated RTT. */
285void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
286{
287 /* Check for valid transport. */
288 SCTP_ASSERT(tp, "NULL transport", return);
289
290 /* We should not be doing any RTO updates unless rto_pending is set. */
291 SCTP_ASSERT(tp->rto_pending, "rto_pending not set", return);
292
293 if (tp->rttvar || tp->srtt) {
294 /* 6.3.1 C3) When a new RTT measurement R' is made, set
295 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
296 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
297 */
298
299 /* Note: The above algorithm has been rewritten to
300 * express rto_beta and rto_alpha as inverse powers
301 * of two.
302 * For example, assuming the default value of RTO.Alpha of
303 * 1/8, rto_alpha would be expressed as 3.
304 */
305 tp->rttvar = tp->rttvar - (tp->rttvar >> sctp_rto_beta)
306 + ((abs(tp->srtt - rtt)) >> sctp_rto_beta);
307 tp->srtt = tp->srtt - (tp->srtt >> sctp_rto_alpha)
308 + (rtt >> sctp_rto_alpha);
309 } else {
310 /* 6.3.1 C2) When the first RTT measurement R is made, set
311 * SRTT <- R, RTTVAR <- R/2.
312 */
313 tp->srtt = rtt;
314 tp->rttvar = rtt >> 1;
315 }
316
317 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
318 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
319 */
320 if (tp->rttvar == 0)
321 tp->rttvar = SCTP_CLOCK_GRANULARITY;
322
323 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
324 tp->rto = tp->srtt + (tp->rttvar << 2);
325
326 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
327 * seconds then it is rounded up to RTO.Min seconds.
328 */
329 if (tp->rto < tp->asoc->rto_min)
330 tp->rto = tp->asoc->rto_min;
331
332 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
333 * at least RTO.max seconds.
334 */
335 if (tp->rto > tp->asoc->rto_max)
336 tp->rto = tp->asoc->rto_max;
337
338 tp->rtt = rtt;
339
340 /* Reset rto_pending so that a new RTT measurement is started when a
341 * new data chunk is sent.
342 */
343 tp->rto_pending = 0;
344
345 SCTP_DEBUG_PRINTK("%s: transport: %p, rtt: %d, srtt: %d "
346 "rttvar: %d, rto: %d\n", __FUNCTION__,
347 tp, rtt, tp->srtt, tp->rttvar, tp->rto);
348}
349
350/* This routine updates the transport's cwnd and partial_bytes_acked
351 * parameters based on the bytes acked in the received SACK.
352 */
353void sctp_transport_raise_cwnd(struct sctp_transport *transport,
354 __u32 sack_ctsn, __u32 bytes_acked)
355{
356 __u32 cwnd, ssthresh, flight_size, pba, pmtu;
357
358 cwnd = transport->cwnd;
359 flight_size = transport->flight_size;
360
361 /* The appropriate cwnd increase algorithm is performed if, and only
362 * if the cumulative TSN has advanced and the congestion window is
363 * being fully utilized.
364 */
365 if ((transport->asoc->ctsn_ack_point >= sack_ctsn) ||
366 (flight_size < cwnd))
367 return;
368
369 ssthresh = transport->ssthresh;
370 pba = transport->partial_bytes_acked;
371 pmtu = transport->asoc->pmtu;
372
373 if (cwnd <= ssthresh) {
374 /* RFC 2960 7.2.1, sctpimpguide-05 2.14.2 When cwnd is less
375 * than or equal to ssthresh an SCTP endpoint MUST use the
376 * slow start algorithm to increase cwnd only if the current
377 * congestion window is being fully utilized and an incoming
378 * SACK advances the Cumulative TSN Ack Point. Only when these
379 * two conditions are met can the cwnd be increased otherwise
380 * the cwnd MUST not be increased. If these conditions are met
381 * then cwnd MUST be increased by at most the lesser of
382 * 1) the total size of the previously outstanding DATA
383 * chunk(s) acknowledged, and 2) the destination's path MTU.
384 */
385 if (bytes_acked > pmtu)
386 cwnd += pmtu;
387 else
388 cwnd += bytes_acked;
389 SCTP_DEBUG_PRINTK("%s: SLOW START: transport: %p, "
390 "bytes_acked: %d, cwnd: %d, ssthresh: %d, "
391 "flight_size: %d, pba: %d\n",
392 __FUNCTION__,
393 transport, bytes_acked, cwnd,
394 ssthresh, flight_size, pba);
395 } else {
396 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
397 * upon each SACK arrival that advances the Cumulative TSN Ack
398 * Point, increase partial_bytes_acked by the total number of
399 * bytes of all new chunks acknowledged in that SACK including
400 * chunks acknowledged by the new Cumulative TSN Ack and by
401 * Gap Ack Blocks.
402 *
403 * When partial_bytes_acked is equal to or greater than cwnd
404 * and before the arrival of the SACK the sender had cwnd or
405 * more bytes of data outstanding (i.e., before arrival of the
406 * SACK, flightsize was greater than or equal to cwnd),
407 * increase cwnd by MTU, and reset partial_bytes_acked to
408 * (partial_bytes_acked - cwnd).
409 */
410 pba += bytes_acked;
411 if (pba >= cwnd) {
412 cwnd += pmtu;
413 pba = ((cwnd < pba) ? (pba - cwnd) : 0);
414 }
415 SCTP_DEBUG_PRINTK("%s: CONGESTION AVOIDANCE: "
416 "transport: %p, bytes_acked: %d, cwnd: %d, "
417 "ssthresh: %d, flight_size: %d, pba: %d\n",
418 __FUNCTION__,
419 transport, bytes_acked, cwnd,
420 ssthresh, flight_size, pba);
421 }
422
423 transport->cwnd = cwnd;
424 transport->partial_bytes_acked = pba;
425}
426
427/* This routine is used to lower the transport's cwnd when congestion is
428 * detected.
429 */
430void sctp_transport_lower_cwnd(struct sctp_transport *transport,
431 sctp_lower_cwnd_t reason)
432{
433 switch (reason) {
434 case SCTP_LOWER_CWND_T3_RTX:
435 /* RFC 2960 Section 7.2.3, sctpimpguide
436 * When the T3-rtx timer expires on an address, SCTP should
437 * perform slow start by:
438 * ssthresh = max(cwnd/2, 4*MTU)
439 * cwnd = 1*MTU
440 * partial_bytes_acked = 0
441 */
442 transport->ssthresh = max(transport->cwnd/2,
443 4*transport->asoc->pmtu);
444 transport->cwnd = transport->asoc->pmtu;
445 break;
446
447 case SCTP_LOWER_CWND_FAST_RTX:
448 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
449 * destination address(es) to which the missing DATA chunks
450 * were last sent, according to the formula described in
451 * Section 7.2.3.
452 *
453 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
454 * losses from SACK (see Section 7.2.4), An endpoint
455 * should do the following:
456 * ssthresh = max(cwnd/2, 4*MTU)
457 * cwnd = ssthresh
458 * partial_bytes_acked = 0
459 */
460 transport->ssthresh = max(transport->cwnd/2,
461 4*transport->asoc->pmtu);
462 transport->cwnd = transport->ssthresh;
463 break;
464
465 case SCTP_LOWER_CWND_ECNE:
466 /* RFC 2481 Section 6.1.2.
467 * If the sender receives an ECN-Echo ACK packet
468 * then the sender knows that congestion was encountered in the
469 * network on the path from the sender to the receiver. The
470 * indication of congestion should be treated just as a
471 * congestion loss in non-ECN Capable TCP. That is, the TCP
472 * source halves the congestion window "cwnd" and reduces the
473 * slow start threshold "ssthresh".
474 * A critical condition is that TCP does not react to
475 * congestion indications more than once every window of
476 * data (or more loosely more than once every round-trip time).
477 */
478 if ((jiffies - transport->last_time_ecne_reduced) >
479 transport->rtt) {
480 transport->ssthresh = max(transport->cwnd/2,
481 4*transport->asoc->pmtu);
482 transport->cwnd = transport->ssthresh;
483 transport->last_time_ecne_reduced = jiffies;
484 }
485 break;
486
487 case SCTP_LOWER_CWND_INACTIVE:
488 /* RFC 2960 Section 7.2.1, sctpimpguide
489 * When the endpoint does not transmit data on a given
490 * transport address, the cwnd of the transport address
491 * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
492 * NOTE: Although the draft recommends that this check needs
493 * to be done every RTO interval, we do it every hearbeat
494 * interval.
495 */
496 if ((jiffies - transport->last_time_used) > transport->rto)
497 transport->cwnd = max(transport->cwnd/2,
498 4*transport->asoc->pmtu);
499 break;
500 };
501
502 transport->partial_bytes_acked = 0;
503 SCTP_DEBUG_PRINTK("%s: transport: %p reason: %d cwnd: "
504 "%d ssthresh: %d\n", __FUNCTION__,
505 transport, reason,
506 transport->cwnd, transport->ssthresh);
507}
508
509/* What is the next timeout value for this transport? */
510unsigned long sctp_transport_timeout(struct sctp_transport *t)
511{
512 unsigned long timeout;
513 timeout = t->hb_interval + t->rto + sctp_jitter(t->rto);
514 timeout += jiffies;
515 return timeout;
516}