aboutsummaryrefslogtreecommitdiffstats
path: root/net/sctp/transport.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
committerLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
commit1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch)
tree0bba044c4ce775e45a88a51686b5d9f90697ea9d /net/sctp/transport.c
Linux-2.6.12-rc2v2.6.12-rc2
Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
Diffstat (limited to 'net/sctp/transport.c')
-rw-r--r--net/sctp/transport.c514
1 files changed, 514 insertions, 0 deletions
diff --git a/net/sctp/transport.c b/net/sctp/transport.c
new file mode 100644
index 000000000000..f30882e1e96a
--- /dev/null
+++ b/net/sctp/transport.c
@@ -0,0 +1,514 @@
1/* 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,
60 int gfp)
61{
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
86 peer->active = SCTP_ACTIVE;
87 peer->hb_allowed = 0;
88
89 /* Initialize the default path max_retrans. */
90 peer->max_retrans = sctp_max_retrans_path;
91 peer->error_count = 0;
92
93 INIT_LIST_HEAD(&peer->transmitted);
94 INIT_LIST_HEAD(&peer->send_ready);
95 INIT_LIST_HEAD(&peer->transports);
96
97 /* Set up the retransmission timer. */
98 init_timer(&peer->T3_rtx_timer);
99 peer->T3_rtx_timer.function = sctp_generate_t3_rtx_event;
100 peer->T3_rtx_timer.data = (unsigned long)peer;
101
102 /* Set up the heartbeat timer. */
103 init_timer(&peer->hb_timer);
104 peer->hb_interval = SCTP_DEFAULT_TIMEOUT_HEARTBEAT;
105 peer->hb_timer.function = sctp_generate_heartbeat_event;
106 peer->hb_timer.data = (unsigned long)peer;
107
108 atomic_set(&peer->refcnt, 1);
109 peer->dead = 0;
110
111 peer->malloced = 0;
112
113 /* Initialize the state information for SFR-CACC */
114 peer->cacc.changeover_active = 0;
115 peer->cacc.cycling_changeover = 0;
116 peer->cacc.next_tsn_at_change = 0;
117 peer->cacc.cacc_saw_newack = 0;
118
119 return peer;
120}
121
122/* Allocate and initialize a new transport. */
123struct sctp_transport *sctp_transport_new(const union sctp_addr *addr, int gfp)
124{
125 struct sctp_transport *transport;
126
127 transport = t_new(struct sctp_transport, gfp);
128 if (!transport)
129 goto fail;
130
131 if (!sctp_transport_init(transport, addr, gfp))
132 goto fail_init;
133
134 transport->malloced = 1;
135 SCTP_DBG_OBJCNT_INC(transport);
136
137 return transport;
138
139fail_init:
140 kfree(transport);
141
142fail:
143 return NULL;
144}
145
146/* This transport is no longer needed. Free up if possible, or
147 * delay until it last reference count.
148 */
149void sctp_transport_free(struct sctp_transport *transport)
150{
151 transport->dead = 1;
152
153 /* Try to delete the heartbeat timer. */
154 if (del_timer(&transport->hb_timer))
155 sctp_transport_put(transport);
156
157 /* Delete the T3_rtx timer if it's active.
158 * There is no point in not doing this now and letting
159 * structure hang around in memory since we know
160 * the tranport is going away.
161 */
162 if (timer_pending(&transport->T3_rtx_timer) &&
163 del_timer(&transport->T3_rtx_timer))
164 sctp_transport_put(transport);
165
166
167 sctp_transport_put(transport);
168}
169
170/* Destroy the transport data structure.
171 * Assumes there are no more users of this structure.
172 */
173static void sctp_transport_destroy(struct sctp_transport *transport)
174{
175 SCTP_ASSERT(transport->dead, "Transport is not dead", return);
176
177 if (transport->asoc)
178 sctp_association_put(transport->asoc);
179
180 sctp_packet_free(&transport->packet);
181
182 dst_release(transport->dst);
183 kfree(transport);
184 SCTP_DBG_OBJCNT_DEC(transport);
185}
186
187/* Start T3_rtx timer if it is not already running and update the heartbeat
188 * timer. This routine is called every time a DATA chunk is sent.
189 */
190void sctp_transport_reset_timers(struct sctp_transport *transport)
191{
192 /* RFC 2960 6.3.2 Retransmission Timer Rules
193 *
194 * R1) Every time a DATA chunk is sent to any address(including a
195 * retransmission), if the T3-rtx timer of that address is not running
196 * start it running so that it will expire after the RTO of that
197 * address.
198 */
199
200 if (!timer_pending(&transport->T3_rtx_timer))
201 if (!mod_timer(&transport->T3_rtx_timer,
202 jiffies + transport->rto))
203 sctp_transport_hold(transport);
204
205 /* When a data chunk is sent, reset the heartbeat interval. */
206 if (!mod_timer(&transport->hb_timer,
207 sctp_transport_timeout(transport)))
208 sctp_transport_hold(transport);
209}
210
211/* This transport has been assigned to an association.
212 * Initialize fields from the association or from the sock itself.
213 * Register the reference count in the association.
214 */
215void sctp_transport_set_owner(struct sctp_transport *transport,
216 struct sctp_association *asoc)
217{
218 transport->asoc = asoc;
219 sctp_association_hold(asoc);
220}
221
222/* Initialize the pmtu of a transport. */
223void sctp_transport_pmtu(struct sctp_transport *transport)
224{
225 struct dst_entry *dst;
226
227 dst = transport->af_specific->get_dst(NULL, &transport->ipaddr, NULL);
228
229 if (dst) {
230 transport->pmtu = dst_mtu(dst);
231 dst_release(dst);
232 } else
233 transport->pmtu = SCTP_DEFAULT_MAXSEGMENT;
234}
235
236/* Caches the dst entry and source address for a transport's destination
237 * address.
238 */
239void sctp_transport_route(struct sctp_transport *transport,
240 union sctp_addr *saddr, struct sctp_sock *opt)
241{
242 struct sctp_association *asoc = transport->asoc;
243 struct sctp_af *af = transport->af_specific;
244 union sctp_addr *daddr = &transport->ipaddr;
245 struct dst_entry *dst;
246
247 dst = af->get_dst(asoc, daddr, saddr);
248
249 if (saddr)
250 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
251 else
252 af->get_saddr(asoc, dst, daddr, &transport->saddr);
253
254 transport->dst = dst;
255 if (dst) {
256 transport->pmtu = dst_mtu(dst);
257
258 /* Initialize sk->sk_rcv_saddr, if the transport is the
259 * association's active path for getsockname().
260 */
261 if (asoc && (transport == asoc->peer.active_path))
262 af->to_sk_saddr(&transport->saddr, asoc->base.sk);
263 } else
264 transport->pmtu = SCTP_DEFAULT_MAXSEGMENT;
265}
266
267/* Hold a reference to a transport. */
268void sctp_transport_hold(struct sctp_transport *transport)
269{
270 atomic_inc(&transport->refcnt);
271}
272
273/* Release a reference to a transport and clean up
274 * if there are no more references.
275 */
276void sctp_transport_put(struct sctp_transport *transport)
277{
278 if (atomic_dec_and_test(&transport->refcnt))
279 sctp_transport_destroy(transport);
280}
281
282/* Update transport's RTO based on the newly calculated RTT. */
283void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
284{
285 /* Check for valid transport. */
286 SCTP_ASSERT(tp, "NULL transport", return);
287
288 /* We should not be doing any RTO updates unless rto_pending is set. */
289 SCTP_ASSERT(tp->rto_pending, "rto_pending not set", return);
290
291 if (tp->rttvar || tp->srtt) {
292 /* 6.3.1 C3) When a new RTT measurement R' is made, set
293 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
294 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
295 */
296
297 /* Note: The above algorithm has been rewritten to
298 * express rto_beta and rto_alpha as inverse powers
299 * of two.
300 * For example, assuming the default value of RTO.Alpha of
301 * 1/8, rto_alpha would be expressed as 3.
302 */
303 tp->rttvar = tp->rttvar - (tp->rttvar >> sctp_rto_beta)
304 + ((abs(tp->srtt - rtt)) >> sctp_rto_beta);
305 tp->srtt = tp->srtt - (tp->srtt >> sctp_rto_alpha)
306 + (rtt >> sctp_rto_alpha);
307 } else {
308 /* 6.3.1 C2) When the first RTT measurement R is made, set
309 * SRTT <- R, RTTVAR <- R/2.
310 */
311 tp->srtt = rtt;
312 tp->rttvar = rtt >> 1;
313 }
314
315 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
316 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
317 */
318 if (tp->rttvar == 0)
319 tp->rttvar = SCTP_CLOCK_GRANULARITY;
320
321 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
322 tp->rto = tp->srtt + (tp->rttvar << 2);
323
324 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
325 * seconds then it is rounded up to RTO.Min seconds.
326 */
327 if (tp->rto < tp->asoc->rto_min)
328 tp->rto = tp->asoc->rto_min;
329
330 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
331 * at least RTO.max seconds.
332 */
333 if (tp->rto > tp->asoc->rto_max)
334 tp->rto = tp->asoc->rto_max;
335
336 tp->rtt = rtt;
337
338 /* Reset rto_pending so that a new RTT measurement is started when a
339 * new data chunk is sent.
340 */
341 tp->rto_pending = 0;
342
343 SCTP_DEBUG_PRINTK("%s: transport: %p, rtt: %d, srtt: %d "
344 "rttvar: %d, rto: %d\n", __FUNCTION__,
345 tp, rtt, tp->srtt, tp->rttvar, tp->rto);
346}
347
348/* This routine updates the transport's cwnd and partial_bytes_acked
349 * parameters based on the bytes acked in the received SACK.
350 */
351void sctp_transport_raise_cwnd(struct sctp_transport *transport,
352 __u32 sack_ctsn, __u32 bytes_acked)
353{
354 __u32 cwnd, ssthresh, flight_size, pba, pmtu;
355
356 cwnd = transport->cwnd;
357 flight_size = transport->flight_size;
358
359 /* The appropriate cwnd increase algorithm is performed if, and only
360 * if the cumulative TSN has advanced and the congestion window is
361 * being fully utilized.
362 */
363 if ((transport->asoc->ctsn_ack_point >= sack_ctsn) ||
364 (flight_size < cwnd))
365 return;
366
367 ssthresh = transport->ssthresh;
368 pba = transport->partial_bytes_acked;
369 pmtu = transport->asoc->pmtu;
370
371 if (cwnd <= ssthresh) {
372 /* RFC 2960 7.2.1, sctpimpguide-05 2.14.2 When cwnd is less
373 * than or equal to ssthresh an SCTP endpoint MUST use the
374 * slow start algorithm to increase cwnd only if the current
375 * congestion window is being fully utilized and an incoming
376 * SACK advances the Cumulative TSN Ack Point. Only when these
377 * two conditions are met can the cwnd be increased otherwise
378 * the cwnd MUST not be increased. If these conditions are met
379 * then cwnd MUST be increased by at most the lesser of
380 * 1) the total size of the previously outstanding DATA
381 * chunk(s) acknowledged, and 2) the destination's path MTU.
382 */
383 if (bytes_acked > pmtu)
384 cwnd += pmtu;
385 else
386 cwnd += bytes_acked;
387 SCTP_DEBUG_PRINTK("%s: SLOW START: transport: %p, "
388 "bytes_acked: %d, cwnd: %d, ssthresh: %d, "
389 "flight_size: %d, pba: %d\n",
390 __FUNCTION__,
391 transport, bytes_acked, cwnd,
392 ssthresh, flight_size, pba);
393 } else {
394 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
395 * upon each SACK arrival that advances the Cumulative TSN Ack
396 * Point, increase partial_bytes_acked by the total number of
397 * bytes of all new chunks acknowledged in that SACK including
398 * chunks acknowledged by the new Cumulative TSN Ack and by
399 * Gap Ack Blocks.
400 *
401 * When partial_bytes_acked is equal to or greater than cwnd
402 * and before the arrival of the SACK the sender had cwnd or
403 * more bytes of data outstanding (i.e., before arrival of the
404 * SACK, flightsize was greater than or equal to cwnd),
405 * increase cwnd by MTU, and reset partial_bytes_acked to
406 * (partial_bytes_acked - cwnd).
407 */
408 pba += bytes_acked;
409 if (pba >= cwnd) {
410 cwnd += pmtu;
411 pba = ((cwnd < pba) ? (pba - cwnd) : 0);
412 }
413 SCTP_DEBUG_PRINTK("%s: CONGESTION AVOIDANCE: "
414 "transport: %p, bytes_acked: %d, cwnd: %d, "
415 "ssthresh: %d, flight_size: %d, pba: %d\n",
416 __FUNCTION__,
417 transport, bytes_acked, cwnd,
418 ssthresh, flight_size, pba);
419 }
420
421 transport->cwnd = cwnd;
422 transport->partial_bytes_acked = pba;
423}
424
425/* This routine is used to lower the transport's cwnd when congestion is
426 * detected.
427 */
428void sctp_transport_lower_cwnd(struct sctp_transport *transport,
429 sctp_lower_cwnd_t reason)
430{
431 switch (reason) {
432 case SCTP_LOWER_CWND_T3_RTX:
433 /* RFC 2960 Section 7.2.3, sctpimpguide
434 * When the T3-rtx timer expires on an address, SCTP should
435 * perform slow start by:
436 * ssthresh = max(cwnd/2, 4*MTU)
437 * cwnd = 1*MTU
438 * partial_bytes_acked = 0
439 */
440 transport->ssthresh = max(transport->cwnd/2,
441 4*transport->asoc->pmtu);
442 transport->cwnd = transport->asoc->pmtu;
443 break;
444
445 case SCTP_LOWER_CWND_FAST_RTX:
446 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
447 * destination address(es) to which the missing DATA chunks
448 * were last sent, according to the formula described in
449 * Section 7.2.3.
450 *
451 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
452 * losses from SACK (see Section 7.2.4), An endpoint
453 * should do the following:
454 * ssthresh = max(cwnd/2, 4*MTU)
455 * cwnd = ssthresh
456 * partial_bytes_acked = 0
457 */
458 transport->ssthresh = max(transport->cwnd/2,
459 4*transport->asoc->pmtu);
460 transport->cwnd = transport->ssthresh;
461 break;
462
463 case SCTP_LOWER_CWND_ECNE:
464 /* RFC 2481 Section 6.1.2.
465 * If the sender receives an ECN-Echo ACK packet
466 * then the sender knows that congestion was encountered in the
467 * network on the path from the sender to the receiver. The
468 * indication of congestion should be treated just as a
469 * congestion loss in non-ECN Capable TCP. That is, the TCP
470 * source halves the congestion window "cwnd" and reduces the
471 * slow start threshold "ssthresh".
472 * A critical condition is that TCP does not react to
473 * congestion indications more than once every window of
474 * data (or more loosely more than once every round-trip time).
475 */
476 if ((jiffies - transport->last_time_ecne_reduced) >
477 transport->rtt) {
478 transport->ssthresh = max(transport->cwnd/2,
479 4*transport->asoc->pmtu);
480 transport->cwnd = transport->ssthresh;
481 transport->last_time_ecne_reduced = jiffies;
482 }
483 break;
484
485 case SCTP_LOWER_CWND_INACTIVE:
486 /* RFC 2960 Section 7.2.1, sctpimpguide
487 * When the endpoint does not transmit data on a given
488 * transport address, the cwnd of the transport address
489 * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
490 * NOTE: Although the draft recommends that this check needs
491 * to be done every RTO interval, we do it every hearbeat
492 * interval.
493 */
494 if ((jiffies - transport->last_time_used) > transport->rto)
495 transport->cwnd = max(transport->cwnd/2,
496 4*transport->asoc->pmtu);
497 break;
498 };
499
500 transport->partial_bytes_acked = 0;
501 SCTP_DEBUG_PRINTK("%s: transport: %p reason: %d cwnd: "
502 "%d ssthresh: %d\n", __FUNCTION__,
503 transport, reason,
504 transport->cwnd, transport->ssthresh);
505}
506
507/* What is the next timeout value for this transport? */
508unsigned long sctp_transport_timeout(struct sctp_transport *t)
509{
510 unsigned long timeout;
511 timeout = t->hb_interval + t->rto + sctp_jitter(t->rto);
512 timeout += jiffies;
513 return timeout;
514}