xref: /openbmc/linux/net/sctp/transport.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
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 <linux/random.h>
53 #include <net/sctp/sctp.h>
54 #include <net/sctp/sm.h>
55 
56 /* 1st Level Abstractions.  */
57 
58 /* Initialize a new transport from provided memory.  */
59 static struct sctp_transport *sctp_transport_init(struct sctp_transport *peer,
60 						  const union sctp_addr *addr,
61 						  gfp_t gfp)
62 {
63 	/* Copy in the address.  */
64 	peer->ipaddr = *addr;
65 	peer->af_specific = sctp_get_af_specific(addr->sa.sa_family);
66 	peer->asoc = NULL;
67 
68 	peer->dst = NULL;
69 	memset(&peer->saddr, 0, sizeof(union sctp_addr));
70 
71 	/* From 6.3.1 RTO Calculation:
72 	 *
73 	 * C1) Until an RTT measurement has been made for a packet sent to the
74 	 * given destination transport address, set RTO to the protocol
75 	 * parameter 'RTO.Initial'.
76 	 */
77 	peer->rtt = 0;
78 	peer->rto = msecs_to_jiffies(sctp_rto_initial);
79 	peer->rttvar = 0;
80 	peer->srtt = 0;
81 	peer->rto_pending = 0;
82 
83 	peer->last_time_heard = jiffies;
84 	peer->last_time_used = jiffies;
85 	peer->last_time_ecne_reduced = jiffies;
86 
87 	peer->init_sent_count = 0;
88 
89 	peer->param_flags = SPP_HB_DISABLE |
90 			    SPP_PMTUD_ENABLE |
91 			    SPP_SACKDELAY_ENABLE;
92 	peer->hbinterval  = 0;
93 
94 	/* Initialize the default path max_retrans.  */
95 	peer->pathmaxrxt  = sctp_max_retrans_path;
96 	peer->error_count = 0;
97 
98 	INIT_LIST_HEAD(&peer->transmitted);
99 	INIT_LIST_HEAD(&peer->send_ready);
100 	INIT_LIST_HEAD(&peer->transports);
101 
102 	/* Set up the retransmission timer.  */
103 	init_timer(&peer->T3_rtx_timer);
104 	peer->T3_rtx_timer.function = sctp_generate_t3_rtx_event;
105 	peer->T3_rtx_timer.data = (unsigned long)peer;
106 
107 	/* Set up the heartbeat timer. */
108 	init_timer(&peer->hb_timer);
109 	peer->hb_timer.function = sctp_generate_heartbeat_event;
110 	peer->hb_timer.data = (unsigned long)peer;
111 
112 	/* Initialize the 64-bit random nonce sent with heartbeat. */
113 	get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce));
114 
115 	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.  */
130 struct sctp_transport *sctp_transport_new(const union sctp_addr *addr,
131 					  gfp_t gfp)
132 {
133 	struct sctp_transport *transport;
134 
135 	transport = t_new(struct sctp_transport, gfp);
136 	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 
147 fail_init:
148 	kfree(transport);
149 
150 fail:
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  */
157 void 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  */
181 static 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 
188 	sctp_packet_free(&transport->packet);
189 
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  */
198 void sctp_transport_reset_timers(struct sctp_transport *transport)
199 {
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 
208 	if (!timer_pending(&transport->T3_rtx_timer))
209 		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  */
223 void 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. */
231 void 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) {
238 		transport->pathmtu = dst_mtu(dst);
239 		dst_release(dst);
240 	} else
241 		transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
242 }
243 
244 /* Caches the dst entry and source address for a transport's destination
245  * address.
246  */
247 void sctp_transport_route(struct sctp_transport *transport,
248 			  union sctp_addr *saddr, struct sctp_sock *opt)
249 {
250 	struct sctp_association *asoc = transport->asoc;
251 	struct sctp_af *af = transport->af_specific;
252 	union sctp_addr *daddr = &transport->ipaddr;
253 	struct dst_entry *dst;
254 
255 	dst = af->get_dst(asoc, daddr, saddr);
256 
257 	if (saddr)
258 		memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
259 	else
260 		af->get_saddr(asoc, dst, daddr, &transport->saddr);
261 
262 	transport->dst = dst;
263 	if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) {
264 		return;
265 	}
266 	if (dst) {
267 		transport->pathmtu = dst_mtu(dst);
268 
269 		/* Initialize sk->sk_rcv_saddr, if the transport is the
270 		 * association's active path for getsockname().
271 		 */
272 		if (asoc && (transport == asoc->peer.active_path))
273 			opt->pf->af->to_sk_saddr(&transport->saddr,
274 						 asoc->base.sk);
275 	} else
276 		transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
277 }
278 
279 /* Hold a reference to a transport.  */
280 void sctp_transport_hold(struct sctp_transport *transport)
281 {
282 	atomic_inc(&transport->refcnt);
283 }
284 
285 /* Release a reference to a transport and clean up
286  * if there are no more references.
287  */
288 void sctp_transport_put(struct sctp_transport *transport)
289 {
290 	if (atomic_dec_and_test(&transport->refcnt))
291 		sctp_transport_destroy(transport);
292 }
293 
294 /* Update transport's RTO based on the newly calculated RTT. */
295 void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
296 {
297 	/* Check for valid transport.  */
298 	SCTP_ASSERT(tp, "NULL transport", return);
299 
300 	/* We should not be doing any RTO updates unless rto_pending is set.  */
301 	SCTP_ASSERT(tp->rto_pending, "rto_pending not set", return);
302 
303 	if (tp->rttvar || tp->srtt) {
304 		/* 6.3.1 C3) When a new RTT measurement R' is made, set
305 		 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
306 		 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
307 		 */
308 
309 		/* Note:  The above algorithm has been rewritten to
310 		 * express rto_beta and rto_alpha as inverse powers
311 		 * of two.
312 		 * For example, assuming the default value of RTO.Alpha of
313 		 * 1/8, rto_alpha would be expressed as 3.
314 		 */
315 		tp->rttvar = tp->rttvar - (tp->rttvar >> sctp_rto_beta)
316 			+ ((abs(tp->srtt - rtt)) >> sctp_rto_beta);
317 		tp->srtt = tp->srtt - (tp->srtt >> sctp_rto_alpha)
318 			+ (rtt >> sctp_rto_alpha);
319 	} else {
320 		/* 6.3.1 C2) When the first RTT measurement R is made, set
321 		 * SRTT <- R, RTTVAR <- R/2.
322 		 */
323 		tp->srtt = rtt;
324 		tp->rttvar = rtt >> 1;
325 	}
326 
327 	/* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
328 	 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
329 	 */
330 	if (tp->rttvar == 0)
331 		tp->rttvar = SCTP_CLOCK_GRANULARITY;
332 
333 	/* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
334 	tp->rto = tp->srtt + (tp->rttvar << 2);
335 
336 	/* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
337 	 * seconds then it is rounded up to RTO.Min seconds.
338 	 */
339 	if (tp->rto < tp->asoc->rto_min)
340 		tp->rto = tp->asoc->rto_min;
341 
342 	/* 6.3.1 C7) A maximum value may be placed on RTO provided it is
343 	 * at least RTO.max seconds.
344 	 */
345 	if (tp->rto > tp->asoc->rto_max)
346 		tp->rto = tp->asoc->rto_max;
347 
348 	tp->rtt = rtt;
349 
350 	/* Reset rto_pending so that a new RTT measurement is started when a
351 	 * new data chunk is sent.
352 	 */
353 	tp->rto_pending = 0;
354 
355 	SCTP_DEBUG_PRINTK("%s: transport: %p, rtt: %d, srtt: %d "
356 			  "rttvar: %d, rto: %ld\n", __FUNCTION__,
357 			  tp, rtt, tp->srtt, tp->rttvar, tp->rto);
358 }
359 
360 /* This routine updates the transport's cwnd and partial_bytes_acked
361  * parameters based on the bytes acked in the received SACK.
362  */
363 void sctp_transport_raise_cwnd(struct sctp_transport *transport,
364 			       __u32 sack_ctsn, __u32 bytes_acked)
365 {
366 	__u32 cwnd, ssthresh, flight_size, pba, pmtu;
367 
368 	cwnd = transport->cwnd;
369 	flight_size = transport->flight_size;
370 
371 	/* The appropriate cwnd increase algorithm is performed if, and only
372 	 * if the cumulative TSN has advanced and the congestion window is
373 	 * being fully utilized.
374 	 */
375 	if ((transport->asoc->ctsn_ack_point >= sack_ctsn) ||
376 	    (flight_size < cwnd))
377 		return;
378 
379 	ssthresh = transport->ssthresh;
380 	pba = transport->partial_bytes_acked;
381 	pmtu = transport->asoc->pathmtu;
382 
383 	if (cwnd <= ssthresh) {
384 		/* RFC 2960 7.2.1, sctpimpguide-05 2.14.2 When cwnd is less
385 		 * than or equal to ssthresh an SCTP endpoint MUST use the
386 		 * slow start algorithm to increase cwnd only if the current
387 		 * congestion window is being fully utilized and an incoming
388 		 * SACK advances the Cumulative TSN Ack Point. Only when these
389 		 * two conditions are met can the cwnd be increased otherwise
390 		 * the cwnd MUST not be increased. If these conditions are met
391 		 * then cwnd MUST be increased by at most the lesser of
392 		 * 1) the total size of the previously outstanding DATA
393 		 * chunk(s) acknowledged, and 2) the destination's path MTU.
394 		 */
395 		if (bytes_acked > pmtu)
396 			cwnd += pmtu;
397 		else
398 			cwnd += bytes_acked;
399 		SCTP_DEBUG_PRINTK("%s: SLOW START: transport: %p, "
400 				  "bytes_acked: %d, cwnd: %d, ssthresh: %d, "
401 				  "flight_size: %d, pba: %d\n",
402 				  __FUNCTION__,
403 				  transport, bytes_acked, cwnd,
404 				  ssthresh, flight_size, pba);
405 	} else {
406 		/* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
407 		 * upon each SACK arrival that advances the Cumulative TSN Ack
408 		 * Point, increase partial_bytes_acked by the total number of
409 		 * bytes of all new chunks acknowledged in that SACK including
410 		 * chunks acknowledged by the new Cumulative TSN Ack and by
411 		 * Gap Ack Blocks.
412 		 *
413 		 * When partial_bytes_acked is equal to or greater than cwnd
414 		 * and before the arrival of the SACK the sender had cwnd or
415 		 * more bytes of data outstanding (i.e., before arrival of the
416 		 * SACK, flightsize was greater than or equal to cwnd),
417 		 * increase cwnd by MTU, and reset partial_bytes_acked to
418 		 * (partial_bytes_acked - cwnd).
419 		 */
420 		pba += bytes_acked;
421 		if (pba >= cwnd) {
422 			cwnd += pmtu;
423 			pba = ((cwnd < pba) ? (pba - cwnd) : 0);
424 		}
425 		SCTP_DEBUG_PRINTK("%s: CONGESTION AVOIDANCE: "
426 				  "transport: %p, bytes_acked: %d, cwnd: %d, "
427 				  "ssthresh: %d, flight_size: %d, pba: %d\n",
428 				  __FUNCTION__,
429 				  transport, bytes_acked, cwnd,
430 				  ssthresh, flight_size, pba);
431 	}
432 
433 	transport->cwnd = cwnd;
434 	transport->partial_bytes_acked = pba;
435 }
436 
437 /* This routine is used to lower the transport's cwnd when congestion is
438  * detected.
439  */
440 void sctp_transport_lower_cwnd(struct sctp_transport *transport,
441 			       sctp_lower_cwnd_t reason)
442 {
443 	switch (reason) {
444 	case SCTP_LOWER_CWND_T3_RTX:
445 		/* RFC 2960 Section 7.2.3, sctpimpguide
446 		 * When the T3-rtx timer expires on an address, SCTP should
447 		 * perform slow start by:
448 		 *      ssthresh = max(cwnd/2, 4*MTU)
449 		 *      cwnd = 1*MTU
450 		 *      partial_bytes_acked = 0
451 		 */
452 		transport->ssthresh = max(transport->cwnd/2,
453 					  4*transport->asoc->pathmtu);
454 		transport->cwnd = transport->asoc->pathmtu;
455 		break;
456 
457 	case SCTP_LOWER_CWND_FAST_RTX:
458 		/* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
459 		 * destination address(es) to which the missing DATA chunks
460 		 * were last sent, according to the formula described in
461 		 * Section 7.2.3.
462 		 *
463 		 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
464 		 * losses from SACK (see Section 7.2.4), An endpoint
465 		 * should do the following:
466 		 *      ssthresh = max(cwnd/2, 4*MTU)
467 		 *      cwnd = ssthresh
468 		 *      partial_bytes_acked = 0
469 		 */
470 		transport->ssthresh = max(transport->cwnd/2,
471 					  4*transport->asoc->pathmtu);
472 		transport->cwnd = transport->ssthresh;
473 		break;
474 
475 	case SCTP_LOWER_CWND_ECNE:
476 		/* RFC 2481 Section 6.1.2.
477 		 * If the sender receives an ECN-Echo ACK packet
478 		 * then the sender knows that congestion was encountered in the
479 		 * network on the path from the sender to the receiver. The
480 		 * indication of congestion should be treated just as a
481 		 * congestion loss in non-ECN Capable TCP. That is, the TCP
482 		 * source halves the congestion window "cwnd" and reduces the
483 		 * slow start threshold "ssthresh".
484 		 * A critical condition is that TCP does not react to
485 		 * congestion indications more than once every window of
486 		 * data (or more loosely more than once every round-trip time).
487 		 */
488 		if ((jiffies - transport->last_time_ecne_reduced) >
489 		    transport->rtt) {
490 			transport->ssthresh = max(transport->cwnd/2,
491 						  4*transport->asoc->pathmtu);
492 			transport->cwnd = transport->ssthresh;
493 			transport->last_time_ecne_reduced = jiffies;
494 		}
495 		break;
496 
497 	case SCTP_LOWER_CWND_INACTIVE:
498 		/* RFC 2960 Section 7.2.1, sctpimpguide
499 		 * When the endpoint does not transmit data on a given
500 		 * transport address, the cwnd of the transport address
501 		 * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
502 		 * NOTE: Although the draft recommends that this check needs
503 		 * to be done every RTO interval, we do it every hearbeat
504 		 * interval.
505 		 */
506 		if ((jiffies - transport->last_time_used) > transport->rto)
507 			transport->cwnd = max(transport->cwnd/2,
508 						 4*transport->asoc->pathmtu);
509 		break;
510 	}
511 
512 	transport->partial_bytes_acked = 0;
513 	SCTP_DEBUG_PRINTK("%s: transport: %p reason: %d cwnd: "
514 			  "%d ssthresh: %d\n", __FUNCTION__,
515 			  transport, reason,
516 			  transport->cwnd, transport->ssthresh);
517 }
518 
519 /* What is the next timeout value for this transport? */
520 unsigned long sctp_transport_timeout(struct sctp_transport *t)
521 {
522 	unsigned long timeout;
523 	timeout = t->rto + sctp_jitter(t->rto);
524 	if (t->state != SCTP_UNCONFIRMED)
525 		timeout += t->hbinterval;
526 	timeout += jiffies;
527 	return timeout;
528 }
529 
530 /* Reset transport variables to their initial values */
531 void sctp_transport_reset(struct sctp_transport *t)
532 {
533 	struct sctp_association *asoc = t->asoc;
534 
535 	/* RFC 2960 (bis), Section 5.2.4
536 	 * All the congestion control parameters (e.g., cwnd, ssthresh)
537 	 * related to this peer MUST be reset to their initial values
538 	 * (see Section 6.2.1)
539 	 */
540 	t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
541 	t->ssthresh = asoc->peer.i.a_rwnd;
542 	t->rto = asoc->rto_initial;
543 	t->rtt = 0;
544 	t->srtt = 0;
545 	t->rttvar = 0;
546 
547 	/* Reset these additional varibles so that we have a clean
548 	 * slate.
549 	 */
550 	t->partial_bytes_acked = 0;
551 	t->flight_size = 0;
552 	t->error_count = 0;
553 	t->rto_pending = 0;
554 
555 	/* Initialize the state information for SFR-CACC */
556 	t->cacc.changeover_active = 0;
557 	t->cacc.cycling_changeover = 0;
558 	t->cacc.next_tsn_at_change = 0;
559 	t->cacc.cacc_saw_newack = 0;
560 }
561