1 /* SCTP kernel 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
8 * This file is part of the SCTP kernel implementation
10 * This module provides the abstraction for an SCTP tranport representing
11 * a remote transport address. For local transport addresses, we just use
14 * This SCTP 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)
20 * This SCTP 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.
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.
31 * Please send any bug reports or fixes you make to the
33 * lksctp developers <lksctp-developers@lists.sourceforge.net>
35 * Or submit a bug report through the following website:
36 * http://www.sf.net/projects/lksctp
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>
47 * Any bugs reported given to us we will try to fix... any fixes shared will
48 * be incorporated into the next SCTP release.
51 #include <linux/types.h>
52 #include <linux/random.h>
53 #include <net/sctp/sctp.h>
54 #include <net/sctp/sm.h>
56 /* 1st Level Abstractions. */
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,
63 /* Copy in the address. */
65 peer->af_specific = sctp_get_af_specific(addr->sa.sa_family);
69 memset(&peer->saddr, 0, sizeof(union sctp_addr));
71 /* From 6.3.1 RTO Calculation:
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'.
77 peer->rto = msecs_to_jiffies(sctp_rto_initial);
81 peer->rto_pending = 0;
83 peer->fast_recovery = 0;
85 peer->last_time_heard = jiffies;
86 peer->last_time_ecne_reduced = jiffies;
88 peer->init_sent_count = 0;
90 peer->param_flags = SPP_HB_DISABLE |
95 /* Initialize the default path max_retrans. */
96 peer->pathmaxrxt = sctp_max_retrans_path;
97 peer->error_count = 0;
99 INIT_LIST_HEAD(&peer->transmitted);
100 INIT_LIST_HEAD(&peer->send_ready);
101 INIT_LIST_HEAD(&peer->transports);
103 peer->T3_rtx_timer.expires = 0;
104 peer->hb_timer.expires = 0;
106 setup_timer(&peer->T3_rtx_timer, sctp_generate_t3_rtx_event,
107 (unsigned long)peer);
108 setup_timer(&peer->hb_timer, sctp_generate_heartbeat_event,
109 (unsigned long)peer);
111 /* Initialize the 64-bit random nonce sent with heartbeat. */
112 get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce));
114 atomic_set(&peer->refcnt, 1);
119 /* Initialize the state information for SFR-CACC */
120 peer->cacc.changeover_active = 0;
121 peer->cacc.cycling_changeover = 0;
122 peer->cacc.next_tsn_at_change = 0;
123 peer->cacc.cacc_saw_newack = 0;
128 /* Allocate and initialize a new transport. */
129 struct sctp_transport *sctp_transport_new(const union sctp_addr *addr,
132 struct sctp_transport *transport;
134 transport = t_new(struct sctp_transport, gfp);
138 if (!sctp_transport_init(transport, addr, gfp))
141 transport->malloced = 1;
142 SCTP_DBG_OBJCNT_INC(transport);
153 /* This transport is no longer needed. Free up if possible, or
154 * delay until it last reference count.
156 void sctp_transport_free(struct sctp_transport *transport)
160 /* Try to delete the heartbeat timer. */
161 if (del_timer(&transport->hb_timer))
162 sctp_transport_put(transport);
164 /* Delete the T3_rtx timer if it's active.
165 * There is no point in not doing this now and letting
166 * structure hang around in memory since we know
167 * the tranport is going away.
169 if (timer_pending(&transport->T3_rtx_timer) &&
170 del_timer(&transport->T3_rtx_timer))
171 sctp_transport_put(transport);
174 sctp_transport_put(transport);
177 /* Destroy the transport data structure.
178 * Assumes there are no more users of this structure.
180 static void sctp_transport_destroy(struct sctp_transport *transport)
182 SCTP_ASSERT(transport->dead, "Transport is not dead", return);
185 sctp_association_put(transport->asoc);
187 sctp_packet_free(&transport->packet);
189 dst_release(transport->dst);
191 SCTP_DBG_OBJCNT_DEC(transport);
194 /* Start T3_rtx timer if it is not already running and update the heartbeat
195 * timer. This routine is called every time a DATA chunk is sent.
197 void sctp_transport_reset_timers(struct sctp_transport *transport, int force)
199 /* RFC 2960 6.3.2 Retransmission Timer Rules
201 * R1) Every time a DATA chunk is sent to any address(including a
202 * retransmission), if the T3-rtx timer of that address is not running
203 * start it running so that it will expire after the RTO of that
207 if (force || !timer_pending(&transport->T3_rtx_timer))
208 if (!mod_timer(&transport->T3_rtx_timer,
209 jiffies + transport->rto))
210 sctp_transport_hold(transport);
212 /* When a data chunk is sent, reset the heartbeat interval. */
213 if (!mod_timer(&transport->hb_timer,
214 sctp_transport_timeout(transport)))
215 sctp_transport_hold(transport);
218 /* This transport has been assigned to an association.
219 * Initialize fields from the association or from the sock itself.
220 * Register the reference count in the association.
222 void sctp_transport_set_owner(struct sctp_transport *transport,
223 struct sctp_association *asoc)
225 transport->asoc = asoc;
226 sctp_association_hold(asoc);
229 /* Initialize the pmtu of a transport. */
230 void sctp_transport_pmtu(struct sctp_transport *transport)
232 struct dst_entry *dst;
234 dst = transport->af_specific->get_dst(NULL, &transport->ipaddr, NULL);
237 transport->pathmtu = dst_mtu(dst);
240 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
243 /* this is a complete rip-off from __sk_dst_check
244 * the cookie is always 0 since this is how it's used in the
247 static struct dst_entry *sctp_transport_dst_check(struct sctp_transport *t)
249 struct dst_entry *dst = t->dst;
251 if (dst && dst->obsolete && dst->ops->check(dst, 0) == NULL) {
260 void sctp_transport_update_pmtu(struct sctp_transport *t, u32 pmtu)
262 struct dst_entry *dst;
264 if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) {
265 printk(KERN_WARNING "%s: Reported pmtu %d too low, "
266 "using default minimum of %d\n",
268 SCTP_DEFAULT_MINSEGMENT);
269 /* Use default minimum segment size and disable
270 * pmtu discovery on this transport.
272 t->pathmtu = SCTP_DEFAULT_MINSEGMENT;
277 dst = sctp_transport_dst_check(t);
279 dst->ops->update_pmtu(dst, pmtu);
282 /* Caches the dst entry and source address for a transport's destination
285 void sctp_transport_route(struct sctp_transport *transport,
286 union sctp_addr *saddr, struct sctp_sock *opt)
288 struct sctp_association *asoc = transport->asoc;
289 struct sctp_af *af = transport->af_specific;
290 union sctp_addr *daddr = &transport->ipaddr;
291 struct dst_entry *dst;
293 dst = af->get_dst(asoc, daddr, saddr);
296 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
298 af->get_saddr(opt, asoc, dst, daddr, &transport->saddr);
300 transport->dst = dst;
301 if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) {
305 transport->pathmtu = dst_mtu(dst);
307 /* Initialize sk->sk_rcv_saddr, if the transport is the
308 * association's active path for getsockname().
310 if (asoc && (!asoc->peer.primary_path ||
311 (transport == asoc->peer.active_path)))
312 opt->pf->af->to_sk_saddr(&transport->saddr,
315 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
318 /* Hold a reference to a transport. */
319 void sctp_transport_hold(struct sctp_transport *transport)
321 atomic_inc(&transport->refcnt);
324 /* Release a reference to a transport and clean up
325 * if there are no more references.
327 void sctp_transport_put(struct sctp_transport *transport)
329 if (atomic_dec_and_test(&transport->refcnt))
330 sctp_transport_destroy(transport);
333 /* Update transport's RTO based on the newly calculated RTT. */
334 void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
336 /* Check for valid transport. */
337 SCTP_ASSERT(tp, "NULL transport", return);
339 /* We should not be doing any RTO updates unless rto_pending is set. */
340 SCTP_ASSERT(tp->rto_pending, "rto_pending not set", return);
342 if (tp->rttvar || tp->srtt) {
343 /* 6.3.1 C3) When a new RTT measurement R' is made, set
344 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
345 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
348 /* Note: The above algorithm has been rewritten to
349 * express rto_beta and rto_alpha as inverse powers
351 * For example, assuming the default value of RTO.Alpha of
352 * 1/8, rto_alpha would be expressed as 3.
354 tp->rttvar = tp->rttvar - (tp->rttvar >> sctp_rto_beta)
355 + ((abs(tp->srtt - rtt)) >> sctp_rto_beta);
356 tp->srtt = tp->srtt - (tp->srtt >> sctp_rto_alpha)
357 + (rtt >> sctp_rto_alpha);
359 /* 6.3.1 C2) When the first RTT measurement R is made, set
360 * SRTT <- R, RTTVAR <- R/2.
363 tp->rttvar = rtt >> 1;
366 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
367 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
370 tp->rttvar = SCTP_CLOCK_GRANULARITY;
372 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
373 tp->rto = tp->srtt + (tp->rttvar << 2);
375 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
376 * seconds then it is rounded up to RTO.Min seconds.
378 if (tp->rto < tp->asoc->rto_min)
379 tp->rto = tp->asoc->rto_min;
381 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
382 * at least RTO.max seconds.
384 if (tp->rto > tp->asoc->rto_max)
385 tp->rto = tp->asoc->rto_max;
389 /* Reset rto_pending so that a new RTT measurement is started when a
390 * new data chunk is sent.
394 SCTP_DEBUG_PRINTK("%s: transport: %p, rtt: %d, srtt: %d "
395 "rttvar: %d, rto: %ld\n", __func__,
396 tp, rtt, tp->srtt, tp->rttvar, tp->rto);
399 /* This routine updates the transport's cwnd and partial_bytes_acked
400 * parameters based on the bytes acked in the received SACK.
402 void sctp_transport_raise_cwnd(struct sctp_transport *transport,
403 __u32 sack_ctsn, __u32 bytes_acked)
405 __u32 cwnd, ssthresh, flight_size, pba, pmtu;
407 cwnd = transport->cwnd;
408 flight_size = transport->flight_size;
410 /* See if we need to exit Fast Recovery first */
411 if (transport->fast_recovery &&
412 TSN_lte(transport->fast_recovery_exit, sack_ctsn))
413 transport->fast_recovery = 0;
415 /* The appropriate cwnd increase algorithm is performed if, and only
416 * if the cumulative TSN whould advanced and the congestion window is
417 * being fully utilized.
419 if (TSN_lte(sack_ctsn, transport->asoc->ctsn_ack_point) ||
420 (flight_size < cwnd))
423 ssthresh = transport->ssthresh;
424 pba = transport->partial_bytes_acked;
425 pmtu = transport->asoc->pathmtu;
427 if (cwnd <= ssthresh) {
429 * o When cwnd is less than or equal to ssthresh, an SCTP
430 * endpoint MUST use the slow-start algorithm to increase
431 * cwnd only if the current congestion window is being fully
432 * utilized, an incoming SACK advances the Cumulative TSN
433 * Ack Point, and the data sender is not in Fast Recovery.
434 * Only when these three conditions are met can the cwnd be
435 * increased; otherwise, the cwnd MUST not be increased.
436 * If these conditions are met, then cwnd MUST be increased
437 * by, at most, the lesser of 1) the total size of the
438 * previously outstanding DATA chunk(s) acknowledged, and
439 * 2) the destination's path MTU. This upper bound protects
440 * against the ACK-Splitting attack outlined in [SAVAGE99].
442 if (transport->fast_recovery)
445 if (bytes_acked > pmtu)
449 SCTP_DEBUG_PRINTK("%s: SLOW START: transport: %p, "
450 "bytes_acked: %d, cwnd: %d, ssthresh: %d, "
451 "flight_size: %d, pba: %d\n",
453 transport, bytes_acked, cwnd,
454 ssthresh, flight_size, pba);
456 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
457 * upon each SACK arrival that advances the Cumulative TSN Ack
458 * Point, increase partial_bytes_acked by the total number of
459 * bytes of all new chunks acknowledged in that SACK including
460 * chunks acknowledged by the new Cumulative TSN Ack and by
463 * When partial_bytes_acked is equal to or greater than cwnd
464 * and before the arrival of the SACK the sender had cwnd or
465 * more bytes of data outstanding (i.e., before arrival of the
466 * SACK, flightsize was greater than or equal to cwnd),
467 * increase cwnd by MTU, and reset partial_bytes_acked to
468 * (partial_bytes_acked - cwnd).
473 pba = ((cwnd < pba) ? (pba - cwnd) : 0);
475 SCTP_DEBUG_PRINTK("%s: CONGESTION AVOIDANCE: "
476 "transport: %p, bytes_acked: %d, cwnd: %d, "
477 "ssthresh: %d, flight_size: %d, pba: %d\n",
479 transport, bytes_acked, cwnd,
480 ssthresh, flight_size, pba);
483 transport->cwnd = cwnd;
484 transport->partial_bytes_acked = pba;
487 /* This routine is used to lower the transport's cwnd when congestion is
490 void sctp_transport_lower_cwnd(struct sctp_transport *transport,
491 sctp_lower_cwnd_t reason)
494 case SCTP_LOWER_CWND_T3_RTX:
495 /* RFC 2960 Section 7.2.3, sctpimpguide
496 * When the T3-rtx timer expires on an address, SCTP should
497 * perform slow start by:
498 * ssthresh = max(cwnd/2, 4*MTU)
500 * partial_bytes_acked = 0
502 transport->ssthresh = max(transport->cwnd/2,
503 4*transport->asoc->pathmtu);
504 transport->cwnd = transport->asoc->pathmtu;
506 /* T3-rtx also clears fast recovery on the transport */
507 transport->fast_recovery = 0;
510 case SCTP_LOWER_CWND_FAST_RTX:
511 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
512 * destination address(es) to which the missing DATA chunks
513 * were last sent, according to the formula described in
516 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
517 * losses from SACK (see Section 7.2.4), An endpoint
518 * should do the following:
519 * ssthresh = max(cwnd/2, 4*MTU)
521 * partial_bytes_acked = 0
523 if (transport->fast_recovery)
526 /* Mark Fast recovery */
527 transport->fast_recovery = 1;
528 transport->fast_recovery_exit = transport->asoc->next_tsn - 1;
530 transport->ssthresh = max(transport->cwnd/2,
531 4*transport->asoc->pathmtu);
532 transport->cwnd = transport->ssthresh;
535 case SCTP_LOWER_CWND_ECNE:
536 /* RFC 2481 Section 6.1.2.
537 * If the sender receives an ECN-Echo ACK packet
538 * then the sender knows that congestion was encountered in the
539 * network on the path from the sender to the receiver. The
540 * indication of congestion should be treated just as a
541 * congestion loss in non-ECN Capable TCP. That is, the TCP
542 * source halves the congestion window "cwnd" and reduces the
543 * slow start threshold "ssthresh".
544 * A critical condition is that TCP does not react to
545 * congestion indications more than once every window of
546 * data (or more loosely more than once every round-trip time).
548 if (time_after(jiffies, transport->last_time_ecne_reduced +
550 transport->ssthresh = max(transport->cwnd/2,
551 4*transport->asoc->pathmtu);
552 transport->cwnd = transport->ssthresh;
553 transport->last_time_ecne_reduced = jiffies;
557 case SCTP_LOWER_CWND_INACTIVE:
558 /* RFC 2960 Section 7.2.1, sctpimpguide
559 * When the endpoint does not transmit data on a given
560 * transport address, the cwnd of the transport address
561 * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
562 * NOTE: Although the draft recommends that this check needs
563 * to be done every RTO interval, we do it every hearbeat
566 transport->cwnd = max(transport->cwnd/2,
567 4*transport->asoc->pathmtu);
571 transport->partial_bytes_acked = 0;
572 SCTP_DEBUG_PRINTK("%s: transport: %p reason: %d cwnd: "
573 "%d ssthresh: %d\n", __func__,
575 transport->cwnd, transport->ssthresh);
578 /* Apply Max.Burst limit to the congestion window:
579 * sctpimpguide-05 2.14.2
580 * D) When the time comes for the sender to
581 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
582 * first be applied to limit how many new DATA chunks may be sent.
583 * The limit is applied by adjusting cwnd as follows:
584 * if ((flightsize+ Max.Burst * MTU) < cwnd)
585 * cwnd = flightsize + Max.Burst * MTU
588 void sctp_transport_burst_limited(struct sctp_transport *t)
590 struct sctp_association *asoc = t->asoc;
591 u32 old_cwnd = t->cwnd;
594 if (t->burst_limited)
597 max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu);
598 if (max_burst_bytes < old_cwnd) {
599 t->cwnd = max_burst_bytes;
600 t->burst_limited = old_cwnd;
604 /* Restore the old cwnd congestion window, after the burst had it's
607 void sctp_transport_burst_reset(struct sctp_transport *t)
609 if (t->burst_limited) {
610 t->cwnd = t->burst_limited;
611 t->burst_limited = 0;
615 /* What is the next timeout value for this transport? */
616 unsigned long sctp_transport_timeout(struct sctp_transport *t)
618 unsigned long timeout;
619 timeout = t->rto + sctp_jitter(t->rto);
620 if (t->state != SCTP_UNCONFIRMED)
621 timeout += t->hbinterval;
626 /* Reset transport variables to their initial values */
627 void sctp_transport_reset(struct sctp_transport *t)
629 struct sctp_association *asoc = t->asoc;
631 /* RFC 2960 (bis), Section 5.2.4
632 * All the congestion control parameters (e.g., cwnd, ssthresh)
633 * related to this peer MUST be reset to their initial values
634 * (see Section 6.2.1)
636 t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
637 t->burst_limited = 0;
638 t->ssthresh = asoc->peer.i.a_rwnd;
639 t->rto = asoc->rto_initial;
644 /* Reset these additional varibles so that we have a clean
647 t->partial_bytes_acked = 0;
652 t->fast_recovery = 0;
654 /* Initialize the state information for SFR-CACC */
655 t->cacc.changeover_active = 0;
656 t->cacc.cycling_changeover = 0;
657 t->cacc.next_tsn_at_change = 0;
658 t->cacc.cacc_saw_newack = 0;