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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
53 #include <linux/slab.h>
54 #include <linux/types.h>
55 #include <linux/random.h>
56 #include <net/sctp/sctp.h>
57 #include <net/sctp/sm.h>
59 /* 1st Level Abstractions. */
61 /* Initialize a new transport from provided memory. */
62 static struct sctp_transport *sctp_transport_init(struct net *net,
63 struct sctp_transport *peer,
64 const union sctp_addr *addr,
67 /* Copy in the address. */
69 peer->af_specific = sctp_get_af_specific(addr->sa.sa_family);
70 memset(&peer->saddr, 0, sizeof(union sctp_addr));
72 peer->sack_generation = 0;
74 /* From 6.3.1 RTO Calculation:
76 * C1) Until an RTT measurement has been made for a packet sent to the
77 * given destination transport address, set RTO to the protocol
78 * parameter 'RTO.Initial'.
80 peer->rto = msecs_to_jiffies(net->sctp.rto_initial);
82 peer->last_time_heard = jiffies;
83 peer->last_time_ecne_reduced = jiffies;
85 peer->param_flags = SPP_HB_DISABLE |
89 /* Initialize the default path max_retrans. */
90 peer->pathmaxrxt = net->sctp.max_retrans_path;
91 peer->pf_retrans = net->sctp.pf_retrans;
93 INIT_LIST_HEAD(&peer->transmitted);
94 INIT_LIST_HEAD(&peer->send_ready);
95 INIT_LIST_HEAD(&peer->transports);
97 setup_timer(&peer->T3_rtx_timer, sctp_generate_t3_rtx_event,
99 setup_timer(&peer->hb_timer, sctp_generate_heartbeat_event,
100 (unsigned long)peer);
101 setup_timer(&peer->proto_unreach_timer,
102 sctp_generate_proto_unreach_event, (unsigned long)peer);
104 /* Initialize the 64-bit random nonce sent with heartbeat. */
105 get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce));
107 atomic_set(&peer->refcnt, 1);
112 /* Allocate and initialize a new transport. */
113 struct sctp_transport *sctp_transport_new(struct net *net,
114 const union sctp_addr *addr,
117 struct sctp_transport *transport;
119 transport = kzalloc(sizeof(*transport), gfp);
123 if (!sctp_transport_init(net, transport, addr, gfp))
126 SCTP_DBG_OBJCNT_INC(transport);
137 /* This transport is no longer needed. Free up if possible, or
138 * delay until it last reference count.
140 void sctp_transport_free(struct sctp_transport *transport)
144 /* Try to delete the heartbeat timer. */
145 if (del_timer(&transport->hb_timer))
146 sctp_transport_put(transport);
148 /* Delete the T3_rtx timer if it's active.
149 * There is no point in not doing this now and letting
150 * structure hang around in memory since we know
151 * the tranport is going away.
153 if (del_timer(&transport->T3_rtx_timer))
154 sctp_transport_put(transport);
156 /* Delete the ICMP proto unreachable timer if it's active. */
157 if (del_timer(&transport->proto_unreach_timer))
158 sctp_association_put(transport->asoc);
160 sctp_transport_put(transport);
163 static void sctp_transport_destroy_rcu(struct rcu_head *head)
165 struct sctp_transport *transport;
167 transport = container_of(head, struct sctp_transport, rcu);
169 dst_release(transport->dst);
171 SCTP_DBG_OBJCNT_DEC(transport);
174 /* Destroy the transport data structure.
175 * Assumes there are no more users of this structure.
177 static void sctp_transport_destroy(struct sctp_transport *transport)
179 if (unlikely(!transport->dead)) {
180 WARN(1, "Attempt to destroy undead transport %p!\n", transport);
184 call_rcu(&transport->rcu, sctp_transport_destroy_rcu);
186 sctp_packet_free(&transport->packet);
189 sctp_association_put(transport->asoc);
192 /* Start T3_rtx timer if it is not already running and update the heartbeat
193 * timer. This routine is called every time a DATA chunk is sent.
195 void sctp_transport_reset_timers(struct sctp_transport *transport)
197 /* RFC 2960 6.3.2 Retransmission Timer Rules
199 * R1) Every time a DATA chunk is sent to any address(including a
200 * retransmission), if the T3-rtx timer of that address is not running
201 * start it running so that it will expire after the RTO of that
205 if (!timer_pending(&transport->T3_rtx_timer))
206 if (!mod_timer(&transport->T3_rtx_timer,
207 jiffies + transport->rto))
208 sctp_transport_hold(transport);
210 /* When a data chunk is sent, reset the heartbeat interval. */
211 if (!mod_timer(&transport->hb_timer,
212 sctp_transport_timeout(transport)))
213 sctp_transport_hold(transport);
216 /* This transport has been assigned to an association.
217 * Initialize fields from the association or from the sock itself.
218 * Register the reference count in the association.
220 void sctp_transport_set_owner(struct sctp_transport *transport,
221 struct sctp_association *asoc)
223 transport->asoc = asoc;
224 sctp_association_hold(asoc);
227 /* Initialize the pmtu of a transport. */
228 void sctp_transport_pmtu(struct sctp_transport *transport, struct sock *sk)
230 /* If we don't have a fresh route, look one up */
231 if (!transport->dst || transport->dst->obsolete) {
232 dst_release(transport->dst);
233 transport->af_specific->get_dst(transport, &transport->saddr,
237 if (transport->dst) {
238 transport->pathmtu = dst_mtu(transport->dst);
240 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
243 void sctp_transport_update_pmtu(struct sock *sk, struct sctp_transport *t, u32 pmtu)
245 struct dst_entry *dst;
247 if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) {
248 pr_warn("%s: Reported pmtu %d too low, using default minimum of %d\n",
250 SCTP_DEFAULT_MINSEGMENT);
251 /* Use default minimum segment size and disable
252 * pmtu discovery on this transport.
254 t->pathmtu = SCTP_DEFAULT_MINSEGMENT;
259 dst = sctp_transport_dst_check(t);
261 t->af_specific->get_dst(t, &t->saddr, &t->fl, sk);
264 dst->ops->update_pmtu(dst, sk, NULL, pmtu);
266 dst = sctp_transport_dst_check(t);
268 t->af_specific->get_dst(t, &t->saddr, &t->fl, sk);
272 /* Caches the dst entry and source address for a transport's destination
275 void sctp_transport_route(struct sctp_transport *transport,
276 union sctp_addr *saddr, struct sctp_sock *opt)
278 struct sctp_association *asoc = transport->asoc;
279 struct sctp_af *af = transport->af_specific;
281 af->get_dst(transport, saddr, &transport->fl, sctp_opt2sk(opt));
284 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
286 af->get_saddr(opt, transport, &transport->fl);
288 if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) {
291 if (transport->dst) {
292 transport->pathmtu = dst_mtu(transport->dst);
294 /* Initialize sk->sk_rcv_saddr, if the transport is the
295 * association's active path for getsockname().
297 if (asoc && (!asoc->peer.primary_path ||
298 (transport == asoc->peer.active_path)))
299 opt->pf->af->to_sk_saddr(&transport->saddr,
302 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
305 /* Hold a reference to a transport. */
306 void sctp_transport_hold(struct sctp_transport *transport)
308 atomic_inc(&transport->refcnt);
311 /* Release a reference to a transport and clean up
312 * if there are no more references.
314 void sctp_transport_put(struct sctp_transport *transport)
316 if (atomic_dec_and_test(&transport->refcnt))
317 sctp_transport_destroy(transport);
320 /* Update transport's RTO based on the newly calculated RTT. */
321 void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
323 if (unlikely(!tp->rto_pending))
324 /* We should not be doing any RTO updates unless rto_pending is set. */
325 pr_debug("%s: rto_pending not set on transport %p!\n", __func__, tp);
327 if (tp->rttvar || tp->srtt) {
328 struct net *net = sock_net(tp->asoc->base.sk);
329 /* 6.3.1 C3) When a new RTT measurement R' is made, set
330 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
331 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
334 /* Note: The above algorithm has been rewritten to
335 * express rto_beta and rto_alpha as inverse powers
337 * For example, assuming the default value of RTO.Alpha of
338 * 1/8, rto_alpha would be expressed as 3.
340 tp->rttvar = tp->rttvar - (tp->rttvar >> net->sctp.rto_beta)
341 + (((__u32)abs64((__s64)tp->srtt - (__s64)rtt)) >> net->sctp.rto_beta);
342 tp->srtt = tp->srtt - (tp->srtt >> net->sctp.rto_alpha)
343 + (rtt >> net->sctp.rto_alpha);
345 /* 6.3.1 C2) When the first RTT measurement R is made, set
346 * SRTT <- R, RTTVAR <- R/2.
349 tp->rttvar = rtt >> 1;
352 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
353 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
356 tp->rttvar = SCTP_CLOCK_GRANULARITY;
358 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
359 tp->rto = tp->srtt + (tp->rttvar << 2);
361 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
362 * seconds then it is rounded up to RTO.Min seconds.
364 if (tp->rto < tp->asoc->rto_min)
365 tp->rto = tp->asoc->rto_min;
367 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
368 * at least RTO.max seconds.
370 if (tp->rto > tp->asoc->rto_max)
371 tp->rto = tp->asoc->rto_max;
373 sctp_max_rto(tp->asoc, tp);
376 /* Reset rto_pending so that a new RTT measurement is started when a
377 * new data chunk is sent.
381 pr_debug("%s: transport:%p, rtt:%d, srtt:%d rttvar:%d, rto:%ld\n",
382 __func__, tp, rtt, tp->srtt, tp->rttvar, tp->rto);
385 /* This routine updates the transport's cwnd and partial_bytes_acked
386 * parameters based on the bytes acked in the received SACK.
388 void sctp_transport_raise_cwnd(struct sctp_transport *transport,
389 __u32 sack_ctsn, __u32 bytes_acked)
391 struct sctp_association *asoc = transport->asoc;
392 __u32 cwnd, ssthresh, flight_size, pba, pmtu;
394 cwnd = transport->cwnd;
395 flight_size = transport->flight_size;
397 /* See if we need to exit Fast Recovery first */
398 if (asoc->fast_recovery &&
399 TSN_lte(asoc->fast_recovery_exit, sack_ctsn))
400 asoc->fast_recovery = 0;
402 /* The appropriate cwnd increase algorithm is performed if, and only
403 * if the cumulative TSN whould advanced and the congestion window is
404 * being fully utilized.
406 if (TSN_lte(sack_ctsn, transport->asoc->ctsn_ack_point) ||
407 (flight_size < cwnd))
410 ssthresh = transport->ssthresh;
411 pba = transport->partial_bytes_acked;
412 pmtu = transport->asoc->pathmtu;
414 if (cwnd <= ssthresh) {
416 * o When cwnd is less than or equal to ssthresh, an SCTP
417 * endpoint MUST use the slow-start algorithm to increase
418 * cwnd only if the current congestion window is being fully
419 * utilized, an incoming SACK advances the Cumulative TSN
420 * Ack Point, and the data sender is not in Fast Recovery.
421 * Only when these three conditions are met can the cwnd be
422 * increased; otherwise, the cwnd MUST not be increased.
423 * If these conditions are met, then cwnd MUST be increased
424 * by, at most, the lesser of 1) the total size of the
425 * previously outstanding DATA chunk(s) acknowledged, and
426 * 2) the destination's path MTU. This upper bound protects
427 * against the ACK-Splitting attack outlined in [SAVAGE99].
429 if (asoc->fast_recovery)
432 if (bytes_acked > pmtu)
437 pr_debug("%s: slow start: transport:%p, bytes_acked:%d, "
438 "cwnd:%d, ssthresh:%d, flight_size:%d, pba:%d\n",
439 __func__, transport, bytes_acked, cwnd, ssthresh,
442 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
443 * upon each SACK arrival that advances the Cumulative TSN Ack
444 * Point, increase partial_bytes_acked by the total number of
445 * bytes of all new chunks acknowledged in that SACK including
446 * chunks acknowledged by the new Cumulative TSN Ack and by
449 * When partial_bytes_acked is equal to or greater than cwnd
450 * and before the arrival of the SACK the sender had cwnd or
451 * more bytes of data outstanding (i.e., before arrival of the
452 * SACK, flightsize was greater than or equal to cwnd),
453 * increase cwnd by MTU, and reset partial_bytes_acked to
454 * (partial_bytes_acked - cwnd).
459 pba = ((cwnd < pba) ? (pba - cwnd) : 0);
462 pr_debug("%s: congestion avoidance: transport:%p, "
463 "bytes_acked:%d, cwnd:%d, ssthresh:%d, "
464 "flight_size:%d, pba:%d\n", __func__,
465 transport, bytes_acked, cwnd, ssthresh,
469 transport->cwnd = cwnd;
470 transport->partial_bytes_acked = pba;
473 /* This routine is used to lower the transport's cwnd when congestion is
476 void sctp_transport_lower_cwnd(struct sctp_transport *transport,
477 sctp_lower_cwnd_t reason)
479 struct sctp_association *asoc = transport->asoc;
482 case SCTP_LOWER_CWND_T3_RTX:
483 /* RFC 2960 Section 7.2.3, sctpimpguide
484 * When the T3-rtx timer expires on an address, SCTP should
485 * perform slow start by:
486 * ssthresh = max(cwnd/2, 4*MTU)
488 * partial_bytes_acked = 0
490 transport->ssthresh = max(transport->cwnd/2,
492 transport->cwnd = asoc->pathmtu;
494 /* T3-rtx also clears fast recovery */
495 asoc->fast_recovery = 0;
498 case SCTP_LOWER_CWND_FAST_RTX:
499 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
500 * destination address(es) to which the missing DATA chunks
501 * were last sent, according to the formula described in
504 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
505 * losses from SACK (see Section 7.2.4), An endpoint
506 * should do the following:
507 * ssthresh = max(cwnd/2, 4*MTU)
509 * partial_bytes_acked = 0
511 if (asoc->fast_recovery)
514 /* Mark Fast recovery */
515 asoc->fast_recovery = 1;
516 asoc->fast_recovery_exit = asoc->next_tsn - 1;
518 transport->ssthresh = max(transport->cwnd/2,
520 transport->cwnd = transport->ssthresh;
523 case SCTP_LOWER_CWND_ECNE:
524 /* RFC 2481 Section 6.1.2.
525 * If the sender receives an ECN-Echo ACK packet
526 * then the sender knows that congestion was encountered in the
527 * network on the path from the sender to the receiver. The
528 * indication of congestion should be treated just as a
529 * congestion loss in non-ECN Capable TCP. That is, the TCP
530 * source halves the congestion window "cwnd" and reduces the
531 * slow start threshold "ssthresh".
532 * A critical condition is that TCP does not react to
533 * congestion indications more than once every window of
534 * data (or more loosely more than once every round-trip time).
536 if (time_after(jiffies, transport->last_time_ecne_reduced +
538 transport->ssthresh = max(transport->cwnd/2,
540 transport->cwnd = transport->ssthresh;
541 transport->last_time_ecne_reduced = jiffies;
545 case SCTP_LOWER_CWND_INACTIVE:
546 /* RFC 2960 Section 7.2.1, sctpimpguide
547 * When the endpoint does not transmit data on a given
548 * transport address, the cwnd of the transport address
549 * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
550 * NOTE: Although the draft recommends that this check needs
551 * to be done every RTO interval, we do it every hearbeat
554 transport->cwnd = max(transport->cwnd/2,
559 transport->partial_bytes_acked = 0;
561 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d\n",
562 __func__, transport, reason, transport->cwnd,
563 transport->ssthresh);
566 /* Apply Max.Burst limit to the congestion window:
567 * sctpimpguide-05 2.14.2
568 * D) When the time comes for the sender to
569 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
570 * first be applied to limit how many new DATA chunks may be sent.
571 * The limit is applied by adjusting cwnd as follows:
572 * if ((flightsize+ Max.Burst * MTU) < cwnd)
573 * cwnd = flightsize + Max.Burst * MTU
576 void sctp_transport_burst_limited(struct sctp_transport *t)
578 struct sctp_association *asoc = t->asoc;
579 u32 old_cwnd = t->cwnd;
582 if (t->burst_limited)
585 max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu);
586 if (max_burst_bytes < old_cwnd) {
587 t->cwnd = max_burst_bytes;
588 t->burst_limited = old_cwnd;
592 /* Restore the old cwnd congestion window, after the burst had it's
595 void sctp_transport_burst_reset(struct sctp_transport *t)
597 if (t->burst_limited) {
598 t->cwnd = t->burst_limited;
599 t->burst_limited = 0;
603 /* What is the next timeout value for this transport? */
604 unsigned long sctp_transport_timeout(struct sctp_transport *t)
606 unsigned long timeout;
607 timeout = t->rto + sctp_jitter(t->rto);
608 if ((t->state != SCTP_UNCONFIRMED) &&
609 (t->state != SCTP_PF))
610 timeout += t->hbinterval;
615 /* Reset transport variables to their initial values */
616 void sctp_transport_reset(struct sctp_transport *t)
618 struct sctp_association *asoc = t->asoc;
620 /* RFC 2960 (bis), Section 5.2.4
621 * All the congestion control parameters (e.g., cwnd, ssthresh)
622 * related to this peer MUST be reset to their initial values
623 * (see Section 6.2.1)
625 t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
626 t->burst_limited = 0;
627 t->ssthresh = asoc->peer.i.a_rwnd;
628 t->rto = asoc->rto_initial;
629 sctp_max_rto(asoc, t);
634 /* Reset these additional varibles so that we have a clean
637 t->partial_bytes_acked = 0;
643 /* Initialize the state information for SFR-CACC */
644 t->cacc.changeover_active = 0;
645 t->cacc.cycling_changeover = 0;
646 t->cacc.next_tsn_at_change = 0;
647 t->cacc.cacc_saw_newack = 0;
650 /* Schedule retransmission on the given transport */
651 void sctp_transport_immediate_rtx(struct sctp_transport *t)
653 /* Stop pending T3_rtx_timer */
654 if (del_timer(&t->T3_rtx_timer))
655 sctp_transport_put(t);
657 sctp_retransmit(&t->asoc->outqueue, t, SCTP_RTXR_T3_RTX);
658 if (!timer_pending(&t->T3_rtx_timer)) {
659 if (!mod_timer(&t->T3_rtx_timer, jiffies + t->rto))
660 sctp_transport_hold(t);