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 = t_new(struct sctp_transport, gfp);
123 if (!sctp_transport_init(net, transport, addr, gfp))
126 transport->malloced = 1;
127 SCTP_DBG_OBJCNT_INC(transport);
138 /* This transport is no longer needed. Free up if possible, or
139 * delay until it last reference count.
141 void sctp_transport_free(struct sctp_transport *transport)
145 /* Try to delete the heartbeat timer. */
146 if (del_timer(&transport->hb_timer))
147 sctp_transport_put(transport);
149 /* Delete the T3_rtx timer if it's active.
150 * There is no point in not doing this now and letting
151 * structure hang around in memory since we know
152 * the tranport is going away.
154 if (timer_pending(&transport->T3_rtx_timer) &&
155 del_timer(&transport->T3_rtx_timer))
156 sctp_transport_put(transport);
158 /* Delete the ICMP proto unreachable timer if it's active. */
159 if (timer_pending(&transport->proto_unreach_timer) &&
160 del_timer(&transport->proto_unreach_timer))
161 sctp_association_put(transport->asoc);
163 sctp_transport_put(transport);
166 /* Destroy the transport data structure.
167 * Assumes there are no more users of this structure.
169 static void sctp_transport_destroy(struct sctp_transport *transport)
171 SCTP_ASSERT(transport->dead, "Transport is not dead", return);
174 sctp_association_put(transport->asoc);
176 sctp_packet_free(&transport->packet);
178 dst_release(transport->dst);
180 SCTP_DBG_OBJCNT_DEC(transport);
183 /* Start T3_rtx timer if it is not already running and update the heartbeat
184 * timer. This routine is called every time a DATA chunk is sent.
186 void sctp_transport_reset_timers(struct sctp_transport *transport)
188 /* RFC 2960 6.3.2 Retransmission Timer Rules
190 * R1) Every time a DATA chunk is sent to any address(including a
191 * retransmission), if the T3-rtx timer of that address is not running
192 * start it running so that it will expire after the RTO of that
196 if (!timer_pending(&transport->T3_rtx_timer))
197 if (!mod_timer(&transport->T3_rtx_timer,
198 jiffies + transport->rto))
199 sctp_transport_hold(transport);
201 /* When a data chunk is sent, reset the heartbeat interval. */
202 if (!mod_timer(&transport->hb_timer,
203 sctp_transport_timeout(transport)))
204 sctp_transport_hold(transport);
207 /* This transport has been assigned to an association.
208 * Initialize fields from the association or from the sock itself.
209 * Register the reference count in the association.
211 void sctp_transport_set_owner(struct sctp_transport *transport,
212 struct sctp_association *asoc)
214 transport->asoc = asoc;
215 sctp_association_hold(asoc);
218 /* Initialize the pmtu of a transport. */
219 void sctp_transport_pmtu(struct sctp_transport *transport, struct sock *sk)
221 /* If we don't have a fresh route, look one up */
222 if (!transport->dst || transport->dst->obsolete) {
223 dst_release(transport->dst);
224 transport->af_specific->get_dst(transport, &transport->saddr,
228 if (transport->dst) {
229 transport->pathmtu = dst_mtu(transport->dst);
231 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
234 void sctp_transport_update_pmtu(struct sock *sk, struct sctp_transport *t, u32 pmtu)
236 struct dst_entry *dst;
238 if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) {
239 pr_warn("%s: Reported pmtu %d too low, using default minimum of %d\n",
241 SCTP_DEFAULT_MINSEGMENT);
242 /* Use default minimum segment size and disable
243 * pmtu discovery on this transport.
245 t->pathmtu = SCTP_DEFAULT_MINSEGMENT;
250 dst = sctp_transport_dst_check(t);
252 t->af_specific->get_dst(t, &t->saddr, &t->fl, sk);
255 dst->ops->update_pmtu(dst, sk, NULL, pmtu);
257 dst = sctp_transport_dst_check(t);
259 t->af_specific->get_dst(t, &t->saddr, &t->fl, sk);
263 /* Caches the dst entry and source address for a transport's destination
266 void sctp_transport_route(struct sctp_transport *transport,
267 union sctp_addr *saddr, struct sctp_sock *opt)
269 struct sctp_association *asoc = transport->asoc;
270 struct sctp_af *af = transport->af_specific;
272 af->get_dst(transport, saddr, &transport->fl, sctp_opt2sk(opt));
275 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
277 af->get_saddr(opt, transport, &transport->fl);
279 if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) {
282 if (transport->dst) {
283 transport->pathmtu = dst_mtu(transport->dst);
285 /* Initialize sk->sk_rcv_saddr, if the transport is the
286 * association's active path for getsockname().
288 if (asoc && (!asoc->peer.primary_path ||
289 (transport == asoc->peer.active_path)))
290 opt->pf->af->to_sk_saddr(&transport->saddr,
293 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
296 /* Hold a reference to a transport. */
297 void sctp_transport_hold(struct sctp_transport *transport)
299 atomic_inc(&transport->refcnt);
302 /* Release a reference to a transport and clean up
303 * if there are no more references.
305 void sctp_transport_put(struct sctp_transport *transport)
307 if (atomic_dec_and_test(&transport->refcnt))
308 sctp_transport_destroy(transport);
311 /* Update transport's RTO based on the newly calculated RTT. */
312 void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
314 /* Check for valid transport. */
315 SCTP_ASSERT(tp, "NULL transport", return);
317 /* We should not be doing any RTO updates unless rto_pending is set. */
318 SCTP_ASSERT(tp->rto_pending, "rto_pending not set", return);
320 if (tp->rttvar || tp->srtt) {
321 struct net *net = sock_net(tp->asoc->base.sk);
322 /* 6.3.1 C3) When a new RTT measurement R' is made, set
323 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
324 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
327 /* Note: The above algorithm has been rewritten to
328 * express rto_beta and rto_alpha as inverse powers
330 * For example, assuming the default value of RTO.Alpha of
331 * 1/8, rto_alpha would be expressed as 3.
333 tp->rttvar = tp->rttvar - (tp->rttvar >> net->sctp.rto_beta)
334 + ((abs(tp->srtt - rtt)) >> net->sctp.rto_beta);
335 tp->srtt = tp->srtt - (tp->srtt >> net->sctp.rto_alpha)
336 + (rtt >> net->sctp.rto_alpha);
338 /* 6.3.1 C2) When the first RTT measurement R is made, set
339 * SRTT <- R, RTTVAR <- R/2.
342 tp->rttvar = rtt >> 1;
345 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
346 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
349 tp->rttvar = SCTP_CLOCK_GRANULARITY;
351 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
352 tp->rto = tp->srtt + (tp->rttvar << 2);
354 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
355 * seconds then it is rounded up to RTO.Min seconds.
357 if (tp->rto < tp->asoc->rto_min)
358 tp->rto = tp->asoc->rto_min;
360 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
361 * at least RTO.max seconds.
363 if (tp->rto > tp->asoc->rto_max)
364 tp->rto = tp->asoc->rto_max;
368 /* Reset rto_pending so that a new RTT measurement is started when a
369 * new data chunk is sent.
373 SCTP_DEBUG_PRINTK("%s: transport: %p, rtt: %d, srtt: %d "
374 "rttvar: %d, rto: %ld\n", __func__,
375 tp, rtt, tp->srtt, tp->rttvar, tp->rto);
378 /* This routine updates the transport's cwnd and partial_bytes_acked
379 * parameters based on the bytes acked in the received SACK.
381 void sctp_transport_raise_cwnd(struct sctp_transport *transport,
382 __u32 sack_ctsn, __u32 bytes_acked)
384 struct sctp_association *asoc = transport->asoc;
385 __u32 cwnd, ssthresh, flight_size, pba, pmtu;
387 cwnd = transport->cwnd;
388 flight_size = transport->flight_size;
390 /* See if we need to exit Fast Recovery first */
391 if (asoc->fast_recovery &&
392 TSN_lte(asoc->fast_recovery_exit, sack_ctsn))
393 asoc->fast_recovery = 0;
395 /* The appropriate cwnd increase algorithm is performed if, and only
396 * if the cumulative TSN whould advanced and the congestion window is
397 * being fully utilized.
399 if (TSN_lte(sack_ctsn, transport->asoc->ctsn_ack_point) ||
400 (flight_size < cwnd))
403 ssthresh = transport->ssthresh;
404 pba = transport->partial_bytes_acked;
405 pmtu = transport->asoc->pathmtu;
407 if (cwnd <= ssthresh) {
409 * o When cwnd is less than or equal to ssthresh, an SCTP
410 * endpoint MUST use the slow-start algorithm to increase
411 * cwnd only if the current congestion window is being fully
412 * utilized, an incoming SACK advances the Cumulative TSN
413 * Ack Point, and the data sender is not in Fast Recovery.
414 * Only when these three conditions are met can the cwnd be
415 * increased; otherwise, the cwnd MUST not be increased.
416 * If these conditions are met, then cwnd MUST be increased
417 * by, at most, the lesser of 1) the total size of the
418 * previously outstanding DATA chunk(s) acknowledged, and
419 * 2) the destination's path MTU. This upper bound protects
420 * against the ACK-Splitting attack outlined in [SAVAGE99].
422 if (asoc->fast_recovery)
425 if (bytes_acked > pmtu)
429 SCTP_DEBUG_PRINTK("%s: SLOW START: transport: %p, "
430 "bytes_acked: %d, cwnd: %d, ssthresh: %d, "
431 "flight_size: %d, pba: %d\n",
433 transport, bytes_acked, cwnd,
434 ssthresh, flight_size, pba);
436 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
437 * upon each SACK arrival that advances the Cumulative TSN Ack
438 * Point, increase partial_bytes_acked by the total number of
439 * bytes of all new chunks acknowledged in that SACK including
440 * chunks acknowledged by the new Cumulative TSN Ack and by
443 * When partial_bytes_acked is equal to or greater than cwnd
444 * and before the arrival of the SACK the sender had cwnd or
445 * more bytes of data outstanding (i.e., before arrival of the
446 * SACK, flightsize was greater than or equal to cwnd),
447 * increase cwnd by MTU, and reset partial_bytes_acked to
448 * (partial_bytes_acked - cwnd).
453 pba = ((cwnd < pba) ? (pba - cwnd) : 0);
455 SCTP_DEBUG_PRINTK("%s: CONGESTION AVOIDANCE: "
456 "transport: %p, bytes_acked: %d, cwnd: %d, "
457 "ssthresh: %d, flight_size: %d, pba: %d\n",
459 transport, bytes_acked, cwnd,
460 ssthresh, flight_size, pba);
463 transport->cwnd = cwnd;
464 transport->partial_bytes_acked = pba;
467 /* This routine is used to lower the transport's cwnd when congestion is
470 void sctp_transport_lower_cwnd(struct sctp_transport *transport,
471 sctp_lower_cwnd_t reason)
473 struct sctp_association *asoc = transport->asoc;
476 case SCTP_LOWER_CWND_T3_RTX:
477 /* RFC 2960 Section 7.2.3, sctpimpguide
478 * When the T3-rtx timer expires on an address, SCTP should
479 * perform slow start by:
480 * ssthresh = max(cwnd/2, 4*MTU)
482 * partial_bytes_acked = 0
484 transport->ssthresh = max(transport->cwnd/2,
486 transport->cwnd = asoc->pathmtu;
488 /* T3-rtx also clears fast recovery */
489 asoc->fast_recovery = 0;
492 case SCTP_LOWER_CWND_FAST_RTX:
493 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
494 * destination address(es) to which the missing DATA chunks
495 * were last sent, according to the formula described in
498 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
499 * losses from SACK (see Section 7.2.4), An endpoint
500 * should do the following:
501 * ssthresh = max(cwnd/2, 4*MTU)
503 * partial_bytes_acked = 0
505 if (asoc->fast_recovery)
508 /* Mark Fast recovery */
509 asoc->fast_recovery = 1;
510 asoc->fast_recovery_exit = asoc->next_tsn - 1;
512 transport->ssthresh = max(transport->cwnd/2,
514 transport->cwnd = transport->ssthresh;
517 case SCTP_LOWER_CWND_ECNE:
518 /* RFC 2481 Section 6.1.2.
519 * If the sender receives an ECN-Echo ACK packet
520 * then the sender knows that congestion was encountered in the
521 * network on the path from the sender to the receiver. The
522 * indication of congestion should be treated just as a
523 * congestion loss in non-ECN Capable TCP. That is, the TCP
524 * source halves the congestion window "cwnd" and reduces the
525 * slow start threshold "ssthresh".
526 * A critical condition is that TCP does not react to
527 * congestion indications more than once every window of
528 * data (or more loosely more than once every round-trip time).
530 if (time_after(jiffies, transport->last_time_ecne_reduced +
532 transport->ssthresh = max(transport->cwnd/2,
534 transport->cwnd = transport->ssthresh;
535 transport->last_time_ecne_reduced = jiffies;
539 case SCTP_LOWER_CWND_INACTIVE:
540 /* RFC 2960 Section 7.2.1, sctpimpguide
541 * When the endpoint does not transmit data on a given
542 * transport address, the cwnd of the transport address
543 * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
544 * NOTE: Although the draft recommends that this check needs
545 * to be done every RTO interval, we do it every hearbeat
548 transport->cwnd = max(transport->cwnd/2,
553 transport->partial_bytes_acked = 0;
554 SCTP_DEBUG_PRINTK("%s: transport: %p reason: %d cwnd: "
555 "%d ssthresh: %d\n", __func__,
557 transport->cwnd, transport->ssthresh);
560 /* Apply Max.Burst limit to the congestion window:
561 * sctpimpguide-05 2.14.2
562 * D) When the time comes for the sender to
563 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
564 * first be applied to limit how many new DATA chunks may be sent.
565 * The limit is applied by adjusting cwnd as follows:
566 * if ((flightsize+ Max.Burst * MTU) < cwnd)
567 * cwnd = flightsize + Max.Burst * MTU
570 void sctp_transport_burst_limited(struct sctp_transport *t)
572 struct sctp_association *asoc = t->asoc;
573 u32 old_cwnd = t->cwnd;
576 if (t->burst_limited)
579 max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu);
580 if (max_burst_bytes < old_cwnd) {
581 t->cwnd = max_burst_bytes;
582 t->burst_limited = old_cwnd;
586 /* Restore the old cwnd congestion window, after the burst had it's
589 void sctp_transport_burst_reset(struct sctp_transport *t)
591 if (t->burst_limited) {
592 t->cwnd = t->burst_limited;
593 t->burst_limited = 0;
597 /* What is the next timeout value for this transport? */
598 unsigned long sctp_transport_timeout(struct sctp_transport *t)
600 unsigned long timeout;
601 timeout = t->rto + sctp_jitter(t->rto);
602 if ((t->state != SCTP_UNCONFIRMED) &&
603 (t->state != SCTP_PF))
604 timeout += t->hbinterval;
609 /* Reset transport variables to their initial values */
610 void sctp_transport_reset(struct sctp_transport *t)
612 struct sctp_association *asoc = t->asoc;
614 /* RFC 2960 (bis), Section 5.2.4
615 * All the congestion control parameters (e.g., cwnd, ssthresh)
616 * related to this peer MUST be reset to their initial values
617 * (see Section 6.2.1)
619 t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
620 t->burst_limited = 0;
621 t->ssthresh = asoc->peer.i.a_rwnd;
622 t->rto = asoc->rto_initial;
627 /* Reset these additional varibles so that we have a clean
630 t->partial_bytes_acked = 0;
636 /* Initialize the state information for SFR-CACC */
637 t->cacc.changeover_active = 0;
638 t->cacc.cycling_changeover = 0;
639 t->cacc.next_tsn_at_change = 0;
640 t->cacc.cacc_saw_newack = 0;
643 /* Schedule retransmission on the given transport */
644 void sctp_transport_immediate_rtx(struct sctp_transport *t)
646 /* Stop pending T3_rtx_timer */
647 if (timer_pending(&t->T3_rtx_timer)) {
648 (void)del_timer(&t->T3_rtx_timer);
649 sctp_transport_put(t);
651 sctp_retransmit(&t->asoc->outqueue, t, SCTP_RTXR_T3_RTX);
652 if (!timer_pending(&t->T3_rtx_timer)) {
653 if (!mod_timer(&t->T3_rtx_timer, jiffies + t->rto))
654 sctp_transport_hold(t);