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 static void sctp_transport_destroy_rcu(struct rcu_head *head)
168 struct sctp_transport *transport;
170 transport = container_of(head, struct sctp_transport, rcu);
172 sctp_association_put(transport->asoc);
174 sctp_packet_free(&transport->packet);
176 dst_release(transport->dst);
178 SCTP_DBG_OBJCNT_DEC(transport);
181 /* Destroy the transport data structure.
182 * Assumes there are no more users of this structure.
184 static void sctp_transport_destroy(struct sctp_transport *transport)
186 SCTP_ASSERT(transport->dead, "Transport is not dead", return);
188 call_rcu(&transport->rcu, sctp_transport_destroy_rcu);
191 /* Start T3_rtx timer if it is not already running and update the heartbeat
192 * timer. This routine is called every time a DATA chunk is sent.
194 void sctp_transport_reset_timers(struct sctp_transport *transport)
196 /* RFC 2960 6.3.2 Retransmission Timer Rules
198 * R1) Every time a DATA chunk is sent to any address(including a
199 * retransmission), if the T3-rtx timer of that address is not running
200 * start it running so that it will expire after the RTO of that
204 if (!timer_pending(&transport->T3_rtx_timer))
205 if (!mod_timer(&transport->T3_rtx_timer,
206 jiffies + transport->rto))
207 sctp_transport_hold(transport);
209 /* When a data chunk is sent, reset the heartbeat interval. */
210 if (!mod_timer(&transport->hb_timer,
211 sctp_transport_timeout(transport)))
212 sctp_transport_hold(transport);
215 /* This transport has been assigned to an association.
216 * Initialize fields from the association or from the sock itself.
217 * Register the reference count in the association.
219 void sctp_transport_set_owner(struct sctp_transport *transport,
220 struct sctp_association *asoc)
222 transport->asoc = asoc;
223 sctp_association_hold(asoc);
226 /* Initialize the pmtu of a transport. */
227 void sctp_transport_pmtu(struct sctp_transport *transport, struct sock *sk)
229 /* If we don't have a fresh route, look one up */
230 if (!transport->dst || transport->dst->obsolete) {
231 dst_release(transport->dst);
232 transport->af_specific->get_dst(transport, &transport->saddr,
236 if (transport->dst) {
237 transport->pathmtu = dst_mtu(transport->dst);
239 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
242 void sctp_transport_update_pmtu(struct sock *sk, struct sctp_transport *t, u32 pmtu)
244 struct dst_entry *dst;
246 if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) {
247 pr_warn("%s: Reported pmtu %d too low, using default minimum of %d\n",
249 SCTP_DEFAULT_MINSEGMENT);
250 /* Use default minimum segment size and disable
251 * pmtu discovery on this transport.
253 t->pathmtu = SCTP_DEFAULT_MINSEGMENT;
258 dst = sctp_transport_dst_check(t);
260 t->af_specific->get_dst(t, &t->saddr, &t->fl, sk);
263 dst->ops->update_pmtu(dst, sk, NULL, pmtu);
265 dst = sctp_transport_dst_check(t);
267 t->af_specific->get_dst(t, &t->saddr, &t->fl, sk);
271 /* Caches the dst entry and source address for a transport's destination
274 void sctp_transport_route(struct sctp_transport *transport,
275 union sctp_addr *saddr, struct sctp_sock *opt)
277 struct sctp_association *asoc = transport->asoc;
278 struct sctp_af *af = transport->af_specific;
280 af->get_dst(transport, saddr, &transport->fl, sctp_opt2sk(opt));
283 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
285 af->get_saddr(opt, transport, &transport->fl);
287 if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) {
290 if (transport->dst) {
291 transport->pathmtu = dst_mtu(transport->dst);
293 /* Initialize sk->sk_rcv_saddr, if the transport is the
294 * association's active path for getsockname().
296 if (asoc && (!asoc->peer.primary_path ||
297 (transport == asoc->peer.active_path)))
298 opt->pf->af->to_sk_saddr(&transport->saddr,
301 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
304 /* Hold a reference to a transport. */
305 void sctp_transport_hold(struct sctp_transport *transport)
307 atomic_inc(&transport->refcnt);
310 /* Release a reference to a transport and clean up
311 * if there are no more references.
313 void sctp_transport_put(struct sctp_transport *transport)
315 if (atomic_dec_and_test(&transport->refcnt))
316 sctp_transport_destroy(transport);
319 /* Update transport's RTO based on the newly calculated RTT. */
320 void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
322 /* Check for valid transport. */
323 SCTP_ASSERT(tp, "NULL transport", return);
325 /* We should not be doing any RTO updates unless rto_pending is set. */
326 SCTP_ASSERT(tp->rto_pending, "rto_pending not set", return);
328 if (tp->rttvar || tp->srtt) {
329 struct net *net = sock_net(tp->asoc->base.sk);
330 /* 6.3.1 C3) When a new RTT measurement R' is made, set
331 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
332 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
335 /* Note: The above algorithm has been rewritten to
336 * express rto_beta and rto_alpha as inverse powers
338 * For example, assuming the default value of RTO.Alpha of
339 * 1/8, rto_alpha would be expressed as 3.
341 tp->rttvar = tp->rttvar - (tp->rttvar >> net->sctp.rto_beta)
342 + (((__u32)abs64((__s64)tp->srtt - (__s64)rtt)) >> net->sctp.rto_beta);
343 tp->srtt = tp->srtt - (tp->srtt >> net->sctp.rto_alpha)
344 + (rtt >> net->sctp.rto_alpha);
346 /* 6.3.1 C2) When the first RTT measurement R is made, set
347 * SRTT <- R, RTTVAR <- R/2.
350 tp->rttvar = rtt >> 1;
353 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
354 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
357 tp->rttvar = SCTP_CLOCK_GRANULARITY;
359 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
360 tp->rto = tp->srtt + (tp->rttvar << 2);
362 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
363 * seconds then it is rounded up to RTO.Min seconds.
365 if (tp->rto < tp->asoc->rto_min)
366 tp->rto = tp->asoc->rto_min;
368 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
369 * at least RTO.max seconds.
371 if (tp->rto > tp->asoc->rto_max)
372 tp->rto = tp->asoc->rto_max;
374 sctp_max_rto(tp->asoc, tp);
377 /* Reset rto_pending so that a new RTT measurement is started when a
378 * new data chunk is sent.
382 SCTP_DEBUG_PRINTK("%s: transport: %p, rtt: %d, srtt: %d "
383 "rttvar: %d, rto: %ld\n", __func__,
384 tp, rtt, tp->srtt, tp->rttvar, tp->rto);
387 /* This routine updates the transport's cwnd and partial_bytes_acked
388 * parameters based on the bytes acked in the received SACK.
390 void sctp_transport_raise_cwnd(struct sctp_transport *transport,
391 __u32 sack_ctsn, __u32 bytes_acked)
393 struct sctp_association *asoc = transport->asoc;
394 __u32 cwnd, ssthresh, flight_size, pba, pmtu;
396 cwnd = transport->cwnd;
397 flight_size = transport->flight_size;
399 /* See if we need to exit Fast Recovery first */
400 if (asoc->fast_recovery &&
401 TSN_lte(asoc->fast_recovery_exit, sack_ctsn))
402 asoc->fast_recovery = 0;
404 /* The appropriate cwnd increase algorithm is performed if, and only
405 * if the cumulative TSN whould advanced and the congestion window is
406 * being fully utilized.
408 if (TSN_lte(sack_ctsn, transport->asoc->ctsn_ack_point) ||
409 (flight_size < cwnd))
412 ssthresh = transport->ssthresh;
413 pba = transport->partial_bytes_acked;
414 pmtu = transport->asoc->pathmtu;
416 if (cwnd <= ssthresh) {
418 * o When cwnd is less than or equal to ssthresh, an SCTP
419 * endpoint MUST use the slow-start algorithm to increase
420 * cwnd only if the current congestion window is being fully
421 * utilized, an incoming SACK advances the Cumulative TSN
422 * Ack Point, and the data sender is not in Fast Recovery.
423 * Only when these three conditions are met can the cwnd be
424 * increased; otherwise, the cwnd MUST not be increased.
425 * If these conditions are met, then cwnd MUST be increased
426 * by, at most, the lesser of 1) the total size of the
427 * previously outstanding DATA chunk(s) acknowledged, and
428 * 2) the destination's path MTU. This upper bound protects
429 * against the ACK-Splitting attack outlined in [SAVAGE99].
431 if (asoc->fast_recovery)
434 if (bytes_acked > pmtu)
438 SCTP_DEBUG_PRINTK("%s: SLOW START: transport: %p, "
439 "bytes_acked: %d, cwnd: %d, ssthresh: %d, "
440 "flight_size: %d, pba: %d\n",
442 transport, bytes_acked, cwnd,
443 ssthresh, flight_size, pba);
445 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
446 * upon each SACK arrival that advances the Cumulative TSN Ack
447 * Point, increase partial_bytes_acked by the total number of
448 * bytes of all new chunks acknowledged in that SACK including
449 * chunks acknowledged by the new Cumulative TSN Ack and by
452 * When partial_bytes_acked is equal to or greater than cwnd
453 * and before the arrival of the SACK the sender had cwnd or
454 * more bytes of data outstanding (i.e., before arrival of the
455 * SACK, flightsize was greater than or equal to cwnd),
456 * increase cwnd by MTU, and reset partial_bytes_acked to
457 * (partial_bytes_acked - cwnd).
462 pba = ((cwnd < pba) ? (pba - cwnd) : 0);
464 SCTP_DEBUG_PRINTK("%s: CONGESTION AVOIDANCE: "
465 "transport: %p, bytes_acked: %d, cwnd: %d, "
466 "ssthresh: %d, flight_size: %d, pba: %d\n",
468 transport, bytes_acked, cwnd,
469 ssthresh, flight_size, pba);
472 transport->cwnd = cwnd;
473 transport->partial_bytes_acked = pba;
476 /* This routine is used to lower the transport's cwnd when congestion is
479 void sctp_transport_lower_cwnd(struct sctp_transport *transport,
480 sctp_lower_cwnd_t reason)
482 struct sctp_association *asoc = transport->asoc;
485 case SCTP_LOWER_CWND_T3_RTX:
486 /* RFC 2960 Section 7.2.3, sctpimpguide
487 * When the T3-rtx timer expires on an address, SCTP should
488 * perform slow start by:
489 * ssthresh = max(cwnd/2, 4*MTU)
491 * partial_bytes_acked = 0
493 transport->ssthresh = max(transport->cwnd/2,
495 transport->cwnd = asoc->pathmtu;
497 /* T3-rtx also clears fast recovery */
498 asoc->fast_recovery = 0;
501 case SCTP_LOWER_CWND_FAST_RTX:
502 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
503 * destination address(es) to which the missing DATA chunks
504 * were last sent, according to the formula described in
507 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
508 * losses from SACK (see Section 7.2.4), An endpoint
509 * should do the following:
510 * ssthresh = max(cwnd/2, 4*MTU)
512 * partial_bytes_acked = 0
514 if (asoc->fast_recovery)
517 /* Mark Fast recovery */
518 asoc->fast_recovery = 1;
519 asoc->fast_recovery_exit = asoc->next_tsn - 1;
521 transport->ssthresh = max(transport->cwnd/2,
523 transport->cwnd = transport->ssthresh;
526 case SCTP_LOWER_CWND_ECNE:
527 /* RFC 2481 Section 6.1.2.
528 * If the sender receives an ECN-Echo ACK packet
529 * then the sender knows that congestion was encountered in the
530 * network on the path from the sender to the receiver. The
531 * indication of congestion should be treated just as a
532 * congestion loss in non-ECN Capable TCP. That is, the TCP
533 * source halves the congestion window "cwnd" and reduces the
534 * slow start threshold "ssthresh".
535 * A critical condition is that TCP does not react to
536 * congestion indications more than once every window of
537 * data (or more loosely more than once every round-trip time).
539 if (time_after(jiffies, transport->last_time_ecne_reduced +
541 transport->ssthresh = max(transport->cwnd/2,
543 transport->cwnd = transport->ssthresh;
544 transport->last_time_ecne_reduced = jiffies;
548 case SCTP_LOWER_CWND_INACTIVE:
549 /* RFC 2960 Section 7.2.1, sctpimpguide
550 * When the endpoint does not transmit data on a given
551 * transport address, the cwnd of the transport address
552 * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
553 * NOTE: Although the draft recommends that this check needs
554 * to be done every RTO interval, we do it every hearbeat
557 transport->cwnd = max(transport->cwnd/2,
562 transport->partial_bytes_acked = 0;
563 SCTP_DEBUG_PRINTK("%s: transport: %p reason: %d cwnd: "
564 "%d ssthresh: %d\n", __func__,
566 transport->cwnd, transport->ssthresh);
569 /* Apply Max.Burst limit to the congestion window:
570 * sctpimpguide-05 2.14.2
571 * D) When the time comes for the sender to
572 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
573 * first be applied to limit how many new DATA chunks may be sent.
574 * The limit is applied by adjusting cwnd as follows:
575 * if ((flightsize+ Max.Burst * MTU) < cwnd)
576 * cwnd = flightsize + Max.Burst * MTU
579 void sctp_transport_burst_limited(struct sctp_transport *t)
581 struct sctp_association *asoc = t->asoc;
582 u32 old_cwnd = t->cwnd;
585 if (t->burst_limited)
588 max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu);
589 if (max_burst_bytes < old_cwnd) {
590 t->cwnd = max_burst_bytes;
591 t->burst_limited = old_cwnd;
595 /* Restore the old cwnd congestion window, after the burst had it's
598 void sctp_transport_burst_reset(struct sctp_transport *t)
600 if (t->burst_limited) {
601 t->cwnd = t->burst_limited;
602 t->burst_limited = 0;
606 /* What is the next timeout value for this transport? */
607 unsigned long sctp_transport_timeout(struct sctp_transport *t)
609 unsigned long timeout;
610 timeout = t->rto + sctp_jitter(t->rto);
611 if ((t->state != SCTP_UNCONFIRMED) &&
612 (t->state != SCTP_PF))
613 timeout += t->hbinterval;
618 /* Reset transport variables to their initial values */
619 void sctp_transport_reset(struct sctp_transport *t)
621 struct sctp_association *asoc = t->asoc;
623 /* RFC 2960 (bis), Section 5.2.4
624 * All the congestion control parameters (e.g., cwnd, ssthresh)
625 * related to this peer MUST be reset to their initial values
626 * (see Section 6.2.1)
628 t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
629 t->burst_limited = 0;
630 t->ssthresh = asoc->peer.i.a_rwnd;
631 t->rto = asoc->rto_initial;
632 sctp_max_rto(asoc, t);
637 /* Reset these additional varibles so that we have a clean
640 t->partial_bytes_acked = 0;
646 /* Initialize the state information for SFR-CACC */
647 t->cacc.changeover_active = 0;
648 t->cacc.cycling_changeover = 0;
649 t->cacc.next_tsn_at_change = 0;
650 t->cacc.cacc_saw_newack = 0;
653 /* Schedule retransmission on the given transport */
654 void sctp_transport_immediate_rtx(struct sctp_transport *t)
656 /* Stop pending T3_rtx_timer */
657 if (timer_pending(&t->T3_rtx_timer)) {
658 (void)del_timer(&t->T3_rtx_timer);
659 sctp_transport_put(t);
661 sctp_retransmit(&t->asoc->outqueue, t, SCTP_RTXR_T3_RTX);
662 if (!timer_pending(&t->T3_rtx_timer)) {
663 if (!mod_timer(&t->T3_rtx_timer, jiffies + t->rto))
664 sctp_transport_hold(t);