1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* SCTP kernel implementation
3 * Copyright (c) 1999-2000 Cisco, Inc.
4 * Copyright (c) 1999-2001 Motorola, Inc.
5 * Copyright (c) 2001-2003 International Business Machines Corp.
6 * Copyright (c) 2001 Intel Corp.
7 * Copyright (c) 2001 La Monte H.P. Yarroll
9 * This file is part of the SCTP kernel implementation
11 * This module provides the abstraction for an SCTP transport representing
12 * a remote transport address. For local transport addresses, we just use
15 * Please send any bug reports or fixes you make to the
17 * lksctp developers <linux-sctp@vger.kernel.org>
19 * Written or modified by:
20 * La Monte H.P. Yarroll <piggy@acm.org>
21 * Karl Knutson <karl@athena.chicago.il.us>
22 * Jon Grimm <jgrimm@us.ibm.com>
23 * Xingang Guo <xingang.guo@intel.com>
24 * Hui Huang <hui.huang@nokia.com>
25 * Sridhar Samudrala <sri@us.ibm.com>
26 * Ardelle Fan <ardelle.fan@intel.com>
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include <linux/slab.h>
32 #include <linux/types.h>
33 #include <linux/random.h>
34 #include <net/sctp/sctp.h>
35 #include <net/sctp/sm.h>
37 /* 1st Level Abstractions. */
39 /* Initialize a new transport from provided memory. */
40 static struct sctp_transport *sctp_transport_init(struct net *net,
41 struct sctp_transport *peer,
42 const union sctp_addr *addr,
45 /* Copy in the address. */
46 peer->af_specific = sctp_get_af_specific(addr->sa.sa_family);
47 memcpy(&peer->ipaddr, addr, peer->af_specific->sockaddr_len);
48 memset(&peer->saddr, 0, sizeof(union sctp_addr));
50 peer->sack_generation = 0;
52 /* From 6.3.1 RTO Calculation:
54 * C1) Until an RTT measurement has been made for a packet sent to the
55 * given destination transport address, set RTO to the protocol
56 * parameter 'RTO.Initial'.
58 peer->rto = msecs_to_jiffies(net->sctp.rto_initial);
60 peer->last_time_heard = 0;
61 peer->last_time_ecne_reduced = jiffies;
63 peer->param_flags = SPP_HB_DISABLE |
67 /* Initialize the default path max_retrans. */
68 peer->pathmaxrxt = net->sctp.max_retrans_path;
69 peer->pf_retrans = net->sctp.pf_retrans;
71 INIT_LIST_HEAD(&peer->transmitted);
72 INIT_LIST_HEAD(&peer->send_ready);
73 INIT_LIST_HEAD(&peer->transports);
75 timer_setup(&peer->T3_rtx_timer, sctp_generate_t3_rtx_event, 0);
76 timer_setup(&peer->hb_timer, sctp_generate_heartbeat_event, 0);
77 timer_setup(&peer->reconf_timer, sctp_generate_reconf_event, 0);
78 timer_setup(&peer->probe_timer, sctp_generate_probe_event, 0);
79 timer_setup(&peer->proto_unreach_timer,
80 sctp_generate_proto_unreach_event, 0);
82 /* Initialize the 64-bit random nonce sent with heartbeat. */
83 get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce));
85 refcount_set(&peer->refcnt, 1);
90 /* Allocate and initialize a new transport. */
91 struct sctp_transport *sctp_transport_new(struct net *net,
92 const union sctp_addr *addr,
95 struct sctp_transport *transport;
97 transport = kzalloc(sizeof(*transport), gfp);
101 if (!sctp_transport_init(net, transport, addr, gfp))
104 SCTP_DBG_OBJCNT_INC(transport);
115 /* This transport is no longer needed. Free up if possible, or
116 * delay until it last reference count.
118 void sctp_transport_free(struct sctp_transport *transport)
120 /* Try to delete the heartbeat timer. */
121 if (del_timer(&transport->hb_timer))
122 sctp_transport_put(transport);
124 /* Delete the T3_rtx timer if it's active.
125 * There is no point in not doing this now and letting
126 * structure hang around in memory since we know
127 * the transport is going away.
129 if (del_timer(&transport->T3_rtx_timer))
130 sctp_transport_put(transport);
132 if (del_timer(&transport->reconf_timer))
133 sctp_transport_put(transport);
135 if (del_timer(&transport->probe_timer))
136 sctp_transport_put(transport);
138 /* Delete the ICMP proto unreachable timer if it's active. */
139 if (del_timer(&transport->proto_unreach_timer))
140 sctp_transport_put(transport);
142 sctp_transport_put(transport);
145 static void sctp_transport_destroy_rcu(struct rcu_head *head)
147 struct sctp_transport *transport;
149 transport = container_of(head, struct sctp_transport, rcu);
151 dst_release(transport->dst);
153 SCTP_DBG_OBJCNT_DEC(transport);
156 /* Destroy the transport data structure.
157 * Assumes there are no more users of this structure.
159 static void sctp_transport_destroy(struct sctp_transport *transport)
161 if (unlikely(refcount_read(&transport->refcnt))) {
162 WARN(1, "Attempt to destroy undead transport %p!\n", transport);
166 sctp_packet_free(&transport->packet);
169 sctp_association_put(transport->asoc);
171 call_rcu(&transport->rcu, sctp_transport_destroy_rcu);
174 /* Start T3_rtx timer if it is not already running and update the heartbeat
175 * timer. This routine is called every time a DATA chunk is sent.
177 void sctp_transport_reset_t3_rtx(struct sctp_transport *transport)
179 /* RFC 2960 6.3.2 Retransmission Timer Rules
181 * R1) Every time a DATA chunk is sent to any address(including a
182 * retransmission), if the T3-rtx timer of that address is not running
183 * start it running so that it will expire after the RTO of that
187 if (!timer_pending(&transport->T3_rtx_timer))
188 if (!mod_timer(&transport->T3_rtx_timer,
189 jiffies + transport->rto))
190 sctp_transport_hold(transport);
193 void sctp_transport_reset_hb_timer(struct sctp_transport *transport)
195 unsigned long expires;
197 /* When a data chunk is sent, reset the heartbeat interval. */
198 expires = jiffies + sctp_transport_timeout(transport);
199 if ((time_before(transport->hb_timer.expires, expires) ||
200 !timer_pending(&transport->hb_timer)) &&
201 !mod_timer(&transport->hb_timer,
202 expires + prandom_u32_max(transport->rto)))
203 sctp_transport_hold(transport);
206 void sctp_transport_reset_reconf_timer(struct sctp_transport *transport)
208 if (!timer_pending(&transport->reconf_timer))
209 if (!mod_timer(&transport->reconf_timer,
210 jiffies + transport->rto))
211 sctp_transport_hold(transport);
214 void sctp_transport_reset_probe_timer(struct sctp_transport *transport)
216 if (timer_pending(&transport->probe_timer))
218 if (!mod_timer(&transport->probe_timer,
219 jiffies + transport->probe_interval))
220 sctp_transport_hold(transport);
223 /* This transport has been assigned to an association.
224 * Initialize fields from the association or from the sock itself.
225 * Register the reference count in the association.
227 void sctp_transport_set_owner(struct sctp_transport *transport,
228 struct sctp_association *asoc)
230 transport->asoc = asoc;
231 sctp_association_hold(asoc);
234 /* Initialize the pmtu of a transport. */
235 void sctp_transport_pmtu(struct sctp_transport *transport, struct sock *sk)
237 /* If we don't have a fresh route, look one up */
238 if (!transport->dst || transport->dst->obsolete) {
239 sctp_transport_dst_release(transport);
240 transport->af_specific->get_dst(transport, &transport->saddr,
244 if (transport->param_flags & SPP_PMTUD_DISABLE) {
245 struct sctp_association *asoc = transport->asoc;
247 if (!transport->pathmtu && asoc && asoc->pathmtu)
248 transport->pathmtu = asoc->pathmtu;
249 if (transport->pathmtu)
254 transport->pathmtu = sctp_dst_mtu(transport->dst);
256 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
258 sctp_transport_pl_update(transport);
261 void sctp_transport_pl_send(struct sctp_transport *t)
263 pr_debug("%s: PLPMTUD: transport: %p, state: %d, pmtu: %d, size: %d, high: %d\n",
264 __func__, t, t->pl.state, t->pl.pmtu, t->pl.probe_size, t->pl.probe_high);
266 if (t->pl.probe_count < SCTP_MAX_PROBES) {
271 if (t->pl.state == SCTP_PL_BASE) {
272 if (t->pl.probe_size == SCTP_BASE_PLPMTU) { /* BASE_PLPMTU Confirmation Failed */
273 t->pl.state = SCTP_PL_ERROR; /* Base -> Error */
275 t->pl.pmtu = SCTP_MIN_PLPMTU;
276 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
277 sctp_assoc_sync_pmtu(t->asoc);
279 } else if (t->pl.state == SCTP_PL_SEARCH) {
280 if (t->pl.pmtu == t->pl.probe_size) { /* Black Hole Detected */
281 t->pl.state = SCTP_PL_BASE; /* Search -> Base */
282 t->pl.probe_size = SCTP_BASE_PLPMTU;
283 t->pl.probe_high = 0;
285 t->pl.pmtu = SCTP_BASE_PLPMTU;
286 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
287 sctp_assoc_sync_pmtu(t->asoc);
288 } else { /* Normal probe failure. */
289 t->pl.probe_high = t->pl.probe_size;
290 t->pl.probe_size = t->pl.pmtu;
292 } else if (t->pl.state == SCTP_PL_COMPLETE) {
293 if (t->pl.pmtu == t->pl.probe_size) { /* Black Hole Detected */
294 t->pl.state = SCTP_PL_BASE; /* Search Complete -> Base */
295 t->pl.probe_size = SCTP_BASE_PLPMTU;
297 t->pl.pmtu = SCTP_BASE_PLPMTU;
298 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
299 sctp_assoc_sync_pmtu(t->asoc);
302 t->pl.probe_count = 1;
305 void sctp_transport_pl_recv(struct sctp_transport *t)
307 pr_debug("%s: PLPMTUD: transport: %p, state: %d, pmtu: %d, size: %d, high: %d\n",
308 __func__, t, t->pl.state, t->pl.pmtu, t->pl.probe_size, t->pl.probe_high);
310 t->pl.pmtu = t->pl.probe_size;
311 t->pl.probe_count = 0;
312 if (t->pl.state == SCTP_PL_BASE) {
313 t->pl.state = SCTP_PL_SEARCH; /* Base -> Search */
314 t->pl.probe_size += SCTP_PL_BIG_STEP;
315 } else if (t->pl.state == SCTP_PL_ERROR) {
316 t->pl.state = SCTP_PL_SEARCH; /* Error -> Search */
318 t->pl.pmtu = t->pl.probe_size;
319 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
320 sctp_assoc_sync_pmtu(t->asoc);
321 t->pl.probe_size += SCTP_PL_BIG_STEP;
322 } else if (t->pl.state == SCTP_PL_SEARCH) {
323 if (!t->pl.probe_high) {
324 t->pl.probe_size = min(t->pl.probe_size + SCTP_PL_BIG_STEP,
328 t->pl.probe_size += SCTP_PL_MIN_STEP;
329 if (t->pl.probe_size >= t->pl.probe_high) {
330 t->pl.probe_high = 0;
331 t->pl.raise_count = 0;
332 t->pl.state = SCTP_PL_COMPLETE; /* Search -> Search Complete */
334 t->pl.probe_size = t->pl.pmtu;
335 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
336 sctp_assoc_sync_pmtu(t->asoc);
338 } else if (t->pl.state == SCTP_PL_COMPLETE) {
340 if (t->pl.raise_count == 30) {
341 /* Raise probe_size again after 30 * interval in Search Complete */
342 t->pl.state = SCTP_PL_SEARCH; /* Search Complete -> Search */
343 t->pl.probe_size += SCTP_PL_MIN_STEP;
348 static bool sctp_transport_pl_toobig(struct sctp_transport *t, u32 pmtu)
350 pr_debug("%s: PLPMTUD: transport: %p, state: %d, pmtu: %d, size: %d, ptb: %d\n",
351 __func__, t, t->pl.state, t->pl.pmtu, t->pl.probe_size, pmtu);
353 if (pmtu < SCTP_MIN_PLPMTU || pmtu >= t->pl.probe_size)
356 if (t->pl.state == SCTP_PL_BASE) {
357 if (pmtu >= SCTP_MIN_PLPMTU && pmtu < SCTP_BASE_PLPMTU) {
358 t->pl.state = SCTP_PL_ERROR; /* Base -> Error */
360 t->pl.pmtu = SCTP_MIN_PLPMTU;
361 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
363 } else if (t->pl.state == SCTP_PL_SEARCH) {
364 if (pmtu >= SCTP_BASE_PLPMTU && pmtu < t->pl.pmtu) {
365 t->pl.state = SCTP_PL_BASE; /* Search -> Base */
366 t->pl.probe_size = SCTP_BASE_PLPMTU;
367 t->pl.probe_count = 0;
369 t->pl.probe_high = 0;
370 t->pl.pmtu = SCTP_BASE_PLPMTU;
371 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
372 } else if (pmtu > t->pl.pmtu && pmtu < t->pl.probe_size) {
373 t->pl.probe_size = pmtu;
374 t->pl.probe_count = 0;
378 } else if (t->pl.state == SCTP_PL_COMPLETE) {
379 if (pmtu >= SCTP_BASE_PLPMTU && pmtu < t->pl.pmtu) {
380 t->pl.state = SCTP_PL_BASE; /* Complete -> Base */
381 t->pl.probe_size = SCTP_BASE_PLPMTU;
382 t->pl.probe_count = 0;
384 t->pl.probe_high = 0;
385 t->pl.pmtu = SCTP_BASE_PLPMTU;
386 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
393 bool sctp_transport_update_pmtu(struct sctp_transport *t, u32 pmtu)
395 struct sock *sk = t->asoc->base.sk;
396 struct dst_entry *dst;
399 if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) {
400 pr_warn_ratelimited("%s: Reported pmtu %d too low, using default minimum of %d\n",
401 __func__, pmtu, SCTP_DEFAULT_MINSEGMENT);
402 /* Use default minimum segment instead */
403 pmtu = SCTP_DEFAULT_MINSEGMENT;
405 pmtu = SCTP_TRUNC4(pmtu);
407 if (sctp_transport_pl_enabled(t))
408 return sctp_transport_pl_toobig(t, pmtu - sctp_transport_pl_hlen(t));
410 dst = sctp_transport_dst_check(t);
412 struct sctp_pf *pf = sctp_get_pf_specific(dst->ops->family);
413 union sctp_addr addr;
415 pf->af->from_sk(&addr, sk);
416 pf->to_sk_daddr(&t->ipaddr, sk);
417 dst->ops->update_pmtu(dst, sk, NULL, pmtu, true);
418 pf->to_sk_daddr(&addr, sk);
420 dst = sctp_transport_dst_check(t);
424 t->af_specific->get_dst(t, &t->saddr, &t->fl, sk);
429 /* Re-fetch, as under layers may have a higher minimum size */
430 pmtu = sctp_dst_mtu(dst);
431 change = t->pathmtu != pmtu;
438 /* Caches the dst entry and source address for a transport's destination
441 void sctp_transport_route(struct sctp_transport *transport,
442 union sctp_addr *saddr, struct sctp_sock *opt)
444 struct sctp_association *asoc = transport->asoc;
445 struct sctp_af *af = transport->af_specific;
447 sctp_transport_dst_release(transport);
448 af->get_dst(transport, saddr, &transport->fl, sctp_opt2sk(opt));
451 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
453 af->get_saddr(opt, transport, &transport->fl);
455 sctp_transport_pmtu(transport, sctp_opt2sk(opt));
457 /* Initialize sk->sk_rcv_saddr, if the transport is the
458 * association's active path for getsockname().
460 if (transport->dst && asoc &&
461 (!asoc->peer.primary_path || transport == asoc->peer.active_path))
462 opt->pf->to_sk_saddr(&transport->saddr, asoc->base.sk);
465 /* Hold a reference to a transport. */
466 int sctp_transport_hold(struct sctp_transport *transport)
468 return refcount_inc_not_zero(&transport->refcnt);
471 /* Release a reference to a transport and clean up
472 * if there are no more references.
474 void sctp_transport_put(struct sctp_transport *transport)
476 if (refcount_dec_and_test(&transport->refcnt))
477 sctp_transport_destroy(transport);
480 /* Update transport's RTO based on the newly calculated RTT. */
481 void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
483 if (unlikely(!tp->rto_pending))
484 /* We should not be doing any RTO updates unless rto_pending is set. */
485 pr_debug("%s: rto_pending not set on transport %p!\n", __func__, tp);
487 if (tp->rttvar || tp->srtt) {
488 struct net *net = tp->asoc->base.net;
489 /* 6.3.1 C3) When a new RTT measurement R' is made, set
490 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
491 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
494 /* Note: The above algorithm has been rewritten to
495 * express rto_beta and rto_alpha as inverse powers
497 * For example, assuming the default value of RTO.Alpha of
498 * 1/8, rto_alpha would be expressed as 3.
500 tp->rttvar = tp->rttvar - (tp->rttvar >> net->sctp.rto_beta)
501 + (((__u32)abs((__s64)tp->srtt - (__s64)rtt)) >> net->sctp.rto_beta);
502 tp->srtt = tp->srtt - (tp->srtt >> net->sctp.rto_alpha)
503 + (rtt >> net->sctp.rto_alpha);
505 /* 6.3.1 C2) When the first RTT measurement R is made, set
506 * SRTT <- R, RTTVAR <- R/2.
509 tp->rttvar = rtt >> 1;
512 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
513 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
516 tp->rttvar = SCTP_CLOCK_GRANULARITY;
518 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
519 tp->rto = tp->srtt + (tp->rttvar << 2);
521 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
522 * seconds then it is rounded up to RTO.Min seconds.
524 if (tp->rto < tp->asoc->rto_min)
525 tp->rto = tp->asoc->rto_min;
527 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
528 * at least RTO.max seconds.
530 if (tp->rto > tp->asoc->rto_max)
531 tp->rto = tp->asoc->rto_max;
533 sctp_max_rto(tp->asoc, tp);
536 /* Reset rto_pending so that a new RTT measurement is started when a
537 * new data chunk is sent.
541 pr_debug("%s: transport:%p, rtt:%d, srtt:%d rttvar:%d, rto:%ld\n",
542 __func__, tp, rtt, tp->srtt, tp->rttvar, tp->rto);
545 /* This routine updates the transport's cwnd and partial_bytes_acked
546 * parameters based on the bytes acked in the received SACK.
548 void sctp_transport_raise_cwnd(struct sctp_transport *transport,
549 __u32 sack_ctsn, __u32 bytes_acked)
551 struct sctp_association *asoc = transport->asoc;
552 __u32 cwnd, ssthresh, flight_size, pba, pmtu;
554 cwnd = transport->cwnd;
555 flight_size = transport->flight_size;
557 /* See if we need to exit Fast Recovery first */
558 if (asoc->fast_recovery &&
559 TSN_lte(asoc->fast_recovery_exit, sack_ctsn))
560 asoc->fast_recovery = 0;
562 ssthresh = transport->ssthresh;
563 pba = transport->partial_bytes_acked;
564 pmtu = transport->asoc->pathmtu;
566 if (cwnd <= ssthresh) {
568 * o When cwnd is less than or equal to ssthresh, an SCTP
569 * endpoint MUST use the slow-start algorithm to increase
570 * cwnd only if the current congestion window is being fully
571 * utilized, an incoming SACK advances the Cumulative TSN
572 * Ack Point, and the data sender is not in Fast Recovery.
573 * Only when these three conditions are met can the cwnd be
574 * increased; otherwise, the cwnd MUST not be increased.
575 * If these conditions are met, then cwnd MUST be increased
576 * by, at most, the lesser of 1) the total size of the
577 * previously outstanding DATA chunk(s) acknowledged, and
578 * 2) the destination's path MTU. This upper bound protects
579 * against the ACK-Splitting attack outlined in [SAVAGE99].
581 if (asoc->fast_recovery)
584 /* The appropriate cwnd increase algorithm is performed
585 * if, and only if the congestion window is being fully
586 * utilized. Note that RFC4960 Errata 3.22 removed the
587 * other condition on ctsn moving.
589 if (flight_size < cwnd)
592 if (bytes_acked > pmtu)
597 pr_debug("%s: slow start: transport:%p, bytes_acked:%d, "
598 "cwnd:%d, ssthresh:%d, flight_size:%d, pba:%d\n",
599 __func__, transport, bytes_acked, cwnd, ssthresh,
602 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
603 * upon each SACK arrival, increase partial_bytes_acked
604 * by the total number of bytes of all new chunks
605 * acknowledged in that SACK including chunks
606 * acknowledged by the new Cumulative TSN Ack and by Gap
607 * Ack Blocks. (updated by RFC4960 Errata 3.22)
609 * When partial_bytes_acked is greater than cwnd and
610 * before the arrival of the SACK the sender had less
611 * bytes of data outstanding than cwnd (i.e., before
612 * arrival of the SACK, flightsize was less than cwnd),
613 * reset partial_bytes_acked to cwnd. (RFC 4960 Errata
616 * When partial_bytes_acked is equal to or greater than
617 * cwnd and before the arrival of the SACK the sender
618 * had cwnd or more bytes of data outstanding (i.e.,
619 * before arrival of the SACK, flightsize was greater
620 * than or equal to cwnd), partial_bytes_acked is reset
621 * to (partial_bytes_acked - cwnd). Next, cwnd is
622 * increased by MTU. (RFC 4960 Errata 3.12)
625 if (pba > cwnd && flight_size < cwnd)
627 if (pba >= cwnd && flight_size >= cwnd) {
632 pr_debug("%s: congestion avoidance: transport:%p, "
633 "bytes_acked:%d, cwnd:%d, ssthresh:%d, "
634 "flight_size:%d, pba:%d\n", __func__,
635 transport, bytes_acked, cwnd, ssthresh,
639 transport->cwnd = cwnd;
640 transport->partial_bytes_acked = pba;
643 /* This routine is used to lower the transport's cwnd when congestion is
646 void sctp_transport_lower_cwnd(struct sctp_transport *transport,
647 enum sctp_lower_cwnd reason)
649 struct sctp_association *asoc = transport->asoc;
652 case SCTP_LOWER_CWND_T3_RTX:
653 /* RFC 2960 Section 7.2.3, sctpimpguide
654 * When the T3-rtx timer expires on an address, SCTP should
655 * perform slow start by:
656 * ssthresh = max(cwnd/2, 4*MTU)
658 * partial_bytes_acked = 0
660 transport->ssthresh = max(transport->cwnd/2,
662 transport->cwnd = asoc->pathmtu;
664 /* T3-rtx also clears fast recovery */
665 asoc->fast_recovery = 0;
668 case SCTP_LOWER_CWND_FAST_RTX:
669 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
670 * destination address(es) to which the missing DATA chunks
671 * were last sent, according to the formula described in
674 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
675 * losses from SACK (see Section 7.2.4), An endpoint
676 * should do the following:
677 * ssthresh = max(cwnd/2, 4*MTU)
679 * partial_bytes_acked = 0
681 if (asoc->fast_recovery)
684 /* Mark Fast recovery */
685 asoc->fast_recovery = 1;
686 asoc->fast_recovery_exit = asoc->next_tsn - 1;
688 transport->ssthresh = max(transport->cwnd/2,
690 transport->cwnd = transport->ssthresh;
693 case SCTP_LOWER_CWND_ECNE:
694 /* RFC 2481 Section 6.1.2.
695 * If the sender receives an ECN-Echo ACK packet
696 * then the sender knows that congestion was encountered in the
697 * network on the path from the sender to the receiver. The
698 * indication of congestion should be treated just as a
699 * congestion loss in non-ECN Capable TCP. That is, the TCP
700 * source halves the congestion window "cwnd" and reduces the
701 * slow start threshold "ssthresh".
702 * A critical condition is that TCP does not react to
703 * congestion indications more than once every window of
704 * data (or more loosely more than once every round-trip time).
706 if (time_after(jiffies, transport->last_time_ecne_reduced +
708 transport->ssthresh = max(transport->cwnd/2,
710 transport->cwnd = transport->ssthresh;
711 transport->last_time_ecne_reduced = jiffies;
715 case SCTP_LOWER_CWND_INACTIVE:
716 /* RFC 2960 Section 7.2.1, sctpimpguide
717 * When the endpoint does not transmit data on a given
718 * transport address, the cwnd of the transport address
719 * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
720 * NOTE: Although the draft recommends that this check needs
721 * to be done every RTO interval, we do it every hearbeat
724 transport->cwnd = max(transport->cwnd/2,
726 /* RFC 4960 Errata 3.27.2: also adjust sshthresh */
727 transport->ssthresh = transport->cwnd;
731 transport->partial_bytes_acked = 0;
733 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d\n",
734 __func__, transport, reason, transport->cwnd,
735 transport->ssthresh);
738 /* Apply Max.Burst limit to the congestion window:
739 * sctpimpguide-05 2.14.2
740 * D) When the time comes for the sender to
741 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
742 * first be applied to limit how many new DATA chunks may be sent.
743 * The limit is applied by adjusting cwnd as follows:
744 * if ((flightsize+ Max.Burst * MTU) < cwnd)
745 * cwnd = flightsize + Max.Burst * MTU
748 void sctp_transport_burst_limited(struct sctp_transport *t)
750 struct sctp_association *asoc = t->asoc;
751 u32 old_cwnd = t->cwnd;
754 if (t->burst_limited || asoc->max_burst == 0)
757 max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu);
758 if (max_burst_bytes < old_cwnd) {
759 t->cwnd = max_burst_bytes;
760 t->burst_limited = old_cwnd;
764 /* Restore the old cwnd congestion window, after the burst had it's
767 void sctp_transport_burst_reset(struct sctp_transport *t)
769 if (t->burst_limited) {
770 t->cwnd = t->burst_limited;
771 t->burst_limited = 0;
775 /* What is the next timeout value for this transport? */
776 unsigned long sctp_transport_timeout(struct sctp_transport *trans)
778 /* RTO + timer slack +/- 50% of RTO */
779 unsigned long timeout = trans->rto >> 1;
781 if (trans->state != SCTP_UNCONFIRMED &&
782 trans->state != SCTP_PF)
783 timeout += trans->hbinterval;
785 return max_t(unsigned long, timeout, HZ / 5);
788 /* Reset transport variables to their initial values */
789 void sctp_transport_reset(struct sctp_transport *t)
791 struct sctp_association *asoc = t->asoc;
793 /* RFC 2960 (bis), Section 5.2.4
794 * All the congestion control parameters (e.g., cwnd, ssthresh)
795 * related to this peer MUST be reset to their initial values
796 * (see Section 6.2.1)
798 t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
799 t->burst_limited = 0;
800 t->ssthresh = asoc->peer.i.a_rwnd;
801 t->rto = asoc->rto_initial;
802 sctp_max_rto(asoc, t);
807 /* Reset these additional variables so that we have a clean slate. */
808 t->partial_bytes_acked = 0;
814 /* Initialize the state information for SFR-CACC */
815 t->cacc.changeover_active = 0;
816 t->cacc.cycling_changeover = 0;
817 t->cacc.next_tsn_at_change = 0;
818 t->cacc.cacc_saw_newack = 0;
821 /* Schedule retransmission on the given transport */
822 void sctp_transport_immediate_rtx(struct sctp_transport *t)
824 /* Stop pending T3_rtx_timer */
825 if (del_timer(&t->T3_rtx_timer))
826 sctp_transport_put(t);
828 sctp_retransmit(&t->asoc->outqueue, t, SCTP_RTXR_T3_RTX);
829 if (!timer_pending(&t->T3_rtx_timer)) {
830 if (!mod_timer(&t->T3_rtx_timer, jiffies + t->rto))
831 sctp_transport_hold(t);
836 void sctp_transport_dst_release(struct sctp_transport *t)
840 t->dst_pending_confirm = 0;
843 /* Schedule neighbour confirm */
844 void sctp_transport_dst_confirm(struct sctp_transport *t)
846 t->dst_pending_confirm = 1;