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, see
28 * <http://www.gnu.org/licenses/>.
30 * Please send any bug reports or fixes you make to the
32 * lksctp developers <linux-sctp@vger.kernel.org>
34 * Written or modified by:
35 * La Monte H.P. Yarroll <piggy@acm.org>
36 * Karl Knutson <karl@athena.chicago.il.us>
37 * Jon Grimm <jgrimm@us.ibm.com>
38 * Xingang Guo <xingang.guo@intel.com>
39 * Hui Huang <hui.huang@nokia.com>
40 * Sridhar Samudrala <sri@us.ibm.com>
41 * Ardelle Fan <ardelle.fan@intel.com>
44 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
46 #include <linux/slab.h>
47 #include <linux/types.h>
48 #include <linux/random.h>
49 #include <net/sctp/sctp.h>
50 #include <net/sctp/sm.h>
52 /* 1st Level Abstractions. */
54 /* Initialize a new transport from provided memory. */
55 static struct sctp_transport *sctp_transport_init(struct net *net,
56 struct sctp_transport *peer,
57 const union sctp_addr *addr,
60 /* Copy in the address. */
62 peer->af_specific = sctp_get_af_specific(addr->sa.sa_family);
63 memset(&peer->saddr, 0, sizeof(union sctp_addr));
65 peer->sack_generation = 0;
67 /* From 6.3.1 RTO Calculation:
69 * C1) Until an RTT measurement has been made for a packet sent to the
70 * given destination transport address, set RTO to the protocol
71 * parameter 'RTO.Initial'.
73 peer->rto = msecs_to_jiffies(net->sctp.rto_initial);
75 peer->last_time_heard = 0;
76 peer->last_time_ecne_reduced = jiffies;
78 peer->param_flags = SPP_HB_DISABLE |
82 /* Initialize the default path max_retrans. */
83 peer->pathmaxrxt = net->sctp.max_retrans_path;
84 peer->pf_retrans = net->sctp.pf_retrans;
86 INIT_LIST_HEAD(&peer->transmitted);
87 INIT_LIST_HEAD(&peer->send_ready);
88 INIT_LIST_HEAD(&peer->transports);
90 setup_timer(&peer->T3_rtx_timer, sctp_generate_t3_rtx_event,
92 setup_timer(&peer->hb_timer, sctp_generate_heartbeat_event,
94 setup_timer(&peer->reconf_timer, sctp_generate_reconf_event,
96 setup_timer(&peer->proto_unreach_timer,
97 sctp_generate_proto_unreach_event, (unsigned long)peer);
99 /* Initialize the 64-bit random nonce sent with heartbeat. */
100 get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce));
102 refcount_set(&peer->refcnt, 1);
107 /* Allocate and initialize a new transport. */
108 struct sctp_transport *sctp_transport_new(struct net *net,
109 const union sctp_addr *addr,
112 struct sctp_transport *transport;
114 transport = kzalloc(sizeof(*transport), gfp);
118 if (!sctp_transport_init(net, transport, addr, gfp))
121 SCTP_DBG_OBJCNT_INC(transport);
132 /* This transport is no longer needed. Free up if possible, or
133 * delay until it last reference count.
135 void sctp_transport_free(struct sctp_transport *transport)
137 /* Try to delete the heartbeat timer. */
138 if (del_timer(&transport->hb_timer))
139 sctp_transport_put(transport);
141 /* Delete the T3_rtx timer if it's active.
142 * There is no point in not doing this now and letting
143 * structure hang around in memory since we know
144 * the tranport is going away.
146 if (del_timer(&transport->T3_rtx_timer))
147 sctp_transport_put(transport);
149 if (del_timer(&transport->reconf_timer))
150 sctp_transport_put(transport);
152 /* Delete the ICMP proto unreachable timer if it's active. */
153 if (del_timer(&transport->proto_unreach_timer))
154 sctp_transport_put(transport);
156 sctp_transport_put(transport);
159 static void sctp_transport_destroy_rcu(struct rcu_head *head)
161 struct sctp_transport *transport;
163 transport = container_of(head, struct sctp_transport, rcu);
165 dst_release(transport->dst);
167 SCTP_DBG_OBJCNT_DEC(transport);
170 /* Destroy the transport data structure.
171 * Assumes there are no more users of this structure.
173 static void sctp_transport_destroy(struct sctp_transport *transport)
175 if (unlikely(refcount_read(&transport->refcnt))) {
176 WARN(1, "Attempt to destroy undead transport %p!\n", transport);
180 sctp_packet_free(&transport->packet);
183 sctp_association_put(transport->asoc);
185 call_rcu(&transport->rcu, sctp_transport_destroy_rcu);
188 /* Start T3_rtx timer if it is not already running and update the heartbeat
189 * timer. This routine is called every time a DATA chunk is sent.
191 void sctp_transport_reset_t3_rtx(struct sctp_transport *transport)
193 /* RFC 2960 6.3.2 Retransmission Timer Rules
195 * R1) Every time a DATA chunk is sent to any address(including a
196 * retransmission), if the T3-rtx timer of that address is not running
197 * start it running so that it will expire after the RTO of that
201 if (!timer_pending(&transport->T3_rtx_timer))
202 if (!mod_timer(&transport->T3_rtx_timer,
203 jiffies + transport->rto))
204 sctp_transport_hold(transport);
207 void sctp_transport_reset_hb_timer(struct sctp_transport *transport)
209 unsigned long expires;
211 /* When a data chunk is sent, reset the heartbeat interval. */
212 expires = jiffies + sctp_transport_timeout(transport);
213 if (!mod_timer(&transport->hb_timer,
214 expires + prandom_u32_max(transport->rto)))
215 sctp_transport_hold(transport);
218 void sctp_transport_reset_reconf_timer(struct sctp_transport *transport)
220 if (!timer_pending(&transport->reconf_timer))
221 if (!mod_timer(&transport->reconf_timer,
222 jiffies + transport->rto))
223 sctp_transport_hold(transport);
226 /* This transport has been assigned to an association.
227 * Initialize fields from the association or from the sock itself.
228 * Register the reference count in the association.
230 void sctp_transport_set_owner(struct sctp_transport *transport,
231 struct sctp_association *asoc)
233 transport->asoc = asoc;
234 sctp_association_hold(asoc);
237 /* Initialize the pmtu of a transport. */
238 void sctp_transport_pmtu(struct sctp_transport *transport, struct sock *sk)
240 /* If we don't have a fresh route, look one up */
241 if (!transport->dst || transport->dst->obsolete) {
242 sctp_transport_dst_release(transport);
243 transport->af_specific->get_dst(transport, &transport->saddr,
247 if (transport->dst) {
248 transport->pathmtu = SCTP_TRUNC4(dst_mtu(transport->dst));
250 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
253 bool sctp_transport_update_pmtu(struct sctp_transport *t, u32 pmtu)
255 struct dst_entry *dst = sctp_transport_dst_check(t);
256 struct sock *sk = t->asoc->base.sk;
259 if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) {
260 pr_warn_ratelimited("%s: Reported pmtu %d too low, using default minimum of %d\n",
261 __func__, pmtu, SCTP_DEFAULT_MINSEGMENT);
262 /* Use default minimum segment instead */
263 pmtu = SCTP_DEFAULT_MINSEGMENT;
265 pmtu = SCTP_TRUNC4(pmtu);
268 struct sctp_pf *pf = sctp_get_pf_specific(dst->ops->family);
269 union sctp_addr addr;
271 pf->af->from_sk(&addr, sk);
272 pf->to_sk_daddr(&t->ipaddr, sk);
273 dst->ops->update_pmtu(dst, sk, NULL, pmtu, true);
274 pf->to_sk_daddr(&addr, sk);
276 dst = sctp_transport_dst_check(t);
280 t->af_specific->get_dst(t, &t->saddr, &t->fl, sk);
285 /* Re-fetch, as under layers may have a higher minimum size */
286 pmtu = SCTP_TRUNC4(dst_mtu(dst));
287 change = t->pathmtu != pmtu;
294 /* Caches the dst entry and source address for a transport's destination
297 void sctp_transport_route(struct sctp_transport *transport,
298 union sctp_addr *saddr, struct sctp_sock *opt)
300 struct sctp_association *asoc = transport->asoc;
301 struct sctp_af *af = transport->af_specific;
303 af->get_dst(transport, saddr, &transport->fl, sctp_opt2sk(opt));
306 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
308 af->get_saddr(opt, transport, &transport->fl);
310 if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) {
313 if (transport->dst) {
314 transport->pathmtu = SCTP_TRUNC4(dst_mtu(transport->dst));
316 /* Initialize sk->sk_rcv_saddr, if the transport is the
317 * association's active path for getsockname().
319 if (asoc && (!asoc->peer.primary_path ||
320 (transport == asoc->peer.active_path)))
321 opt->pf->to_sk_saddr(&transport->saddr,
324 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
327 /* Hold a reference to a transport. */
328 int sctp_transport_hold(struct sctp_transport *transport)
330 return refcount_inc_not_zero(&transport->refcnt);
333 /* Release a reference to a transport and clean up
334 * if there are no more references.
336 void sctp_transport_put(struct sctp_transport *transport)
338 if (refcount_dec_and_test(&transport->refcnt))
339 sctp_transport_destroy(transport);
342 /* Update transport's RTO based on the newly calculated RTT. */
343 void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
345 if (unlikely(!tp->rto_pending))
346 /* We should not be doing any RTO updates unless rto_pending is set. */
347 pr_debug("%s: rto_pending not set on transport %p!\n", __func__, tp);
349 if (tp->rttvar || tp->srtt) {
350 struct net *net = sock_net(tp->asoc->base.sk);
351 /* 6.3.1 C3) When a new RTT measurement R' is made, set
352 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
353 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
356 /* Note: The above algorithm has been rewritten to
357 * express rto_beta and rto_alpha as inverse powers
359 * For example, assuming the default value of RTO.Alpha of
360 * 1/8, rto_alpha would be expressed as 3.
362 tp->rttvar = tp->rttvar - (tp->rttvar >> net->sctp.rto_beta)
363 + (((__u32)abs((__s64)tp->srtt - (__s64)rtt)) >> net->sctp.rto_beta);
364 tp->srtt = tp->srtt - (tp->srtt >> net->sctp.rto_alpha)
365 + (rtt >> net->sctp.rto_alpha);
367 /* 6.3.1 C2) When the first RTT measurement R is made, set
368 * SRTT <- R, RTTVAR <- R/2.
371 tp->rttvar = rtt >> 1;
374 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
375 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
378 tp->rttvar = SCTP_CLOCK_GRANULARITY;
380 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
381 tp->rto = tp->srtt + (tp->rttvar << 2);
383 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
384 * seconds then it is rounded up to RTO.Min seconds.
386 if (tp->rto < tp->asoc->rto_min)
387 tp->rto = tp->asoc->rto_min;
389 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
390 * at least RTO.max seconds.
392 if (tp->rto > tp->asoc->rto_max)
393 tp->rto = tp->asoc->rto_max;
395 sctp_max_rto(tp->asoc, tp);
398 /* Reset rto_pending so that a new RTT measurement is started when a
399 * new data chunk is sent.
403 pr_debug("%s: transport:%p, rtt:%d, srtt:%d rttvar:%d, rto:%ld\n",
404 __func__, tp, rtt, tp->srtt, tp->rttvar, tp->rto);
407 /* This routine updates the transport's cwnd and partial_bytes_acked
408 * parameters based on the bytes acked in the received SACK.
410 void sctp_transport_raise_cwnd(struct sctp_transport *transport,
411 __u32 sack_ctsn, __u32 bytes_acked)
413 struct sctp_association *asoc = transport->asoc;
414 __u32 cwnd, ssthresh, flight_size, pba, pmtu;
416 cwnd = transport->cwnd;
417 flight_size = transport->flight_size;
419 /* See if we need to exit Fast Recovery first */
420 if (asoc->fast_recovery &&
421 TSN_lte(asoc->fast_recovery_exit, sack_ctsn))
422 asoc->fast_recovery = 0;
424 ssthresh = transport->ssthresh;
425 pba = transport->partial_bytes_acked;
426 pmtu = transport->asoc->pathmtu;
428 if (cwnd <= ssthresh) {
430 * o When cwnd is less than or equal to ssthresh, an SCTP
431 * endpoint MUST use the slow-start algorithm to increase
432 * cwnd only if the current congestion window is being fully
433 * utilized, an incoming SACK advances the Cumulative TSN
434 * Ack Point, and the data sender is not in Fast Recovery.
435 * Only when these three conditions are met can the cwnd be
436 * increased; otherwise, the cwnd MUST not be increased.
437 * If these conditions are met, then cwnd MUST be increased
438 * by, at most, the lesser of 1) the total size of the
439 * previously outstanding DATA chunk(s) acknowledged, and
440 * 2) the destination's path MTU. This upper bound protects
441 * against the ACK-Splitting attack outlined in [SAVAGE99].
443 if (asoc->fast_recovery)
446 /* The appropriate cwnd increase algorithm is performed
447 * if, and only if the congestion window is being fully
448 * utilized. Note that RFC4960 Errata 3.22 removed the
449 * other condition on ctsn moving.
451 if (flight_size < cwnd)
454 if (bytes_acked > pmtu)
459 pr_debug("%s: slow start: transport:%p, bytes_acked:%d, "
460 "cwnd:%d, ssthresh:%d, flight_size:%d, pba:%d\n",
461 __func__, transport, bytes_acked, cwnd, ssthresh,
464 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
465 * upon each SACK arrival, increase partial_bytes_acked
466 * by the total number of bytes of all new chunks
467 * acknowledged in that SACK including chunks
468 * acknowledged by the new Cumulative TSN Ack and by Gap
469 * Ack Blocks. (updated by RFC4960 Errata 3.22)
471 * When partial_bytes_acked is greater than cwnd and
472 * before the arrival of the SACK the sender had less
473 * bytes of data outstanding than cwnd (i.e., before
474 * arrival of the SACK, flightsize was less than cwnd),
475 * reset partial_bytes_acked to cwnd. (RFC 4960 Errata
478 * When partial_bytes_acked is equal to or greater than
479 * cwnd and before the arrival of the SACK the sender
480 * had cwnd or more bytes of data outstanding (i.e.,
481 * before arrival of the SACK, flightsize was greater
482 * than or equal to cwnd), partial_bytes_acked is reset
483 * to (partial_bytes_acked - cwnd). Next, cwnd is
484 * increased by MTU. (RFC 4960 Errata 3.12)
487 if (pba > cwnd && flight_size < cwnd)
489 if (pba >= cwnd && flight_size >= cwnd) {
494 pr_debug("%s: congestion avoidance: transport:%p, "
495 "bytes_acked:%d, cwnd:%d, ssthresh:%d, "
496 "flight_size:%d, pba:%d\n", __func__,
497 transport, bytes_acked, cwnd, ssthresh,
501 transport->cwnd = cwnd;
502 transport->partial_bytes_acked = pba;
505 /* This routine is used to lower the transport's cwnd when congestion is
508 void sctp_transport_lower_cwnd(struct sctp_transport *transport,
509 enum sctp_lower_cwnd reason)
511 struct sctp_association *asoc = transport->asoc;
514 case SCTP_LOWER_CWND_T3_RTX:
515 /* RFC 2960 Section 7.2.3, sctpimpguide
516 * When the T3-rtx timer expires on an address, SCTP should
517 * perform slow start by:
518 * ssthresh = max(cwnd/2, 4*MTU)
520 * partial_bytes_acked = 0
522 transport->ssthresh = max(transport->cwnd/2,
524 transport->cwnd = asoc->pathmtu;
526 /* T3-rtx also clears fast recovery */
527 asoc->fast_recovery = 0;
530 case SCTP_LOWER_CWND_FAST_RTX:
531 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
532 * destination address(es) to which the missing DATA chunks
533 * were last sent, according to the formula described in
536 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
537 * losses from SACK (see Section 7.2.4), An endpoint
538 * should do the following:
539 * ssthresh = max(cwnd/2, 4*MTU)
541 * partial_bytes_acked = 0
543 if (asoc->fast_recovery)
546 /* Mark Fast recovery */
547 asoc->fast_recovery = 1;
548 asoc->fast_recovery_exit = asoc->next_tsn - 1;
550 transport->ssthresh = max(transport->cwnd/2,
552 transport->cwnd = transport->ssthresh;
555 case SCTP_LOWER_CWND_ECNE:
556 /* RFC 2481 Section 6.1.2.
557 * If the sender receives an ECN-Echo ACK packet
558 * then the sender knows that congestion was encountered in the
559 * network on the path from the sender to the receiver. The
560 * indication of congestion should be treated just as a
561 * congestion loss in non-ECN Capable TCP. That is, the TCP
562 * source halves the congestion window "cwnd" and reduces the
563 * slow start threshold "ssthresh".
564 * A critical condition is that TCP does not react to
565 * congestion indications more than once every window of
566 * data (or more loosely more than once every round-trip time).
568 if (time_after(jiffies, transport->last_time_ecne_reduced +
570 transport->ssthresh = max(transport->cwnd/2,
572 transport->cwnd = transport->ssthresh;
573 transport->last_time_ecne_reduced = jiffies;
577 case SCTP_LOWER_CWND_INACTIVE:
578 /* RFC 2960 Section 7.2.1, sctpimpguide
579 * When the endpoint does not transmit data on a given
580 * transport address, the cwnd of the transport address
581 * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
582 * NOTE: Although the draft recommends that this check needs
583 * to be done every RTO interval, we do it every hearbeat
586 transport->cwnd = max(transport->cwnd/2,
588 /* RFC 4960 Errata 3.27.2: also adjust sshthresh */
589 transport->ssthresh = transport->cwnd;
593 transport->partial_bytes_acked = 0;
595 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d\n",
596 __func__, transport, reason, transport->cwnd,
597 transport->ssthresh);
600 /* Apply Max.Burst limit to the congestion window:
601 * sctpimpguide-05 2.14.2
602 * D) When the time comes for the sender to
603 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
604 * first be applied to limit how many new DATA chunks may be sent.
605 * The limit is applied by adjusting cwnd as follows:
606 * if ((flightsize+ Max.Burst * MTU) < cwnd)
607 * cwnd = flightsize + Max.Burst * MTU
610 void sctp_transport_burst_limited(struct sctp_transport *t)
612 struct sctp_association *asoc = t->asoc;
613 u32 old_cwnd = t->cwnd;
616 if (t->burst_limited || asoc->max_burst == 0)
619 max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu);
620 if (max_burst_bytes < old_cwnd) {
621 t->cwnd = max_burst_bytes;
622 t->burst_limited = old_cwnd;
626 /* Restore the old cwnd congestion window, after the burst had it's
629 void sctp_transport_burst_reset(struct sctp_transport *t)
631 if (t->burst_limited) {
632 t->cwnd = t->burst_limited;
633 t->burst_limited = 0;
637 /* What is the next timeout value for this transport? */
638 unsigned long sctp_transport_timeout(struct sctp_transport *trans)
640 /* RTO + timer slack +/- 50% of RTO */
641 unsigned long timeout = trans->rto >> 1;
643 if (trans->state != SCTP_UNCONFIRMED &&
644 trans->state != SCTP_PF)
645 timeout += trans->hbinterval;
647 return max_t(unsigned long, timeout, HZ / 5);
650 /* Reset transport variables to their initial values */
651 void sctp_transport_reset(struct sctp_transport *t)
653 struct sctp_association *asoc = t->asoc;
655 /* RFC 2960 (bis), Section 5.2.4
656 * All the congestion control parameters (e.g., cwnd, ssthresh)
657 * related to this peer MUST be reset to their initial values
658 * (see Section 6.2.1)
660 t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
661 t->burst_limited = 0;
662 t->ssthresh = asoc->peer.i.a_rwnd;
663 t->rto = asoc->rto_initial;
664 sctp_max_rto(asoc, t);
669 /* Reset these additional variables so that we have a clean slate. */
670 t->partial_bytes_acked = 0;
676 /* Initialize the state information for SFR-CACC */
677 t->cacc.changeover_active = 0;
678 t->cacc.cycling_changeover = 0;
679 t->cacc.next_tsn_at_change = 0;
680 t->cacc.cacc_saw_newack = 0;
683 /* Schedule retransmission on the given transport */
684 void sctp_transport_immediate_rtx(struct sctp_transport *t)
686 /* Stop pending T3_rtx_timer */
687 if (del_timer(&t->T3_rtx_timer))
688 sctp_transport_put(t);
690 sctp_retransmit(&t->asoc->outqueue, t, SCTP_RTXR_T3_RTX);
691 if (!timer_pending(&t->T3_rtx_timer)) {
692 if (!mod_timer(&t->T3_rtx_timer, jiffies + t->rto))
693 sctp_transport_hold(t);
698 void sctp_transport_dst_release(struct sctp_transport *t)
702 t->dst_pending_confirm = 0;
705 /* Schedule neighbour confirm */
706 void sctp_transport_dst_confirm(struct sctp_transport *t)
708 t->dst_pending_confirm = 1;