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 #include <linux/slab.h>
52 #include <linux/types.h>
53 #include <linux/random.h>
54 #include <net/sctp/sctp.h>
55 #include <net/sctp/sm.h>
57 /* 1st Level Abstractions. */
59 /* Initialize a new transport from provided memory. */
60 static struct sctp_transport *sctp_transport_init(struct sctp_transport *peer,
61 const union sctp_addr *addr,
64 /* Copy in the address. */
66 peer->af_specific = sctp_get_af_specific(addr->sa.sa_family);
67 memset(&peer->saddr, 0, sizeof(union sctp_addr));
69 /* From 6.3.1 RTO Calculation:
71 * C1) Until an RTT measurement has been made for a packet sent to the
72 * given destination transport address, set RTO to the protocol
73 * parameter 'RTO.Initial'.
75 peer->rto = msecs_to_jiffies(sctp_rto_initial);
77 peer->last_time_heard = jiffies;
78 peer->last_time_ecne_reduced = jiffies;
80 peer->param_flags = SPP_HB_DISABLE |
84 /* Initialize the default path max_retrans. */
85 peer->pathmaxrxt = sctp_max_retrans_path;
87 INIT_LIST_HEAD(&peer->transmitted);
88 INIT_LIST_HEAD(&peer->send_ready);
89 INIT_LIST_HEAD(&peer->transports);
91 setup_timer(&peer->T3_rtx_timer, sctp_generate_t3_rtx_event,
93 setup_timer(&peer->hb_timer, sctp_generate_heartbeat_event,
95 setup_timer(&peer->proto_unreach_timer,
96 sctp_generate_proto_unreach_event, (unsigned long)peer);
98 /* Initialize the 64-bit random nonce sent with heartbeat. */
99 get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce));
101 atomic_set(&peer->refcnt, 1);
106 /* Allocate and initialize a new transport. */
107 struct sctp_transport *sctp_transport_new(const union sctp_addr *addr,
110 struct sctp_transport *transport;
112 transport = t_new(struct sctp_transport, gfp);
116 if (!sctp_transport_init(transport, addr, gfp))
119 transport->malloced = 1;
120 SCTP_DBG_OBJCNT_INC(transport);
131 /* This transport is no longer needed. Free up if possible, or
132 * delay until it last reference count.
134 void sctp_transport_free(struct sctp_transport *transport)
138 /* Try to delete the heartbeat timer. */
139 if (del_timer(&transport->hb_timer))
140 sctp_transport_put(transport);
142 /* Delete the T3_rtx timer if it's active.
143 * There is no point in not doing this now and letting
144 * structure hang around in memory since we know
145 * the tranport is going away.
147 if (timer_pending(&transport->T3_rtx_timer) &&
148 del_timer(&transport->T3_rtx_timer))
149 sctp_transport_put(transport);
152 sctp_transport_put(transport);
155 /* Destroy the transport data structure.
156 * Assumes there are no more users of this structure.
158 static void sctp_transport_destroy(struct sctp_transport *transport)
160 SCTP_ASSERT(transport->dead, "Transport is not dead", return);
163 sctp_association_put(transport->asoc);
165 sctp_packet_free(&transport->packet);
167 dst_release(transport->dst);
169 SCTP_DBG_OBJCNT_DEC(transport);
172 /* Start T3_rtx timer if it is not already running and update the heartbeat
173 * timer. This routine is called every time a DATA chunk is sent.
175 void sctp_transport_reset_timers(struct sctp_transport *transport)
177 /* RFC 2960 6.3.2 Retransmission Timer Rules
179 * R1) Every time a DATA chunk is sent to any address(including a
180 * retransmission), if the T3-rtx timer of that address is not running
181 * start it running so that it will expire after the RTO of that
185 if (!timer_pending(&transport->T3_rtx_timer))
186 if (!mod_timer(&transport->T3_rtx_timer,
187 jiffies + transport->rto))
188 sctp_transport_hold(transport);
190 /* When a data chunk is sent, reset the heartbeat interval. */
191 if (!mod_timer(&transport->hb_timer,
192 sctp_transport_timeout(transport)))
193 sctp_transport_hold(transport);
196 /* This transport has been assigned to an association.
197 * Initialize fields from the association or from the sock itself.
198 * Register the reference count in the association.
200 void sctp_transport_set_owner(struct sctp_transport *transport,
201 struct sctp_association *asoc)
203 transport->asoc = asoc;
204 sctp_association_hold(asoc);
207 /* Initialize the pmtu of a transport. */
208 void sctp_transport_pmtu(struct sctp_transport *transport)
210 struct dst_entry *dst;
212 dst = transport->af_specific->get_dst(NULL, &transport->ipaddr, NULL);
215 transport->pathmtu = dst_mtu(dst);
218 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
221 /* this is a complete rip-off from __sk_dst_check
222 * the cookie is always 0 since this is how it's used in the
225 static struct dst_entry *sctp_transport_dst_check(struct sctp_transport *t)
227 struct dst_entry *dst = t->dst;
229 if (dst && dst->obsolete && dst->ops->check(dst, 0) == NULL) {
238 void sctp_transport_update_pmtu(struct sctp_transport *t, u32 pmtu)
240 struct dst_entry *dst;
242 if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) {
243 printk(KERN_WARNING "%s: Reported pmtu %d too low, "
244 "using default minimum of %d\n",
246 SCTP_DEFAULT_MINSEGMENT);
247 /* Use default minimum segment size and disable
248 * pmtu discovery on this transport.
250 t->pathmtu = SCTP_DEFAULT_MINSEGMENT;
255 dst = sctp_transport_dst_check(t);
257 dst->ops->update_pmtu(dst, pmtu);
260 /* Caches the dst entry and source address for a transport's destination
263 void sctp_transport_route(struct sctp_transport *transport,
264 union sctp_addr *saddr, struct sctp_sock *opt)
266 struct sctp_association *asoc = transport->asoc;
267 struct sctp_af *af = transport->af_specific;
268 union sctp_addr *daddr = &transport->ipaddr;
269 struct dst_entry *dst;
271 dst = af->get_dst(asoc, daddr, saddr);
274 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
276 af->get_saddr(opt, asoc, dst, daddr, &transport->saddr);
278 transport->dst = dst;
279 if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) {
283 transport->pathmtu = dst_mtu(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 /* 6.3.1 C3) When a new RTT measurement R' is made, set
322 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
323 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
326 /* Note: The above algorithm has been rewritten to
327 * express rto_beta and rto_alpha as inverse powers
329 * For example, assuming the default value of RTO.Alpha of
330 * 1/8, rto_alpha would be expressed as 3.
332 tp->rttvar = tp->rttvar - (tp->rttvar >> sctp_rto_beta)
333 + ((abs(tp->srtt - rtt)) >> sctp_rto_beta);
334 tp->srtt = tp->srtt - (tp->srtt >> sctp_rto_alpha)
335 + (rtt >> sctp_rto_alpha);
337 /* 6.3.1 C2) When the first RTT measurement R is made, set
338 * SRTT <- R, RTTVAR <- R/2.
341 tp->rttvar = rtt >> 1;
344 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
345 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
348 tp->rttvar = SCTP_CLOCK_GRANULARITY;
350 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
351 tp->rto = tp->srtt + (tp->rttvar << 2);
353 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
354 * seconds then it is rounded up to RTO.Min seconds.
356 if (tp->rto < tp->asoc->rto_min)
357 tp->rto = tp->asoc->rto_min;
359 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
360 * at least RTO.max seconds.
362 if (tp->rto > tp->asoc->rto_max)
363 tp->rto = tp->asoc->rto_max;
367 /* Reset rto_pending so that a new RTT measurement is started when a
368 * new data chunk is sent.
372 SCTP_DEBUG_PRINTK("%s: transport: %p, rtt: %d, srtt: %d "
373 "rttvar: %d, rto: %ld\n", __func__,
374 tp, rtt, tp->srtt, tp->rttvar, tp->rto);
377 /* This routine updates the transport's cwnd and partial_bytes_acked
378 * parameters based on the bytes acked in the received SACK.
380 void sctp_transport_raise_cwnd(struct sctp_transport *transport,
381 __u32 sack_ctsn, __u32 bytes_acked)
383 struct sctp_association *asoc = transport->asoc;
384 __u32 cwnd, ssthresh, flight_size, pba, pmtu;
386 cwnd = transport->cwnd;
387 flight_size = transport->flight_size;
389 /* See if we need to exit Fast Recovery first */
390 if (asoc->fast_recovery &&
391 TSN_lte(asoc->fast_recovery_exit, sack_ctsn))
392 asoc->fast_recovery = 0;
394 /* The appropriate cwnd increase algorithm is performed if, and only
395 * if the cumulative TSN whould advanced and the congestion window is
396 * being fully utilized.
398 if (TSN_lte(sack_ctsn, transport->asoc->ctsn_ack_point) ||
399 (flight_size < cwnd))
402 ssthresh = transport->ssthresh;
403 pba = transport->partial_bytes_acked;
404 pmtu = transport->asoc->pathmtu;
406 if (cwnd <= ssthresh) {
408 * o When cwnd is less than or equal to ssthresh, an SCTP
409 * endpoint MUST use the slow-start algorithm to increase
410 * cwnd only if the current congestion window is being fully
411 * utilized, an incoming SACK advances the Cumulative TSN
412 * Ack Point, and the data sender is not in Fast Recovery.
413 * Only when these three conditions are met can the cwnd be
414 * increased; otherwise, the cwnd MUST not be increased.
415 * If these conditions are met, then cwnd MUST be increased
416 * by, at most, the lesser of 1) the total size of the
417 * previously outstanding DATA chunk(s) acknowledged, and
418 * 2) the destination's path MTU. This upper bound protects
419 * against the ACK-Splitting attack outlined in [SAVAGE99].
421 if (asoc->fast_recovery)
424 if (bytes_acked > pmtu)
428 SCTP_DEBUG_PRINTK("%s: SLOW START: transport: %p, "
429 "bytes_acked: %d, cwnd: %d, ssthresh: %d, "
430 "flight_size: %d, pba: %d\n",
432 transport, bytes_acked, cwnd,
433 ssthresh, flight_size, pba);
435 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
436 * upon each SACK arrival that advances the Cumulative TSN Ack
437 * Point, increase partial_bytes_acked by the total number of
438 * bytes of all new chunks acknowledged in that SACK including
439 * chunks acknowledged by the new Cumulative TSN Ack and by
442 * When partial_bytes_acked is equal to or greater than cwnd
443 * and before the arrival of the SACK the sender had cwnd or
444 * more bytes of data outstanding (i.e., before arrival of the
445 * SACK, flightsize was greater than or equal to cwnd),
446 * increase cwnd by MTU, and reset partial_bytes_acked to
447 * (partial_bytes_acked - cwnd).
452 pba = ((cwnd < pba) ? (pba - cwnd) : 0);
454 SCTP_DEBUG_PRINTK("%s: CONGESTION AVOIDANCE: "
455 "transport: %p, bytes_acked: %d, cwnd: %d, "
456 "ssthresh: %d, flight_size: %d, pba: %d\n",
458 transport, bytes_acked, cwnd,
459 ssthresh, flight_size, pba);
462 transport->cwnd = cwnd;
463 transport->partial_bytes_acked = pba;
466 /* This routine is used to lower the transport's cwnd when congestion is
469 void sctp_transport_lower_cwnd(struct sctp_transport *transport,
470 sctp_lower_cwnd_t reason)
472 struct sctp_association *asoc = transport->asoc;
475 case SCTP_LOWER_CWND_T3_RTX:
476 /* RFC 2960 Section 7.2.3, sctpimpguide
477 * When the T3-rtx timer expires on an address, SCTP should
478 * perform slow start by:
479 * ssthresh = max(cwnd/2, 4*MTU)
481 * partial_bytes_acked = 0
483 transport->ssthresh = max(transport->cwnd/2,
485 transport->cwnd = asoc->pathmtu;
487 /* T3-rtx also clears fast recovery */
488 asoc->fast_recovery = 0;
491 case SCTP_LOWER_CWND_FAST_RTX:
492 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
493 * destination address(es) to which the missing DATA chunks
494 * were last sent, according to the formula described in
497 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
498 * losses from SACK (see Section 7.2.4), An endpoint
499 * should do the following:
500 * ssthresh = max(cwnd/2, 4*MTU)
502 * partial_bytes_acked = 0
504 if (asoc->fast_recovery)
507 /* Mark Fast recovery */
508 asoc->fast_recovery = 1;
509 asoc->fast_recovery_exit = asoc->next_tsn - 1;
511 transport->ssthresh = max(transport->cwnd/2,
513 transport->cwnd = transport->ssthresh;
516 case SCTP_LOWER_CWND_ECNE:
517 /* RFC 2481 Section 6.1.2.
518 * If the sender receives an ECN-Echo ACK packet
519 * then the sender knows that congestion was encountered in the
520 * network on the path from the sender to the receiver. The
521 * indication of congestion should be treated just as a
522 * congestion loss in non-ECN Capable TCP. That is, the TCP
523 * source halves the congestion window "cwnd" and reduces the
524 * slow start threshold "ssthresh".
525 * A critical condition is that TCP does not react to
526 * congestion indications more than once every window of
527 * data (or more loosely more than once every round-trip time).
529 if (time_after(jiffies, transport->last_time_ecne_reduced +
531 transport->ssthresh = max(transport->cwnd/2,
533 transport->cwnd = transport->ssthresh;
534 transport->last_time_ecne_reduced = jiffies;
538 case SCTP_LOWER_CWND_INACTIVE:
539 /* RFC 2960 Section 7.2.1, sctpimpguide
540 * When the endpoint does not transmit data on a given
541 * transport address, the cwnd of the transport address
542 * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
543 * NOTE: Although the draft recommends that this check needs
544 * to be done every RTO interval, we do it every hearbeat
547 transport->cwnd = max(transport->cwnd/2,
552 transport->partial_bytes_acked = 0;
553 SCTP_DEBUG_PRINTK("%s: transport: %p reason: %d cwnd: "
554 "%d ssthresh: %d\n", __func__,
556 transport->cwnd, transport->ssthresh);
559 /* Apply Max.Burst limit to the congestion window:
560 * sctpimpguide-05 2.14.2
561 * D) When the time comes for the sender to
562 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
563 * first be applied to limit how many new DATA chunks may be sent.
564 * The limit is applied by adjusting cwnd as follows:
565 * if ((flightsize+ Max.Burst * MTU) < cwnd)
566 * cwnd = flightsize + Max.Burst * MTU
569 void sctp_transport_burst_limited(struct sctp_transport *t)
571 struct sctp_association *asoc = t->asoc;
572 u32 old_cwnd = t->cwnd;
575 if (t->burst_limited)
578 max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu);
579 if (max_burst_bytes < old_cwnd) {
580 t->cwnd = max_burst_bytes;
581 t->burst_limited = old_cwnd;
585 /* Restore the old cwnd congestion window, after the burst had it's
588 void sctp_transport_burst_reset(struct sctp_transport *t)
590 if (t->burst_limited) {
591 t->cwnd = t->burst_limited;
592 t->burst_limited = 0;
596 /* What is the next timeout value for this transport? */
597 unsigned long sctp_transport_timeout(struct sctp_transport *t)
599 unsigned long timeout;
600 timeout = t->rto + sctp_jitter(t->rto);
601 if (t->state != SCTP_UNCONFIRMED)
602 timeout += t->hbinterval;
607 /* Reset transport variables to their initial values */
608 void sctp_transport_reset(struct sctp_transport *t)
610 struct sctp_association *asoc = t->asoc;
612 /* RFC 2960 (bis), Section 5.2.4
613 * All the congestion control parameters (e.g., cwnd, ssthresh)
614 * related to this peer MUST be reset to their initial values
615 * (see Section 6.2.1)
617 t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
618 t->burst_limited = 0;
619 t->ssthresh = asoc->peer.i.a_rwnd;
620 t->rto = asoc->rto_initial;
625 /* Reset these additional varibles so that we have a clean
628 t->partial_bytes_acked = 0;
634 /* Initialize the state information for SFR-CACC */
635 t->cacc.changeover_active = 0;
636 t->cacc.cycling_changeover = 0;
637 t->cacc.next_tsn_at_change = 0;
638 t->cacc.cacc_saw_newack = 0;