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,
96 /* Initialize the 64-bit random nonce sent with heartbeat. */
97 get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce));
99 atomic_set(&peer->refcnt, 1);
104 /* Allocate and initialize a new transport. */
105 struct sctp_transport *sctp_transport_new(const union sctp_addr *addr,
108 struct sctp_transport *transport;
110 transport = t_new(struct sctp_transport, gfp);
114 if (!sctp_transport_init(transport, addr, gfp))
117 transport->malloced = 1;
118 SCTP_DBG_OBJCNT_INC(transport);
129 /* This transport is no longer needed. Free up if possible, or
130 * delay until it last reference count.
132 void sctp_transport_free(struct sctp_transport *transport)
136 /* Try to delete the heartbeat timer. */
137 if (del_timer(&transport->hb_timer))
138 sctp_transport_put(transport);
140 /* Delete the T3_rtx timer if it's active.
141 * There is no point in not doing this now and letting
142 * structure hang around in memory since we know
143 * the tranport is going away.
145 if (timer_pending(&transport->T3_rtx_timer) &&
146 del_timer(&transport->T3_rtx_timer))
147 sctp_transport_put(transport);
150 sctp_transport_put(transport);
153 /* Destroy the transport data structure.
154 * Assumes there are no more users of this structure.
156 static void sctp_transport_destroy(struct sctp_transport *transport)
158 SCTP_ASSERT(transport->dead, "Transport is not dead", return);
161 sctp_association_put(transport->asoc);
163 sctp_packet_free(&transport->packet);
165 dst_release(transport->dst);
167 SCTP_DBG_OBJCNT_DEC(transport);
170 /* Start T3_rtx timer if it is not already running and update the heartbeat
171 * timer. This routine is called every time a DATA chunk is sent.
173 void sctp_transport_reset_timers(struct sctp_transport *transport)
175 /* RFC 2960 6.3.2 Retransmission Timer Rules
177 * R1) Every time a DATA chunk is sent to any address(including a
178 * retransmission), if the T3-rtx timer of that address is not running
179 * start it running so that it will expire after the RTO of that
183 if (!timer_pending(&transport->T3_rtx_timer))
184 if (!mod_timer(&transport->T3_rtx_timer,
185 jiffies + transport->rto))
186 sctp_transport_hold(transport);
188 /* When a data chunk is sent, reset the heartbeat interval. */
189 if (!mod_timer(&transport->hb_timer,
190 sctp_transport_timeout(transport)))
191 sctp_transport_hold(transport);
194 /* This transport has been assigned to an association.
195 * Initialize fields from the association or from the sock itself.
196 * Register the reference count in the association.
198 void sctp_transport_set_owner(struct sctp_transport *transport,
199 struct sctp_association *asoc)
201 transport->asoc = asoc;
202 sctp_association_hold(asoc);
205 /* Initialize the pmtu of a transport. */
206 void sctp_transport_pmtu(struct sctp_transport *transport)
208 struct dst_entry *dst;
210 dst = transport->af_specific->get_dst(NULL, &transport->ipaddr, NULL);
213 transport->pathmtu = dst_mtu(dst);
216 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
219 /* this is a complete rip-off from __sk_dst_check
220 * the cookie is always 0 since this is how it's used in the
223 static struct dst_entry *sctp_transport_dst_check(struct sctp_transport *t)
225 struct dst_entry *dst = t->dst;
227 if (dst && dst->obsolete && dst->ops->check(dst, 0) == NULL) {
236 void sctp_transport_update_pmtu(struct sctp_transport *t, u32 pmtu)
238 struct dst_entry *dst;
240 if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) {
241 printk(KERN_WARNING "%s: Reported pmtu %d too low, "
242 "using default minimum of %d\n",
244 SCTP_DEFAULT_MINSEGMENT);
245 /* Use default minimum segment size and disable
246 * pmtu discovery on this transport.
248 t->pathmtu = SCTP_DEFAULT_MINSEGMENT;
253 dst = sctp_transport_dst_check(t);
255 dst->ops->update_pmtu(dst, pmtu);
258 /* Caches the dst entry and source address for a transport's destination
261 void sctp_transport_route(struct sctp_transport *transport,
262 union sctp_addr *saddr, struct sctp_sock *opt)
264 struct sctp_association *asoc = transport->asoc;
265 struct sctp_af *af = transport->af_specific;
266 union sctp_addr *daddr = &transport->ipaddr;
267 struct dst_entry *dst;
269 dst = af->get_dst(asoc, daddr, saddr);
272 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
274 af->get_saddr(opt, asoc, dst, daddr, &transport->saddr);
276 transport->dst = dst;
277 if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) {
281 transport->pathmtu = dst_mtu(dst);
283 /* Initialize sk->sk_rcv_saddr, if the transport is the
284 * association's active path for getsockname().
286 if (asoc && (!asoc->peer.primary_path ||
287 (transport == asoc->peer.active_path)))
288 opt->pf->af->to_sk_saddr(&transport->saddr,
291 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
294 /* Hold a reference to a transport. */
295 void sctp_transport_hold(struct sctp_transport *transport)
297 atomic_inc(&transport->refcnt);
300 /* Release a reference to a transport and clean up
301 * if there are no more references.
303 void sctp_transport_put(struct sctp_transport *transport)
305 if (atomic_dec_and_test(&transport->refcnt))
306 sctp_transport_destroy(transport);
309 /* Update transport's RTO based on the newly calculated RTT. */
310 void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
312 /* Check for valid transport. */
313 SCTP_ASSERT(tp, "NULL transport", return);
315 /* We should not be doing any RTO updates unless rto_pending is set. */
316 SCTP_ASSERT(tp->rto_pending, "rto_pending not set", return);
318 if (tp->rttvar || tp->srtt) {
319 /* 6.3.1 C3) When a new RTT measurement R' is made, set
320 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
321 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
324 /* Note: The above algorithm has been rewritten to
325 * express rto_beta and rto_alpha as inverse powers
327 * For example, assuming the default value of RTO.Alpha of
328 * 1/8, rto_alpha would be expressed as 3.
330 tp->rttvar = tp->rttvar - (tp->rttvar >> sctp_rto_beta)
331 + ((abs(tp->srtt - rtt)) >> sctp_rto_beta);
332 tp->srtt = tp->srtt - (tp->srtt >> sctp_rto_alpha)
333 + (rtt >> sctp_rto_alpha);
335 /* 6.3.1 C2) When the first RTT measurement R is made, set
336 * SRTT <- R, RTTVAR <- R/2.
339 tp->rttvar = rtt >> 1;
342 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
343 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
346 tp->rttvar = SCTP_CLOCK_GRANULARITY;
348 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
349 tp->rto = tp->srtt + (tp->rttvar << 2);
351 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
352 * seconds then it is rounded up to RTO.Min seconds.
354 if (tp->rto < tp->asoc->rto_min)
355 tp->rto = tp->asoc->rto_min;
357 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
358 * at least RTO.max seconds.
360 if (tp->rto > tp->asoc->rto_max)
361 tp->rto = tp->asoc->rto_max;
365 /* Reset rto_pending so that a new RTT measurement is started when a
366 * new data chunk is sent.
370 SCTP_DEBUG_PRINTK("%s: transport: %p, rtt: %d, srtt: %d "
371 "rttvar: %d, rto: %ld\n", __func__,
372 tp, rtt, tp->srtt, tp->rttvar, tp->rto);
375 /* This routine updates the transport's cwnd and partial_bytes_acked
376 * parameters based on the bytes acked in the received SACK.
378 void sctp_transport_raise_cwnd(struct sctp_transport *transport,
379 __u32 sack_ctsn, __u32 bytes_acked)
381 struct sctp_association *asoc = transport->asoc;
382 __u32 cwnd, ssthresh, flight_size, pba, pmtu;
384 cwnd = transport->cwnd;
385 flight_size = transport->flight_size;
387 /* See if we need to exit Fast Recovery first */
388 if (asoc->fast_recovery &&
389 TSN_lte(asoc->fast_recovery_exit, sack_ctsn))
390 asoc->fast_recovery = 0;
392 /* The appropriate cwnd increase algorithm is performed if, and only
393 * if the cumulative TSN whould advanced and the congestion window is
394 * being fully utilized.
396 if (TSN_lte(sack_ctsn, transport->asoc->ctsn_ack_point) ||
397 (flight_size < cwnd))
400 ssthresh = transport->ssthresh;
401 pba = transport->partial_bytes_acked;
402 pmtu = transport->asoc->pathmtu;
404 if (cwnd <= ssthresh) {
406 * o When cwnd is less than or equal to ssthresh, an SCTP
407 * endpoint MUST use the slow-start algorithm to increase
408 * cwnd only if the current congestion window is being fully
409 * utilized, an incoming SACK advances the Cumulative TSN
410 * Ack Point, and the data sender is not in Fast Recovery.
411 * Only when these three conditions are met can the cwnd be
412 * increased; otherwise, the cwnd MUST not be increased.
413 * If these conditions are met, then cwnd MUST be increased
414 * by, at most, the lesser of 1) the total size of the
415 * previously outstanding DATA chunk(s) acknowledged, and
416 * 2) the destination's path MTU. This upper bound protects
417 * against the ACK-Splitting attack outlined in [SAVAGE99].
419 if (asoc->fast_recovery)
422 if (bytes_acked > pmtu)
426 SCTP_DEBUG_PRINTK("%s: SLOW START: transport: %p, "
427 "bytes_acked: %d, cwnd: %d, ssthresh: %d, "
428 "flight_size: %d, pba: %d\n",
430 transport, bytes_acked, cwnd,
431 ssthresh, flight_size, pba);
433 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
434 * upon each SACK arrival that advances the Cumulative TSN Ack
435 * Point, increase partial_bytes_acked by the total number of
436 * bytes of all new chunks acknowledged in that SACK including
437 * chunks acknowledged by the new Cumulative TSN Ack and by
440 * When partial_bytes_acked is equal to or greater than cwnd
441 * and before the arrival of the SACK the sender had cwnd or
442 * more bytes of data outstanding (i.e., before arrival of the
443 * SACK, flightsize was greater than or equal to cwnd),
444 * increase cwnd by MTU, and reset partial_bytes_acked to
445 * (partial_bytes_acked - cwnd).
450 pba = ((cwnd < pba) ? (pba - cwnd) : 0);
452 SCTP_DEBUG_PRINTK("%s: CONGESTION AVOIDANCE: "
453 "transport: %p, bytes_acked: %d, cwnd: %d, "
454 "ssthresh: %d, flight_size: %d, pba: %d\n",
456 transport, bytes_acked, cwnd,
457 ssthresh, flight_size, pba);
460 transport->cwnd = cwnd;
461 transport->partial_bytes_acked = pba;
464 /* This routine is used to lower the transport's cwnd when congestion is
467 void sctp_transport_lower_cwnd(struct sctp_transport *transport,
468 sctp_lower_cwnd_t reason)
470 struct sctp_association *asoc = transport->asoc;
473 case SCTP_LOWER_CWND_T3_RTX:
474 /* RFC 2960 Section 7.2.3, sctpimpguide
475 * When the T3-rtx timer expires on an address, SCTP should
476 * perform slow start by:
477 * ssthresh = max(cwnd/2, 4*MTU)
479 * partial_bytes_acked = 0
481 transport->ssthresh = max(transport->cwnd/2,
483 transport->cwnd = asoc->pathmtu;
485 /* T3-rtx also clears fast recovery */
486 asoc->fast_recovery = 0;
489 case SCTP_LOWER_CWND_FAST_RTX:
490 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
491 * destination address(es) to which the missing DATA chunks
492 * were last sent, according to the formula described in
495 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
496 * losses from SACK (see Section 7.2.4), An endpoint
497 * should do the following:
498 * ssthresh = max(cwnd/2, 4*MTU)
500 * partial_bytes_acked = 0
502 if (asoc->fast_recovery)
505 /* Mark Fast recovery */
506 asoc->fast_recovery = 1;
507 asoc->fast_recovery_exit = asoc->next_tsn - 1;
509 transport->ssthresh = max(transport->cwnd/2,
511 transport->cwnd = transport->ssthresh;
514 case SCTP_LOWER_CWND_ECNE:
515 /* RFC 2481 Section 6.1.2.
516 * If the sender receives an ECN-Echo ACK packet
517 * then the sender knows that congestion was encountered in the
518 * network on the path from the sender to the receiver. The
519 * indication of congestion should be treated just as a
520 * congestion loss in non-ECN Capable TCP. That is, the TCP
521 * source halves the congestion window "cwnd" and reduces the
522 * slow start threshold "ssthresh".
523 * A critical condition is that TCP does not react to
524 * congestion indications more than once every window of
525 * data (or more loosely more than once every round-trip time).
527 if (time_after(jiffies, transport->last_time_ecne_reduced +
529 transport->ssthresh = max(transport->cwnd/2,
531 transport->cwnd = transport->ssthresh;
532 transport->last_time_ecne_reduced = jiffies;
536 case SCTP_LOWER_CWND_INACTIVE:
537 /* RFC 2960 Section 7.2.1, sctpimpguide
538 * When the endpoint does not transmit data on a given
539 * transport address, the cwnd of the transport address
540 * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
541 * NOTE: Although the draft recommends that this check needs
542 * to be done every RTO interval, we do it every hearbeat
545 transport->cwnd = max(transport->cwnd/2,
550 transport->partial_bytes_acked = 0;
551 SCTP_DEBUG_PRINTK("%s: transport: %p reason: %d cwnd: "
552 "%d ssthresh: %d\n", __func__,
554 transport->cwnd, transport->ssthresh);
557 /* Apply Max.Burst limit to the congestion window:
558 * sctpimpguide-05 2.14.2
559 * D) When the time comes for the sender to
560 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
561 * first be applied to limit how many new DATA chunks may be sent.
562 * The limit is applied by adjusting cwnd as follows:
563 * if ((flightsize+ Max.Burst * MTU) < cwnd)
564 * cwnd = flightsize + Max.Burst * MTU
567 void sctp_transport_burst_limited(struct sctp_transport *t)
569 struct sctp_association *asoc = t->asoc;
570 u32 old_cwnd = t->cwnd;
573 if (t->burst_limited)
576 max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu);
577 if (max_burst_bytes < old_cwnd) {
578 t->cwnd = max_burst_bytes;
579 t->burst_limited = old_cwnd;
583 /* Restore the old cwnd congestion window, after the burst had it's
586 void sctp_transport_burst_reset(struct sctp_transport *t)
588 if (t->burst_limited) {
589 t->cwnd = t->burst_limited;
590 t->burst_limited = 0;
594 /* What is the next timeout value for this transport? */
595 unsigned long sctp_transport_timeout(struct sctp_transport *t)
597 unsigned long timeout;
598 timeout = t->rto + sctp_jitter(t->rto);
599 if (t->state != SCTP_UNCONFIRMED)
600 timeout += t->hbinterval;
605 /* Reset transport variables to their initial values */
606 void sctp_transport_reset(struct sctp_transport *t)
608 struct sctp_association *asoc = t->asoc;
610 /* RFC 2960 (bis), Section 5.2.4
611 * All the congestion control parameters (e.g., cwnd, ssthresh)
612 * related to this peer MUST be reset to their initial values
613 * (see Section 6.2.1)
615 t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
616 t->burst_limited = 0;
617 t->ssthresh = asoc->peer.i.a_rwnd;
618 t->rto = asoc->rto_initial;
623 /* Reset these additional varibles so that we have a clean
626 t->partial_bytes_acked = 0;
632 /* Initialize the state information for SFR-CACC */
633 t->cacc.changeover_active = 0;
634 t->cacc.cycling_changeover = 0;
635 t->cacc.next_tsn_at_change = 0;
636 t->cacc.cacc_saw_newack = 0;