Merge branch 'master' of github.com:davem330/net-next
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / sctp / transport.c
blob394c57ca2f54210e4060654fb72f1937aee3bc75
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
12 * union sctp_addr.
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)
18 * any later version.
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
32 * email address(es):
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 sctp_transport *peer,
63 const union sctp_addr *addr,
64 gfp_t gfp)
66 /* Copy in the address. */
67 peer->ipaddr = *addr;
68 peer->af_specific = sctp_get_af_specific(addr->sa.sa_family);
69 memset(&peer->saddr, 0, sizeof(union sctp_addr));
71 /* From 6.3.1 RTO Calculation:
73 * C1) Until an RTT measurement has been made for a packet sent to the
74 * given destination transport address, set RTO to the protocol
75 * parameter 'RTO.Initial'.
77 peer->rto = msecs_to_jiffies(sctp_rto_initial);
79 peer->last_time_heard = jiffies;
80 peer->last_time_ecne_reduced = jiffies;
82 peer->param_flags = SPP_HB_DISABLE |
83 SPP_PMTUD_ENABLE |
84 SPP_SACKDELAY_ENABLE;
86 /* Initialize the default path max_retrans. */
87 peer->pathmaxrxt = sctp_max_retrans_path;
89 INIT_LIST_HEAD(&peer->transmitted);
90 INIT_LIST_HEAD(&peer->send_ready);
91 INIT_LIST_HEAD(&peer->transports);
93 setup_timer(&peer->T3_rtx_timer, sctp_generate_t3_rtx_event,
94 (unsigned long)peer);
95 setup_timer(&peer->hb_timer, sctp_generate_heartbeat_event,
96 (unsigned long)peer);
97 setup_timer(&peer->proto_unreach_timer,
98 sctp_generate_proto_unreach_event, (unsigned long)peer);
100 /* Initialize the 64-bit random nonce sent with heartbeat. */
101 get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce));
103 atomic_set(&peer->refcnt, 1);
105 return peer;
108 /* Allocate and initialize a new transport. */
109 struct sctp_transport *sctp_transport_new(const union sctp_addr *addr,
110 gfp_t gfp)
112 struct sctp_transport *transport;
114 transport = t_new(struct sctp_transport, gfp);
115 if (!transport)
116 goto fail;
118 if (!sctp_transport_init(transport, addr, gfp))
119 goto fail_init;
121 transport->malloced = 1;
122 SCTP_DBG_OBJCNT_INC(transport);
124 return transport;
126 fail_init:
127 kfree(transport);
129 fail:
130 return NULL;
133 /* This transport is no longer needed. Free up if possible, or
134 * delay until it last reference count.
136 void sctp_transport_free(struct sctp_transport *transport)
138 transport->dead = 1;
140 /* Try to delete the heartbeat timer. */
141 if (del_timer(&transport->hb_timer))
142 sctp_transport_put(transport);
144 /* Delete the T3_rtx timer if it's active.
145 * There is no point in not doing this now and letting
146 * structure hang around in memory since we know
147 * the tranport is going away.
149 if (timer_pending(&transport->T3_rtx_timer) &&
150 del_timer(&transport->T3_rtx_timer))
151 sctp_transport_put(transport);
153 /* Delete the ICMP proto unreachable timer if it's active. */
154 if (timer_pending(&transport->proto_unreach_timer) &&
155 del_timer(&transport->proto_unreach_timer))
156 sctp_association_put(transport->asoc);
158 sctp_transport_put(transport);
161 /* Destroy the transport data structure.
162 * Assumes there are no more users of this structure.
164 static void sctp_transport_destroy(struct sctp_transport *transport)
166 SCTP_ASSERT(transport->dead, "Transport is not dead", return);
168 if (transport->asoc)
169 sctp_association_put(transport->asoc);
171 sctp_packet_free(&transport->packet);
173 dst_release(transport->dst);
174 kfree(transport);
175 SCTP_DBG_OBJCNT_DEC(transport);
178 /* Start T3_rtx timer if it is not already running and update the heartbeat
179 * timer. This routine is called every time a DATA chunk is sent.
181 void sctp_transport_reset_timers(struct sctp_transport *transport)
183 /* RFC 2960 6.3.2 Retransmission Timer Rules
185 * R1) Every time a DATA chunk is sent to any address(including a
186 * retransmission), if the T3-rtx timer of that address is not running
187 * start it running so that it will expire after the RTO of that
188 * address.
191 if (!timer_pending(&transport->T3_rtx_timer))
192 if (!mod_timer(&transport->T3_rtx_timer,
193 jiffies + transport->rto))
194 sctp_transport_hold(transport);
196 /* When a data chunk is sent, reset the heartbeat interval. */
197 if (!mod_timer(&transport->hb_timer,
198 sctp_transport_timeout(transport)))
199 sctp_transport_hold(transport);
202 /* This transport has been assigned to an association.
203 * Initialize fields from the association or from the sock itself.
204 * Register the reference count in the association.
206 void sctp_transport_set_owner(struct sctp_transport *transport,
207 struct sctp_association *asoc)
209 transport->asoc = asoc;
210 sctp_association_hold(asoc);
213 /* Initialize the pmtu of a transport. */
214 void sctp_transport_pmtu(struct sctp_transport *transport, struct sock *sk)
216 /* If we don't have a fresh route, look one up */
217 if (!transport->dst || transport->dst->obsolete > 1) {
218 dst_release(transport->dst);
219 transport->af_specific->get_dst(transport, &transport->saddr,
220 &transport->fl, sk);
223 if (transport->dst) {
224 transport->pathmtu = dst_mtu(transport->dst);
225 } else
226 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
229 /* this is a complete rip-off from __sk_dst_check
230 * the cookie is always 0 since this is how it's used in the
231 * pmtu code
233 static struct dst_entry *sctp_transport_dst_check(struct sctp_transport *t)
235 struct dst_entry *dst = t->dst;
237 if (dst && dst->obsolete && dst->ops->check(dst, 0) == NULL) {
238 dst_release(t->dst);
239 t->dst = NULL;
240 return NULL;
243 return dst;
246 void sctp_transport_update_pmtu(struct sctp_transport *t, u32 pmtu)
248 struct dst_entry *dst;
250 if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) {
251 pr_warn("%s: Reported pmtu %d too low, using default minimum of %d\n",
252 __func__, pmtu,
253 SCTP_DEFAULT_MINSEGMENT);
254 /* Use default minimum segment size and disable
255 * pmtu discovery on this transport.
257 t->pathmtu = SCTP_DEFAULT_MINSEGMENT;
258 } else {
259 t->pathmtu = pmtu;
262 dst = sctp_transport_dst_check(t);
263 if (dst)
264 dst->ops->update_pmtu(dst, pmtu);
267 /* Caches the dst entry and source address for a transport's destination
268 * address.
270 void sctp_transport_route(struct sctp_transport *transport,
271 union sctp_addr *saddr, struct sctp_sock *opt)
273 struct sctp_association *asoc = transport->asoc;
274 struct sctp_af *af = transport->af_specific;
276 af->get_dst(transport, saddr, &transport->fl, sctp_opt2sk(opt));
278 if (saddr)
279 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
280 else
281 af->get_saddr(opt, transport, &transport->fl);
283 if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) {
284 return;
286 if (transport->dst) {
287 transport->pathmtu = dst_mtu(transport->dst);
289 /* Initialize sk->sk_rcv_saddr, if the transport is the
290 * association's active path for getsockname().
292 if (asoc && (!asoc->peer.primary_path ||
293 (transport == asoc->peer.active_path)))
294 opt->pf->af->to_sk_saddr(&transport->saddr,
295 asoc->base.sk);
296 } else
297 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
300 /* Hold a reference to a transport. */
301 void sctp_transport_hold(struct sctp_transport *transport)
303 atomic_inc(&transport->refcnt);
306 /* Release a reference to a transport and clean up
307 * if there are no more references.
309 void sctp_transport_put(struct sctp_transport *transport)
311 if (atomic_dec_and_test(&transport->refcnt))
312 sctp_transport_destroy(transport);
315 /* Update transport's RTO based on the newly calculated RTT. */
316 void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
318 /* Check for valid transport. */
319 SCTP_ASSERT(tp, "NULL transport", return);
321 /* We should not be doing any RTO updates unless rto_pending is set. */
322 SCTP_ASSERT(tp->rto_pending, "rto_pending not set", return);
324 if (tp->rttvar || tp->srtt) {
325 /* 6.3.1 C3) When a new RTT measurement R' is made, set
326 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
327 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
330 /* Note: The above algorithm has been rewritten to
331 * express rto_beta and rto_alpha as inverse powers
332 * of two.
333 * For example, assuming the default value of RTO.Alpha of
334 * 1/8, rto_alpha would be expressed as 3.
336 tp->rttvar = tp->rttvar - (tp->rttvar >> sctp_rto_beta)
337 + ((abs(tp->srtt - rtt)) >> sctp_rto_beta);
338 tp->srtt = tp->srtt - (tp->srtt >> sctp_rto_alpha)
339 + (rtt >> sctp_rto_alpha);
340 } else {
341 /* 6.3.1 C2) When the first RTT measurement R is made, set
342 * SRTT <- R, RTTVAR <- R/2.
344 tp->srtt = rtt;
345 tp->rttvar = rtt >> 1;
348 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
349 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
351 if (tp->rttvar == 0)
352 tp->rttvar = SCTP_CLOCK_GRANULARITY;
354 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
355 tp->rto = tp->srtt + (tp->rttvar << 2);
357 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
358 * seconds then it is rounded up to RTO.Min seconds.
360 if (tp->rto < tp->asoc->rto_min)
361 tp->rto = tp->asoc->rto_min;
363 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
364 * at least RTO.max seconds.
366 if (tp->rto > tp->asoc->rto_max)
367 tp->rto = tp->asoc->rto_max;
369 tp->rtt = rtt;
371 /* Reset rto_pending so that a new RTT measurement is started when a
372 * new data chunk is sent.
374 tp->rto_pending = 0;
376 SCTP_DEBUG_PRINTK("%s: transport: %p, rtt: %d, srtt: %d "
377 "rttvar: %d, rto: %ld\n", __func__,
378 tp, rtt, tp->srtt, tp->rttvar, tp->rto);
381 /* This routine updates the transport's cwnd and partial_bytes_acked
382 * parameters based on the bytes acked in the received SACK.
384 void sctp_transport_raise_cwnd(struct sctp_transport *transport,
385 __u32 sack_ctsn, __u32 bytes_acked)
387 struct sctp_association *asoc = transport->asoc;
388 __u32 cwnd, ssthresh, flight_size, pba, pmtu;
390 cwnd = transport->cwnd;
391 flight_size = transport->flight_size;
393 /* See if we need to exit Fast Recovery first */
394 if (asoc->fast_recovery &&
395 TSN_lte(asoc->fast_recovery_exit, sack_ctsn))
396 asoc->fast_recovery = 0;
398 /* The appropriate cwnd increase algorithm is performed if, and only
399 * if the cumulative TSN whould advanced and the congestion window is
400 * being fully utilized.
402 if (TSN_lte(sack_ctsn, transport->asoc->ctsn_ack_point) ||
403 (flight_size < cwnd))
404 return;
406 ssthresh = transport->ssthresh;
407 pba = transport->partial_bytes_acked;
408 pmtu = transport->asoc->pathmtu;
410 if (cwnd <= ssthresh) {
411 /* RFC 4960 7.2.1
412 * o When cwnd is less than or equal to ssthresh, an SCTP
413 * endpoint MUST use the slow-start algorithm to increase
414 * cwnd only if the current congestion window is being fully
415 * utilized, an incoming SACK advances the Cumulative TSN
416 * Ack Point, and the data sender is not in Fast Recovery.
417 * Only when these three conditions are met can the cwnd be
418 * increased; otherwise, the cwnd MUST not be increased.
419 * If these conditions are met, then cwnd MUST be increased
420 * by, at most, the lesser of 1) the total size of the
421 * previously outstanding DATA chunk(s) acknowledged, and
422 * 2) the destination's path MTU. This upper bound protects
423 * against the ACK-Splitting attack outlined in [SAVAGE99].
425 if (asoc->fast_recovery)
426 return;
428 if (bytes_acked > pmtu)
429 cwnd += pmtu;
430 else
431 cwnd += bytes_acked;
432 SCTP_DEBUG_PRINTK("%s: SLOW START: transport: %p, "
433 "bytes_acked: %d, cwnd: %d, ssthresh: %d, "
434 "flight_size: %d, pba: %d\n",
435 __func__,
436 transport, bytes_acked, cwnd,
437 ssthresh, flight_size, pba);
438 } else {
439 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
440 * upon each SACK arrival that advances the Cumulative TSN Ack
441 * Point, increase partial_bytes_acked by the total number of
442 * bytes of all new chunks acknowledged in that SACK including
443 * chunks acknowledged by the new Cumulative TSN Ack and by
444 * Gap Ack Blocks.
446 * When partial_bytes_acked is equal to or greater than cwnd
447 * and before the arrival of the SACK the sender had cwnd or
448 * more bytes of data outstanding (i.e., before arrival of the
449 * SACK, flightsize was greater than or equal to cwnd),
450 * increase cwnd by MTU, and reset partial_bytes_acked to
451 * (partial_bytes_acked - cwnd).
453 pba += bytes_acked;
454 if (pba >= cwnd) {
455 cwnd += pmtu;
456 pba = ((cwnd < pba) ? (pba - cwnd) : 0);
458 SCTP_DEBUG_PRINTK("%s: CONGESTION AVOIDANCE: "
459 "transport: %p, bytes_acked: %d, cwnd: %d, "
460 "ssthresh: %d, flight_size: %d, pba: %d\n",
461 __func__,
462 transport, bytes_acked, cwnd,
463 ssthresh, flight_size, pba);
466 transport->cwnd = cwnd;
467 transport->partial_bytes_acked = pba;
470 /* This routine is used to lower the transport's cwnd when congestion is
471 * detected.
473 void sctp_transport_lower_cwnd(struct sctp_transport *transport,
474 sctp_lower_cwnd_t reason)
476 struct sctp_association *asoc = transport->asoc;
478 switch (reason) {
479 case SCTP_LOWER_CWND_T3_RTX:
480 /* RFC 2960 Section 7.2.3, sctpimpguide
481 * When the T3-rtx timer expires on an address, SCTP should
482 * perform slow start by:
483 * ssthresh = max(cwnd/2, 4*MTU)
484 * cwnd = 1*MTU
485 * partial_bytes_acked = 0
487 transport->ssthresh = max(transport->cwnd/2,
488 4*asoc->pathmtu);
489 transport->cwnd = asoc->pathmtu;
491 /* T3-rtx also clears fast recovery */
492 asoc->fast_recovery = 0;
493 break;
495 case SCTP_LOWER_CWND_FAST_RTX:
496 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
497 * destination address(es) to which the missing DATA chunks
498 * were last sent, according to the formula described in
499 * Section 7.2.3.
501 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
502 * losses from SACK (see Section 7.2.4), An endpoint
503 * should do the following:
504 * ssthresh = max(cwnd/2, 4*MTU)
505 * cwnd = ssthresh
506 * partial_bytes_acked = 0
508 if (asoc->fast_recovery)
509 return;
511 /* Mark Fast recovery */
512 asoc->fast_recovery = 1;
513 asoc->fast_recovery_exit = asoc->next_tsn - 1;
515 transport->ssthresh = max(transport->cwnd/2,
516 4*asoc->pathmtu);
517 transport->cwnd = transport->ssthresh;
518 break;
520 case SCTP_LOWER_CWND_ECNE:
521 /* RFC 2481 Section 6.1.2.
522 * If the sender receives an ECN-Echo ACK packet
523 * then the sender knows that congestion was encountered in the
524 * network on the path from the sender to the receiver. The
525 * indication of congestion should be treated just as a
526 * congestion loss in non-ECN Capable TCP. That is, the TCP
527 * source halves the congestion window "cwnd" and reduces the
528 * slow start threshold "ssthresh".
529 * A critical condition is that TCP does not react to
530 * congestion indications more than once every window of
531 * data (or more loosely more than once every round-trip time).
533 if (time_after(jiffies, transport->last_time_ecne_reduced +
534 transport->rtt)) {
535 transport->ssthresh = max(transport->cwnd/2,
536 4*asoc->pathmtu);
537 transport->cwnd = transport->ssthresh;
538 transport->last_time_ecne_reduced = jiffies;
540 break;
542 case SCTP_LOWER_CWND_INACTIVE:
543 /* RFC 2960 Section 7.2.1, sctpimpguide
544 * When the endpoint does not transmit data on a given
545 * transport address, the cwnd of the transport address
546 * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
547 * NOTE: Although the draft recommends that this check needs
548 * to be done every RTO interval, we do it every hearbeat
549 * interval.
551 transport->cwnd = max(transport->cwnd/2,
552 4*asoc->pathmtu);
553 break;
556 transport->partial_bytes_acked = 0;
557 SCTP_DEBUG_PRINTK("%s: transport: %p reason: %d cwnd: "
558 "%d ssthresh: %d\n", __func__,
559 transport, reason,
560 transport->cwnd, transport->ssthresh);
563 /* Apply Max.Burst limit to the congestion window:
564 * sctpimpguide-05 2.14.2
565 * D) When the time comes for the sender to
566 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
567 * first be applied to limit how many new DATA chunks may be sent.
568 * The limit is applied by adjusting cwnd as follows:
569 * if ((flightsize+ Max.Burst * MTU) < cwnd)
570 * cwnd = flightsize + Max.Burst * MTU
573 void sctp_transport_burst_limited(struct sctp_transport *t)
575 struct sctp_association *asoc = t->asoc;
576 u32 old_cwnd = t->cwnd;
577 u32 max_burst_bytes;
579 if (t->burst_limited)
580 return;
582 max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu);
583 if (max_burst_bytes < old_cwnd) {
584 t->cwnd = max_burst_bytes;
585 t->burst_limited = old_cwnd;
589 /* Restore the old cwnd congestion window, after the burst had it's
590 * desired effect.
592 void sctp_transport_burst_reset(struct sctp_transport *t)
594 if (t->burst_limited) {
595 t->cwnd = t->burst_limited;
596 t->burst_limited = 0;
600 /* What is the next timeout value for this transport? */
601 unsigned long sctp_transport_timeout(struct sctp_transport *t)
603 unsigned long timeout;
604 timeout = t->rto + sctp_jitter(t->rto);
605 if (t->state != SCTP_UNCONFIRMED)
606 timeout += t->hbinterval;
607 timeout += jiffies;
608 return timeout;
611 /* Reset transport variables to their initial values */
612 void sctp_transport_reset(struct sctp_transport *t)
614 struct sctp_association *asoc = t->asoc;
616 /* RFC 2960 (bis), Section 5.2.4
617 * All the congestion control parameters (e.g., cwnd, ssthresh)
618 * related to this peer MUST be reset to their initial values
619 * (see Section 6.2.1)
621 t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
622 t->burst_limited = 0;
623 t->ssthresh = asoc->peer.i.a_rwnd;
624 t->rto = asoc->rto_initial;
625 t->rtt = 0;
626 t->srtt = 0;
627 t->rttvar = 0;
629 /* Reset these additional varibles so that we have a clean
630 * slate.
632 t->partial_bytes_acked = 0;
633 t->flight_size = 0;
634 t->error_count = 0;
635 t->rto_pending = 0;
636 t->hb_sent = 0;
638 /* Initialize the state information for SFR-CACC */
639 t->cacc.changeover_active = 0;
640 t->cacc.cycling_changeover = 0;
641 t->cacc.next_tsn_at_change = 0;
642 t->cacc.cacc_saw_newack = 0;