e1000: Add device IDs of blade version of the 82571 quad port
[pv_ops_mirror.git] / net / sctp / transport.c
blob5f467c914f8025fe595cf2a87bc355308abb587f
1 /* SCTP kernel reference 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 reference 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 * The SCTP reference 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 * The SCTP reference 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 #include <linux/types.h>
52 #include <linux/random.h>
53 #include <net/sctp/sctp.h>
54 #include <net/sctp/sm.h>
56 /* 1st Level Abstractions. */
58 /* Initialize a new transport from provided memory. */
59 static struct sctp_transport *sctp_transport_init(struct sctp_transport *peer,
60 const union sctp_addr *addr,
61 gfp_t gfp)
63 /* Copy in the address. */
64 peer->ipaddr = *addr;
65 peer->af_specific = sctp_get_af_specific(addr->sa.sa_family);
66 peer->asoc = NULL;
68 peer->dst = NULL;
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->rtt = 0;
78 peer->rto = msecs_to_jiffies(sctp_rto_initial);
79 peer->rttvar = 0;
80 peer->srtt = 0;
81 peer->rto_pending = 0;
83 peer->last_time_heard = jiffies;
84 peer->last_time_used = jiffies;
85 peer->last_time_ecne_reduced = jiffies;
87 peer->init_sent_count = 0;
89 peer->param_flags = SPP_HB_DISABLE |
90 SPP_PMTUD_ENABLE |
91 SPP_SACKDELAY_ENABLE;
92 peer->hbinterval = 0;
94 /* Initialize the default path max_retrans. */
95 peer->pathmaxrxt = sctp_max_retrans_path;
96 peer->error_count = 0;
98 INIT_LIST_HEAD(&peer->transmitted);
99 INIT_LIST_HEAD(&peer->send_ready);
100 INIT_LIST_HEAD(&peer->transports);
102 /* Set up the retransmission timer. */
103 init_timer(&peer->T3_rtx_timer);
104 peer->T3_rtx_timer.function = sctp_generate_t3_rtx_event;
105 peer->T3_rtx_timer.data = (unsigned long)peer;
107 /* Set up the heartbeat timer. */
108 init_timer(&peer->hb_timer);
109 peer->hb_timer.function = sctp_generate_heartbeat_event;
110 peer->hb_timer.data = (unsigned long)peer;
112 /* Initialize the 64-bit random nonce sent with heartbeat. */
113 get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce));
115 atomic_set(&peer->refcnt, 1);
116 peer->dead = 0;
118 peer->malloced = 0;
120 /* Initialize the state information for SFR-CACC */
121 peer->cacc.changeover_active = 0;
122 peer->cacc.cycling_changeover = 0;
123 peer->cacc.next_tsn_at_change = 0;
124 peer->cacc.cacc_saw_newack = 0;
126 return peer;
129 /* Allocate and initialize a new transport. */
130 struct sctp_transport *sctp_transport_new(const union sctp_addr *addr,
131 gfp_t gfp)
133 struct sctp_transport *transport;
135 transport = t_new(struct sctp_transport, gfp);
136 if (!transport)
137 goto fail;
139 if (!sctp_transport_init(transport, addr, gfp))
140 goto fail_init;
142 transport->malloced = 1;
143 SCTP_DBG_OBJCNT_INC(transport);
145 return transport;
147 fail_init:
148 kfree(transport);
150 fail:
151 return NULL;
154 /* This transport is no longer needed. Free up if possible, or
155 * delay until it last reference count.
157 void sctp_transport_free(struct sctp_transport *transport)
159 transport->dead = 1;
161 /* Try to delete the heartbeat timer. */
162 if (del_timer(&transport->hb_timer))
163 sctp_transport_put(transport);
165 /* Delete the T3_rtx timer if it's active.
166 * There is no point in not doing this now and letting
167 * structure hang around in memory since we know
168 * the tranport is going away.
170 if (timer_pending(&transport->T3_rtx_timer) &&
171 del_timer(&transport->T3_rtx_timer))
172 sctp_transport_put(transport);
175 sctp_transport_put(transport);
178 /* Destroy the transport data structure.
179 * Assumes there are no more users of this structure.
181 static void sctp_transport_destroy(struct sctp_transport *transport)
183 SCTP_ASSERT(transport->dead, "Transport is not dead", return);
185 if (transport->asoc)
186 sctp_association_put(transport->asoc);
188 sctp_packet_free(&transport->packet);
190 dst_release(transport->dst);
191 kfree(transport);
192 SCTP_DBG_OBJCNT_DEC(transport);
195 /* Start T3_rtx timer if it is not already running and update the heartbeat
196 * timer. This routine is called every time a DATA chunk is sent.
198 void sctp_transport_reset_timers(struct sctp_transport *transport)
200 /* RFC 2960 6.3.2 Retransmission Timer Rules
202 * R1) Every time a DATA chunk is sent to any address(including a
203 * retransmission), if the T3-rtx timer of that address is not running
204 * start it running so that it will expire after the RTO of that
205 * address.
208 if (!timer_pending(&transport->T3_rtx_timer))
209 if (!mod_timer(&transport->T3_rtx_timer,
210 jiffies + transport->rto))
211 sctp_transport_hold(transport);
213 /* When a data chunk is sent, reset the heartbeat interval. */
214 if (!mod_timer(&transport->hb_timer,
215 sctp_transport_timeout(transport)))
216 sctp_transport_hold(transport);
219 /* This transport has been assigned to an association.
220 * Initialize fields from the association or from the sock itself.
221 * Register the reference count in the association.
223 void sctp_transport_set_owner(struct sctp_transport *transport,
224 struct sctp_association *asoc)
226 transport->asoc = asoc;
227 sctp_association_hold(asoc);
230 /* Initialize the pmtu of a transport. */
231 void sctp_transport_pmtu(struct sctp_transport *transport)
233 struct dst_entry *dst;
235 dst = transport->af_specific->get_dst(NULL, &transport->ipaddr, NULL);
237 if (dst) {
238 transport->pathmtu = dst_mtu(dst);
239 dst_release(dst);
240 } else
241 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
244 /* this is a complete rip-off from __sk_dst_check
245 * the cookie is always 0 since this is how it's used in the
246 * pmtu code
248 static struct dst_entry *sctp_transport_dst_check(struct sctp_transport *t)
250 struct dst_entry *dst = t->dst;
252 if (dst && dst->obsolete && dst->ops->check(dst, 0) == NULL) {
253 dst_release(t->dst);
254 t->dst = NULL;
255 return NULL;
258 return dst;
261 void sctp_transport_update_pmtu(struct sctp_transport *t, u32 pmtu)
263 struct dst_entry *dst;
265 if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) {
266 printk(KERN_WARNING "%s: Reported pmtu %d too low, "
267 "using default minimum of %d\n",
268 __FUNCTION__, pmtu,
269 SCTP_DEFAULT_MINSEGMENT);
270 /* Use default minimum segment size and disable
271 * pmtu discovery on this transport.
273 t->pathmtu = SCTP_DEFAULT_MINSEGMENT;
274 } else {
275 t->pathmtu = pmtu;
278 dst = sctp_transport_dst_check(t);
279 if (dst)
280 dst->ops->update_pmtu(dst, pmtu);
283 /* Caches the dst entry and source address for a transport's destination
284 * address.
286 void sctp_transport_route(struct sctp_transport *transport,
287 union sctp_addr *saddr, struct sctp_sock *opt)
289 struct sctp_association *asoc = transport->asoc;
290 struct sctp_af *af = transport->af_specific;
291 union sctp_addr *daddr = &transport->ipaddr;
292 struct dst_entry *dst;
294 dst = af->get_dst(asoc, daddr, saddr);
296 if (saddr)
297 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
298 else
299 af->get_saddr(asoc, dst, daddr, &transport->saddr);
301 transport->dst = dst;
302 if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) {
303 return;
305 if (dst) {
306 transport->pathmtu = dst_mtu(dst);
308 /* Initialize sk->sk_rcv_saddr, if the transport is the
309 * association's active path for getsockname().
311 if (asoc && (transport == asoc->peer.active_path))
312 opt->pf->af->to_sk_saddr(&transport->saddr,
313 asoc->base.sk);
314 } else
315 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
318 /* Hold a reference to a transport. */
319 void sctp_transport_hold(struct sctp_transport *transport)
321 atomic_inc(&transport->refcnt);
324 /* Release a reference to a transport and clean up
325 * if there are no more references.
327 void sctp_transport_put(struct sctp_transport *transport)
329 if (atomic_dec_and_test(&transport->refcnt))
330 sctp_transport_destroy(transport);
333 /* Update transport's RTO based on the newly calculated RTT. */
334 void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
336 /* Check for valid transport. */
337 SCTP_ASSERT(tp, "NULL transport", return);
339 /* We should not be doing any RTO updates unless rto_pending is set. */
340 SCTP_ASSERT(tp->rto_pending, "rto_pending not set", return);
342 if (tp->rttvar || tp->srtt) {
343 /* 6.3.1 C3) When a new RTT measurement R' is made, set
344 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
345 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
348 /* Note: The above algorithm has been rewritten to
349 * express rto_beta and rto_alpha as inverse powers
350 * of two.
351 * For example, assuming the default value of RTO.Alpha of
352 * 1/8, rto_alpha would be expressed as 3.
354 tp->rttvar = tp->rttvar - (tp->rttvar >> sctp_rto_beta)
355 + ((abs(tp->srtt - rtt)) >> sctp_rto_beta);
356 tp->srtt = tp->srtt - (tp->srtt >> sctp_rto_alpha)
357 + (rtt >> sctp_rto_alpha);
358 } else {
359 /* 6.3.1 C2) When the first RTT measurement R is made, set
360 * SRTT <- R, RTTVAR <- R/2.
362 tp->srtt = rtt;
363 tp->rttvar = rtt >> 1;
366 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
367 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
369 if (tp->rttvar == 0)
370 tp->rttvar = SCTP_CLOCK_GRANULARITY;
372 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
373 tp->rto = tp->srtt + (tp->rttvar << 2);
375 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
376 * seconds then it is rounded up to RTO.Min seconds.
378 if (tp->rto < tp->asoc->rto_min)
379 tp->rto = tp->asoc->rto_min;
381 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
382 * at least RTO.max seconds.
384 if (tp->rto > tp->asoc->rto_max)
385 tp->rto = tp->asoc->rto_max;
387 tp->rtt = rtt;
389 /* Reset rto_pending so that a new RTT measurement is started when a
390 * new data chunk is sent.
392 tp->rto_pending = 0;
394 SCTP_DEBUG_PRINTK("%s: transport: %p, rtt: %d, srtt: %d "
395 "rttvar: %d, rto: %ld\n", __FUNCTION__,
396 tp, rtt, tp->srtt, tp->rttvar, tp->rto);
399 /* This routine updates the transport's cwnd and partial_bytes_acked
400 * parameters based on the bytes acked in the received SACK.
402 void sctp_transport_raise_cwnd(struct sctp_transport *transport,
403 __u32 sack_ctsn, __u32 bytes_acked)
405 __u32 cwnd, ssthresh, flight_size, pba, pmtu;
407 cwnd = transport->cwnd;
408 flight_size = transport->flight_size;
410 /* The appropriate cwnd increase algorithm is performed if, and only
411 * if the cumulative TSN has advanced and the congestion window is
412 * being fully utilized.
414 if ((transport->asoc->ctsn_ack_point >= sack_ctsn) ||
415 (flight_size < cwnd))
416 return;
418 ssthresh = transport->ssthresh;
419 pba = transport->partial_bytes_acked;
420 pmtu = transport->asoc->pathmtu;
422 if (cwnd <= ssthresh) {
423 /* RFC 2960 7.2.1, sctpimpguide-05 2.14.2 When cwnd is less
424 * than or equal to ssthresh an SCTP endpoint MUST use the
425 * slow start algorithm to increase cwnd only if the current
426 * congestion window is being fully utilized and an incoming
427 * SACK advances the Cumulative TSN Ack Point. Only when these
428 * two conditions are met can the cwnd be increased otherwise
429 * the cwnd MUST not be increased. If these conditions are met
430 * then cwnd MUST be increased by at most the lesser of
431 * 1) the total size of the previously outstanding DATA
432 * chunk(s) acknowledged, and 2) the destination's path MTU.
434 if (bytes_acked > pmtu)
435 cwnd += pmtu;
436 else
437 cwnd += bytes_acked;
438 SCTP_DEBUG_PRINTK("%s: SLOW START: transport: %p, "
439 "bytes_acked: %d, cwnd: %d, ssthresh: %d, "
440 "flight_size: %d, pba: %d\n",
441 __FUNCTION__,
442 transport, bytes_acked, cwnd,
443 ssthresh, flight_size, pba);
444 } else {
445 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
446 * upon each SACK arrival that advances the Cumulative TSN Ack
447 * Point, increase partial_bytes_acked by the total number of
448 * bytes of all new chunks acknowledged in that SACK including
449 * chunks acknowledged by the new Cumulative TSN Ack and by
450 * Gap Ack Blocks.
452 * When partial_bytes_acked is equal to or greater than cwnd
453 * and before the arrival of the SACK the sender had cwnd or
454 * more bytes of data outstanding (i.e., before arrival of the
455 * SACK, flightsize was greater than or equal to cwnd),
456 * increase cwnd by MTU, and reset partial_bytes_acked to
457 * (partial_bytes_acked - cwnd).
459 pba += bytes_acked;
460 if (pba >= cwnd) {
461 cwnd += pmtu;
462 pba = ((cwnd < pba) ? (pba - cwnd) : 0);
464 SCTP_DEBUG_PRINTK("%s: CONGESTION AVOIDANCE: "
465 "transport: %p, bytes_acked: %d, cwnd: %d, "
466 "ssthresh: %d, flight_size: %d, pba: %d\n",
467 __FUNCTION__,
468 transport, bytes_acked, cwnd,
469 ssthresh, flight_size, pba);
472 transport->cwnd = cwnd;
473 transport->partial_bytes_acked = pba;
476 /* This routine is used to lower the transport's cwnd when congestion is
477 * detected.
479 void sctp_transport_lower_cwnd(struct sctp_transport *transport,
480 sctp_lower_cwnd_t reason)
482 switch (reason) {
483 case SCTP_LOWER_CWND_T3_RTX:
484 /* RFC 2960 Section 7.2.3, sctpimpguide
485 * When the T3-rtx timer expires on an address, SCTP should
486 * perform slow start by:
487 * ssthresh = max(cwnd/2, 4*MTU)
488 * cwnd = 1*MTU
489 * partial_bytes_acked = 0
491 transport->ssthresh = max(transport->cwnd/2,
492 4*transport->asoc->pathmtu);
493 transport->cwnd = transport->asoc->pathmtu;
494 break;
496 case SCTP_LOWER_CWND_FAST_RTX:
497 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
498 * destination address(es) to which the missing DATA chunks
499 * were last sent, according to the formula described in
500 * Section 7.2.3.
502 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
503 * losses from SACK (see Section 7.2.4), An endpoint
504 * should do the following:
505 * ssthresh = max(cwnd/2, 4*MTU)
506 * cwnd = ssthresh
507 * partial_bytes_acked = 0
509 transport->ssthresh = max(transport->cwnd/2,
510 4*transport->asoc->pathmtu);
511 transport->cwnd = transport->ssthresh;
512 break;
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 ((jiffies - transport->last_time_ecne_reduced) >
528 transport->rtt) {
529 transport->ssthresh = max(transport->cwnd/2,
530 4*transport->asoc->pathmtu);
531 transport->cwnd = transport->ssthresh;
532 transport->last_time_ecne_reduced = jiffies;
534 break;
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
543 * interval.
545 if ((jiffies - transport->last_time_used) > transport->rto)
546 transport->cwnd = max(transport->cwnd/2,
547 4*transport->asoc->pathmtu);
548 break;
551 transport->partial_bytes_acked = 0;
552 SCTP_DEBUG_PRINTK("%s: transport: %p reason: %d cwnd: "
553 "%d ssthresh: %d\n", __FUNCTION__,
554 transport, reason,
555 transport->cwnd, transport->ssthresh);
558 /* What is the next timeout value for this transport? */
559 unsigned long sctp_transport_timeout(struct sctp_transport *t)
561 unsigned long timeout;
562 timeout = t->rto + sctp_jitter(t->rto);
563 if (t->state != SCTP_UNCONFIRMED)
564 timeout += t->hbinterval;
565 timeout += jiffies;
566 return timeout;
569 /* Reset transport variables to their initial values */
570 void sctp_transport_reset(struct sctp_transport *t)
572 struct sctp_association *asoc = t->asoc;
574 /* RFC 2960 (bis), Section 5.2.4
575 * All the congestion control parameters (e.g., cwnd, ssthresh)
576 * related to this peer MUST be reset to their initial values
577 * (see Section 6.2.1)
579 t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
580 t->ssthresh = asoc->peer.i.a_rwnd;
581 t->rto = asoc->rto_initial;
582 t->rtt = 0;
583 t->srtt = 0;
584 t->rttvar = 0;
586 /* Reset these additional varibles so that we have a clean
587 * slate.
589 t->partial_bytes_acked = 0;
590 t->flight_size = 0;
591 t->error_count = 0;
592 t->rto_pending = 0;
594 /* Initialize the state information for SFR-CACC */
595 t->cacc.changeover_active = 0;
596 t->cacc.cycling_changeover = 0;
597 t->cacc.next_tsn_at_change = 0;
598 t->cacc.cacc_saw_newack = 0;