blob: a0a431824f63a731728bd36279dc76b86cc2471b [file] [log] [blame]
Vlad Yasevich60c778b2008-01-11 09:57:09 -05001/* SCTP kernel implementation
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * 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
7 *
Vlad Yasevich60c778b2008-01-11 09:57:09 -05008 * This file is part of the SCTP kernel implementation
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 *
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.
13 *
Vlad Yasevich60c778b2008-01-11 09:57:09 -050014 * This SCTP implementation is free software;
Linus Torvalds1da177e2005-04-16 15:20:36 -070015 * 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.
19 *
Vlad Yasevich60c778b2008-01-11 09:57:09 -050020 * This SCTP implementation is distributed in the hope that it
Linus Torvalds1da177e2005-04-16 15:20:36 -070021 * 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.
25 *
26 * You should have received a copy of the GNU General Public License
Jeff Kirsher4b2f13a2013-12-06 06:28:48 -080027 * along with GNU CC; see the file COPYING. If not, see
28 * <http://www.gnu.org/licenses/>.
Linus Torvalds1da177e2005-04-16 15:20:36 -070029 *
30 * Please send any bug reports or fixes you make to the
31 * email address(es):
Daniel Borkmann91705c62013-07-23 14:51:47 +020032 * lksctp developers <linux-sctp@vger.kernel.org>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 * 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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 */
43
Joe Perches145ce502010-08-24 13:21:08 +000044#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
45
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090046#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/types.h>
Sridhar Samudralaad8fec12006-07-21 14:48:50 -070048#include <linux/random.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <net/sctp/sctp.h>
50#include <net/sctp/sm.h>
51
52/* 1st Level Abstractions. */
53
54/* Initialize a new transport from provided memory. */
Eric W. Biederman89bf3452012-08-07 07:26:14 +000055static struct sctp_transport *sctp_transport_init(struct net *net,
56 struct sctp_transport *peer,
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 const union sctp_addr *addr,
Al Virodd0fc662005-10-07 07:46:04 +010058 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -070059{
60 /* Copy in the address. */
61 peer->ipaddr = *addr;
62 peer->af_specific = sctp_get_af_specific(addr->sa.sa_family);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 memset(&peer->saddr, 0, sizeof(union sctp_addr));
64
Neil Horman42448542012-06-30 03:04:26 +000065 peer->sack_generation = 0;
66
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 /* From 6.3.1 RTO Calculation:
68 *
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'.
72 */
Eric W. Biedermane1fc3b12012-08-07 07:29:57 +000073 peer->rto = msecs_to_jiffies(net->sctp.rto_initial);
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
Daniel Borkmanne575235f2014-06-11 18:19:30 +020075 peer->last_time_heard = ktime_get();
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 peer->last_time_ecne_reduced = jiffies;
77
Frank Filz52ccb8e2005-12-22 11:36:46 -080078 peer->param_flags = SPP_HB_DISABLE |
79 SPP_PMTUD_ENABLE |
80 SPP_SACKDELAY_ENABLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
82 /* Initialize the default path max_retrans. */
Eric W. Biedermane1fc3b12012-08-07 07:29:57 +000083 peer->pathmaxrxt = net->sctp.max_retrans_path;
84 peer->pf_retrans = net->sctp.pf_retrans;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
86 INIT_LIST_HEAD(&peer->transmitted);
87 INIT_LIST_HEAD(&peer->send_ready);
88 INIT_LIST_HEAD(&peer->transports);
89
Pavel Emelyanovb24b8a22008-01-23 21:20:07 -080090 setup_timer(&peer->T3_rtx_timer, sctp_generate_t3_rtx_event,
91 (unsigned long)peer);
92 setup_timer(&peer->hb_timer, sctp_generate_heartbeat_event,
93 (unsigned long)peer);
Vlad Yasevich50b5d6a2010-05-06 00:56:07 -070094 setup_timer(&peer->proto_unreach_timer,
95 sctp_generate_proto_unreach_event, (unsigned long)peer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
Sridhar Samudralaad8fec12006-07-21 14:48:50 -070097 /* Initialize the 64-bit random nonce sent with heartbeat. */
98 get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce));
99
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 atomic_set(&peer->refcnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
102 return peer;
103}
104
105/* Allocate and initialize a new transport. */
Eric W. Biederman89bf3452012-08-07 07:26:14 +0000106struct sctp_transport *sctp_transport_new(struct net *net,
107 const union sctp_addr *addr,
Al Virodd0fc662005-10-07 07:46:04 +0100108 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109{
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900110 struct sctp_transport *transport;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
Daniel Borkmann939cfa72013-06-17 11:40:04 +0200112 transport = kzalloc(sizeof(*transport), gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 if (!transport)
114 goto fail;
115
Eric W. Biederman89bf3452012-08-07 07:26:14 +0000116 if (!sctp_transport_init(net, transport, addr, gfp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 goto fail_init;
118
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 SCTP_DBG_OBJCNT_INC(transport);
120
121 return transport;
122
123fail_init:
124 kfree(transport);
125
126fail:
127 return NULL;
128}
129
130/* This transport is no longer needed. Free up if possible, or
131 * delay until it last reference count.
132 */
133void sctp_transport_free(struct sctp_transport *transport)
134{
135 transport->dead = 1;
136
137 /* Try to delete the heartbeat timer. */
138 if (del_timer(&transport->hb_timer))
139 sctp_transport_put(transport);
140
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.
145 */
Ying Xue25cc4ae2013-02-03 20:32:57 +0000146 if (del_timer(&transport->T3_rtx_timer))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 sctp_transport_put(transport);
148
Wei Yongjun55fa0cf2010-05-09 16:56:07 +0000149 /* Delete the ICMP proto unreachable timer if it's active. */
Ying Xue25cc4ae2013-02-03 20:32:57 +0000150 if (del_timer(&transport->proto_unreach_timer))
Wei Yongjun55fa0cf2010-05-09 16:56:07 +0000151 sctp_association_put(transport->asoc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
153 sctp_transport_put(transport);
154}
155
Thomas Graf45122ca262012-12-06 09:25:05 +0000156static void sctp_transport_destroy_rcu(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157{
Thomas Graf45122ca262012-12-06 09:25:05 +0000158 struct sctp_transport *transport;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
Thomas Graf45122ca262012-12-06 09:25:05 +0000160 transport = container_of(head, struct sctp_transport, rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162 dst_release(transport->dst);
163 kfree(transport);
164 SCTP_DBG_OBJCNT_DEC(transport);
165}
166
Thomas Graf45122ca262012-12-06 09:25:05 +0000167/* Destroy the transport data structure.
168 * Assumes there are no more users of this structure.
169 */
170static void sctp_transport_destroy(struct sctp_transport *transport)
171{
Daniel Borkmannbb333812013-06-28 19:49:40 +0200172 if (unlikely(!transport->dead)) {
173 WARN(1, "Attempt to destroy undead transport %p!\n", transport);
174 return;
175 }
Thomas Graf45122ca262012-12-06 09:25:05 +0000176
Daniel Borkmann8c986532013-02-01 04:37:43 +0000177 sctp_packet_free(&transport->packet);
178
179 if (transport->asoc)
180 sctp_association_put(transport->asoc);
Daniel Borkmann771085d2013-08-09 16:25:21 +0200181
182 call_rcu(&transport->rcu, sctp_transport_destroy_rcu);
Thomas Graf45122ca262012-12-06 09:25:05 +0000183}
184
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185/* Start T3_rtx timer if it is not already running and update the heartbeat
186 * timer. This routine is called every time a DATA chunk is sent.
187 */
Vlad Yasevichd9efc222010-04-30 22:41:09 -0400188void sctp_transport_reset_timers(struct sctp_transport *transport)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189{
190 /* RFC 2960 6.3.2 Retransmission Timer Rules
191 *
192 * R1) Every time a DATA chunk is sent to any address(including a
193 * retransmission), if the T3-rtx timer of that address is not running
194 * start it running so that it will expire after the RTO of that
195 * address.
196 */
197
Vlad Yasevichd9efc222010-04-30 22:41:09 -0400198 if (!timer_pending(&transport->T3_rtx_timer))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 if (!mod_timer(&transport->T3_rtx_timer,
200 jiffies + transport->rto))
201 sctp_transport_hold(transport);
202
203 /* When a data chunk is sent, reset the heartbeat interval. */
204 if (!mod_timer(&transport->hb_timer,
205 sctp_transport_timeout(transport)))
206 sctp_transport_hold(transport);
207}
208
209/* This transport has been assigned to an association.
210 * Initialize fields from the association or from the sock itself.
211 * Register the reference count in the association.
212 */
213void sctp_transport_set_owner(struct sctp_transport *transport,
214 struct sctp_association *asoc)
215{
216 transport->asoc = asoc;
217 sctp_association_hold(asoc);
218}
219
220/* Initialize the pmtu of a transport. */
Vlad Yasevich9914ae32011-04-26 21:51:31 +0000221void sctp_transport_pmtu(struct sctp_transport *transport, struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222{
Vlad Yasevichda0420b2011-04-26 21:54:17 +0000223 /* If we don't have a fresh route, look one up */
David S. Millerf5b0a872012-07-19 12:31:33 -0700224 if (!transport->dst || transport->dst->obsolete) {
Vlad Yasevichda0420b2011-04-26 21:54:17 +0000225 dst_release(transport->dst);
226 transport->af_specific->get_dst(transport, &transport->saddr,
David S. Miller8663c932011-05-06 16:32:47 -0700227 &transport->fl, sk);
Vlad Yasevichda0420b2011-04-26 21:54:17 +0000228 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Vlad Yasevichda0420b2011-04-26 21:54:17 +0000230 if (transport->dst) {
231 transport->pathmtu = dst_mtu(transport->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 } else
Frank Filz52ccb8e2005-12-22 11:36:46 -0800233 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234}
235
David S. Miller02f3d4c2012-07-16 03:57:14 -0700236void sctp_transport_update_pmtu(struct sock *sk, struct sctp_transport *t, u32 pmtu)
Vlad Yasevichc910b472007-06-07 13:47:03 -0400237{
238 struct dst_entry *dst;
239
240 if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) {
Joe Perches145ce502010-08-24 13:21:08 +0000241 pr_warn("%s: Reported pmtu %d too low, using default minimum of %d\n",
242 __func__, pmtu,
243 SCTP_DEFAULT_MINSEGMENT);
Vlad Yasevichc910b472007-06-07 13:47:03 -0400244 /* Use default minimum segment size and disable
245 * pmtu discovery on this transport.
246 */
247 t->pathmtu = SCTP_DEFAULT_MINSEGMENT;
Vlad Yasevichc910b472007-06-07 13:47:03 -0400248 } else {
249 t->pathmtu = pmtu;
250 }
251
252 dst = sctp_transport_dst_check(t);
David S. Miller02f3d4c2012-07-16 03:57:14 -0700253 if (!dst)
254 t->af_specific->get_dst(t, &t->saddr, &t->fl, sk);
255
256 if (dst) {
David S. Miller6700c272012-07-17 03:29:28 -0700257 dst->ops->update_pmtu(dst, sk, NULL, pmtu);
David S. Miller02f3d4c2012-07-16 03:57:14 -0700258
259 dst = sctp_transport_dst_check(t);
260 if (!dst)
261 t->af_specific->get_dst(t, &t->saddr, &t->fl, sk);
262 }
Vlad Yasevichc910b472007-06-07 13:47:03 -0400263}
264
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265/* Caches the dst entry and source address for a transport's destination
266 * address.
267 */
268void sctp_transport_route(struct sctp_transport *transport,
269 union sctp_addr *saddr, struct sctp_sock *opt)
270{
271 struct sctp_association *asoc = transport->asoc;
272 struct sctp_af *af = transport->af_specific;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
David S. Miller8663c932011-05-06 16:32:47 -0700274 af->get_dst(transport, saddr, &transport->fl, sctp_opt2sk(opt));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
276 if (saddr)
277 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
278 else
David S. Miller8663c932011-05-06 16:32:47 -0700279 af->get_saddr(opt, transport, &transport->fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
Frank Filz52ccb8e2005-12-22 11:36:46 -0800281 if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) {
282 return;
283 }
Vlad Yasevichda0420b2011-04-26 21:54:17 +0000284 if (transport->dst) {
285 transport->pathmtu = dst_mtu(transport->dst);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
287 /* Initialize sk->sk_rcv_saddr, if the transport is the
288 * association's active path for getsockname().
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900289 */
Vlad Yasevicha78102e2009-11-11 11:54:37 +0000290 if (asoc && (!asoc->peer.primary_path ||
291 (transport == asoc->peer.active_path)))
Jason Gunthorpe299ee122014-07-30 12:40:53 -0600292 opt->pf->to_sk_saddr(&transport->saddr,
293 asoc->base.sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 } else
Frank Filz52ccb8e2005-12-22 11:36:46 -0800295 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296}
297
298/* Hold a reference to a transport. */
299void sctp_transport_hold(struct sctp_transport *transport)
300{
301 atomic_inc(&transport->refcnt);
302}
303
304/* Release a reference to a transport and clean up
305 * if there are no more references.
306 */
307void sctp_transport_put(struct sctp_transport *transport)
308{
309 if (atomic_dec_and_test(&transport->refcnt))
310 sctp_transport_destroy(transport);
311}
312
313/* Update transport's RTO based on the newly calculated RTT. */
314void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
315{
Daniel Borkmannbb333812013-06-28 19:49:40 +0200316 if (unlikely(!tp->rto_pending))
317 /* We should not be doing any RTO updates unless rto_pending is set. */
318 pr_debug("%s: rto_pending not set on transport %p!\n", __func__, tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319
320 if (tp->rttvar || tp->srtt) {
Eric W. Biedermane1fc3b12012-08-07 07:29:57 +0000321 struct net *net = sock_net(tp->asoc->base.sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 /* 6.3.1 C3) When a new RTT measurement R' is made, set
323 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
324 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
325 */
326
327 /* Note: The above algorithm has been rewritten to
328 * express rto_beta and rto_alpha as inverse powers
329 * of two.
330 * For example, assuming the default value of RTO.Alpha of
331 * 1/8, rto_alpha would be expressed as 3.
332 */
Eric W. Biedermane1fc3b12012-08-07 07:29:57 +0000333 tp->rttvar = tp->rttvar - (tp->rttvar >> net->sctp.rto_beta)
Schoch Christian92d64c22012-11-28 05:18:29 +0000334 + (((__u32)abs64((__s64)tp->srtt - (__s64)rtt)) >> net->sctp.rto_beta);
Eric W. Biedermane1fc3b12012-08-07 07:29:57 +0000335 tp->srtt = tp->srtt - (tp->srtt >> net->sctp.rto_alpha)
336 + (rtt >> net->sctp.rto_alpha);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 } else {
338 /* 6.3.1 C2) When the first RTT measurement R is made, set
339 * SRTT <- R, RTTVAR <- R/2.
340 */
341 tp->srtt = rtt;
342 tp->rttvar = rtt >> 1;
343 }
344
345 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
346 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
347 */
348 if (tp->rttvar == 0)
349 tp->rttvar = SCTP_CLOCK_GRANULARITY;
350
351 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
352 tp->rto = tp->srtt + (tp->rttvar << 2);
353
354 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
355 * seconds then it is rounded up to RTO.Min seconds.
356 */
357 if (tp->rto < tp->asoc->rto_min)
358 tp->rto = tp->asoc->rto_min;
359
360 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
361 * at least RTO.max seconds.
362 */
363 if (tp->rto > tp->asoc->rto_max)
364 tp->rto = tp->asoc->rto_max;
365
Michele Baldessari196d6752012-12-01 04:49:42 +0000366 sctp_max_rto(tp->asoc, tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 tp->rtt = rtt;
368
369 /* Reset rto_pending so that a new RTT measurement is started when a
370 * new data chunk is sent.
371 */
372 tp->rto_pending = 0;
373
Daniel Borkmannbb333812013-06-28 19:49:40 +0200374 pr_debug("%s: transport:%p, rtt:%d, srtt:%d rttvar:%d, rto:%ld\n",
375 __func__, tp, rtt, tp->srtt, tp->rttvar, tp->rto);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376}
377
378/* This routine updates the transport's cwnd and partial_bytes_acked
379 * parameters based on the bytes acked in the received SACK.
380 */
381void sctp_transport_raise_cwnd(struct sctp_transport *transport,
382 __u32 sack_ctsn, __u32 bytes_acked)
383{
Vlad Yasevichcf9b4812010-04-30 22:41:10 -0400384 struct sctp_association *asoc = transport->asoc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 __u32 cwnd, ssthresh, flight_size, pba, pmtu;
386
387 cwnd = transport->cwnd;
388 flight_size = transport->flight_size;
389
Vlad Yasevicha6465232008-06-04 12:38:43 -0700390 /* See if we need to exit Fast Recovery first */
Vlad Yasevichcf9b4812010-04-30 22:41:10 -0400391 if (asoc->fast_recovery &&
392 TSN_lte(asoc->fast_recovery_exit, sack_ctsn))
393 asoc->fast_recovery = 0;
Vlad Yasevicha6465232008-06-04 12:38:43 -0700394
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 /* The appropriate cwnd increase algorithm is performed if, and only
Vlad Yasevicha6465232008-06-04 12:38:43 -0700396 * if the cumulative TSN whould advanced and the congestion window is
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 * being fully utilized.
398 */
Vlad Yasevicha6465232008-06-04 12:38:43 -0700399 if (TSN_lte(sack_ctsn, transport->asoc->ctsn_ack_point) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 (flight_size < cwnd))
401 return;
402
403 ssthresh = transport->ssthresh;
404 pba = transport->partial_bytes_acked;
Frank Filz52ccb8e2005-12-22 11:36:46 -0800405 pmtu = transport->asoc->pathmtu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407 if (cwnd <= ssthresh) {
Vlad Yasevicha6465232008-06-04 12:38:43 -0700408 /* RFC 4960 7.2.1
409 * o When cwnd is less than or equal to ssthresh, an SCTP
410 * endpoint MUST use the slow-start algorithm to increase
411 * cwnd only if the current congestion window is being fully
412 * utilized, an incoming SACK advances the Cumulative TSN
413 * Ack Point, and the data sender is not in Fast Recovery.
414 * Only when these three conditions are met can the cwnd be
415 * increased; otherwise, the cwnd MUST not be increased.
416 * If these conditions are met, then cwnd MUST be increased
417 * by, at most, the lesser of 1) the total size of the
418 * previously outstanding DATA chunk(s) acknowledged, and
419 * 2) the destination's path MTU. This upper bound protects
420 * against the ACK-Splitting attack outlined in [SAVAGE99].
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 */
Vlad Yasevichcf9b4812010-04-30 22:41:10 -0400422 if (asoc->fast_recovery)
Vlad Yasevicha6465232008-06-04 12:38:43 -0700423 return;
424
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 if (bytes_acked > pmtu)
426 cwnd += pmtu;
427 else
428 cwnd += bytes_acked;
Daniel Borkmannbb333812013-06-28 19:49:40 +0200429
430 pr_debug("%s: slow start: transport:%p, bytes_acked:%d, "
431 "cwnd:%d, ssthresh:%d, flight_size:%d, pba:%d\n",
432 __func__, transport, bytes_acked, cwnd, ssthresh,
433 flight_size, pba);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 } else {
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
440 * Gap Ack Blocks.
441 *
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).
448 */
449 pba += bytes_acked;
450 if (pba >= cwnd) {
451 cwnd += pmtu;
452 pba = ((cwnd < pba) ? (pba - cwnd) : 0);
453 }
Daniel Borkmannbb333812013-06-28 19:49:40 +0200454
455 pr_debug("%s: congestion avoidance: transport:%p, "
456 "bytes_acked:%d, cwnd:%d, ssthresh:%d, "
457 "flight_size:%d, pba:%d\n", __func__,
458 transport, bytes_acked, cwnd, ssthresh,
459 flight_size, pba);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 }
461
462 transport->cwnd = cwnd;
463 transport->partial_bytes_acked = pba;
464}
465
466/* This routine is used to lower the transport's cwnd when congestion is
467 * detected.
468 */
469void sctp_transport_lower_cwnd(struct sctp_transport *transport,
470 sctp_lower_cwnd_t reason)
471{
Vlad Yasevichcf9b4812010-04-30 22:41:10 -0400472 struct sctp_association *asoc = transport->asoc;
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 switch (reason) {
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)
480 * cwnd = 1*MTU
481 * partial_bytes_acked = 0
482 */
483 transport->ssthresh = max(transport->cwnd/2,
Vlad Yasevichcf9b4812010-04-30 22:41:10 -0400484 4*asoc->pathmtu);
485 transport->cwnd = asoc->pathmtu;
Vlad Yasevich33ce8282009-09-04 18:20:58 -0400486
Vlad Yasevichcf9b4812010-04-30 22:41:10 -0400487 /* T3-rtx also clears fast recovery */
488 asoc->fast_recovery = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 break;
490
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
495 * Section 7.2.3.
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900496 *
497 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 * losses from SACK (see Section 7.2.4), An endpoint
499 * should do the following:
500 * ssthresh = max(cwnd/2, 4*MTU)
501 * cwnd = ssthresh
502 * partial_bytes_acked = 0
503 */
Vlad Yasevichcf9b4812010-04-30 22:41:10 -0400504 if (asoc->fast_recovery)
Vlad Yasevicha6465232008-06-04 12:38:43 -0700505 return;
506
507 /* Mark Fast recovery */
Vlad Yasevichcf9b4812010-04-30 22:41:10 -0400508 asoc->fast_recovery = 1;
509 asoc->fast_recovery_exit = asoc->next_tsn - 1;
Vlad Yasevicha6465232008-06-04 12:38:43 -0700510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 transport->ssthresh = max(transport->cwnd/2,
Vlad Yasevichcf9b4812010-04-30 22:41:10 -0400512 4*asoc->pathmtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 transport->cwnd = transport->ssthresh;
514 break;
515
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).
528 */
Wei Yongjunf61f6f82009-03-02 09:46:13 +0000529 if (time_after(jiffies, transport->last_time_ecne_reduced +
530 transport->rtt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 transport->ssthresh = max(transport->cwnd/2,
Vlad Yasevichcf9b4812010-04-30 22:41:10 -0400532 4*asoc->pathmtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 transport->cwnd = transport->ssthresh;
534 transport->last_time_ecne_reduced = jiffies;
535 }
536 break;
537
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
545 * interval.
546 */
Vlad Yasevich245cba72009-11-23 15:53:58 -0500547 transport->cwnd = max(transport->cwnd/2,
Vlad Yasevichcf9b4812010-04-30 22:41:10 -0400548 4*asoc->pathmtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700550 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
552 transport->partial_bytes_acked = 0;
Daniel Borkmannbb333812013-06-28 19:49:40 +0200553
554 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d\n",
555 __func__, transport, reason, transport->cwnd,
556 transport->ssthresh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557}
558
Vlad Yasevich46d5a802009-11-23 15:54:00 -0500559/* 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
567 */
568
569void sctp_transport_burst_limited(struct sctp_transport *t)
570{
571 struct sctp_association *asoc = t->asoc;
572 u32 old_cwnd = t->cwnd;
573 u32 max_burst_bytes;
574
wangweidong78ac8142013-12-04 17:32:39 +0800575 if (t->burst_limited || asoc->max_burst == 0)
Vlad Yasevich46d5a802009-11-23 15:54:00 -0500576 return;
577
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;
582 }
583}
584
585/* Restore the old cwnd congestion window, after the burst had it's
586 * desired effect.
587 */
588void sctp_transport_burst_reset(struct sctp_transport *t)
589{
590 if (t->burst_limited) {
591 t->cwnd = t->burst_limited;
592 t->burst_limited = 0;
593 }
594}
595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596/* What is the next timeout value for this transport? */
Daniel Borkmann8f610592014-06-30 13:52:08 +0200597unsigned long sctp_transport_timeout(struct sctp_transport *trans)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598{
Daniel Borkmann8f610592014-06-30 13:52:08 +0200599 /* RTO + timer slack +/- 50% of RTO */
600 unsigned long timeout = (trans->rto >> 1) + prandom_u32_max(trans->rto);
601
602 if (trans->state != SCTP_UNCONFIRMED &&
603 trans->state != SCTP_PF)
604 timeout += trans->hbinterval;
605
606 return timeout + jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607}
Vlad Yasevich749bf922007-03-19 17:02:30 -0700608
609/* Reset transport variables to their initial values */
610void sctp_transport_reset(struct sctp_transport *t)
611{
612 struct sctp_association *asoc = t->asoc;
613
614 /* RFC 2960 (bis), Section 5.2.4
615 * All the congestion control parameters (e.g., cwnd, ssthresh)
616 * related to this peer MUST be reset to their initial values
617 * (see Section 6.2.1)
618 */
619 t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
Vlad Yasevich46d5a802009-11-23 15:54:00 -0500620 t->burst_limited = 0;
Vlad Yasevich289f4242007-03-22 12:26:25 -0700621 t->ssthresh = asoc->peer.i.a_rwnd;
Andrei Pelinescu-Onciul5fdd4ba2009-11-29 00:14:02 -0800622 t->rto = asoc->rto_initial;
Michele Baldessari196d6752012-12-01 04:49:42 +0000623 sctp_max_rto(asoc, t);
Vlad Yasevich749bf922007-03-19 17:02:30 -0700624 t->rtt = 0;
625 t->srtt = 0;
626 t->rttvar = 0;
627
628 /* Reset these additional varibles so that we have a clean
629 * slate.
630 */
631 t->partial_bytes_acked = 0;
632 t->flight_size = 0;
633 t->error_count = 0;
634 t->rto_pending = 0;
Vlad Yasevichfaee47c2009-02-13 08:33:43 +0000635 t->hb_sent = 0;
Vlad Yasevich749bf922007-03-19 17:02:30 -0700636
637 /* Initialize the state information for SFR-CACC */
638 t->cacc.changeover_active = 0;
639 t->cacc.cycling_changeover = 0;
640 t->cacc.next_tsn_at_change = 0;
641 t->cacc.cacc_saw_newack = 0;
642}
Michio Hondaddc4bbe2011-06-17 11:03:23 +0900643
644/* Schedule retransmission on the given transport */
645void sctp_transport_immediate_rtx(struct sctp_transport *t)
646{
647 /* Stop pending T3_rtx_timer */
Ying Xue25cc4ae2013-02-03 20:32:57 +0000648 if (del_timer(&t->T3_rtx_timer))
Michio Hondaddc4bbe2011-06-17 11:03:23 +0900649 sctp_transport_put(t);
Ying Xue25cc4ae2013-02-03 20:32:57 +0000650
Michio Hondaddc4bbe2011-06-17 11:03:23 +0900651 sctp_retransmit(&t->asoc->outqueue, t, SCTP_RTXR_T3_RTX);
652 if (!timer_pending(&t->T3_rtx_timer)) {
653 if (!mod_timer(&t->T3_rtx_timer, jiffies + t->rto))
654 sctp_transport_hold(t);
655 }
Michio Hondaddc4bbe2011-06-17 11:03:23 +0900656}