1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* SCTP kernel implementation
3 * Copyright (c) 1999-2000 Cisco, Inc.
4 * Copyright (c) 1999-2001 Motorola, Inc.
5 * Copyright (c) 2001-2003 International Business Machines Corp.
6 * Copyright (c) 2001 Intel Corp.
7 * Copyright (c) 2001 La Monte H.P. Yarroll
8 *
9 * This file is part of the SCTP kernel implementation
10 *
11 * This module provides the abstraction for an SCTP transport representing
12 * a remote transport address. For local transport addresses, we just use
13 * union sctp_addr.
14 *
15 * Please send any bug reports or fixes you make to the
16 * email address(es):
17 * lksctp developers <linux-sctp@vger.kernel.org>
18 *
19 * Written or modified by:
20 * La Monte H.P. Yarroll <piggy@acm.org>
21 * Karl Knutson <karl@athena.chicago.il.us>
22 * Jon Grimm <jgrimm@us.ibm.com>
23 * Xingang Guo <xingang.guo@intel.com>
24 * Hui Huang <hui.huang@nokia.com>
25 * Sridhar Samudrala <sri@us.ibm.com>
26 * Ardelle Fan <ardelle.fan@intel.com>
27 */
28
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30
31 #include <linux/slab.h>
32 #include <linux/types.h>
33 #include <linux/random.h>
34 #include <net/sctp/sctp.h>
35 #include <net/sctp/sm.h>
36
37 /* 1st Level Abstractions. */
38
39 /* Initialize a new transport from provided memory. */
sctp_transport_init(struct net * net,struct sctp_transport * peer,const union sctp_addr * addr,gfp_t gfp)40 static struct sctp_transport *sctp_transport_init(struct net *net,
41 struct sctp_transport *peer,
42 const union sctp_addr *addr,
43 gfp_t gfp)
44 {
45 /* Copy in the address. */
46 peer->af_specific = sctp_get_af_specific(addr->sa.sa_family);
47 memcpy(&peer->ipaddr, addr, peer->af_specific->sockaddr_len);
48 memset(&peer->saddr, 0, sizeof(union sctp_addr));
49
50 peer->sack_generation = 0;
51
52 /* From 6.3.1 RTO Calculation:
53 *
54 * C1) Until an RTT measurement has been made for a packet sent to the
55 * given destination transport address, set RTO to the protocol
56 * parameter 'RTO.Initial'.
57 */
58 peer->rto = msecs_to_jiffies(net->sctp.rto_initial);
59
60 peer->last_time_heard = 0;
61 peer->last_time_ecne_reduced = jiffies;
62
63 peer->param_flags = SPP_HB_DISABLE |
64 SPP_PMTUD_ENABLE |
65 SPP_SACKDELAY_ENABLE;
66
67 /* Initialize the default path max_retrans. */
68 peer->pathmaxrxt = net->sctp.max_retrans_path;
69 peer->pf_retrans = net->sctp.pf_retrans;
70
71 INIT_LIST_HEAD(&peer->transmitted);
72 INIT_LIST_HEAD(&peer->send_ready);
73 INIT_LIST_HEAD(&peer->transports);
74
75 timer_setup(&peer->T3_rtx_timer, sctp_generate_t3_rtx_event, 0);
76 timer_setup(&peer->hb_timer, sctp_generate_heartbeat_event, 0);
77 timer_setup(&peer->reconf_timer, sctp_generate_reconf_event, 0);
78 timer_setup(&peer->probe_timer, sctp_generate_probe_event, 0);
79 timer_setup(&peer->proto_unreach_timer,
80 sctp_generate_proto_unreach_event, 0);
81
82 /* Initialize the 64-bit random nonce sent with heartbeat. */
83 get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce));
84
85 refcount_set(&peer->refcnt, 1);
86
87 return peer;
88 }
89
90 /* Allocate and initialize a new transport. */
sctp_transport_new(struct net * net,const union sctp_addr * addr,gfp_t gfp)91 struct sctp_transport *sctp_transport_new(struct net *net,
92 const union sctp_addr *addr,
93 gfp_t gfp)
94 {
95 struct sctp_transport *transport;
96
97 transport = kzalloc(sizeof(*transport), gfp);
98 if (!transport)
99 goto fail;
100
101 if (!sctp_transport_init(net, transport, addr, gfp))
102 goto fail_init;
103
104 SCTP_DBG_OBJCNT_INC(transport);
105
106 return transport;
107
108 fail_init:
109 kfree(transport);
110
111 fail:
112 return NULL;
113 }
114
115 /* This transport is no longer needed. Free up if possible, or
116 * delay until it last reference count.
117 */
sctp_transport_free(struct sctp_transport * transport)118 void sctp_transport_free(struct sctp_transport *transport)
119 {
120 /* Try to delete the heartbeat timer. */
121 if (del_timer(&transport->hb_timer))
122 sctp_transport_put(transport);
123
124 /* Delete the T3_rtx timer if it's active.
125 * There is no point in not doing this now and letting
126 * structure hang around in memory since we know
127 * the transport is going away.
128 */
129 if (del_timer(&transport->T3_rtx_timer))
130 sctp_transport_put(transport);
131
132 if (del_timer(&transport->reconf_timer))
133 sctp_transport_put(transport);
134
135 if (del_timer(&transport->probe_timer))
136 sctp_transport_put(transport);
137
138 /* Delete the ICMP proto unreachable timer if it's active. */
139 if (del_timer(&transport->proto_unreach_timer))
140 sctp_transport_put(transport);
141
142 sctp_transport_put(transport);
143 }
144
sctp_transport_destroy_rcu(struct rcu_head * head)145 static void sctp_transport_destroy_rcu(struct rcu_head *head)
146 {
147 struct sctp_transport *transport;
148
149 transport = container_of(head, struct sctp_transport, rcu);
150
151 dst_release(transport->dst);
152 kfree(transport);
153 SCTP_DBG_OBJCNT_DEC(transport);
154 }
155
156 /* Destroy the transport data structure.
157 * Assumes there are no more users of this structure.
158 */
sctp_transport_destroy(struct sctp_transport * transport)159 static void sctp_transport_destroy(struct sctp_transport *transport)
160 {
161 if (unlikely(refcount_read(&transport->refcnt))) {
162 WARN(1, "Attempt to destroy undead transport %p!\n", transport);
163 return;
164 }
165
166 sctp_packet_free(&transport->packet);
167
168 if (transport->asoc)
169 sctp_association_put(transport->asoc);
170
171 call_rcu(&transport->rcu, sctp_transport_destroy_rcu);
172 }
173
174 /* Start T3_rtx timer if it is not already running and update the heartbeat
175 * timer. This routine is called every time a DATA chunk is sent.
176 */
sctp_transport_reset_t3_rtx(struct sctp_transport * transport)177 void sctp_transport_reset_t3_rtx(struct sctp_transport *transport)
178 {
179 /* RFC 2960 6.3.2 Retransmission Timer Rules
180 *
181 * R1) Every time a DATA chunk is sent to any address(including a
182 * retransmission), if the T3-rtx timer of that address is not running
183 * start it running so that it will expire after the RTO of that
184 * address.
185 */
186
187 if (!timer_pending(&transport->T3_rtx_timer))
188 if (!mod_timer(&transport->T3_rtx_timer,
189 jiffies + transport->rto))
190 sctp_transport_hold(transport);
191 }
192
sctp_transport_reset_hb_timer(struct sctp_transport * transport)193 void sctp_transport_reset_hb_timer(struct sctp_transport *transport)
194 {
195 unsigned long expires;
196
197 /* When a data chunk is sent, reset the heartbeat interval. */
198 expires = jiffies + sctp_transport_timeout(transport);
199 if ((time_before(transport->hb_timer.expires, expires) ||
200 !timer_pending(&transport->hb_timer)) &&
201 !mod_timer(&transport->hb_timer,
202 expires + prandom_u32_max(transport->rto)))
203 sctp_transport_hold(transport);
204 }
205
sctp_transport_reset_reconf_timer(struct sctp_transport * transport)206 void sctp_transport_reset_reconf_timer(struct sctp_transport *transport)
207 {
208 if (!timer_pending(&transport->reconf_timer))
209 if (!mod_timer(&transport->reconf_timer,
210 jiffies + transport->rto))
211 sctp_transport_hold(transport);
212 }
213
sctp_transport_reset_probe_timer(struct sctp_transport * transport)214 void sctp_transport_reset_probe_timer(struct sctp_transport *transport)
215 {
216 if (!mod_timer(&transport->probe_timer,
217 jiffies + transport->probe_interval))
218 sctp_transport_hold(transport);
219 }
220
sctp_transport_reset_raise_timer(struct sctp_transport * transport)221 void sctp_transport_reset_raise_timer(struct sctp_transport *transport)
222 {
223 if (!mod_timer(&transport->probe_timer,
224 jiffies + transport->probe_interval * 30))
225 sctp_transport_hold(transport);
226 }
227
228 /* This transport has been assigned to an association.
229 * Initialize fields from the association or from the sock itself.
230 * Register the reference count in the association.
231 */
sctp_transport_set_owner(struct sctp_transport * transport,struct sctp_association * asoc)232 void sctp_transport_set_owner(struct sctp_transport *transport,
233 struct sctp_association *asoc)
234 {
235 transport->asoc = asoc;
236 sctp_association_hold(asoc);
237 }
238
239 /* Initialize the pmtu of a transport. */
sctp_transport_pmtu(struct sctp_transport * transport,struct sock * sk)240 void sctp_transport_pmtu(struct sctp_transport *transport, struct sock *sk)
241 {
242 /* If we don't have a fresh route, look one up */
243 if (!transport->dst || transport->dst->obsolete) {
244 sctp_transport_dst_release(transport);
245 transport->af_specific->get_dst(transport, &transport->saddr,
246 &transport->fl, sk);
247 }
248
249 if (transport->param_flags & SPP_PMTUD_DISABLE) {
250 struct sctp_association *asoc = transport->asoc;
251
252 if (!transport->pathmtu && asoc && asoc->pathmtu)
253 transport->pathmtu = asoc->pathmtu;
254 if (transport->pathmtu)
255 return;
256 }
257
258 if (transport->dst)
259 transport->pathmtu = sctp_dst_mtu(transport->dst);
260 else
261 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
262
263 sctp_transport_pl_update(transport);
264 }
265
sctp_transport_pl_send(struct sctp_transport * t)266 void sctp_transport_pl_send(struct sctp_transport *t)
267 {
268 if (t->pl.probe_count < SCTP_MAX_PROBES)
269 goto out;
270
271 t->pl.probe_count = 0;
272 if (t->pl.state == SCTP_PL_BASE) {
273 if (t->pl.probe_size == SCTP_BASE_PLPMTU) { /* BASE_PLPMTU Confirmation Failed */
274 t->pl.state = SCTP_PL_ERROR; /* Base -> Error */
275
276 t->pl.pmtu = SCTP_BASE_PLPMTU;
277 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
278 sctp_assoc_sync_pmtu(t->asoc);
279 }
280 } else if (t->pl.state == SCTP_PL_SEARCH) {
281 if (t->pl.pmtu == t->pl.probe_size) { /* Black Hole Detected */
282 t->pl.state = SCTP_PL_BASE; /* Search -> Base */
283 t->pl.probe_size = SCTP_BASE_PLPMTU;
284 t->pl.probe_high = 0;
285
286 t->pl.pmtu = SCTP_BASE_PLPMTU;
287 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
288 sctp_assoc_sync_pmtu(t->asoc);
289 } else { /* Normal probe failure. */
290 t->pl.probe_high = t->pl.probe_size;
291 t->pl.probe_size = t->pl.pmtu;
292 }
293 } else if (t->pl.state == SCTP_PL_COMPLETE) {
294 if (t->pl.pmtu == t->pl.probe_size) { /* Black Hole Detected */
295 t->pl.state = SCTP_PL_BASE; /* Search Complete -> Base */
296 t->pl.probe_size = SCTP_BASE_PLPMTU;
297
298 t->pl.pmtu = SCTP_BASE_PLPMTU;
299 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
300 sctp_assoc_sync_pmtu(t->asoc);
301 }
302 }
303
304 out:
305 pr_debug("%s: PLPMTUD: transport: %p, state: %d, pmtu: %d, size: %d, high: %d\n",
306 __func__, t, t->pl.state, t->pl.pmtu, t->pl.probe_size, t->pl.probe_high);
307 t->pl.probe_count++;
308 }
309
sctp_transport_pl_recv(struct sctp_transport * t)310 bool sctp_transport_pl_recv(struct sctp_transport *t)
311 {
312 pr_debug("%s: PLPMTUD: transport: %p, state: %d, pmtu: %d, size: %d, high: %d\n",
313 __func__, t, t->pl.state, t->pl.pmtu, t->pl.probe_size, t->pl.probe_high);
314
315 t->pl.pmtu = t->pl.probe_size;
316 t->pl.probe_count = 0;
317 if (t->pl.state == SCTP_PL_BASE) {
318 t->pl.state = SCTP_PL_SEARCH; /* Base -> Search */
319 t->pl.probe_size += SCTP_PL_BIG_STEP;
320 } else if (t->pl.state == SCTP_PL_ERROR) {
321 t->pl.state = SCTP_PL_SEARCH; /* Error -> Search */
322
323 t->pl.pmtu = t->pl.probe_size;
324 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
325 sctp_assoc_sync_pmtu(t->asoc);
326 t->pl.probe_size += SCTP_PL_BIG_STEP;
327 } else if (t->pl.state == SCTP_PL_SEARCH) {
328 if (!t->pl.probe_high) {
329 t->pl.probe_size = min(t->pl.probe_size + SCTP_PL_BIG_STEP,
330 SCTP_MAX_PLPMTU);
331 return false;
332 }
333 t->pl.probe_size += SCTP_PL_MIN_STEP;
334 if (t->pl.probe_size >= t->pl.probe_high) {
335 t->pl.probe_high = 0;
336 t->pl.state = SCTP_PL_COMPLETE; /* Search -> Search Complete */
337
338 t->pl.probe_size = t->pl.pmtu;
339 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
340 sctp_assoc_sync_pmtu(t->asoc);
341 sctp_transport_reset_raise_timer(t);
342 }
343 } else if (t->pl.state == SCTP_PL_COMPLETE) {
344 /* Raise probe_size again after 30 * interval in Search Complete */
345 t->pl.state = SCTP_PL_SEARCH; /* Search Complete -> Search */
346 t->pl.probe_size += SCTP_PL_MIN_STEP;
347 }
348
349 return t->pl.state == SCTP_PL_COMPLETE;
350 }
351
sctp_transport_pl_toobig(struct sctp_transport * t,u32 pmtu)352 static bool sctp_transport_pl_toobig(struct sctp_transport *t, u32 pmtu)
353 {
354 pr_debug("%s: PLPMTUD: transport: %p, state: %d, pmtu: %d, size: %d, ptb: %d\n",
355 __func__, t, t->pl.state, t->pl.pmtu, t->pl.probe_size, pmtu);
356
357 if (pmtu < SCTP_MIN_PLPMTU || pmtu >= t->pl.probe_size)
358 return false;
359
360 if (t->pl.state == SCTP_PL_BASE) {
361 if (pmtu >= SCTP_MIN_PLPMTU && pmtu < SCTP_BASE_PLPMTU) {
362 t->pl.state = SCTP_PL_ERROR; /* Base -> Error */
363
364 t->pl.pmtu = SCTP_BASE_PLPMTU;
365 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
366 return true;
367 }
368 } else if (t->pl.state == SCTP_PL_SEARCH) {
369 if (pmtu >= SCTP_BASE_PLPMTU && pmtu < t->pl.pmtu) {
370 t->pl.state = SCTP_PL_BASE; /* Search -> Base */
371 t->pl.probe_size = SCTP_BASE_PLPMTU;
372 t->pl.probe_count = 0;
373
374 t->pl.probe_high = 0;
375 t->pl.pmtu = SCTP_BASE_PLPMTU;
376 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
377 return true;
378 } else if (pmtu > t->pl.pmtu && pmtu < t->pl.probe_size) {
379 t->pl.probe_size = pmtu;
380 t->pl.probe_count = 0;
381 }
382 } else if (t->pl.state == SCTP_PL_COMPLETE) {
383 if (pmtu >= SCTP_BASE_PLPMTU && pmtu < t->pl.pmtu) {
384 t->pl.state = SCTP_PL_BASE; /* Complete -> Base */
385 t->pl.probe_size = SCTP_BASE_PLPMTU;
386 t->pl.probe_count = 0;
387
388 t->pl.probe_high = 0;
389 t->pl.pmtu = SCTP_BASE_PLPMTU;
390 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
391 sctp_transport_reset_probe_timer(t);
392 return true;
393 }
394 }
395
396 return false;
397 }
398
sctp_transport_update_pmtu(struct sctp_transport * t,u32 pmtu)399 bool sctp_transport_update_pmtu(struct sctp_transport *t, u32 pmtu)
400 {
401 struct sock *sk = t->asoc->base.sk;
402 struct dst_entry *dst;
403 bool change = true;
404
405 if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) {
406 pr_warn_ratelimited("%s: Reported pmtu %d too low, using default minimum of %d\n",
407 __func__, pmtu, SCTP_DEFAULT_MINSEGMENT);
408 /* Use default minimum segment instead */
409 pmtu = SCTP_DEFAULT_MINSEGMENT;
410 }
411 pmtu = SCTP_TRUNC4(pmtu);
412
413 if (sctp_transport_pl_enabled(t))
414 return sctp_transport_pl_toobig(t, pmtu - sctp_transport_pl_hlen(t));
415
416 dst = sctp_transport_dst_check(t);
417 if (dst) {
418 struct sctp_pf *pf = sctp_get_pf_specific(dst->ops->family);
419 union sctp_addr addr;
420
421 pf->af->from_sk(&addr, sk);
422 pf->to_sk_daddr(&t->ipaddr, sk);
423 dst->ops->update_pmtu(dst, sk, NULL, pmtu, true);
424 pf->to_sk_daddr(&addr, sk);
425
426 dst = sctp_transport_dst_check(t);
427 }
428
429 if (!dst) {
430 t->af_specific->get_dst(t, &t->saddr, &t->fl, sk);
431 dst = t->dst;
432 }
433
434 if (dst) {
435 /* Re-fetch, as under layers may have a higher minimum size */
436 pmtu = sctp_dst_mtu(dst);
437 change = t->pathmtu != pmtu;
438 }
439 t->pathmtu = pmtu;
440
441 return change;
442 }
443
444 /* Caches the dst entry and source address for a transport's destination
445 * address.
446 */
sctp_transport_route(struct sctp_transport * transport,union sctp_addr * saddr,struct sctp_sock * opt)447 void sctp_transport_route(struct sctp_transport *transport,
448 union sctp_addr *saddr, struct sctp_sock *opt)
449 {
450 struct sctp_association *asoc = transport->asoc;
451 struct sctp_af *af = transport->af_specific;
452
453 sctp_transport_dst_release(transport);
454 af->get_dst(transport, saddr, &transport->fl, sctp_opt2sk(opt));
455
456 if (saddr)
457 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
458 else
459 af->get_saddr(opt, transport, &transport->fl);
460
461 sctp_transport_pmtu(transport, sctp_opt2sk(opt));
462
463 /* Initialize sk->sk_rcv_saddr, if the transport is the
464 * association's active path for getsockname().
465 */
466 if (transport->dst && asoc &&
467 (!asoc->peer.primary_path || transport == asoc->peer.active_path))
468 opt->pf->to_sk_saddr(&transport->saddr, asoc->base.sk);
469 }
470
471 /* Hold a reference to a transport. */
sctp_transport_hold(struct sctp_transport * transport)472 int sctp_transport_hold(struct sctp_transport *transport)
473 {
474 return refcount_inc_not_zero(&transport->refcnt);
475 }
476
477 /* Release a reference to a transport and clean up
478 * if there are no more references.
479 */
sctp_transport_put(struct sctp_transport * transport)480 void sctp_transport_put(struct sctp_transport *transport)
481 {
482 if (refcount_dec_and_test(&transport->refcnt))
483 sctp_transport_destroy(transport);
484 }
485
486 /* Update transport's RTO based on the newly calculated RTT. */
sctp_transport_update_rto(struct sctp_transport * tp,__u32 rtt)487 void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
488 {
489 if (unlikely(!tp->rto_pending))
490 /* We should not be doing any RTO updates unless rto_pending is set. */
491 pr_debug("%s: rto_pending not set on transport %p!\n", __func__, tp);
492
493 if (tp->rttvar || tp->srtt) {
494 struct net *net = tp->asoc->base.net;
495 /* 6.3.1 C3) When a new RTT measurement R' is made, set
496 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
497 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
498 */
499
500 /* Note: The above algorithm has been rewritten to
501 * express rto_beta and rto_alpha as inverse powers
502 * of two.
503 * For example, assuming the default value of RTO.Alpha of
504 * 1/8, rto_alpha would be expressed as 3.
505 */
506 tp->rttvar = tp->rttvar - (tp->rttvar >> net->sctp.rto_beta)
507 + (((__u32)abs((__s64)tp->srtt - (__s64)rtt)) >> net->sctp.rto_beta);
508 tp->srtt = tp->srtt - (tp->srtt >> net->sctp.rto_alpha)
509 + (rtt >> net->sctp.rto_alpha);
510 } else {
511 /* 6.3.1 C2) When the first RTT measurement R is made, set
512 * SRTT <- R, RTTVAR <- R/2.
513 */
514 tp->srtt = rtt;
515 tp->rttvar = rtt >> 1;
516 }
517
518 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
519 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
520 */
521 if (tp->rttvar == 0)
522 tp->rttvar = SCTP_CLOCK_GRANULARITY;
523
524 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
525 tp->rto = tp->srtt + (tp->rttvar << 2);
526
527 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
528 * seconds then it is rounded up to RTO.Min seconds.
529 */
530 if (tp->rto < tp->asoc->rto_min)
531 tp->rto = tp->asoc->rto_min;
532
533 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
534 * at least RTO.max seconds.
535 */
536 if (tp->rto > tp->asoc->rto_max)
537 tp->rto = tp->asoc->rto_max;
538
539 sctp_max_rto(tp->asoc, tp);
540 tp->rtt = rtt;
541
542 /* Reset rto_pending so that a new RTT measurement is started when a
543 * new data chunk is sent.
544 */
545 tp->rto_pending = 0;
546
547 pr_debug("%s: transport:%p, rtt:%d, srtt:%d rttvar:%d, rto:%ld\n",
548 __func__, tp, rtt, tp->srtt, tp->rttvar, tp->rto);
549 }
550
551 /* This routine updates the transport's cwnd and partial_bytes_acked
552 * parameters based on the bytes acked in the received SACK.
553 */
sctp_transport_raise_cwnd(struct sctp_transport * transport,__u32 sack_ctsn,__u32 bytes_acked)554 void sctp_transport_raise_cwnd(struct sctp_transport *transport,
555 __u32 sack_ctsn, __u32 bytes_acked)
556 {
557 struct sctp_association *asoc = transport->asoc;
558 __u32 cwnd, ssthresh, flight_size, pba, pmtu;
559
560 cwnd = transport->cwnd;
561 flight_size = transport->flight_size;
562
563 /* See if we need to exit Fast Recovery first */
564 if (asoc->fast_recovery &&
565 TSN_lte(asoc->fast_recovery_exit, sack_ctsn))
566 asoc->fast_recovery = 0;
567
568 ssthresh = transport->ssthresh;
569 pba = transport->partial_bytes_acked;
570 pmtu = transport->asoc->pathmtu;
571
572 if (cwnd <= ssthresh) {
573 /* RFC 4960 7.2.1
574 * o When cwnd is less than or equal to ssthresh, an SCTP
575 * endpoint MUST use the slow-start algorithm to increase
576 * cwnd only if the current congestion window is being fully
577 * utilized, an incoming SACK advances the Cumulative TSN
578 * Ack Point, and the data sender is not in Fast Recovery.
579 * Only when these three conditions are met can the cwnd be
580 * increased; otherwise, the cwnd MUST not be increased.
581 * If these conditions are met, then cwnd MUST be increased
582 * by, at most, the lesser of 1) the total size of the
583 * previously outstanding DATA chunk(s) acknowledged, and
584 * 2) the destination's path MTU. This upper bound protects
585 * against the ACK-Splitting attack outlined in [SAVAGE99].
586 */
587 if (asoc->fast_recovery)
588 return;
589
590 /* The appropriate cwnd increase algorithm is performed
591 * if, and only if the congestion window is being fully
592 * utilized. Note that RFC4960 Errata 3.22 removed the
593 * other condition on ctsn moving.
594 */
595 if (flight_size < cwnd)
596 return;
597
598 if (bytes_acked > pmtu)
599 cwnd += pmtu;
600 else
601 cwnd += bytes_acked;
602
603 pr_debug("%s: slow start: transport:%p, bytes_acked:%d, "
604 "cwnd:%d, ssthresh:%d, flight_size:%d, pba:%d\n",
605 __func__, transport, bytes_acked, cwnd, ssthresh,
606 flight_size, pba);
607 } else {
608 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
609 * upon each SACK arrival, increase partial_bytes_acked
610 * by the total number of bytes of all new chunks
611 * acknowledged in that SACK including chunks
612 * acknowledged by the new Cumulative TSN Ack and by Gap
613 * Ack Blocks. (updated by RFC4960 Errata 3.22)
614 *
615 * When partial_bytes_acked is greater than cwnd and
616 * before the arrival of the SACK the sender had less
617 * bytes of data outstanding than cwnd (i.e., before
618 * arrival of the SACK, flightsize was less than cwnd),
619 * reset partial_bytes_acked to cwnd. (RFC 4960 Errata
620 * 3.26)
621 *
622 * When partial_bytes_acked is equal to or greater than
623 * cwnd and before the arrival of the SACK the sender
624 * had cwnd or more bytes of data outstanding (i.e.,
625 * before arrival of the SACK, flightsize was greater
626 * than or equal to cwnd), partial_bytes_acked is reset
627 * to (partial_bytes_acked - cwnd). Next, cwnd is
628 * increased by MTU. (RFC 4960 Errata 3.12)
629 */
630 pba += bytes_acked;
631 if (pba > cwnd && flight_size < cwnd)
632 pba = cwnd;
633 if (pba >= cwnd && flight_size >= cwnd) {
634 pba = pba - cwnd;
635 cwnd += pmtu;
636 }
637
638 pr_debug("%s: congestion avoidance: transport:%p, "
639 "bytes_acked:%d, cwnd:%d, ssthresh:%d, "
640 "flight_size:%d, pba:%d\n", __func__,
641 transport, bytes_acked, cwnd, ssthresh,
642 flight_size, pba);
643 }
644
645 transport->cwnd = cwnd;
646 transport->partial_bytes_acked = pba;
647 }
648
649 /* This routine is used to lower the transport's cwnd when congestion is
650 * detected.
651 */
sctp_transport_lower_cwnd(struct sctp_transport * transport,enum sctp_lower_cwnd reason)652 void sctp_transport_lower_cwnd(struct sctp_transport *transport,
653 enum sctp_lower_cwnd reason)
654 {
655 struct sctp_association *asoc = transport->asoc;
656
657 switch (reason) {
658 case SCTP_LOWER_CWND_T3_RTX:
659 /* RFC 2960 Section 7.2.3, sctpimpguide
660 * When the T3-rtx timer expires on an address, SCTP should
661 * perform slow start by:
662 * ssthresh = max(cwnd/2, 4*MTU)
663 * cwnd = 1*MTU
664 * partial_bytes_acked = 0
665 */
666 transport->ssthresh = max(transport->cwnd/2,
667 4*asoc->pathmtu);
668 transport->cwnd = asoc->pathmtu;
669
670 /* T3-rtx also clears fast recovery */
671 asoc->fast_recovery = 0;
672 break;
673
674 case SCTP_LOWER_CWND_FAST_RTX:
675 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
676 * destination address(es) to which the missing DATA chunks
677 * were last sent, according to the formula described in
678 * Section 7.2.3.
679 *
680 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
681 * losses from SACK (see Section 7.2.4), An endpoint
682 * should do the following:
683 * ssthresh = max(cwnd/2, 4*MTU)
684 * cwnd = ssthresh
685 * partial_bytes_acked = 0
686 */
687 if (asoc->fast_recovery)
688 return;
689
690 /* Mark Fast recovery */
691 asoc->fast_recovery = 1;
692 asoc->fast_recovery_exit = asoc->next_tsn - 1;
693
694 transport->ssthresh = max(transport->cwnd/2,
695 4*asoc->pathmtu);
696 transport->cwnd = transport->ssthresh;
697 break;
698
699 case SCTP_LOWER_CWND_ECNE:
700 /* RFC 2481 Section 6.1.2.
701 * If the sender receives an ECN-Echo ACK packet
702 * then the sender knows that congestion was encountered in the
703 * network on the path from the sender to the receiver. The
704 * indication of congestion should be treated just as a
705 * congestion loss in non-ECN Capable TCP. That is, the TCP
706 * source halves the congestion window "cwnd" and reduces the
707 * slow start threshold "ssthresh".
708 * A critical condition is that TCP does not react to
709 * congestion indications more than once every window of
710 * data (or more loosely more than once every round-trip time).
711 */
712 if (time_after(jiffies, transport->last_time_ecne_reduced +
713 transport->rtt)) {
714 transport->ssthresh = max(transport->cwnd/2,
715 4*asoc->pathmtu);
716 transport->cwnd = transport->ssthresh;
717 transport->last_time_ecne_reduced = jiffies;
718 }
719 break;
720
721 case SCTP_LOWER_CWND_INACTIVE:
722 /* RFC 2960 Section 7.2.1, sctpimpguide
723 * When the endpoint does not transmit data on a given
724 * transport address, the cwnd of the transport address
725 * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
726 * NOTE: Although the draft recommends that this check needs
727 * to be done every RTO interval, we do it every hearbeat
728 * interval.
729 */
730 transport->cwnd = max(transport->cwnd/2,
731 4*asoc->pathmtu);
732 /* RFC 4960 Errata 3.27.2: also adjust sshthresh */
733 transport->ssthresh = transport->cwnd;
734 break;
735 }
736
737 transport->partial_bytes_acked = 0;
738
739 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d\n",
740 __func__, transport, reason, transport->cwnd,
741 transport->ssthresh);
742 }
743
744 /* Apply Max.Burst limit to the congestion window:
745 * sctpimpguide-05 2.14.2
746 * D) When the time comes for the sender to
747 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
748 * first be applied to limit how many new DATA chunks may be sent.
749 * The limit is applied by adjusting cwnd as follows:
750 * if ((flightsize+ Max.Burst * MTU) < cwnd)
751 * cwnd = flightsize + Max.Burst * MTU
752 */
753
sctp_transport_burst_limited(struct sctp_transport * t)754 void sctp_transport_burst_limited(struct sctp_transport *t)
755 {
756 struct sctp_association *asoc = t->asoc;
757 u32 old_cwnd = t->cwnd;
758 u32 max_burst_bytes;
759
760 if (t->burst_limited || asoc->max_burst == 0)
761 return;
762
763 max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu);
764 if (max_burst_bytes < old_cwnd) {
765 t->cwnd = max_burst_bytes;
766 t->burst_limited = old_cwnd;
767 }
768 }
769
770 /* Restore the old cwnd congestion window, after the burst had it's
771 * desired effect.
772 */
sctp_transport_burst_reset(struct sctp_transport * t)773 void sctp_transport_burst_reset(struct sctp_transport *t)
774 {
775 if (t->burst_limited) {
776 t->cwnd = t->burst_limited;
777 t->burst_limited = 0;
778 }
779 }
780
781 /* What is the next timeout value for this transport? */
sctp_transport_timeout(struct sctp_transport * trans)782 unsigned long sctp_transport_timeout(struct sctp_transport *trans)
783 {
784 /* RTO + timer slack +/- 50% of RTO */
785 unsigned long timeout = trans->rto >> 1;
786
787 if (trans->state != SCTP_UNCONFIRMED &&
788 trans->state != SCTP_PF)
789 timeout += trans->hbinterval;
790
791 return max_t(unsigned long, timeout, HZ / 5);
792 }
793
794 /* Reset transport variables to their initial values */
sctp_transport_reset(struct sctp_transport * t)795 void sctp_transport_reset(struct sctp_transport *t)
796 {
797 struct sctp_association *asoc = t->asoc;
798
799 /* RFC 2960 (bis), Section 5.2.4
800 * All the congestion control parameters (e.g., cwnd, ssthresh)
801 * related to this peer MUST be reset to their initial values
802 * (see Section 6.2.1)
803 */
804 t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
805 t->burst_limited = 0;
806 t->ssthresh = asoc->peer.i.a_rwnd;
807 t->rto = asoc->rto_initial;
808 sctp_max_rto(asoc, t);
809 t->rtt = 0;
810 t->srtt = 0;
811 t->rttvar = 0;
812
813 /* Reset these additional variables so that we have a clean slate. */
814 t->partial_bytes_acked = 0;
815 t->flight_size = 0;
816 t->error_count = 0;
817 t->rto_pending = 0;
818 t->hb_sent = 0;
819
820 /* Initialize the state information for SFR-CACC */
821 t->cacc.changeover_active = 0;
822 t->cacc.cycling_changeover = 0;
823 t->cacc.next_tsn_at_change = 0;
824 t->cacc.cacc_saw_newack = 0;
825 }
826
827 /* Schedule retransmission on the given transport */
sctp_transport_immediate_rtx(struct sctp_transport * t)828 void sctp_transport_immediate_rtx(struct sctp_transport *t)
829 {
830 /* Stop pending T3_rtx_timer */
831 if (del_timer(&t->T3_rtx_timer))
832 sctp_transport_put(t);
833
834 sctp_retransmit(&t->asoc->outqueue, t, SCTP_RTXR_T3_RTX);
835 if (!timer_pending(&t->T3_rtx_timer)) {
836 if (!mod_timer(&t->T3_rtx_timer, jiffies + t->rto))
837 sctp_transport_hold(t);
838 }
839 }
840
841 /* Drop dst */
sctp_transport_dst_release(struct sctp_transport * t)842 void sctp_transport_dst_release(struct sctp_transport *t)
843 {
844 dst_release(t->dst);
845 t->dst = NULL;
846 t->dst_pending_confirm = 0;
847 }
848
849 /* Schedule neighbour confirm */
sctp_transport_dst_confirm(struct sctp_transport * t)850 void sctp_transport_dst_confirm(struct sctp_transport *t)
851 {
852 t->dst_pending_confirm = 1;
853 }
854