]>
Commit | Line | Data |
---|---|---|
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 | |
7 | * | |
8 | * This file is part of the SCTP kernel implementation | |
9 | * | |
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 | * | |
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. | |
19 | * | |
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. | |
25 | * | |
26 | * You should have received a copy of the GNU General Public License | |
27 | * along with GNU CC; see the file COPYING. If not, see | |
28 | * <http://www.gnu.org/licenses/>. | |
29 | * | |
30 | * Please send any bug reports or fixes you make to the | |
31 | * email address(es): | |
32 | * lksctp developers <linux-sctp@vger.kernel.org> | |
33 | * | |
34 | * 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> | |
42 | */ | |
43 | ||
44 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
45 | ||
46 | #include <linux/slab.h> | |
47 | #include <linux/types.h> | |
48 | #include <linux/random.h> | |
49 | #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. */ | |
55 | static struct sctp_transport *sctp_transport_init(struct net *net, | |
56 | struct sctp_transport *peer, | |
57 | const union sctp_addr *addr, | |
58 | gfp_t gfp) | |
59 | { | |
60 | /* Copy in the address. */ | |
61 | peer->ipaddr = *addr; | |
62 | peer->af_specific = sctp_get_af_specific(addr->sa.sa_family); | |
63 | memset(&peer->saddr, 0, sizeof(union sctp_addr)); | |
64 | ||
65 | peer->sack_generation = 0; | |
66 | ||
67 | /* 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 | */ | |
73 | peer->rto = msecs_to_jiffies(net->sctp.rto_initial); | |
74 | ||
75 | peer->last_time_heard = 0; | |
76 | peer->last_time_ecne_reduced = jiffies; | |
77 | ||
78 | peer->param_flags = SPP_HB_DISABLE | | |
79 | SPP_PMTUD_ENABLE | | |
80 | SPP_SACKDELAY_ENABLE; | |
81 | ||
82 | /* Initialize the default path max_retrans. */ | |
83 | peer->pathmaxrxt = net->sctp.max_retrans_path; | |
84 | peer->pf_retrans = net->sctp.pf_retrans; | |
85 | ||
86 | INIT_LIST_HEAD(&peer->transmitted); | |
87 | INIT_LIST_HEAD(&peer->send_ready); | |
88 | INIT_LIST_HEAD(&peer->transports); | |
89 | ||
90 | 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); | |
94 | setup_timer(&peer->proto_unreach_timer, | |
95 | sctp_generate_proto_unreach_event, (unsigned long)peer); | |
96 | ||
97 | /* Initialize the 64-bit random nonce sent with heartbeat. */ | |
98 | get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce)); | |
99 | ||
100 | atomic_set(&peer->refcnt, 1); | |
101 | ||
102 | return peer; | |
103 | } | |
104 | ||
105 | /* Allocate and initialize a new transport. */ | |
106 | struct sctp_transport *sctp_transport_new(struct net *net, | |
107 | const union sctp_addr *addr, | |
108 | gfp_t gfp) | |
109 | { | |
110 | struct sctp_transport *transport; | |
111 | ||
112 | transport = kzalloc(sizeof(*transport), gfp); | |
113 | if (!transport) | |
114 | goto fail; | |
115 | ||
116 | if (!sctp_transport_init(net, transport, addr, gfp)) | |
117 | goto fail_init; | |
118 | ||
119 | SCTP_DBG_OBJCNT_INC(transport); | |
120 | ||
121 | return transport; | |
122 | ||
123 | fail_init: | |
124 | kfree(transport); | |
125 | ||
126 | fail: | |
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 | */ | |
133 | void sctp_transport_free(struct sctp_transport *transport) | |
134 | { | |
135 | /* Try to delete the heartbeat timer. */ | |
136 | if (del_timer(&transport->hb_timer)) | |
137 | sctp_transport_put(transport); | |
138 | ||
139 | /* Delete the T3_rtx timer if it's active. | |
140 | * There is no point in not doing this now and letting | |
141 | * structure hang around in memory since we know | |
142 | * the tranport is going away. | |
143 | */ | |
144 | if (del_timer(&transport->T3_rtx_timer)) | |
145 | sctp_transport_put(transport); | |
146 | ||
147 | /* Delete the ICMP proto unreachable timer if it's active. */ | |
148 | if (del_timer(&transport->proto_unreach_timer)) | |
149 | sctp_association_put(transport->asoc); | |
150 | ||
151 | sctp_transport_put(transport); | |
152 | } | |
153 | ||
154 | static void sctp_transport_destroy_rcu(struct rcu_head *head) | |
155 | { | |
156 | struct sctp_transport *transport; | |
157 | ||
158 | transport = container_of(head, struct sctp_transport, rcu); | |
159 | ||
160 | dst_release(transport->dst); | |
161 | kfree(transport); | |
162 | SCTP_DBG_OBJCNT_DEC(transport); | |
163 | } | |
164 | ||
165 | /* Destroy the transport data structure. | |
166 | * Assumes there are no more users of this structure. | |
167 | */ | |
168 | static void sctp_transport_destroy(struct sctp_transport *transport) | |
169 | { | |
170 | if (unlikely(atomic_read(&transport->refcnt))) { | |
171 | WARN(1, "Attempt to destroy undead transport %p!\n", transport); | |
172 | return; | |
173 | } | |
174 | ||
175 | sctp_packet_free(&transport->packet); | |
176 | ||
177 | if (transport->asoc) | |
178 | sctp_association_put(transport->asoc); | |
179 | ||
180 | call_rcu(&transport->rcu, sctp_transport_destroy_rcu); | |
181 | } | |
182 | ||
183 | /* Start T3_rtx timer if it is not already running and update the heartbeat | |
184 | * timer. This routine is called every time a DATA chunk is sent. | |
185 | */ | |
186 | void sctp_transport_reset_t3_rtx(struct sctp_transport *transport) | |
187 | { | |
188 | /* RFC 2960 6.3.2 Retransmission Timer Rules | |
189 | * | |
190 | * R1) Every time a DATA chunk is sent to any address(including a | |
191 | * retransmission), if the T3-rtx timer of that address is not running | |
192 | * start it running so that it will expire after the RTO of that | |
193 | * address. | |
194 | */ | |
195 | ||
196 | if (!timer_pending(&transport->T3_rtx_timer)) | |
197 | if (!mod_timer(&transport->T3_rtx_timer, | |
198 | jiffies + transport->rto)) | |
199 | sctp_transport_hold(transport); | |
200 | } | |
201 | ||
202 | void sctp_transport_reset_hb_timer(struct sctp_transport *transport) | |
203 | { | |
204 | unsigned long expires; | |
205 | ||
206 | /* When a data chunk is sent, reset the heartbeat interval. */ | |
207 | expires = jiffies + sctp_transport_timeout(transport); | |
208 | if (time_before(transport->hb_timer.expires, expires) && | |
209 | !mod_timer(&transport->hb_timer, | |
210 | expires + prandom_u32_max(transport->rto))) | |
211 | sctp_transport_hold(transport); | |
212 | } | |
213 | ||
214 | /* This transport has been assigned to an association. | |
215 | * Initialize fields from the association or from the sock itself. | |
216 | * Register the reference count in the association. | |
217 | */ | |
218 | void sctp_transport_set_owner(struct sctp_transport *transport, | |
219 | struct sctp_association *asoc) | |
220 | { | |
221 | transport->asoc = asoc; | |
222 | sctp_association_hold(asoc); | |
223 | } | |
224 | ||
225 | /* Initialize the pmtu of a transport. */ | |
226 | void sctp_transport_pmtu(struct sctp_transport *transport, struct sock *sk) | |
227 | { | |
228 | /* If we don't have a fresh route, look one up */ | |
229 | if (!transport->dst || transport->dst->obsolete) { | |
230 | sctp_transport_dst_release(transport); | |
231 | transport->af_specific->get_dst(transport, &transport->saddr, | |
232 | &transport->fl, sk); | |
233 | } | |
234 | ||
235 | if (transport->dst) { | |
236 | transport->pathmtu = SCTP_TRUNC4(dst_mtu(transport->dst)); | |
237 | } else | |
238 | transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT; | |
239 | } | |
240 | ||
241 | void sctp_transport_update_pmtu(struct sock *sk, struct sctp_transport *t, u32 pmtu) | |
242 | { | |
243 | struct dst_entry *dst; | |
244 | ||
245 | if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) { | |
246 | pr_warn("%s: Reported pmtu %d too low, using default minimum of %d\n", | |
247 | __func__, pmtu, | |
248 | SCTP_DEFAULT_MINSEGMENT); | |
249 | /* Use default minimum segment size and disable | |
250 | * pmtu discovery on this transport. | |
251 | */ | |
252 | t->pathmtu = SCTP_DEFAULT_MINSEGMENT; | |
253 | } else { | |
254 | t->pathmtu = pmtu; | |
255 | } | |
256 | ||
257 | dst = sctp_transport_dst_check(t); | |
258 | if (!dst) | |
259 | t->af_specific->get_dst(t, &t->saddr, &t->fl, sk); | |
260 | ||
261 | if (dst) { | |
262 | dst->ops->update_pmtu(dst, sk, NULL, pmtu); | |
263 | ||
264 | dst = sctp_transport_dst_check(t); | |
265 | if (!dst) | |
266 | t->af_specific->get_dst(t, &t->saddr, &t->fl, sk); | |
267 | } | |
268 | } | |
269 | ||
270 | /* Caches the dst entry and source address for a transport's destination | |
271 | * address. | |
272 | */ | |
273 | void sctp_transport_route(struct sctp_transport *transport, | |
274 | union sctp_addr *saddr, struct sctp_sock *opt) | |
275 | { | |
276 | struct sctp_association *asoc = transport->asoc; | |
277 | struct sctp_af *af = transport->af_specific; | |
278 | ||
279 | af->get_dst(transport, saddr, &transport->fl, sctp_opt2sk(opt)); | |
280 | ||
281 | if (saddr) | |
282 | memcpy(&transport->saddr, saddr, sizeof(union sctp_addr)); | |
283 | else | |
284 | af->get_saddr(opt, transport, &transport->fl); | |
285 | ||
286 | if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) { | |
287 | return; | |
288 | } | |
289 | if (transport->dst) { | |
290 | transport->pathmtu = SCTP_TRUNC4(dst_mtu(transport->dst)); | |
291 | ||
292 | /* Initialize sk->sk_rcv_saddr, if the transport is the | |
293 | * association's active path for getsockname(). | |
294 | */ | |
295 | if (asoc && (!asoc->peer.primary_path || | |
296 | (transport == asoc->peer.active_path))) | |
297 | opt->pf->to_sk_saddr(&transport->saddr, | |
298 | asoc->base.sk); | |
299 | } else | |
300 | transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT; | |
301 | } | |
302 | ||
303 | /* Hold a reference to a transport. */ | |
304 | int sctp_transport_hold(struct sctp_transport *transport) | |
305 | { | |
306 | return atomic_add_unless(&transport->refcnt, 1, 0); | |
307 | } | |
308 | ||
309 | /* Release a reference to a transport and clean up | |
310 | * if there are no more references. | |
311 | */ | |
312 | void sctp_transport_put(struct sctp_transport *transport) | |
313 | { | |
314 | if (atomic_dec_and_test(&transport->refcnt)) | |
315 | sctp_transport_destroy(transport); | |
316 | } | |
317 | ||
318 | /* Update transport's RTO based on the newly calculated RTT. */ | |
319 | void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt) | |
320 | { | |
321 | if (unlikely(!tp->rto_pending)) | |
322 | /* We should not be doing any RTO updates unless rto_pending is set. */ | |
323 | pr_debug("%s: rto_pending not set on transport %p!\n", __func__, tp); | |
324 | ||
325 | if (tp->rttvar || tp->srtt) { | |
326 | struct net *net = sock_net(tp->asoc->base.sk); | |
327 | /* 6.3.1 C3) When a new RTT measurement R' is made, set | |
328 | * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'| | |
329 | * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R' | |
330 | */ | |
331 | ||
332 | /* Note: The above algorithm has been rewritten to | |
333 | * express rto_beta and rto_alpha as inverse powers | |
334 | * of two. | |
335 | * For example, assuming the default value of RTO.Alpha of | |
336 | * 1/8, rto_alpha would be expressed as 3. | |
337 | */ | |
338 | tp->rttvar = tp->rttvar - (tp->rttvar >> net->sctp.rto_beta) | |
339 | + (((__u32)abs((__s64)tp->srtt - (__s64)rtt)) >> net->sctp.rto_beta); | |
340 | tp->srtt = tp->srtt - (tp->srtt >> net->sctp.rto_alpha) | |
341 | + (rtt >> net->sctp.rto_alpha); | |
342 | } else { | |
343 | /* 6.3.1 C2) When the first RTT measurement R is made, set | |
344 | * SRTT <- R, RTTVAR <- R/2. | |
345 | */ | |
346 | tp->srtt = rtt; | |
347 | tp->rttvar = rtt >> 1; | |
348 | } | |
349 | ||
350 | /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then | |
351 | * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY. | |
352 | */ | |
353 | if (tp->rttvar == 0) | |
354 | tp->rttvar = SCTP_CLOCK_GRANULARITY; | |
355 | ||
356 | /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */ | |
357 | tp->rto = tp->srtt + (tp->rttvar << 2); | |
358 | ||
359 | /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min | |
360 | * seconds then it is rounded up to RTO.Min seconds. | |
361 | */ | |
362 | if (tp->rto < tp->asoc->rto_min) | |
363 | tp->rto = tp->asoc->rto_min; | |
364 | ||
365 | /* 6.3.1 C7) A maximum value may be placed on RTO provided it is | |
366 | * at least RTO.max seconds. | |
367 | */ | |
368 | if (tp->rto > tp->asoc->rto_max) | |
369 | tp->rto = tp->asoc->rto_max; | |
370 | ||
371 | sctp_max_rto(tp->asoc, tp); | |
372 | tp->rtt = rtt; | |
373 | ||
374 | /* Reset rto_pending so that a new RTT measurement is started when a | |
375 | * new data chunk is sent. | |
376 | */ | |
377 | tp->rto_pending = 0; | |
378 | ||
379 | pr_debug("%s: transport:%p, rtt:%d, srtt:%d rttvar:%d, rto:%ld\n", | |
380 | __func__, tp, rtt, tp->srtt, tp->rttvar, tp->rto); | |
381 | } | |
382 | ||
383 | /* This routine updates the transport's cwnd and partial_bytes_acked | |
384 | * parameters based on the bytes acked in the received SACK. | |
385 | */ | |
386 | void sctp_transport_raise_cwnd(struct sctp_transport *transport, | |
387 | __u32 sack_ctsn, __u32 bytes_acked) | |
388 | { | |
389 | struct sctp_association *asoc = transport->asoc; | |
390 | __u32 cwnd, ssthresh, flight_size, pba, pmtu; | |
391 | ||
392 | cwnd = transport->cwnd; | |
393 | flight_size = transport->flight_size; | |
394 | ||
395 | /* See if we need to exit Fast Recovery first */ | |
396 | if (asoc->fast_recovery && | |
397 | TSN_lte(asoc->fast_recovery_exit, sack_ctsn)) | |
398 | asoc->fast_recovery = 0; | |
399 | ||
400 | /* The appropriate cwnd increase algorithm is performed if, and only | |
401 | * if the cumulative TSN whould advanced and the congestion window is | |
402 | * being fully utilized. | |
403 | */ | |
404 | if (TSN_lte(sack_ctsn, transport->asoc->ctsn_ack_point) || | |
405 | (flight_size < cwnd)) | |
406 | return; | |
407 | ||
408 | ssthresh = transport->ssthresh; | |
409 | pba = transport->partial_bytes_acked; | |
410 | pmtu = transport->asoc->pathmtu; | |
411 | ||
412 | if (cwnd <= ssthresh) { | |
413 | /* RFC 4960 7.2.1 | |
414 | * o When cwnd is less than or equal to ssthresh, an SCTP | |
415 | * endpoint MUST use the slow-start algorithm to increase | |
416 | * cwnd only if the current congestion window is being fully | |
417 | * utilized, an incoming SACK advances the Cumulative TSN | |
418 | * Ack Point, and the data sender is not in Fast Recovery. | |
419 | * Only when these three conditions are met can the cwnd be | |
420 | * increased; otherwise, the cwnd MUST not be increased. | |
421 | * If these conditions are met, then cwnd MUST be increased | |
422 | * by, at most, the lesser of 1) the total size of the | |
423 | * previously outstanding DATA chunk(s) acknowledged, and | |
424 | * 2) the destination's path MTU. This upper bound protects | |
425 | * against the ACK-Splitting attack outlined in [SAVAGE99]. | |
426 | */ | |
427 | if (asoc->fast_recovery) | |
428 | return; | |
429 | ||
430 | if (bytes_acked > pmtu) | |
431 | cwnd += pmtu; | |
432 | else | |
433 | cwnd += bytes_acked; | |
434 | ||
435 | pr_debug("%s: slow start: transport:%p, bytes_acked:%d, " | |
436 | "cwnd:%d, ssthresh:%d, flight_size:%d, pba:%d\n", | |
437 | __func__, transport, bytes_acked, cwnd, ssthresh, | |
438 | flight_size, pba); | |
439 | } else { | |
440 | /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh, | |
441 | * upon each SACK arrival that advances the Cumulative TSN Ack | |
442 | * Point, increase partial_bytes_acked by the total number of | |
443 | * bytes of all new chunks acknowledged in that SACK including | |
444 | * chunks acknowledged by the new Cumulative TSN Ack and by | |
445 | * Gap Ack Blocks. | |
446 | * | |
447 | * When partial_bytes_acked is equal to or greater than cwnd | |
448 | * and before the arrival of the SACK the sender had cwnd or | |
449 | * more bytes of data outstanding (i.e., before arrival of the | |
450 | * SACK, flightsize was greater than or equal to cwnd), | |
451 | * increase cwnd by MTU, and reset partial_bytes_acked to | |
452 | * (partial_bytes_acked - cwnd). | |
453 | */ | |
454 | pba += bytes_acked; | |
455 | if (pba >= cwnd) { | |
456 | cwnd += pmtu; | |
457 | pba = ((cwnd < pba) ? (pba - cwnd) : 0); | |
458 | } | |
459 | ||
460 | pr_debug("%s: congestion avoidance: transport:%p, " | |
461 | "bytes_acked:%d, cwnd:%d, ssthresh:%d, " | |
462 | "flight_size:%d, pba:%d\n", __func__, | |
463 | transport, bytes_acked, cwnd, ssthresh, | |
464 | flight_size, pba); | |
465 | } | |
466 | ||
467 | transport->cwnd = cwnd; | |
468 | transport->partial_bytes_acked = pba; | |
469 | } | |
470 | ||
471 | /* This routine is used to lower the transport's cwnd when congestion is | |
472 | * detected. | |
473 | */ | |
474 | void sctp_transport_lower_cwnd(struct sctp_transport *transport, | |
475 | sctp_lower_cwnd_t reason) | |
476 | { | |
477 | struct sctp_association *asoc = transport->asoc; | |
478 | ||
479 | switch (reason) { | |
480 | case SCTP_LOWER_CWND_T3_RTX: | |
481 | /* RFC 2960 Section 7.2.3, sctpimpguide | |
482 | * When the T3-rtx timer expires on an address, SCTP should | |
483 | * perform slow start by: | |
484 | * ssthresh = max(cwnd/2, 4*MTU) | |
485 | * cwnd = 1*MTU | |
486 | * partial_bytes_acked = 0 | |
487 | */ | |
488 | transport->ssthresh = max(transport->cwnd/2, | |
489 | 4*asoc->pathmtu); | |
490 | transport->cwnd = asoc->pathmtu; | |
491 | ||
492 | /* T3-rtx also clears fast recovery */ | |
493 | asoc->fast_recovery = 0; | |
494 | break; | |
495 | ||
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. | |
501 | * | |
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 | |
508 | */ | |
509 | if (asoc->fast_recovery) | |
510 | return; | |
511 | ||
512 | /* Mark Fast recovery */ | |
513 | asoc->fast_recovery = 1; | |
514 | asoc->fast_recovery_exit = asoc->next_tsn - 1; | |
515 | ||
516 | transport->ssthresh = max(transport->cwnd/2, | |
517 | 4*asoc->pathmtu); | |
518 | transport->cwnd = transport->ssthresh; | |
519 | break; | |
520 | ||
521 | case SCTP_LOWER_CWND_ECNE: | |
522 | /* RFC 2481 Section 6.1.2. | |
523 | * If the sender receives an ECN-Echo ACK packet | |
524 | * then the sender knows that congestion was encountered in the | |
525 | * network on the path from the sender to the receiver. The | |
526 | * indication of congestion should be treated just as a | |
527 | * congestion loss in non-ECN Capable TCP. That is, the TCP | |
528 | * source halves the congestion window "cwnd" and reduces the | |
529 | * slow start threshold "ssthresh". | |
530 | * A critical condition is that TCP does not react to | |
531 | * congestion indications more than once every window of | |
532 | * data (or more loosely more than once every round-trip time). | |
533 | */ | |
534 | if (time_after(jiffies, transport->last_time_ecne_reduced + | |
535 | transport->rtt)) { | |
536 | transport->ssthresh = max(transport->cwnd/2, | |
537 | 4*asoc->pathmtu); | |
538 | transport->cwnd = transport->ssthresh; | |
539 | transport->last_time_ecne_reduced = jiffies; | |
540 | } | |
541 | break; | |
542 | ||
543 | case SCTP_LOWER_CWND_INACTIVE: | |
544 | /* RFC 2960 Section 7.2.1, sctpimpguide | |
545 | * When the endpoint does not transmit data on a given | |
546 | * transport address, the cwnd of the transport address | |
547 | * should be adjusted to max(cwnd/2, 4*MTU) per RTO. | |
548 | * NOTE: Although the draft recommends that this check needs | |
549 | * to be done every RTO interval, we do it every hearbeat | |
550 | * interval. | |
551 | */ | |
552 | transport->cwnd = max(transport->cwnd/2, | |
553 | 4*asoc->pathmtu); | |
554 | break; | |
555 | } | |
556 | ||
557 | transport->partial_bytes_acked = 0; | |
558 | ||
559 | pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d\n", | |
560 | __func__, transport, reason, transport->cwnd, | |
561 | transport->ssthresh); | |
562 | } | |
563 | ||
564 | /* Apply Max.Burst limit to the congestion window: | |
565 | * sctpimpguide-05 2.14.2 | |
566 | * D) When the time comes for the sender to | |
567 | * transmit new DATA chunks, the protocol parameter Max.Burst MUST | |
568 | * first be applied to limit how many new DATA chunks may be sent. | |
569 | * The limit is applied by adjusting cwnd as follows: | |
570 | * if ((flightsize+ Max.Burst * MTU) < cwnd) | |
571 | * cwnd = flightsize + Max.Burst * MTU | |
572 | */ | |
573 | ||
574 | void sctp_transport_burst_limited(struct sctp_transport *t) | |
575 | { | |
576 | struct sctp_association *asoc = t->asoc; | |
577 | u32 old_cwnd = t->cwnd; | |
578 | u32 max_burst_bytes; | |
579 | ||
580 | if (t->burst_limited || asoc->max_burst == 0) | |
581 | return; | |
582 | ||
583 | max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu); | |
584 | if (max_burst_bytes < old_cwnd) { | |
585 | t->cwnd = max_burst_bytes; | |
586 | t->burst_limited = old_cwnd; | |
587 | } | |
588 | } | |
589 | ||
590 | /* Restore the old cwnd congestion window, after the burst had it's | |
591 | * desired effect. | |
592 | */ | |
593 | void sctp_transport_burst_reset(struct sctp_transport *t) | |
594 | { | |
595 | if (t->burst_limited) { | |
596 | t->cwnd = t->burst_limited; | |
597 | t->burst_limited = 0; | |
598 | } | |
599 | } | |
600 | ||
601 | /* What is the next timeout value for this transport? */ | |
602 | unsigned long sctp_transport_timeout(struct sctp_transport *trans) | |
603 | { | |
604 | /* RTO + timer slack +/- 50% of RTO */ | |
605 | unsigned long timeout = trans->rto >> 1; | |
606 | ||
607 | if (trans->state != SCTP_UNCONFIRMED && | |
608 | trans->state != SCTP_PF) | |
609 | timeout += trans->hbinterval; | |
610 | ||
611 | return timeout; | |
612 | } | |
613 | ||
614 | /* Reset transport variables to their initial values */ | |
615 | void sctp_transport_reset(struct sctp_transport *t) | |
616 | { | |
617 | struct sctp_association *asoc = t->asoc; | |
618 | ||
619 | /* RFC 2960 (bis), Section 5.2.4 | |
620 | * All the congestion control parameters (e.g., cwnd, ssthresh) | |
621 | * related to this peer MUST be reset to their initial values | |
622 | * (see Section 6.2.1) | |
623 | */ | |
624 | t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380)); | |
625 | t->burst_limited = 0; | |
626 | t->ssthresh = asoc->peer.i.a_rwnd; | |
627 | t->rto = asoc->rto_initial; | |
628 | sctp_max_rto(asoc, t); | |
629 | t->rtt = 0; | |
630 | t->srtt = 0; | |
631 | t->rttvar = 0; | |
632 | ||
633 | /* Reset these additional varibles so that we have a clean | |
634 | * slate. | |
635 | */ | |
636 | t->partial_bytes_acked = 0; | |
637 | t->flight_size = 0; | |
638 | t->error_count = 0; | |
639 | t->rto_pending = 0; | |
640 | t->hb_sent = 0; | |
641 | ||
642 | /* Initialize the state information for SFR-CACC */ | |
643 | t->cacc.changeover_active = 0; | |
644 | t->cacc.cycling_changeover = 0; | |
645 | t->cacc.next_tsn_at_change = 0; | |
646 | t->cacc.cacc_saw_newack = 0; | |
647 | } | |
648 | ||
649 | /* Schedule retransmission on the given transport */ | |
650 | void sctp_transport_immediate_rtx(struct sctp_transport *t) | |
651 | { | |
652 | /* Stop pending T3_rtx_timer */ | |
653 | if (del_timer(&t->T3_rtx_timer)) | |
654 | sctp_transport_put(t); | |
655 | ||
656 | sctp_retransmit(&t->asoc->outqueue, t, SCTP_RTXR_T3_RTX); | |
657 | if (!timer_pending(&t->T3_rtx_timer)) { | |
658 | if (!mod_timer(&t->T3_rtx_timer, jiffies + t->rto)) | |
659 | sctp_transport_hold(t); | |
660 | } | |
661 | } | |
662 | ||
663 | /* Drop dst */ | |
664 | void sctp_transport_dst_release(struct sctp_transport *t) | |
665 | { | |
666 | dst_release(t->dst); | |
667 | t->dst = NULL; | |
668 | t->dst_pending_confirm = 0; | |
669 | } | |
670 | ||
671 | /* Schedule neighbour confirm */ | |
672 | void sctp_transport_dst_confirm(struct sctp_transport *t) | |
673 | { | |
674 | t->dst_pending_confirm = 1; | |
675 | } |