]>
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 | timer_setup(&peer->T3_rtx_timer, sctp_generate_t3_rtx_event, 0); | |
91 | timer_setup(&peer->hb_timer, sctp_generate_heartbeat_event, 0); | |
92 | timer_setup(&peer->reconf_timer, sctp_generate_reconf_event, 0); | |
93 | timer_setup(&peer->proto_unreach_timer, | |
94 | sctp_generate_proto_unreach_event, 0); | |
95 | ||
96 | /* Initialize the 64-bit random nonce sent with heartbeat. */ | |
97 | get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce)); | |
98 | ||
99 | refcount_set(&peer->refcnt, 1); | |
100 | ||
101 | return peer; | |
102 | } | |
103 | ||
104 | /* Allocate and initialize a new transport. */ | |
105 | struct sctp_transport *sctp_transport_new(struct net *net, | |
106 | const union sctp_addr *addr, | |
107 | gfp_t gfp) | |
108 | { | |
109 | struct sctp_transport *transport; | |
110 | ||
111 | transport = kzalloc(sizeof(*transport), gfp); | |
112 | if (!transport) | |
113 | goto fail; | |
114 | ||
115 | if (!sctp_transport_init(net, transport, addr, gfp)) | |
116 | goto fail_init; | |
117 | ||
118 | SCTP_DBG_OBJCNT_INC(transport); | |
119 | ||
120 | return transport; | |
121 | ||
122 | fail_init: | |
123 | kfree(transport); | |
124 | ||
125 | fail: | |
126 | return NULL; | |
127 | } | |
128 | ||
129 | /* This transport is no longer needed. Free up if possible, or | |
130 | * delay until it last reference count. | |
131 | */ | |
132 | void sctp_transport_free(struct sctp_transport *transport) | |
133 | { | |
134 | /* Try to delete the heartbeat timer. */ | |
135 | if (del_timer(&transport->hb_timer)) | |
136 | sctp_transport_put(transport); | |
137 | ||
138 | /* Delete the T3_rtx timer if it's active. | |
139 | * There is no point in not doing this now and letting | |
140 | * structure hang around in memory since we know | |
141 | * the tranport is going away. | |
142 | */ | |
143 | if (del_timer(&transport->T3_rtx_timer)) | |
144 | sctp_transport_put(transport); | |
145 | ||
146 | if (del_timer(&transport->reconf_timer)) | |
147 | sctp_transport_put(transport); | |
148 | ||
149 | /* Delete the ICMP proto unreachable timer if it's active. */ | |
150 | if (del_timer(&transport->proto_unreach_timer)) | |
151 | sctp_association_put(transport->asoc); | |
152 | ||
153 | sctp_transport_put(transport); | |
154 | } | |
155 | ||
156 | static void sctp_transport_destroy_rcu(struct rcu_head *head) | |
157 | { | |
158 | struct sctp_transport *transport; | |
159 | ||
160 | transport = container_of(head, struct sctp_transport, rcu); | |
161 | ||
162 | dst_release(transport->dst); | |
163 | kfree(transport); | |
164 | SCTP_DBG_OBJCNT_DEC(transport); | |
165 | } | |
166 | ||
167 | /* Destroy the transport data structure. | |
168 | * Assumes there are no more users of this structure. | |
169 | */ | |
170 | static void sctp_transport_destroy(struct sctp_transport *transport) | |
171 | { | |
172 | if (unlikely(refcount_read(&transport->refcnt))) { | |
173 | WARN(1, "Attempt to destroy undead transport %p!\n", transport); | |
174 | return; | |
175 | } | |
176 | ||
177 | sctp_packet_free(&transport->packet); | |
178 | ||
179 | if (transport->asoc) | |
180 | sctp_association_put(transport->asoc); | |
181 | ||
182 | call_rcu(&transport->rcu, sctp_transport_destroy_rcu); | |
183 | } | |
184 | ||
185 | /* 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 | */ | |
188 | void sctp_transport_reset_t3_rtx(struct sctp_transport *transport) | |
189 | { | |
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 | ||
198 | if (!timer_pending(&transport->T3_rtx_timer)) | |
199 | if (!mod_timer(&transport->T3_rtx_timer, | |
200 | jiffies + transport->rto)) | |
201 | sctp_transport_hold(transport); | |
202 | } | |
203 | ||
204 | void sctp_transport_reset_hb_timer(struct sctp_transport *transport) | |
205 | { | |
206 | unsigned long expires; | |
207 | ||
208 | /* When a data chunk is sent, reset the heartbeat interval. */ | |
209 | expires = jiffies + sctp_transport_timeout(transport); | |
210 | if ((time_before(transport->hb_timer.expires, expires) || | |
211 | !timer_pending(&transport->hb_timer)) && | |
212 | !mod_timer(&transport->hb_timer, | |
213 | expires + prandom_u32_max(transport->rto))) | |
214 | sctp_transport_hold(transport); | |
215 | } | |
216 | ||
217 | void sctp_transport_reset_reconf_timer(struct sctp_transport *transport) | |
218 | { | |
219 | if (!timer_pending(&transport->reconf_timer)) | |
220 | if (!mod_timer(&transport->reconf_timer, | |
221 | jiffies + transport->rto)) | |
222 | sctp_transport_hold(transport); | |
223 | } | |
224 | ||
225 | /* This transport has been assigned to an association. | |
226 | * Initialize fields from the association or from the sock itself. | |
227 | * Register the reference count in the association. | |
228 | */ | |
229 | void sctp_transport_set_owner(struct sctp_transport *transport, | |
230 | struct sctp_association *asoc) | |
231 | { | |
232 | transport->asoc = asoc; | |
233 | sctp_association_hold(asoc); | |
234 | } | |
235 | ||
236 | /* Initialize the pmtu of a transport. */ | |
237 | void sctp_transport_pmtu(struct sctp_transport *transport, struct sock *sk) | |
238 | { | |
239 | /* If we don't have a fresh route, look one up */ | |
240 | if (!transport->dst || transport->dst->obsolete) { | |
241 | sctp_transport_dst_release(transport); | |
242 | transport->af_specific->get_dst(transport, &transport->saddr, | |
243 | &transport->fl, sk); | |
244 | } | |
245 | ||
246 | if (transport->param_flags & SPP_PMTUD_DISABLE) { | |
247 | struct sctp_association *asoc = transport->asoc; | |
248 | ||
249 | if (!transport->pathmtu && asoc && asoc->pathmtu) | |
250 | transport->pathmtu = asoc->pathmtu; | |
251 | if (transport->pathmtu) | |
252 | return; | |
253 | } | |
254 | ||
255 | if (transport->dst) | |
256 | transport->pathmtu = sctp_dst_mtu(transport->dst); | |
257 | else | |
258 | transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT; | |
259 | } | |
260 | ||
261 | bool sctp_transport_update_pmtu(struct sctp_transport *t, u32 pmtu) | |
262 | { | |
263 | struct dst_entry *dst = sctp_transport_dst_check(t); | |
264 | struct sock *sk = t->asoc->base.sk; | |
265 | bool change = true; | |
266 | ||
267 | if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) { | |
268 | pr_warn_ratelimited("%s: Reported pmtu %d too low, using default minimum of %d\n", | |
269 | __func__, pmtu, SCTP_DEFAULT_MINSEGMENT); | |
270 | /* Use default minimum segment instead */ | |
271 | pmtu = SCTP_DEFAULT_MINSEGMENT; | |
272 | } | |
273 | pmtu = SCTP_TRUNC4(pmtu); | |
274 | ||
275 | if (dst) { | |
276 | struct sctp_pf *pf = sctp_get_pf_specific(dst->ops->family); | |
277 | union sctp_addr addr; | |
278 | ||
279 | pf->af->from_sk(&addr, sk); | |
280 | pf->to_sk_daddr(&t->ipaddr, sk); | |
281 | dst->ops->update_pmtu(dst, sk, NULL, pmtu); | |
282 | pf->to_sk_daddr(&addr, sk); | |
283 | ||
284 | dst = sctp_transport_dst_check(t); | |
285 | } | |
286 | ||
287 | if (!dst) { | |
288 | t->af_specific->get_dst(t, &t->saddr, &t->fl, sk); | |
289 | dst = t->dst; | |
290 | } | |
291 | ||
292 | if (dst) { | |
293 | /* Re-fetch, as under layers may have a higher minimum size */ | |
294 | pmtu = sctp_dst_mtu(dst); | |
295 | change = t->pathmtu != pmtu; | |
296 | } | |
297 | t->pathmtu = pmtu; | |
298 | ||
299 | return change; | |
300 | } | |
301 | ||
302 | /* Caches the dst entry and source address for a transport's destination | |
303 | * address. | |
304 | */ | |
305 | void sctp_transport_route(struct sctp_transport *transport, | |
306 | union sctp_addr *saddr, struct sctp_sock *opt) | |
307 | { | |
308 | struct sctp_association *asoc = transport->asoc; | |
309 | struct sctp_af *af = transport->af_specific; | |
310 | ||
311 | sctp_transport_dst_release(transport); | |
312 | af->get_dst(transport, saddr, &transport->fl, sctp_opt2sk(opt)); | |
313 | ||
314 | if (saddr) | |
315 | memcpy(&transport->saddr, saddr, sizeof(union sctp_addr)); | |
316 | else | |
317 | af->get_saddr(opt, transport, &transport->fl); | |
318 | ||
319 | sctp_transport_pmtu(transport, sctp_opt2sk(opt)); | |
320 | ||
321 | /* Initialize sk->sk_rcv_saddr, if the transport is the | |
322 | * association's active path for getsockname(). | |
323 | */ | |
324 | if (transport->dst && asoc && | |
325 | (!asoc->peer.primary_path || transport == asoc->peer.active_path)) | |
326 | opt->pf->to_sk_saddr(&transport->saddr, asoc->base.sk); | |
327 | } | |
328 | ||
329 | /* Hold a reference to a transport. */ | |
330 | int sctp_transport_hold(struct sctp_transport *transport) | |
331 | { | |
332 | return refcount_inc_not_zero(&transport->refcnt); | |
333 | } | |
334 | ||
335 | /* Release a reference to a transport and clean up | |
336 | * if there are no more references. | |
337 | */ | |
338 | void sctp_transport_put(struct sctp_transport *transport) | |
339 | { | |
340 | if (refcount_dec_and_test(&transport->refcnt)) | |
341 | sctp_transport_destroy(transport); | |
342 | } | |
343 | ||
344 | /* Update transport's RTO based on the newly calculated RTT. */ | |
345 | void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt) | |
346 | { | |
347 | if (unlikely(!tp->rto_pending)) | |
348 | /* We should not be doing any RTO updates unless rto_pending is set. */ | |
349 | pr_debug("%s: rto_pending not set on transport %p!\n", __func__, tp); | |
350 | ||
351 | if (tp->rttvar || tp->srtt) { | |
352 | struct net *net = sock_net(tp->asoc->base.sk); | |
353 | /* 6.3.1 C3) When a new RTT measurement R' is made, set | |
354 | * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'| | |
355 | * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R' | |
356 | */ | |
357 | ||
358 | /* Note: The above algorithm has been rewritten to | |
359 | * express rto_beta and rto_alpha as inverse powers | |
360 | * of two. | |
361 | * For example, assuming the default value of RTO.Alpha of | |
362 | * 1/8, rto_alpha would be expressed as 3. | |
363 | */ | |
364 | tp->rttvar = tp->rttvar - (tp->rttvar >> net->sctp.rto_beta) | |
365 | + (((__u32)abs((__s64)tp->srtt - (__s64)rtt)) >> net->sctp.rto_beta); | |
366 | tp->srtt = tp->srtt - (tp->srtt >> net->sctp.rto_alpha) | |
367 | + (rtt >> net->sctp.rto_alpha); | |
368 | } else { | |
369 | /* 6.3.1 C2) When the first RTT measurement R is made, set | |
370 | * SRTT <- R, RTTVAR <- R/2. | |
371 | */ | |
372 | tp->srtt = rtt; | |
373 | tp->rttvar = rtt >> 1; | |
374 | } | |
375 | ||
376 | /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then | |
377 | * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY. | |
378 | */ | |
379 | if (tp->rttvar == 0) | |
380 | tp->rttvar = SCTP_CLOCK_GRANULARITY; | |
381 | ||
382 | /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */ | |
383 | tp->rto = tp->srtt + (tp->rttvar << 2); | |
384 | ||
385 | /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min | |
386 | * seconds then it is rounded up to RTO.Min seconds. | |
387 | */ | |
388 | if (tp->rto < tp->asoc->rto_min) | |
389 | tp->rto = tp->asoc->rto_min; | |
390 | ||
391 | /* 6.3.1 C7) A maximum value may be placed on RTO provided it is | |
392 | * at least RTO.max seconds. | |
393 | */ | |
394 | if (tp->rto > tp->asoc->rto_max) | |
395 | tp->rto = tp->asoc->rto_max; | |
396 | ||
397 | sctp_max_rto(tp->asoc, tp); | |
398 | tp->rtt = rtt; | |
399 | ||
400 | /* Reset rto_pending so that a new RTT measurement is started when a | |
401 | * new data chunk is sent. | |
402 | */ | |
403 | tp->rto_pending = 0; | |
404 | ||
405 | pr_debug("%s: transport:%p, rtt:%d, srtt:%d rttvar:%d, rto:%ld\n", | |
406 | __func__, tp, rtt, tp->srtt, tp->rttvar, tp->rto); | |
407 | } | |
408 | ||
409 | /* This routine updates the transport's cwnd and partial_bytes_acked | |
410 | * parameters based on the bytes acked in the received SACK. | |
411 | */ | |
412 | void sctp_transport_raise_cwnd(struct sctp_transport *transport, | |
413 | __u32 sack_ctsn, __u32 bytes_acked) | |
414 | { | |
415 | struct sctp_association *asoc = transport->asoc; | |
416 | __u32 cwnd, ssthresh, flight_size, pba, pmtu; | |
417 | ||
418 | cwnd = transport->cwnd; | |
419 | flight_size = transport->flight_size; | |
420 | ||
421 | /* See if we need to exit Fast Recovery first */ | |
422 | if (asoc->fast_recovery && | |
423 | TSN_lte(asoc->fast_recovery_exit, sack_ctsn)) | |
424 | asoc->fast_recovery = 0; | |
425 | ||
426 | ssthresh = transport->ssthresh; | |
427 | pba = transport->partial_bytes_acked; | |
428 | pmtu = transport->asoc->pathmtu; | |
429 | ||
430 | if (cwnd <= ssthresh) { | |
431 | /* RFC 4960 7.2.1 | |
432 | * o When cwnd is less than or equal to ssthresh, an SCTP | |
433 | * endpoint MUST use the slow-start algorithm to increase | |
434 | * cwnd only if the current congestion window is being fully | |
435 | * utilized, an incoming SACK advances the Cumulative TSN | |
436 | * Ack Point, and the data sender is not in Fast Recovery. | |
437 | * Only when these three conditions are met can the cwnd be | |
438 | * increased; otherwise, the cwnd MUST not be increased. | |
439 | * If these conditions are met, then cwnd MUST be increased | |
440 | * by, at most, the lesser of 1) the total size of the | |
441 | * previously outstanding DATA chunk(s) acknowledged, and | |
442 | * 2) the destination's path MTU. This upper bound protects | |
443 | * against the ACK-Splitting attack outlined in [SAVAGE99]. | |
444 | */ | |
445 | if (asoc->fast_recovery) | |
446 | return; | |
447 | ||
448 | /* The appropriate cwnd increase algorithm is performed | |
449 | * if, and only if the congestion window is being fully | |
450 | * utilized. Note that RFC4960 Errata 3.22 removed the | |
451 | * other condition on ctsn moving. | |
452 | */ | |
453 | if (flight_size < cwnd) | |
454 | return; | |
455 | ||
456 | if (bytes_acked > pmtu) | |
457 | cwnd += pmtu; | |
458 | else | |
459 | cwnd += bytes_acked; | |
460 | ||
461 | pr_debug("%s: slow start: transport:%p, bytes_acked:%d, " | |
462 | "cwnd:%d, ssthresh:%d, flight_size:%d, pba:%d\n", | |
463 | __func__, transport, bytes_acked, cwnd, ssthresh, | |
464 | flight_size, pba); | |
465 | } else { | |
466 | /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh, | |
467 | * upon each SACK arrival, increase partial_bytes_acked | |
468 | * by the total number of bytes of all new chunks | |
469 | * acknowledged in that SACK including chunks | |
470 | * acknowledged by the new Cumulative TSN Ack and by Gap | |
471 | * Ack Blocks. (updated by RFC4960 Errata 3.22) | |
472 | * | |
473 | * When partial_bytes_acked is greater than cwnd and | |
474 | * before the arrival of the SACK the sender had less | |
475 | * bytes of data outstanding than cwnd (i.e., before | |
476 | * arrival of the SACK, flightsize was less than cwnd), | |
477 | * reset partial_bytes_acked to cwnd. (RFC 4960 Errata | |
478 | * 3.26) | |
479 | * | |
480 | * When partial_bytes_acked is equal to or greater than | |
481 | * cwnd and before the arrival of the SACK the sender | |
482 | * had cwnd or more bytes of data outstanding (i.e., | |
483 | * before arrival of the SACK, flightsize was greater | |
484 | * than or equal to cwnd), partial_bytes_acked is reset | |
485 | * to (partial_bytes_acked - cwnd). Next, cwnd is | |
486 | * increased by MTU. (RFC 4960 Errata 3.12) | |
487 | */ | |
488 | pba += bytes_acked; | |
489 | if (pba > cwnd && flight_size < cwnd) | |
490 | pba = cwnd; | |
491 | if (pba >= cwnd && flight_size >= cwnd) { | |
492 | pba = pba - cwnd; | |
493 | cwnd += pmtu; | |
494 | } | |
495 | ||
496 | pr_debug("%s: congestion avoidance: transport:%p, " | |
497 | "bytes_acked:%d, cwnd:%d, ssthresh:%d, " | |
498 | "flight_size:%d, pba:%d\n", __func__, | |
499 | transport, bytes_acked, cwnd, ssthresh, | |
500 | flight_size, pba); | |
501 | } | |
502 | ||
503 | transport->cwnd = cwnd; | |
504 | transport->partial_bytes_acked = pba; | |
505 | } | |
506 | ||
507 | /* This routine is used to lower the transport's cwnd when congestion is | |
508 | * detected. | |
509 | */ | |
510 | void sctp_transport_lower_cwnd(struct sctp_transport *transport, | |
511 | enum sctp_lower_cwnd reason) | |
512 | { | |
513 | struct sctp_association *asoc = transport->asoc; | |
514 | ||
515 | switch (reason) { | |
516 | case SCTP_LOWER_CWND_T3_RTX: | |
517 | /* RFC 2960 Section 7.2.3, sctpimpguide | |
518 | * When the T3-rtx timer expires on an address, SCTP should | |
519 | * perform slow start by: | |
520 | * ssthresh = max(cwnd/2, 4*MTU) | |
521 | * cwnd = 1*MTU | |
522 | * partial_bytes_acked = 0 | |
523 | */ | |
524 | transport->ssthresh = max(transport->cwnd/2, | |
525 | 4*asoc->pathmtu); | |
526 | transport->cwnd = asoc->pathmtu; | |
527 | ||
528 | /* T3-rtx also clears fast recovery */ | |
529 | asoc->fast_recovery = 0; | |
530 | break; | |
531 | ||
532 | case SCTP_LOWER_CWND_FAST_RTX: | |
533 | /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the | |
534 | * destination address(es) to which the missing DATA chunks | |
535 | * were last sent, according to the formula described in | |
536 | * Section 7.2.3. | |
537 | * | |
538 | * RFC 2960 7.2.3, sctpimpguide Upon detection of packet | |
539 | * losses from SACK (see Section 7.2.4), An endpoint | |
540 | * should do the following: | |
541 | * ssthresh = max(cwnd/2, 4*MTU) | |
542 | * cwnd = ssthresh | |
543 | * partial_bytes_acked = 0 | |
544 | */ | |
545 | if (asoc->fast_recovery) | |
546 | return; | |
547 | ||
548 | /* Mark Fast recovery */ | |
549 | asoc->fast_recovery = 1; | |
550 | asoc->fast_recovery_exit = asoc->next_tsn - 1; | |
551 | ||
552 | transport->ssthresh = max(transport->cwnd/2, | |
553 | 4*asoc->pathmtu); | |
554 | transport->cwnd = transport->ssthresh; | |
555 | break; | |
556 | ||
557 | case SCTP_LOWER_CWND_ECNE: | |
558 | /* RFC 2481 Section 6.1.2. | |
559 | * If the sender receives an ECN-Echo ACK packet | |
560 | * then the sender knows that congestion was encountered in the | |
561 | * network on the path from the sender to the receiver. The | |
562 | * indication of congestion should be treated just as a | |
563 | * congestion loss in non-ECN Capable TCP. That is, the TCP | |
564 | * source halves the congestion window "cwnd" and reduces the | |
565 | * slow start threshold "ssthresh". | |
566 | * A critical condition is that TCP does not react to | |
567 | * congestion indications more than once every window of | |
568 | * data (or more loosely more than once every round-trip time). | |
569 | */ | |
570 | if (time_after(jiffies, transport->last_time_ecne_reduced + | |
571 | transport->rtt)) { | |
572 | transport->ssthresh = max(transport->cwnd/2, | |
573 | 4*asoc->pathmtu); | |
574 | transport->cwnd = transport->ssthresh; | |
575 | transport->last_time_ecne_reduced = jiffies; | |
576 | } | |
577 | break; | |
578 | ||
579 | case SCTP_LOWER_CWND_INACTIVE: | |
580 | /* RFC 2960 Section 7.2.1, sctpimpguide | |
581 | * When the endpoint does not transmit data on a given | |
582 | * transport address, the cwnd of the transport address | |
583 | * should be adjusted to max(cwnd/2, 4*MTU) per RTO. | |
584 | * NOTE: Although the draft recommends that this check needs | |
585 | * to be done every RTO interval, we do it every hearbeat | |
586 | * interval. | |
587 | */ | |
588 | transport->cwnd = max(transport->cwnd/2, | |
589 | 4*asoc->pathmtu); | |
590 | /* RFC 4960 Errata 3.27.2: also adjust sshthresh */ | |
591 | transport->ssthresh = transport->cwnd; | |
592 | break; | |
593 | } | |
594 | ||
595 | transport->partial_bytes_acked = 0; | |
596 | ||
597 | pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d\n", | |
598 | __func__, transport, reason, transport->cwnd, | |
599 | transport->ssthresh); | |
600 | } | |
601 | ||
602 | /* Apply Max.Burst limit to the congestion window: | |
603 | * sctpimpguide-05 2.14.2 | |
604 | * D) When the time comes for the sender to | |
605 | * transmit new DATA chunks, the protocol parameter Max.Burst MUST | |
606 | * first be applied to limit how many new DATA chunks may be sent. | |
607 | * The limit is applied by adjusting cwnd as follows: | |
608 | * if ((flightsize+ Max.Burst * MTU) < cwnd) | |
609 | * cwnd = flightsize + Max.Burst * MTU | |
610 | */ | |
611 | ||
612 | void sctp_transport_burst_limited(struct sctp_transport *t) | |
613 | { | |
614 | struct sctp_association *asoc = t->asoc; | |
615 | u32 old_cwnd = t->cwnd; | |
616 | u32 max_burst_bytes; | |
617 | ||
618 | if (t->burst_limited || asoc->max_burst == 0) | |
619 | return; | |
620 | ||
621 | max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu); | |
622 | if (max_burst_bytes < old_cwnd) { | |
623 | t->cwnd = max_burst_bytes; | |
624 | t->burst_limited = old_cwnd; | |
625 | } | |
626 | } | |
627 | ||
628 | /* Restore the old cwnd congestion window, after the burst had it's | |
629 | * desired effect. | |
630 | */ | |
631 | void sctp_transport_burst_reset(struct sctp_transport *t) | |
632 | { | |
633 | if (t->burst_limited) { | |
634 | t->cwnd = t->burst_limited; | |
635 | t->burst_limited = 0; | |
636 | } | |
637 | } | |
638 | ||
639 | /* What is the next timeout value for this transport? */ | |
640 | unsigned long sctp_transport_timeout(struct sctp_transport *trans) | |
641 | { | |
642 | /* RTO + timer slack +/- 50% of RTO */ | |
643 | unsigned long timeout = trans->rto >> 1; | |
644 | ||
645 | if (trans->state != SCTP_UNCONFIRMED && | |
646 | trans->state != SCTP_PF) | |
647 | timeout += trans->hbinterval; | |
648 | ||
649 | return max_t(unsigned long, timeout, HZ / 5); | |
650 | } | |
651 | ||
652 | /* Reset transport variables to their initial values */ | |
653 | void sctp_transport_reset(struct sctp_transport *t) | |
654 | { | |
655 | struct sctp_association *asoc = t->asoc; | |
656 | ||
657 | /* RFC 2960 (bis), Section 5.2.4 | |
658 | * All the congestion control parameters (e.g., cwnd, ssthresh) | |
659 | * related to this peer MUST be reset to their initial values | |
660 | * (see Section 6.2.1) | |
661 | */ | |
662 | t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380)); | |
663 | t->burst_limited = 0; | |
664 | t->ssthresh = asoc->peer.i.a_rwnd; | |
665 | t->rto = asoc->rto_initial; | |
666 | sctp_max_rto(asoc, t); | |
667 | t->rtt = 0; | |
668 | t->srtt = 0; | |
669 | t->rttvar = 0; | |
670 | ||
671 | /* Reset these additional variables so that we have a clean slate. */ | |
672 | t->partial_bytes_acked = 0; | |
673 | t->flight_size = 0; | |
674 | t->error_count = 0; | |
675 | t->rto_pending = 0; | |
676 | t->hb_sent = 0; | |
677 | ||
678 | /* Initialize the state information for SFR-CACC */ | |
679 | t->cacc.changeover_active = 0; | |
680 | t->cacc.cycling_changeover = 0; | |
681 | t->cacc.next_tsn_at_change = 0; | |
682 | t->cacc.cacc_saw_newack = 0; | |
683 | } | |
684 | ||
685 | /* Schedule retransmission on the given transport */ | |
686 | void sctp_transport_immediate_rtx(struct sctp_transport *t) | |
687 | { | |
688 | /* Stop pending T3_rtx_timer */ | |
689 | if (del_timer(&t->T3_rtx_timer)) | |
690 | sctp_transport_put(t); | |
691 | ||
692 | sctp_retransmit(&t->asoc->outqueue, t, SCTP_RTXR_T3_RTX); | |
693 | if (!timer_pending(&t->T3_rtx_timer)) { | |
694 | if (!mod_timer(&t->T3_rtx_timer, jiffies + t->rto)) | |
695 | sctp_transport_hold(t); | |
696 | } | |
697 | } | |
698 | ||
699 | /* Drop dst */ | |
700 | void sctp_transport_dst_release(struct sctp_transport *t) | |
701 | { | |
702 | dst_release(t->dst); | |
703 | t->dst = NULL; | |
704 | t->dst_pending_confirm = 0; | |
705 | } | |
706 | ||
707 | /* Schedule neighbour confirm */ | |
708 | void sctp_transport_dst_confirm(struct sctp_transport *t) | |
709 | { | |
710 | t->dst_pending_confirm = 1; | |
711 | } |