1 /*-
2 * Copyright (c) 2001-2007, by Cisco Systems, Inc. All rights reserved.
3 * Copyright (c) 2008-2012, by Randall Stewart. All rights reserved.
4 * Copyright (c) 2008-2012, by Michael Tuexen. All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions are met:
8 *
9 * a) Redistributions of source code must retain the above copyright notice,
10 * this list of conditions and the following disclaimer.
11 *
12 * b) Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in
14 * the documentation and/or other materials provided with the distribution.
15 *
16 * c) Neither the name of Cisco Systems, Inc. nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
22 * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
24 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
30 * THE POSSIBILITY OF SUCH DAMAGE.
31 */
32
33 #ifdef __FreeBSD__
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD: head/sys/netinet/sctp_cc_functions.c 240158 2012-09-06 07:03:56Z tuexen $");
36 #endif
37
38 #include <netinet/sctp_os.h>
39 #include <netinet/sctp_var.h>
40 #include <netinet/sctp_sysctl.h>
41 #include <netinet/sctp_pcb.h>
42 #include <netinet/sctp_header.h>
43 #include <netinet/sctputil.h>
44 #include <netinet/sctp_output.h>
45 #include <netinet/sctp_input.h>
46 #include <netinet/sctp_indata.h>
47 #include <netinet/sctp_uio.h>
48 #include <netinet/sctp_timer.h>
49 #include <netinet/sctp_auth.h>
50 #include <netinet/sctp_asconf.h>
51 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
52 #include <netinet/sctp_dtrace_declare.h>
53 #endif
54
55 #define SHIFT_MPTCP_MULTI_N 40
56 #define SHIFT_MPTCP_MULTI_Z 16
57 #define SHIFT_MPTCP_MULTI 8
58
59 static void
sctp_set_initial_cc_param(struct sctp_tcb * stcb,struct sctp_nets * net)60 sctp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net)
61 {
62 struct sctp_association *assoc;
63 uint32_t cwnd_in_mtu;
64
65 assoc = &stcb->asoc;
66 cwnd_in_mtu = SCTP_BASE_SYSCTL(sctp_initial_cwnd);
67 if (cwnd_in_mtu == 0) {
68 /* Using 0 means that the value of RFC 4960 is used. */
69 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND));
70 } else {
71 /*
72 * We take the minimum of the burst limit and the
73 * initial congestion window.
74 */
75 if ((assoc->max_burst > 0) && (cwnd_in_mtu > assoc->max_burst))
76 cwnd_in_mtu = assoc->max_burst;
77 net->cwnd = (net->mtu - sizeof(struct sctphdr)) * cwnd_in_mtu;
78 }
79 if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) ||
80 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2)) {
81 /* In case of resource pooling initialize appropriately */
82 net->cwnd /= assoc->numnets;
83 if (net->cwnd < (net->mtu - sizeof(struct sctphdr))) {
84 net->cwnd = net->mtu - sizeof(struct sctphdr);
85 }
86 }
87 net->ssthresh = assoc->peers_rwnd;
88 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
89 SDT_PROBE(sctp, cwnd, net, init,
90 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net,
91 0, net->cwnd);
92 #endif
93 if (SCTP_BASE_SYSCTL(sctp_logging_level) &
94 (SCTP_CWND_MONITOR_ENABLE|SCTP_CWND_LOGGING_ENABLE)) {
95 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION);
96 }
97 }
98
99 static void
sctp_cwnd_update_after_fr(struct sctp_tcb * stcb,struct sctp_association * asoc)100 sctp_cwnd_update_after_fr(struct sctp_tcb *stcb,
101 struct sctp_association *asoc)
102 {
103 struct sctp_nets *net;
104 uint32_t t_ssthresh, t_cwnd;
105 uint64_t t_ucwnd_sbw;
106
107 /* MT FIXME: Don't compute this over and over again */
108 t_ssthresh = 0;
109 t_cwnd = 0;
110 t_ucwnd_sbw = 0;
111 if ((asoc->sctp_cmt_on_off == SCTP_CMT_RPV1) ||
112 (asoc->sctp_cmt_on_off == SCTP_CMT_RPV2)) {
113 TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
114 t_ssthresh += net->ssthresh;
115 t_cwnd += net->cwnd;
116 if (net->lastsa > 0) {
117 t_ucwnd_sbw += (uint64_t)net->cwnd / (uint64_t)net->lastsa;
118 }
119 }
120 if (t_ucwnd_sbw == 0) {
121 t_ucwnd_sbw = 1;
122 }
123 }
124
125 /*-
126 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) &&
127 * (net->fast_retran_loss_recovery == 0)))
128 */
129 TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
130 if ((asoc->fast_retran_loss_recovery == 0) ||
131 (asoc->sctp_cmt_on_off > 0)) {
132 /* out of a RFC2582 Fast recovery window? */
133 if (net->net_ack > 0) {
134 /*
135 * per section 7.2.3, are there any
136 * destinations that had a fast retransmit
137 * to them. If so what we need to do is
138 * adjust ssthresh and cwnd.
139 */
140 struct sctp_tmit_chunk *lchk;
141 int old_cwnd = net->cwnd;
142
143 if ((asoc->sctp_cmt_on_off == SCTP_CMT_RPV1) ||
144 (asoc->sctp_cmt_on_off == SCTP_CMT_RPV2)) {
145 if (asoc->sctp_cmt_on_off == SCTP_CMT_RPV1) {
146 net->ssthresh = (uint32_t)(((uint64_t)4 *
147 (uint64_t)net->mtu *
148 (uint64_t)net->ssthresh) /
149 (uint64_t)t_ssthresh);
150
151 }
152 if (asoc->sctp_cmt_on_off == SCTP_CMT_RPV2) {
153 uint32_t srtt;
154
155 srtt = net->lastsa;
156 /* lastsa>>3; we don't need to devide ...*/
157 if (srtt == 0) {
158 srtt = 1;
159 }
160 /* Short Version => Equal to Contel Version MBe */
161 net->ssthresh = (uint32_t) (((uint64_t)4 *
162 (uint64_t)net->mtu *
163 (uint64_t)net->cwnd) /
164 ((uint64_t)srtt *
165 t_ucwnd_sbw));
166 /* INCREASE FACTOR */;
167 }
168 if ((net->cwnd > t_cwnd / 2) &&
169 (net->ssthresh < net->cwnd - t_cwnd / 2)) {
170 net->ssthresh = net->cwnd - t_cwnd / 2;
171 }
172 if (net->ssthresh < net->mtu) {
173 net->ssthresh = net->mtu;
174 }
175 } else {
176 net->ssthresh = net->cwnd / 2;
177 if (net->ssthresh < (net->mtu * 2)) {
178 net->ssthresh = 2 * net->mtu;
179 }
180 }
181 net->cwnd = net->ssthresh;
182 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
183 SDT_PROBE(sctp, cwnd, net, fr,
184 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net,
185 old_cwnd, net->cwnd);
186 #endif
187 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
188 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd),
189 SCTP_CWND_LOG_FROM_FR);
190 }
191 lchk = TAILQ_FIRST(&asoc->send_queue);
192
193 net->partial_bytes_acked = 0;
194 /* Turn on fast recovery window */
195 asoc->fast_retran_loss_recovery = 1;
196 if (lchk == NULL) {
197 /* Mark end of the window */
198 asoc->fast_recovery_tsn = asoc->sending_seq - 1;
199 } else {
200 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
201 }
202
203 /*
204 * CMT fast recovery -- per destination
205 * recovery variable.
206 */
207 net->fast_retran_loss_recovery = 1;
208
209 if (lchk == NULL) {
210 /* Mark end of the window */
211 net->fast_recovery_tsn = asoc->sending_seq - 1;
212 } else {
213 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
214 }
215
216 sctp_timer_stop(SCTP_TIMER_TYPE_SEND,
217 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA+SCTP_LOC_32 );
218 sctp_timer_start(SCTP_TIMER_TYPE_SEND,
219 stcb->sctp_ep, stcb, net);
220 }
221 } else if (net->net_ack > 0) {
222 /*
223 * Mark a peg that we WOULD have done a cwnd
224 * reduction but RFC2582 prevented this action.
225 */
226 SCTP_STAT_INCR(sctps_fastretransinrtt);
227 }
228 }
229 }
230
231 /* Defines for instantaneous bw decisions */
232 #define SCTP_INST_LOOSING 1 /* Loosing to other flows */
233 #define SCTP_INST_NEUTRAL 2 /* Neutral, no indication */
234 #define SCTP_INST_GAINING 3 /* Gaining, step down possible */
235
236
237 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
238 static int
cc_bw_same(struct sctp_tcb * stcb,struct sctp_nets * net,uint64_t nbw,uint64_t rtt_offset,uint64_t vtag,uint8_t inst_ind)239 cc_bw_same(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw,
240 uint64_t rtt_offset, uint64_t vtag, uint8_t inst_ind)
241 #else
242 static int
243 cc_bw_same(struct sctp_tcb *stcb SCTP_UNUSED, struct sctp_nets *net, uint64_t nbw,
244 uint64_t rtt_offset, uint8_t inst_ind)
245 #endif
246 {
247 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
248 uint64_t oth, probepoint;
249 #endif
250
251 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
252 probepoint = (((uint64_t)net->cwnd) << 32);
253 #endif
254 if (net->rtt > net->cc_mod.rtcc.lbw_rtt + rtt_offset) {
255 /*
256 * rtt increased
257 * we don't update bw.. so we don't
258 * update the rtt either.
259 */
260 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
261 /* Probe point 5 */
262 probepoint |= ((5 << 16) | 1);
263 SDT_PROBE(sctp, cwnd, net, rttvar,
264 vtag,
265 ((net->cc_mod.rtcc.lbw << 32) | nbw),
266 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
267 net->flight_size,
268 probepoint);
269 #endif
270 if ((net->cc_mod.rtcc.steady_step) && (inst_ind != SCTP_INST_LOOSING)) {
271 if (net->cc_mod.rtcc.last_step_state == 5)
272 net->cc_mod.rtcc.step_cnt++;
273 else
274 net->cc_mod.rtcc.step_cnt = 1;
275 net->cc_mod.rtcc.last_step_state = 5;
276 if ((net->cc_mod.rtcc.step_cnt == net->cc_mod.rtcc.steady_step) ||
277 ((net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step) &&
278 ((net->cc_mod.rtcc.step_cnt % net->cc_mod.rtcc.steady_step) == 0))) {
279 /* Try a step down */
280 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
281 oth = net->cc_mod.rtcc.vol_reduce;
282 oth <<= 16;
283 oth |= net->cc_mod.rtcc.step_cnt;
284 oth <<= 16;
285 oth |= net->cc_mod.rtcc.last_step_state;
286 SDT_PROBE(sctp, cwnd, net, rttstep,
287 vtag,
288 ((net->cc_mod.rtcc.lbw << 32) | nbw),
289 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
290 oth,
291 probepoint);
292 #endif
293 if (net->cwnd > (4 * net->mtu)) {
294 net->cwnd -= net->mtu;
295 net->cc_mod.rtcc.vol_reduce++;
296 } else {
297 net->cc_mod.rtcc.step_cnt = 0;
298 }
299 }
300 }
301 return (1);
302 }
303 if (net->rtt < net->cc_mod.rtcc.lbw_rtt-rtt_offset) {
304 /*
305 * rtt decreased, there could be more room.
306 * we update both the bw and the rtt here to
307 * lock this in as a good step down.
308 */
309 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
310 /* Probe point 6 */
311 probepoint |= ((6 << 16) | 0);
312 SDT_PROBE(sctp, cwnd, net, rttvar,
313 vtag,
314 ((net->cc_mod.rtcc.lbw << 32) | nbw),
315 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
316 net->flight_size,
317 probepoint);
318 #endif
319 if (net->cc_mod.rtcc.steady_step) {
320 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
321 oth = net->cc_mod.rtcc.vol_reduce;
322 oth <<= 16;
323 oth |= net->cc_mod.rtcc.step_cnt;
324 oth <<= 16;
325 oth |= net->cc_mod.rtcc.last_step_state;
326 SDT_PROBE(sctp, cwnd, net, rttstep,
327 vtag,
328 ((net->cc_mod.rtcc.lbw << 32) | nbw),
329 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
330 oth,
331 probepoint);
332 #endif
333 if ((net->cc_mod.rtcc.last_step_state == 5) &&
334 (net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step)) {
335 /* Step down worked */
336 net->cc_mod.rtcc.step_cnt = 0;
337 return (1);
338 } else {
339 net->cc_mod.rtcc.last_step_state = 6;
340 net->cc_mod.rtcc.step_cnt = 0;
341 }
342 }
343 net->cc_mod.rtcc.lbw = nbw;
344 net->cc_mod.rtcc.lbw_rtt = net->rtt;
345 net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd;
346 if (inst_ind == SCTP_INST_GAINING)
347 return (1);
348 else if (inst_ind == SCTP_INST_NEUTRAL)
349 return (1);
350 else
351 return (0);
352 }
353 /* Ok bw and rtt remained the same .. no update to any
354 */
355 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
356 /* Probe point 7 */
357 probepoint |= ((7 << 16) | net->cc_mod.rtcc.ret_from_eq);
358 SDT_PROBE(sctp, cwnd, net, rttvar,
359 vtag,
360 ((net->cc_mod.rtcc.lbw << 32) | nbw),
361 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
362 net->flight_size,
363 probepoint);
364 #endif
365 if ((net->cc_mod.rtcc.steady_step) && (inst_ind != SCTP_INST_LOOSING)) {
366 if (net->cc_mod.rtcc.last_step_state == 5)
367 net->cc_mod.rtcc.step_cnt++;
368 else
369 net->cc_mod.rtcc.step_cnt = 1;
370 net->cc_mod.rtcc.last_step_state = 5;
371 if ((net->cc_mod.rtcc.step_cnt == net->cc_mod.rtcc.steady_step) ||
372 ((net->cc_mod.rtcc.step_cnt > net->cc_mod.rtcc.steady_step) &&
373 ((net->cc_mod.rtcc.step_cnt % net->cc_mod.rtcc.steady_step) == 0))) {
374 /* Try a step down */
375 if (net->cwnd > (4 * net->mtu)) {
376 net->cwnd -= net->mtu;
377 net->cc_mod.rtcc.vol_reduce++;
378 return (1);
379 } else {
380 net->cc_mod.rtcc.step_cnt = 0;
381 }
382 }
383 }
384 if (inst_ind == SCTP_INST_GAINING)
385 return (1);
386 else if (inst_ind == SCTP_INST_NEUTRAL)
387 return (1);
388 else
389 return ((int)net->cc_mod.rtcc.ret_from_eq);
390 }
391
392 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
393 static int
cc_bw_decrease(struct sctp_tcb * stcb,struct sctp_nets * net,uint64_t nbw,uint64_t rtt_offset,uint64_t vtag,uint8_t inst_ind)394 cc_bw_decrease(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, uint64_t rtt_offset,
395 uint64_t vtag, uint8_t inst_ind)
396 #else
397 static int
398 cc_bw_decrease(struct sctp_tcb *stcb SCTP_UNUSED, struct sctp_nets *net, uint64_t nbw, uint64_t rtt_offset,
399 uint8_t inst_ind)
400 #endif
401 {
402 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
403 uint64_t oth, probepoint;
404 #endif
405
406 /* Bandwidth decreased.*/
407 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
408 probepoint = (((uint64_t)net->cwnd) << 32);
409 #endif
410 if (net->rtt > net->cc_mod.rtcc.lbw_rtt+rtt_offset) {
411 /* rtt increased */
412 /* Did we add more */
413 if ((net->cwnd > net->cc_mod.rtcc.cwnd_at_bw_set) &&
414 (inst_ind != SCTP_INST_LOOSING)) {
415 /* We caused it maybe.. back off? */
416 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
417 /* PROBE POINT 1 */
418 probepoint |= ((1 << 16) | 1);
419 SDT_PROBE(sctp, cwnd, net, rttvar,
420 vtag,
421 ((net->cc_mod.rtcc.lbw << 32) | nbw),
422 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
423 net->flight_size,
424 probepoint);
425 #endif
426 if (net->cc_mod.rtcc.ret_from_eq) {
427 /* Switch over to CA if we are less aggressive */
428 net->ssthresh = net->cwnd-1;
429 net->partial_bytes_acked = 0;
430 }
431 return (1);
432 }
433 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
434 /* Probe point 2 */
435 probepoint |= ((2 << 16) | 0);
436 SDT_PROBE(sctp, cwnd, net, rttvar,
437 vtag,
438 ((net->cc_mod.rtcc.lbw << 32) | nbw),
439 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
440 net->flight_size,
441 probepoint);
442 #endif
443 /* Someone else - fight for more? */
444 if (net->cc_mod.rtcc.steady_step) {
445 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
446 oth = net->cc_mod.rtcc.vol_reduce;
447 oth <<= 16;
448 oth |= net->cc_mod.rtcc.step_cnt;
449 oth <<= 16;
450 oth |= net->cc_mod.rtcc.last_step_state;
451 SDT_PROBE(sctp, cwnd, net, rttstep,
452 vtag,
453 ((net->cc_mod.rtcc.lbw << 32) | nbw),
454 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
455 oth,
456 probepoint);
457 #endif
458 /* Did we voluntarily give up some? if so take
459 * one back please
460 */
461 if ((net->cc_mod.rtcc.vol_reduce) &&
462 (inst_ind != SCTP_INST_GAINING)) {
463 net->cwnd += net->mtu;
464 net->cc_mod.rtcc.vol_reduce--;
465 }
466 net->cc_mod.rtcc.last_step_state = 2;
467 net->cc_mod.rtcc.step_cnt = 0;
468 }
469 goto out_decision;
470 } else if (net->rtt < net->cc_mod.rtcc.lbw_rtt-rtt_offset) {
471 /* bw & rtt decreased */
472 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
473 /* Probe point 3 */
474 probepoint |= ((3 << 16) | 0);
475 SDT_PROBE(sctp, cwnd, net, rttvar,
476 vtag,
477 ((net->cc_mod.rtcc.lbw << 32) | nbw),
478 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
479 net->flight_size,
480 probepoint);
481 #endif
482 if (net->cc_mod.rtcc.steady_step) {
483 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
484 oth = net->cc_mod.rtcc.vol_reduce;
485 oth <<= 16;
486 oth |= net->cc_mod.rtcc.step_cnt;
487 oth <<= 16;
488 oth |= net->cc_mod.rtcc.last_step_state;
489 SDT_PROBE(sctp, cwnd, net, rttstep,
490 vtag,
491 ((net->cc_mod.rtcc.lbw << 32) | nbw),
492 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
493 oth,
494 probepoint);
495 #endif
496 if ((net->cc_mod.rtcc.vol_reduce) &&
497 (inst_ind != SCTP_INST_GAINING)) {
498 net->cwnd += net->mtu;
499 net->cc_mod.rtcc.vol_reduce--;
500 }
501 net->cc_mod.rtcc.last_step_state = 3;
502 net->cc_mod.rtcc.step_cnt = 0;
503 }
504 goto out_decision;
505 }
506 /* The bw decreased but rtt stayed the same */
507 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
508 /* Probe point 4 */
509 probepoint |= ((4 << 16) | 0);
510 SDT_PROBE(sctp, cwnd, net, rttvar,
511 vtag,
512 ((net->cc_mod.rtcc.lbw << 32) | nbw),
513 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
514 net->flight_size,
515 probepoint);
516 #endif
517 if (net->cc_mod.rtcc.steady_step) {
518 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
519 oth = net->cc_mod.rtcc.vol_reduce;
520 oth <<= 16;
521 oth |= net->cc_mod.rtcc.step_cnt;
522 oth <<= 16;
523 oth |= net->cc_mod.rtcc.last_step_state;
524 SDT_PROBE(sctp, cwnd, net, rttstep,
525 vtag,
526 ((net->cc_mod.rtcc.lbw << 32) | nbw),
527 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
528 oth,
529 probepoint);
530 #endif
531 if ((net->cc_mod.rtcc.vol_reduce) &&
532 (inst_ind != SCTP_INST_GAINING)) {
533 net->cwnd += net->mtu;
534 net->cc_mod.rtcc.vol_reduce--;
535 }
536 net->cc_mod.rtcc.last_step_state = 4;
537 net->cc_mod.rtcc.step_cnt = 0;
538 }
539 out_decision:
540 net->cc_mod.rtcc.lbw = nbw;
541 net->cc_mod.rtcc.lbw_rtt = net->rtt;
542 net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd;
543 if (inst_ind == SCTP_INST_GAINING) {
544 return (1);
545 } else {
546 return (0);
547 }
548 }
549
550 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
551 static int
cc_bw_increase(struct sctp_tcb * stcb,struct sctp_nets * net,uint64_t nbw,uint64_t vtag)552 cc_bw_increase(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw, uint64_t vtag)
553 #else
554 static int
555 cc_bw_increase(struct sctp_tcb *stcb SCTP_UNUSED, struct sctp_nets *net, uint64_t nbw)
556 #endif
557 {
558 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
559 uint64_t oth, probepoint;
560
561 #endif
562 /* BW increased, so update and
563 * return 0, since all actions in
564 * our table say to do the normal CC
565 * update. Note that we pay no attention to
566 * the inst_ind since our overall sum is increasing.
567 */
568 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
569 /* PROBE POINT 0 */
570 probepoint = (((uint64_t)net->cwnd) << 32);
571 SDT_PROBE(sctp, cwnd, net, rttvar,
572 vtag,
573 ((net->cc_mod.rtcc.lbw << 32) | nbw),
574 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
575 net->flight_size,
576 probepoint);
577 #endif
578 if (net->cc_mod.rtcc.steady_step) {
579 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
580 oth = net->cc_mod.rtcc.vol_reduce;
581 oth <<= 16;
582 oth |= net->cc_mod.rtcc.step_cnt;
583 oth <<= 16;
584 oth |= net->cc_mod.rtcc.last_step_state;
585 SDT_PROBE(sctp, cwnd, net, rttstep,
586 vtag,
587 ((net->cc_mod.rtcc.lbw << 32) | nbw),
588 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
589 oth,
590 probepoint);
591 #endif
592 net->cc_mod.rtcc.last_step_state = 0;
593 net->cc_mod.rtcc.step_cnt = 0;
594 net->cc_mod.rtcc.vol_reduce = 0;
595 }
596 net->cc_mod.rtcc.lbw = nbw;
597 net->cc_mod.rtcc.lbw_rtt = net->rtt;
598 net->cc_mod.rtcc.cwnd_at_bw_set = net->cwnd;
599 return (0);
600 }
601
602 /* RTCC Algoritm to limit growth of cwnd, return
603 * true if you want to NOT allow cwnd growth
604 */
605 static int
cc_bw_limit(struct sctp_tcb * stcb,struct sctp_nets * net,uint64_t nbw)606 cc_bw_limit(struct sctp_tcb *stcb, struct sctp_nets *net, uint64_t nbw)
607 {
608 uint64_t bw_offset, rtt_offset;
609 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
610 uint64_t probepoint, rtt, vtag;
611 #endif
612 uint64_t bytes_for_this_rtt, inst_bw;
613 uint64_t div, inst_off;
614 int bw_shift;
615 uint8_t inst_ind;
616 int ret;
617 /*-
618 * Here we need to see if we want
619 * to limit cwnd growth due to increase
620 * in overall rtt but no increase in bw.
621 * We use the following table to figure
622 * out what we should do. When we return
623 * 0, cc update goes on as planned. If we
624 * return 1, then no cc update happens and cwnd
625 * stays where it is at.
626 * ----------------------------------
627 * BW | RTT | Action
628 * *********************************
629 * INC | INC | return 0
630 * ----------------------------------
631 * INC | SAME | return 0
632 * ----------------------------------
633 * INC | DECR | return 0
634 * ----------------------------------
635 * SAME | INC | return 1
636 * ----------------------------------
637 * SAME | SAME | return 1
638 * ----------------------------------
639 * SAME | DECR | return 0
640 * ----------------------------------
641 * DECR | INC | return 0 or 1 based on if we caused.
642 * ----------------------------------
643 * DECR | SAME | return 0
644 * ----------------------------------
645 * DECR | DECR | return 0
646 * ----------------------------------
647 *
648 * We are a bit fuzz on what an increase or
649 * decrease is. For BW it is the same if
650 * it did not change within 1/64th. For
651 * RTT it stayed the same if it did not
652 * change within 1/32nd
653 */
654 bw_shift = SCTP_BASE_SYSCTL(sctp_rttvar_bw);
655 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
656 rtt = stcb->asoc.my_vtag;
657 vtag = (rtt << 32) | (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) | (stcb->rport);
658 probepoint = (((uint64_t)net->cwnd) << 32);
659 rtt = net->rtt;
660 #endif
661 if (net->cc_mod.rtcc.rtt_set_this_sack) {
662 net->cc_mod.rtcc.rtt_set_this_sack = 0;
663 bytes_for_this_rtt = net->cc_mod.rtcc.bw_bytes - net->cc_mod.rtcc.bw_bytes_at_last_rttc;
664 net->cc_mod.rtcc.bw_bytes_at_last_rttc = net->cc_mod.rtcc.bw_bytes;
665 if (net->rtt) {
666 div = net->rtt / 1000;
667 if (div) {
668 inst_bw = bytes_for_this_rtt / div;
669 inst_off = inst_bw >> bw_shift;
670 if (inst_bw > nbw)
671 inst_ind = SCTP_INST_GAINING;
672 else if ((inst_bw+inst_off) < nbw)
673 inst_ind = SCTP_INST_LOOSING;
674 else
675 inst_ind = SCTP_INST_NEUTRAL;
676 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
677 probepoint |= ((0xb << 16) | inst_ind);
678 #endif
679 } else {
680 inst_ind = net->cc_mod.rtcc.last_inst_ind;
681 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
682 inst_bw = bytes_for_this_rtt / (uint64_t)(net->rtt);
683 /* Can't determine do not change */
684 probepoint |= ((0xc << 16) | inst_ind);
685 #endif
686 }
687 } else {
688 inst_ind = net->cc_mod.rtcc.last_inst_ind;
689 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
690 inst_bw = bytes_for_this_rtt;
691 /* Can't determine do not change */
692 probepoint |= ((0xd << 16) | inst_ind);
693 #endif
694 }
695 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
696 SDT_PROBE(sctp, cwnd, net, rttvar,
697 vtag,
698 ((nbw << 32) | inst_bw),
699 ((net->cc_mod.rtcc.lbw_rtt << 32) | rtt),
700 net->flight_size,
701 probepoint);
702 #endif
703 } else {
704 /* No rtt measurement, use last one */
705 inst_ind = net->cc_mod.rtcc.last_inst_ind;
706 }
707 bw_offset = net->cc_mod.rtcc.lbw >> bw_shift;
708 if (nbw > net->cc_mod.rtcc.lbw + bw_offset) {
709 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
710 ret = cc_bw_increase(stcb, net, nbw, vtag);
711 #else
712 ret = cc_bw_increase(stcb, net, nbw);
713 #endif
714 goto out;
715 }
716 rtt_offset = net->cc_mod.rtcc.lbw_rtt >> SCTP_BASE_SYSCTL(sctp_rttvar_rtt);
717 if (nbw < net->cc_mod.rtcc.lbw - bw_offset) {
718 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
719 ret = cc_bw_decrease(stcb, net, nbw, rtt_offset, vtag, inst_ind);
720 #else
721 ret = cc_bw_decrease(stcb, net, nbw, rtt_offset, inst_ind);
722 #endif
723 goto out;
724 }
725 /* If we reach here then
726 * we are in a situation where
727 * the bw stayed the same.
728 */
729 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
730 ret = cc_bw_same(stcb, net, nbw, rtt_offset, vtag, inst_ind);
731 #else
732 ret = cc_bw_same(stcb, net, nbw, rtt_offset, inst_ind);
733 #endif
734 out:
735 net->cc_mod.rtcc.last_inst_ind = inst_ind;
736 return (ret);
737 }
738
739 static void
sctp_cwnd_update_after_sack_common(struct sctp_tcb * stcb,struct sctp_association * asoc,int accum_moved,int reneged_all SCTP_UNUSED,int will_exit,int use_rtcc)740 sctp_cwnd_update_after_sack_common(struct sctp_tcb *stcb,
741 struct sctp_association *asoc,
742 int accum_moved, int reneged_all SCTP_UNUSED, int will_exit, int use_rtcc)
743 {
744 struct sctp_nets *net;
745 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
746 int old_cwnd;
747 #endif
748 uint32_t t_ssthresh, t_cwnd, incr;
749 uint64_t t_ucwnd_sbw;
750 uint64_t t_path_mptcp;
751 uint64_t mptcp_like_alpha;
752 uint32_t srtt;
753 uint64_t max_path;
754
755 /* MT FIXME: Don't compute this over and over again */
756 t_ssthresh = 0;
757 t_cwnd = 0;
758 t_ucwnd_sbw = 0;
759 t_path_mptcp = 0;
760 mptcp_like_alpha = 1;
761 if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) ||
762 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2) ||
763 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_MPTCP)) {
764 max_path = 0;
765 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) {
766 t_ssthresh += net->ssthresh;
767 t_cwnd += net->cwnd;
768 /* lastsa>>3; we don't need to devide ...*/
769 srtt = net->lastsa;
770 if (srtt > 0) {
771 uint64_t tmp;
772
773 t_ucwnd_sbw += (uint64_t)net->cwnd / (uint64_t)srtt;
774 t_path_mptcp += (((uint64_t)net->cwnd) << SHIFT_MPTCP_MULTI_Z) /
775 (((uint64_t)net->mtu) * (uint64_t)srtt);
776 tmp = (((uint64_t)net->cwnd) << SHIFT_MPTCP_MULTI_N) /
777 ((uint64_t)net->mtu * (uint64_t)(srtt * srtt));
778 if (tmp > max_path) {
779 max_path = tmp;
780 }
781 }
782 }
783 if (t_path_mptcp > 0) {
784 mptcp_like_alpha = max_path / (t_path_mptcp * t_path_mptcp);
785 } else {
786 mptcp_like_alpha = 1;
787 }
788 }
789 if (t_ssthresh == 0) {
790 t_ssthresh = 1;
791 }
792 if (t_ucwnd_sbw == 0) {
793 t_ucwnd_sbw = 1;
794 }
795 /******************************/
796 /* update cwnd and Early FR */
797 /******************************/
798 TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
799
800 #ifdef JANA_CMT_FAST_RECOVERY
801 /*
802 * CMT fast recovery code. Need to debug.
803 */
804 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) {
805 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) ||
806 SCTP_TSN_GE(net->pseudo_cumack,net->fast_recovery_tsn)) {
807 net->will_exit_fast_recovery = 1;
808 }
809 }
810 #endif
811 /* if nothing was acked on this destination skip it */
812 if (net->net_ack == 0) {
813 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
814 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK);
815 }
816 continue;
817 }
818 #ifdef JANA_CMT_FAST_RECOVERY
819 /* CMT fast recovery code
820 */
821 /*
822 if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery && net->will_exit_fast_recovery == 0) {
823 @@@ Do something
824 }
825 else if (sctp_cmt_on_off == 0 && asoc->fast_retran_loss_recovery && will_exit == 0) {
826 */
827 #endif
828
829 if (asoc->fast_retran_loss_recovery &&
830 (will_exit == 0) &&
831 (asoc->sctp_cmt_on_off == 0)) {
832 /*
833 * If we are in loss recovery we skip any cwnd
834 * update
835 */
836 return;
837 }
838 /*
839 * Did any measurements go on for this network?
840 */
841 if (use_rtcc && (net->cc_mod.rtcc.tls_needs_set > 0)) {
842 uint64_t nbw;
843 /*
844 * At this point our bw_bytes has been updated
845 * by incoming sack information.
846 *
847 * But our bw may not yet be set.
848 *
849 */
850 if ((net->cc_mod.rtcc.new_tot_time/1000) > 0) {
851 nbw = net->cc_mod.rtcc.bw_bytes/(net->cc_mod.rtcc.new_tot_time/1000);
852 } else {
853 nbw = net->cc_mod.rtcc.bw_bytes;
854 }
855 if (net->cc_mod.rtcc.lbw) {
856 if (cc_bw_limit(stcb, net, nbw)) {
857 /* Hold here, no update */
858 continue;
859 }
860 } else {
861 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
862 uint64_t vtag, probepoint;
863
864 probepoint = (((uint64_t)net->cwnd) << 32);
865 probepoint |= ((0xa << 16) | 0);
866 vtag = (net->rtt << 32) |
867 (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) |
868 (stcb->rport);
869
870 SDT_PROBE(sctp, cwnd, net, rttvar,
871 vtag,
872 nbw,
873 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
874 net->flight_size,
875 probepoint);
876 #endif
877 net->cc_mod.rtcc.lbw = nbw;
878 net->cc_mod.rtcc.lbw_rtt = net->rtt;
879 if (net->cc_mod.rtcc.rtt_set_this_sack) {
880 net->cc_mod.rtcc.rtt_set_this_sack = 0;
881 net->cc_mod.rtcc.bw_bytes_at_last_rttc = net->cc_mod.rtcc.bw_bytes;
882 }
883 }
884 }
885 /*
886 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has
887 * moved.
888 */
889 if (accum_moved ||
890 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) {
891 /* If the cumulative ack moved we can proceed */
892 if (net->cwnd <= net->ssthresh) {
893 /* We are in slow start */
894 if (net->flight_size + net->net_ack >= net->cwnd) {
895 uint32_t limit;
896
897 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
898 old_cwnd = net->cwnd;
899 #endif
900 switch (asoc->sctp_cmt_on_off) {
901 case SCTP_CMT_RPV1:
902 limit = (uint32_t)(((uint64_t)net->mtu *
903 (uint64_t)SCTP_BASE_SYSCTL(sctp_L2_abc_variable) *
904 (uint64_t)net->ssthresh) /
905 (uint64_t)t_ssthresh);
906 incr = (uint32_t)(((uint64_t)net->net_ack *
907 (uint64_t)net->ssthresh) /
908 (uint64_t)t_ssthresh);
909 if (incr > limit) {
910 incr = limit;
911 }
912 if (incr == 0) {
913 incr = 1;
914 }
915 break;
916 case SCTP_CMT_RPV2:
917 /* lastsa>>3; we don't need to divide ...*/
918 srtt = net->lastsa;
919 if (srtt == 0) {
920 srtt = 1;
921 }
922 limit = (uint32_t)(((uint64_t)net->mtu *
923 (uint64_t)SCTP_BASE_SYSCTL(sctp_L2_abc_variable) *
924 (uint64_t)net->cwnd) /
925 ((uint64_t)srtt * t_ucwnd_sbw));
926 /* INCREASE FACTOR */
927 incr = (uint32_t)(((uint64_t)net->net_ack *
928 (uint64_t)net->cwnd) /
929 ((uint64_t)srtt * t_ucwnd_sbw));
930 /* INCREASE FACTOR */
931 if (incr > limit) {
932 incr = limit;
933 }
934 if (incr == 0) {
935 incr = 1;
936 }
937 break;
938 case SCTP_CMT_MPTCP:
939 limit = (uint32_t)(((uint64_t)net->mtu *
940 mptcp_like_alpha *
941 (uint64_t)SCTP_BASE_SYSCTL(sctp_L2_abc_variable)) >>
942 SHIFT_MPTCP_MULTI);
943 incr = (uint32_t)(((uint64_t)net->net_ack *
944 mptcp_like_alpha) >>
945 SHIFT_MPTCP_MULTI);
946 if (incr > limit) {
947 incr = limit;
948 }
949 if (incr > net->net_ack) {
950 incr = net->net_ack;
951 }
952 if (incr > net->mtu) {
953 incr = net->mtu;
954 }
955 break;
956 default:
957 incr = net->net_ack;
958 if (incr > net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable)) {
959 incr = net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable);
960 }
961 break;
962 }
963 net->cwnd += incr;
964 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
965 sctp_log_cwnd(stcb, net, incr,
966 SCTP_CWND_LOG_FROM_SS);
967 }
968 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
969 SDT_PROBE(sctp, cwnd, net, ack,
970 stcb->asoc.my_vtag,
971 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)),
972 net,
973 old_cwnd, net->cwnd);
974 #endif
975 } else {
976 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
977 sctp_log_cwnd(stcb, net, net->net_ack,
978 SCTP_CWND_LOG_NOADV_SS);
979 }
980 }
981 } else {
982 /* We are in congestion avoidance */
983 /*
984 * Add to pba
985 */
986 net->partial_bytes_acked += net->net_ack;
987
988 if ((net->flight_size + net->net_ack >= net->cwnd) &&
989 (net->partial_bytes_acked >= net->cwnd)) {
990 net->partial_bytes_acked -= net->cwnd;
991 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
992 old_cwnd = net->cwnd;
993 #endif
994 switch (asoc->sctp_cmt_on_off) {
995 case SCTP_CMT_RPV1:
996 incr = (uint32_t)(((uint64_t)net->mtu *
997 (uint64_t)net->ssthresh) /
998 (uint64_t)t_ssthresh);
999 if (incr == 0) {
1000 incr = 1;
1001 }
1002 break;
1003 case SCTP_CMT_RPV2:
1004 /* lastsa>>3; we don't need to divide ... */
1005 srtt = net->lastsa;
1006 if (srtt == 0) {
1007 srtt = 1;
1008 }
1009 incr = (uint32_t)((uint64_t)net->mtu *
1010 (uint64_t)net->cwnd /
1011 ((uint64_t)srtt *
1012 t_ucwnd_sbw));
1013 /* INCREASE FACTOR */
1014 if (incr == 0) {
1015 incr = 1;
1016 }
1017 break;
1018 case SCTP_CMT_MPTCP:
1019 incr = (uint32_t)((mptcp_like_alpha *
1020 (uint64_t) net->cwnd) >>
1021 SHIFT_MPTCP_MULTI);
1022 if (incr > net->mtu) {
1023 incr = net->mtu;
1024 }
1025 break;
1026 default:
1027 incr = net->mtu;
1028 break;
1029 }
1030 net->cwnd += incr;
1031 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
1032 SDT_PROBE(sctp, cwnd, net, ack,
1033 stcb->asoc.my_vtag,
1034 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)),
1035 net,
1036 old_cwnd, net->cwnd);
1037 #endif
1038 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
1039 sctp_log_cwnd(stcb, net, net->mtu,
1040 SCTP_CWND_LOG_FROM_CA);
1041 }
1042 } else {
1043 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
1044 sctp_log_cwnd(stcb, net, net->net_ack,
1045 SCTP_CWND_LOG_NOADV_CA);
1046 }
1047 }
1048 }
1049 } else {
1050 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
1051 sctp_log_cwnd(stcb, net, net->mtu,
1052 SCTP_CWND_LOG_NO_CUMACK);
1053 }
1054 }
1055 }
1056 }
1057
1058 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
1059 static void
sctp_cwnd_update_exit_pf_common(struct sctp_tcb * stcb,struct sctp_nets * net)1060 sctp_cwnd_update_exit_pf_common(struct sctp_tcb *stcb, struct sctp_nets *net)
1061 #else
1062 static void
1063 sctp_cwnd_update_exit_pf_common(struct sctp_tcb *stcb SCTP_UNUSED, struct sctp_nets *net)
1064 #endif
1065 {
1066 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
1067 int old_cwnd;
1068
1069 old_cwnd = net->cwnd;
1070 #endif
1071 net->cwnd = net->mtu;
1072 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
1073 SDT_PROBE(sctp, cwnd, net, ack,
1074 stcb->asoc.my_vtag, ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)), net,
1075 old_cwnd, net->cwnd);
1076 #endif
1077 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n",
1078 (void *)net, net->cwnd);
1079 }
1080
1081
1082 static void
sctp_cwnd_update_after_timeout(struct sctp_tcb * stcb,struct sctp_nets * net)1083 sctp_cwnd_update_after_timeout(struct sctp_tcb *stcb, struct sctp_nets *net)
1084 {
1085 int old_cwnd = net->cwnd;
1086 uint32_t t_ssthresh, t_cwnd;
1087 uint64_t t_ucwnd_sbw;
1088
1089 /* MT FIXME: Don't compute this over and over again */
1090 t_ssthresh = 0;
1091 t_cwnd = 0;
1092 if ((stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) ||
1093 (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV2)) {
1094 struct sctp_nets *lnet;
1095 uint32_t srtt;
1096
1097 t_ucwnd_sbw = 0;
1098 TAILQ_FOREACH(lnet, &stcb->asoc.nets, sctp_next) {
1099 t_ssthresh += lnet->ssthresh;
1100 t_cwnd += lnet->cwnd;
1101 srtt = lnet->lastsa;
1102 /* lastsa>>3; we don't need to divide ... */
1103 if (srtt > 0) {
1104 t_ucwnd_sbw += (uint64_t)lnet->cwnd / (uint64_t)srtt;
1105 }
1106 }
1107 if (t_ssthresh < 1) {
1108 t_ssthresh = 1;
1109 }
1110 if (t_ucwnd_sbw < 1) {
1111 t_ucwnd_sbw = 1;
1112 }
1113 if (stcb->asoc.sctp_cmt_on_off == SCTP_CMT_RPV1) {
1114 net->ssthresh = (uint32_t)(((uint64_t)4 *
1115 (uint64_t)net->mtu *
1116 (uint64_t)net->ssthresh) /
1117 (uint64_t)t_ssthresh);
1118 } else {
1119 uint64_t cc_delta;
1120
1121 srtt = net->lastsa;
1122 /* lastsa>>3; we don't need to divide ... */
1123 if (srtt == 0) {
1124 srtt = 1;
1125 }
1126 cc_delta = t_ucwnd_sbw * (uint64_t)srtt / 2;
1127 if (cc_delta < t_cwnd) {
1128 net->ssthresh = (uint32_t)((uint64_t)t_cwnd - cc_delta);
1129 } else {
1130 net->ssthresh = net->mtu;
1131 }
1132 }
1133 if ((net->cwnd > t_cwnd / 2) &&
1134 (net->ssthresh < net->cwnd - t_cwnd / 2)) {
1135 net->ssthresh = net->cwnd - t_cwnd / 2;
1136 }
1137 if (net->ssthresh < net->mtu) {
1138 net->ssthresh = net->mtu;
1139 }
1140 } else {
1141 net->ssthresh = max(net->cwnd / 2, 4 * net->mtu);
1142 }
1143 net->cwnd = net->mtu;
1144 net->partial_bytes_acked = 0;
1145 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
1146 SDT_PROBE(sctp, cwnd, net, to,
1147 stcb->asoc.my_vtag,
1148 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)),
1149 net,
1150 old_cwnd, net->cwnd);
1151 #endif
1152 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
1153 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX);
1154 }
1155 }
1156
1157 static void
sctp_cwnd_update_after_ecn_echo_common(struct sctp_tcb * stcb,struct sctp_nets * net,int in_window,int num_pkt_lost,int use_rtcc)1158 sctp_cwnd_update_after_ecn_echo_common(struct sctp_tcb *stcb, struct sctp_nets *net,
1159 int in_window, int num_pkt_lost, int use_rtcc)
1160 {
1161 int old_cwnd = net->cwnd;
1162 if ((use_rtcc) && (net->lan_type == SCTP_LAN_LOCAL) && (net->cc_mod.rtcc.use_dccc_ecn)) {
1163 /* Data center Congestion Control */
1164 if (in_window == 0) {
1165 /* Go to CA with the cwnd at the point we sent
1166 * the TSN that was marked with a CE.
1167 */
1168 if (net->ecn_prev_cwnd < net->cwnd) {
1169 /* Restore to prev cwnd */
1170 net->cwnd = net->ecn_prev_cwnd - (net->mtu * num_pkt_lost);
1171 } else {
1172 /* Just cut in 1/2 */
1173 net->cwnd /= 2;
1174 }
1175 /* Drop to CA */
1176 net->ssthresh = net->cwnd - (num_pkt_lost * net->mtu);
1177 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
1178 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT);
1179 }
1180 } else {
1181 /* Further tuning down required over the drastic orginal cut */
1182 net->ssthresh -= (net->mtu * num_pkt_lost);
1183 net->cwnd -= (net->mtu * num_pkt_lost);
1184 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
1185 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT);
1186 }
1187
1188 }
1189 SCTP_STAT_INCR(sctps_ecnereducedcwnd);
1190 } else {
1191 if (in_window == 0) {
1192 SCTP_STAT_INCR(sctps_ecnereducedcwnd);
1193 net->ssthresh = net->cwnd / 2;
1194 if (net->ssthresh < net->mtu) {
1195 net->ssthresh = net->mtu;
1196 /* here back off the timer as well, to slow us down */
1197 net->RTO <<= 1;
1198 }
1199 net->cwnd = net->ssthresh;
1200 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
1201 SDT_PROBE(sctp, cwnd, net, ecn,
1202 stcb->asoc.my_vtag,
1203 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)),
1204 net,
1205 old_cwnd, net->cwnd);
1206 #endif
1207 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
1208 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT);
1209 }
1210 }
1211 }
1212
1213 }
1214
1215 static void
sctp_cwnd_update_after_packet_dropped(struct sctp_tcb * stcb,struct sctp_nets * net,struct sctp_pktdrop_chunk * cp,uint32_t * bottle_bw,uint32_t * on_queue)1216 sctp_cwnd_update_after_packet_dropped(struct sctp_tcb *stcb,
1217 struct sctp_nets *net, struct sctp_pktdrop_chunk *cp,
1218 uint32_t *bottle_bw, uint32_t *on_queue)
1219 {
1220 uint32_t bw_avail;
1221 int rtt;
1222 unsigned int incr;
1223 int old_cwnd = net->cwnd;
1224
1225 /* need real RTT in msd for this calc */
1226 rtt = net->rtt / 1000;
1227 /* get bottle neck bw */
1228 *bottle_bw = ntohl(cp->bottle_bw);
1229 /* and whats on queue */
1230 *on_queue = ntohl(cp->current_onq);
1231 /*
1232 * adjust the on-queue if our flight is more it could be
1233 * that the router has not yet gotten data "in-flight" to it
1234 */
1235 if (*on_queue < net->flight_size)
1236 *on_queue = net->flight_size;
1237 /* calculate the available space */
1238 bw_avail = (*bottle_bw * rtt) / 1000;
1239 if (bw_avail > *bottle_bw) {
1240 /*
1241 * Cap the growth to no more than the bottle neck.
1242 * This can happen as RTT slides up due to queues.
1243 * It also means if you have more than a 1 second
1244 * RTT with a empty queue you will be limited to the
1245 * bottle_bw per second no matter if other points
1246 * have 1/2 the RTT and you could get more out...
1247 */
1248 bw_avail = *bottle_bw;
1249 }
1250 if (*on_queue > bw_avail) {
1251 /*
1252 * No room for anything else don't allow anything
1253 * else to be "added to the fire".
1254 */
1255 int seg_inflight, seg_onqueue, my_portion;
1256 net->partial_bytes_acked = 0;
1257
1258 /* how much are we over queue size? */
1259 incr = *on_queue - bw_avail;
1260 if (stcb->asoc.seen_a_sack_this_pkt) {
1261 /*
1262 * undo any cwnd adjustment that the sack
1263 * might have made
1264 */
1265 net->cwnd = net->prev_cwnd;
1266 }
1267 /* Now how much of that is mine? */
1268 seg_inflight = net->flight_size / net->mtu;
1269 seg_onqueue = *on_queue / net->mtu;
1270 my_portion = (incr * seg_inflight) / seg_onqueue;
1271
1272 /* Have I made an adjustment already */
1273 if (net->cwnd > net->flight_size) {
1274 /*
1275 * for this flight I made an adjustment we
1276 * need to decrease the portion by a share
1277 * our previous adjustment.
1278 */
1279 int diff_adj;
1280
1281 diff_adj = net->cwnd - net->flight_size;
1282 if (diff_adj > my_portion)
1283 my_portion = 0;
1284 else
1285 my_portion -= diff_adj;
1286 }
1287 /*
1288 * back down to the previous cwnd (assume we have
1289 * had a sack before this packet). minus what ever
1290 * portion of the overage is my fault.
1291 */
1292 net->cwnd -= my_portion;
1293
1294 /* we will NOT back down more than 1 MTU */
1295 if (net->cwnd <= net->mtu) {
1296 net->cwnd = net->mtu;
1297 }
1298 /* force into CA */
1299 net->ssthresh = net->cwnd - 1;
1300 } else {
1301 /*
1302 * Take 1/4 of the space left or max burst up ..
1303 * whichever is less.
1304 */
1305 incr = (bw_avail - *on_queue) >> 2;
1306 if ((stcb->asoc.max_burst > 0) &&
1307 (stcb->asoc.max_burst * net->mtu < incr)) {
1308 incr = stcb->asoc.max_burst * net->mtu;
1309 }
1310 net->cwnd += incr;
1311 }
1312 if (net->cwnd > bw_avail) {
1313 /* We can't exceed the pipe size */
1314 net->cwnd = bw_avail;
1315 }
1316 if (net->cwnd < net->mtu) {
1317 /* We always have 1 MTU */
1318 net->cwnd = net->mtu;
1319 }
1320
1321 if (net->cwnd - old_cwnd != 0) {
1322 /* log only changes */
1323 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
1324 SDT_PROBE(sctp, cwnd, net, pd,
1325 stcb->asoc.my_vtag,
1326 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)),
1327 net,
1328 old_cwnd, net->cwnd);
1329 #endif
1330 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
1331 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd),
1332 SCTP_CWND_LOG_FROM_SAT);
1333 }
1334 }
1335 }
1336
1337 static void
sctp_cwnd_update_after_output(struct sctp_tcb * stcb,struct sctp_nets * net,int burst_limit)1338 sctp_cwnd_update_after_output(struct sctp_tcb *stcb,
1339 struct sctp_nets *net, int burst_limit)
1340 {
1341 int old_cwnd = net->cwnd;
1342
1343 if (net->ssthresh < net->cwnd)
1344 net->ssthresh = net->cwnd;
1345 if (burst_limit) {
1346 net->cwnd = (net->flight_size + (burst_limit * net->mtu));
1347 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
1348 SDT_PROBE(sctp, cwnd, net, bl,
1349 stcb->asoc.my_vtag,
1350 ((stcb->sctp_ep->sctp_lport << 16) | (stcb->rport)),
1351 net,
1352 old_cwnd, net->cwnd);
1353 #endif
1354 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
1355 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_BRST);
1356 }
1357 }
1358 }
1359
1360 static void
sctp_cwnd_update_after_sack(struct sctp_tcb * stcb,struct sctp_association * asoc,int accum_moved,int reneged_all,int will_exit)1361 sctp_cwnd_update_after_sack(struct sctp_tcb *stcb,
1362 struct sctp_association *asoc,
1363 int accum_moved, int reneged_all, int will_exit)
1364 {
1365 /* Passing a zero argument in last disables the rtcc algoritm */
1366 sctp_cwnd_update_after_sack_common(stcb, asoc, accum_moved, reneged_all, will_exit, 0);
1367 }
1368
1369 static void
sctp_cwnd_update_after_ecn_echo(struct sctp_tcb * stcb,struct sctp_nets * net,int in_window,int num_pkt_lost)1370 sctp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net,
1371 int in_window, int num_pkt_lost)
1372 {
1373 /* Passing a zero argument in last disables the rtcc algoritm */
1374 sctp_cwnd_update_after_ecn_echo_common(stcb, net, in_window, num_pkt_lost, 0);
1375 }
1376
1377 /* Here starts the RTCCVAR type CC invented by RRS which
1378 * is a slight mod to RFC2581. We reuse a common routine or
1379 * two since these algoritms are so close and need to
1380 * remain the same.
1381 */
1382 static void
sctp_cwnd_update_rtcc_after_ecn_echo(struct sctp_tcb * stcb,struct sctp_nets * net,int in_window,int num_pkt_lost)1383 sctp_cwnd_update_rtcc_after_ecn_echo(struct sctp_tcb *stcb, struct sctp_nets *net,
1384 int in_window, int num_pkt_lost)
1385 {
1386 sctp_cwnd_update_after_ecn_echo_common(stcb, net, in_window, num_pkt_lost, 1);
1387 }
1388
1389
1390 static
sctp_cwnd_update_rtcc_tsn_acknowledged(struct sctp_nets * net,struct sctp_tmit_chunk * tp1)1391 void sctp_cwnd_update_rtcc_tsn_acknowledged(struct sctp_nets *net,
1392 struct sctp_tmit_chunk *tp1)
1393 {
1394 net->cc_mod.rtcc.bw_bytes += tp1->send_size;
1395 }
1396
1397 static void
sctp_cwnd_prepare_rtcc_net_for_sack(struct sctp_tcb * stcb SCTP_UNUSED,struct sctp_nets * net)1398 sctp_cwnd_prepare_rtcc_net_for_sack(struct sctp_tcb *stcb SCTP_UNUSED,
1399 struct sctp_nets *net)
1400 {
1401 if (net->cc_mod.rtcc.tls_needs_set > 0) {
1402 /* We had a bw measurment going on */
1403 struct timeval ltls;
1404 SCTP_GETPTIME_TIMEVAL(<ls);
1405 timevalsub(<ls, &net->cc_mod.rtcc.tls);
1406 net->cc_mod.rtcc.new_tot_time = (ltls.tv_sec * 1000000) + ltls.tv_usec;
1407 }
1408 }
1409
1410 static void
sctp_cwnd_new_rtcc_transmission_begins(struct sctp_tcb * stcb,struct sctp_nets * net)1411 sctp_cwnd_new_rtcc_transmission_begins(struct sctp_tcb *stcb,
1412 struct sctp_nets *net)
1413 {
1414 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
1415 uint64_t vtag, probepoint;
1416
1417 #endif
1418 if (net->cc_mod.rtcc.lbw) {
1419 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
1420 /* Clear the old bw.. we went to 0 in-flight */
1421 vtag = (net->rtt << 32) | (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) |
1422 (stcb->rport);
1423 probepoint = (((uint64_t)net->cwnd) << 32);
1424 /* Probe point 8 */
1425 probepoint |= ((8 << 16) | 0);
1426 SDT_PROBE(sctp, cwnd, net, rttvar,
1427 vtag,
1428 ((net->cc_mod.rtcc.lbw << 32) | 0),
1429 ((net->cc_mod.rtcc.lbw_rtt << 32) | net->rtt),
1430 net->flight_size,
1431 probepoint);
1432 #endif
1433 net->cc_mod.rtcc.lbw_rtt = 0;
1434 net->cc_mod.rtcc.cwnd_at_bw_set = 0;
1435 net->cc_mod.rtcc.lbw = 0;
1436 net->cc_mod.rtcc.bw_bytes_at_last_rttc = 0;
1437 net->cc_mod.rtcc.vol_reduce = 0;
1438 net->cc_mod.rtcc.bw_tot_time = 0;
1439 net->cc_mod.rtcc.bw_bytes = 0;
1440 net->cc_mod.rtcc.tls_needs_set = 0;
1441 if (net->cc_mod.rtcc.steady_step) {
1442 net->cc_mod.rtcc.vol_reduce = 0;
1443 net->cc_mod.rtcc.step_cnt = 0;
1444 net->cc_mod.rtcc.last_step_state = 0;
1445 }
1446 if (net->cc_mod.rtcc.ret_from_eq) {
1447 /* less aggressive one - reset cwnd too */
1448 uint32_t cwnd_in_mtu, cwnd;
1449
1450 cwnd_in_mtu = SCTP_BASE_SYSCTL(sctp_initial_cwnd);
1451 if (cwnd_in_mtu == 0) {
1452 /* Using 0 means that the value of RFC 4960 is used. */
1453 cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND));
1454 } else {
1455 /*
1456 * We take the minimum of the burst limit and the
1457 * initial congestion window.
1458 */
1459 if ((stcb->asoc.max_burst > 0) && (cwnd_in_mtu > stcb->asoc.max_burst))
1460 cwnd_in_mtu = stcb->asoc.max_burst;
1461 cwnd = (net->mtu - sizeof(struct sctphdr)) * cwnd_in_mtu;
1462 }
1463 if (net->cwnd > cwnd) {
1464 /* Only set if we are not a timeout (i.e. down to 1 mtu) */
1465 net->cwnd = cwnd;
1466 }
1467 }
1468 }
1469 }
1470
1471 static void
sctp_set_rtcc_initial_cc_param(struct sctp_tcb * stcb,struct sctp_nets * net)1472 sctp_set_rtcc_initial_cc_param(struct sctp_tcb *stcb,
1473 struct sctp_nets *net)
1474 {
1475 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
1476 uint64_t vtag, probepoint;
1477
1478 #endif
1479 sctp_set_initial_cc_param(stcb, net);
1480 stcb->asoc.use_precise_time = 1;
1481 #if defined(__FreeBSD__) && __FreeBSD_version >= 803000
1482 probepoint = (((uint64_t)net->cwnd) << 32);
1483 probepoint |= ((9 << 16) | 0);
1484 vtag = (net->rtt << 32) |
1485 (((uint32_t)(stcb->sctp_ep->sctp_lport)) << 16) |
1486 (stcb->rport);
1487 SDT_PROBE(sctp, cwnd, net, rttvar,
1488 vtag,
1489 0,
1490 0,
1491 0,
1492 probepoint);
1493 #endif
1494 net->cc_mod.rtcc.lbw_rtt = 0;
1495 net->cc_mod.rtcc.cwnd_at_bw_set = 0;
1496 net->cc_mod.rtcc.vol_reduce = 0;
1497 net->cc_mod.rtcc.lbw = 0;
1498 net->cc_mod.rtcc.vol_reduce = 0;
1499 net->cc_mod.rtcc.bw_bytes_at_last_rttc = 0;
1500 net->cc_mod.rtcc.bw_tot_time = 0;
1501 net->cc_mod.rtcc.bw_bytes = 0;
1502 net->cc_mod.rtcc.tls_needs_set = 0;
1503 net->cc_mod.rtcc.ret_from_eq = SCTP_BASE_SYSCTL(sctp_rttvar_eqret);
1504 net->cc_mod.rtcc.steady_step = SCTP_BASE_SYSCTL(sctp_steady_step);
1505 net->cc_mod.rtcc.use_dccc_ecn = SCTP_BASE_SYSCTL(sctp_use_dccc_ecn);
1506 net->cc_mod.rtcc.step_cnt = 0;
1507 net->cc_mod.rtcc.last_step_state = 0;
1508
1509
1510 }
1511
1512 static int
sctp_cwnd_rtcc_socket_option(struct sctp_tcb * stcb,int setorget,struct sctp_cc_option * cc_opt)1513 sctp_cwnd_rtcc_socket_option(struct sctp_tcb *stcb, int setorget,
1514 struct sctp_cc_option *cc_opt)
1515 {
1516 struct sctp_nets *net;
1517 if (setorget == 1) {
1518 /* a set */
1519 if (cc_opt->option == SCTP_CC_OPT_RTCC_SETMODE) {
1520 if ((cc_opt->aid_value.assoc_value != 0) &&
1521 (cc_opt->aid_value.assoc_value != 1)) {
1522 return (EINVAL);
1523 }
1524 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) {
1525 net->cc_mod.rtcc.ret_from_eq = cc_opt->aid_value.assoc_value;
1526 }
1527 } else if (cc_opt->option == SCTP_CC_OPT_USE_DCCC_ECN) {
1528 if ((cc_opt->aid_value.assoc_value != 0) &&
1529 (cc_opt->aid_value.assoc_value != 1)) {
1530 return (EINVAL);
1531 }
1532 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) {
1533 net->cc_mod.rtcc.use_dccc_ecn = cc_opt->aid_value.assoc_value;
1534 }
1535 } else if (cc_opt->option == SCTP_CC_OPT_STEADY_STEP) {
1536 TAILQ_FOREACH(net, &stcb->asoc.nets, sctp_next) {
1537 net->cc_mod.rtcc.steady_step = cc_opt->aid_value.assoc_value;
1538 }
1539 } else {
1540 return (EINVAL);
1541 }
1542 } else {
1543 /* a get */
1544 if (cc_opt->option == SCTP_CC_OPT_RTCC_SETMODE) {
1545 net = TAILQ_FIRST(&stcb->asoc.nets);
1546 if (net == NULL) {
1547 return (EFAULT);
1548 }
1549 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.ret_from_eq;
1550 } else if (cc_opt->option == SCTP_CC_OPT_USE_DCCC_ECN) {
1551 net = TAILQ_FIRST(&stcb->asoc.nets);
1552 if (net == NULL) {
1553 return (EFAULT);
1554 }
1555 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.use_dccc_ecn;
1556 } else if (cc_opt->option == SCTP_CC_OPT_STEADY_STEP) {
1557 net = TAILQ_FIRST(&stcb->asoc.nets);
1558 if (net == NULL) {
1559 return (EFAULT);
1560 }
1561 cc_opt->aid_value.assoc_value = net->cc_mod.rtcc.steady_step;
1562 } else {
1563 return (EINVAL);
1564 }
1565 }
1566 return (0);
1567 }
1568
1569 static void
sctp_cwnd_update_rtcc_packet_transmitted(struct sctp_tcb * stcb SCTP_UNUSED,struct sctp_nets * net)1570 sctp_cwnd_update_rtcc_packet_transmitted(struct sctp_tcb *stcb SCTP_UNUSED,
1571 struct sctp_nets *net)
1572 {
1573 if (net->cc_mod.rtcc.tls_needs_set == 0) {
1574 SCTP_GETPTIME_TIMEVAL(&net->cc_mod.rtcc.tls);
1575 net->cc_mod.rtcc.tls_needs_set = 2;
1576 }
1577 }
1578
1579 static void
sctp_cwnd_update_rtcc_after_sack(struct sctp_tcb * stcb,struct sctp_association * asoc,int accum_moved,int reneged_all,int will_exit)1580 sctp_cwnd_update_rtcc_after_sack(struct sctp_tcb *stcb,
1581 struct sctp_association *asoc,
1582 int accum_moved, int reneged_all, int will_exit)
1583 {
1584 /* Passing a one argument at the last enables the rtcc algoritm */
1585 sctp_cwnd_update_after_sack_common(stcb, asoc, accum_moved, reneged_all, will_exit, 1);
1586 }
1587
1588 static void
sctp_rtt_rtcc_calculated(struct sctp_tcb * stcb SCTP_UNUSED,struct sctp_nets * net,struct timeval * now SCTP_UNUSED)1589 sctp_rtt_rtcc_calculated(struct sctp_tcb *stcb SCTP_UNUSED,
1590 struct sctp_nets *net,
1591 struct timeval *now SCTP_UNUSED)
1592 {
1593 net->cc_mod.rtcc.rtt_set_this_sack = 1;
1594 }
1595
1596 /* Here starts Sally Floyds HS-TCP */
1597
1598 struct sctp_hs_raise_drop {
1599 int32_t cwnd;
1600 int32_t increase;
1601 int32_t drop_percent;
1602 };
1603
1604 #define SCTP_HS_TABLE_SIZE 73
1605
1606 struct sctp_hs_raise_drop sctp_cwnd_adjust[SCTP_HS_TABLE_SIZE] = {
1607 {38, 1, 50}, /* 0 */
1608 {118, 2, 44}, /* 1 */
1609 {221, 3, 41}, /* 2 */
1610 {347, 4, 38}, /* 3 */
1611 {495, 5, 37}, /* 4 */
1612 {663, 6, 35}, /* 5 */
1613 {851, 7, 34}, /* 6 */
1614 {1058, 8, 33}, /* 7 */
1615 {1284, 9, 32}, /* 8 */
1616 {1529, 10, 31}, /* 9 */
1617 {1793, 11, 30}, /* 10 */
1618 {2076, 12, 29}, /* 11 */
1619 {2378, 13, 28}, /* 12 */
1620 {2699, 14, 28}, /* 13 */
1621 {3039, 15, 27}, /* 14 */
1622 {3399, 16, 27}, /* 15 */
1623 {3778, 17, 26}, /* 16 */
1624 {4177, 18, 26}, /* 17 */
1625 {4596, 19, 25}, /* 18 */
1626 {5036, 20, 25}, /* 19 */
1627 {5497, 21, 24}, /* 20 */
1628 {5979, 22, 24}, /* 21 */
1629 {6483, 23, 23}, /* 22 */
1630 {7009, 24, 23}, /* 23 */
1631 {7558, 25, 22}, /* 24 */
1632 {8130, 26, 22}, /* 25 */
1633 {8726, 27, 22}, /* 26 */
1634 {9346, 28, 21}, /* 27 */
1635 {9991, 29, 21}, /* 28 */
1636 {10661, 30, 21}, /* 29 */
1637 {11358, 31, 20}, /* 30 */
1638 {12082, 32, 20}, /* 31 */
1639 {12834, 33, 20}, /* 32 */
1640 {13614, 34, 19}, /* 33 */
1641 {14424, 35, 19}, /* 34 */
1642 {15265, 36, 19}, /* 35 */
1643 {16137, 37, 19}, /* 36 */
1644 {17042, 38, 18}, /* 37 */
1645 {17981, 39, 18}, /* 38 */
1646 {18955, 40, 18}, /* 39 */
1647 {19965, 41, 17}, /* 40 */
1648 {21013, 42, 17}, /* 41 */
1649 {22101, 43, 17}, /* 42 */
1650 {23230, 44, 17}, /* 43 */
1651 {24402, 45, 16}, /* 44 */
1652 {25618, 46, 16}, /* 45 */
1653 {26881, 47, 16}, /* 46 */
1654 {28193, 48, 16}, /* 47 */
1655 {29557, 49, 15}, /* 48 */
1656 {30975, 50, 15}, /* 49 */
1657 {32450, 51, 15}, /* 50 */
1658 {33986, 52, 15}, /* 51 */
1659 {35586, 53, 14}, /* 52 */
1660 {37253, 54, 14}, /* 53 */
1661 {38992, 55, 14}, /* 54 */
1662 {40808, 56, 14}, /* 55 */
1663 {42707, 57, 13}, /* 56 */
1664 {44694, 58, 13}, /* 57 */
1665 {46776, 59, 13}, /* 58 */
1666 {48961, 60, 13}, /* 59 */
1667 {51258, 61, 13}, /* 60 */
1668 {53677, 62, 12}, /* 61 */
1669 {56230, 63, 12}, /* 62 */
1670 {58932, 64, 12}, /* 63 */
1671 {61799, 65, 12}, /* 64 */
1672 {64851, 66, 11}, /* 65 */
1673 {68113, 67, 11}, /* 66 */
1674 {71617, 68, 11}, /* 67 */
1675 {75401, 69, 10}, /* 68 */
1676 {79517, 70, 10}, /* 69 */
1677 {84035, 71, 10}, /* 70 */
1678 {89053, 72, 10}, /* 71 */
1679 {94717, 73, 9} /* 72 */
1680 };
1681
1682 static void
sctp_hs_cwnd_increase(struct sctp_tcb * stcb,struct sctp_nets * net)1683 sctp_hs_cwnd_increase(struct sctp_tcb *stcb, struct sctp_nets *net)
1684 {
1685 int cur_val, i, indx, incr;
1686
1687 cur_val = net->cwnd >> 10;
1688 indx = SCTP_HS_TABLE_SIZE - 1;
1689
1690 if (cur_val < sctp_cwnd_adjust[0].cwnd) {
1691 /* normal mode */
1692 if (net->net_ack > net->mtu) {
1693 net->cwnd += net->mtu;
1694 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
1695 sctp_log_cwnd(stcb, net, net->mtu, SCTP_CWND_LOG_FROM_SS);
1696 }
1697 } else {
1698 net->cwnd += net->net_ack;
1699 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
1700 sctp_log_cwnd(stcb, net, net->net_ack, SCTP_CWND_LOG_FROM_SS);
1701 }
1702 }
1703 } else {
1704 for (i = net->last_hs_used; i < SCTP_HS_TABLE_SIZE; i++) {
1705 if (cur_val < sctp_cwnd_adjust[i].cwnd) {
1706 indx = i;
1707 break;
1708 }
1709 }
1710 net->last_hs_used = indx;
1711 incr = ((sctp_cwnd_adjust[indx].increase) << 10);
1712 net->cwnd += incr;
1713 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
1714 sctp_log_cwnd(stcb, net, incr, SCTP_CWND_LOG_FROM_SS);
1715 }
1716 }
1717 }
1718
1719 static void
sctp_hs_cwnd_decrease(struct sctp_tcb * stcb,struct sctp_nets * net)1720 sctp_hs_cwnd_decrease(struct sctp_tcb *stcb, struct sctp_nets *net)
1721 {
1722 int cur_val, i, indx;
1723 int old_cwnd = net->cwnd;
1724
1725 cur_val = net->cwnd >> 10;
1726 if (cur_val < sctp_cwnd_adjust[0].cwnd) {
1727 /* normal mode */
1728 net->ssthresh = net->cwnd / 2;
1729 if (net->ssthresh < (net->mtu * 2)) {
1730 net->ssthresh = 2 * net->mtu;
1731 }
1732 net->cwnd = net->ssthresh;
1733 } else {
1734 /* drop by the proper amount */
1735 net->ssthresh = net->cwnd - (int)((net->cwnd / 100) *
1736 sctp_cwnd_adjust[net->last_hs_used].drop_percent);
1737 net->cwnd = net->ssthresh;
1738 /* now where are we */
1739 indx = net->last_hs_used;
1740 cur_val = net->cwnd >> 10;
1741 /* reset where we are in the table */
1742 if (cur_val < sctp_cwnd_adjust[0].cwnd) {
1743 /* feel out of hs */
1744 net->last_hs_used = 0;
1745 } else {
1746 for (i = indx; i >= 1; i--) {
1747 if (cur_val > sctp_cwnd_adjust[i - 1].cwnd) {
1748 break;
1749 }
1750 }
1751 net->last_hs_used = indx;
1752 }
1753 }
1754 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
1755 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_FR);
1756 }
1757 }
1758
1759 static void
sctp_hs_cwnd_update_after_fr(struct sctp_tcb * stcb,struct sctp_association * asoc)1760 sctp_hs_cwnd_update_after_fr(struct sctp_tcb *stcb,
1761 struct sctp_association *asoc)
1762 {
1763 struct sctp_nets *net;
1764 /*
1765 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) &&
1766 * (net->fast_retran_loss_recovery == 0)))
1767 */
1768 TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
1769 if ((asoc->fast_retran_loss_recovery == 0) ||
1770 (asoc->sctp_cmt_on_off > 0)) {
1771 /* out of a RFC2582 Fast recovery window? */
1772 if (net->net_ack > 0) {
1773 /*
1774 * per section 7.2.3, are there any
1775 * destinations that had a fast retransmit
1776 * to them. If so what we need to do is
1777 * adjust ssthresh and cwnd.
1778 */
1779 struct sctp_tmit_chunk *lchk;
1780
1781 sctp_hs_cwnd_decrease(stcb, net);
1782
1783 lchk = TAILQ_FIRST(&asoc->send_queue);
1784
1785 net->partial_bytes_acked = 0;
1786 /* Turn on fast recovery window */
1787 asoc->fast_retran_loss_recovery = 1;
1788 if (lchk == NULL) {
1789 /* Mark end of the window */
1790 asoc->fast_recovery_tsn = asoc->sending_seq - 1;
1791 } else {
1792 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
1793 }
1794
1795 /*
1796 * CMT fast recovery -- per destination
1797 * recovery variable.
1798 */
1799 net->fast_retran_loss_recovery = 1;
1800
1801 if (lchk == NULL) {
1802 /* Mark end of the window */
1803 net->fast_recovery_tsn = asoc->sending_seq - 1;
1804 } else {
1805 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
1806 }
1807
1808 sctp_timer_stop(SCTP_TIMER_TYPE_SEND,
1809 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA+SCTP_LOC_32);
1810 sctp_timer_start(SCTP_TIMER_TYPE_SEND,
1811 stcb->sctp_ep, stcb, net);
1812 }
1813 } else if (net->net_ack > 0) {
1814 /*
1815 * Mark a peg that we WOULD have done a cwnd
1816 * reduction but RFC2582 prevented this action.
1817 */
1818 SCTP_STAT_INCR(sctps_fastretransinrtt);
1819 }
1820 }
1821 }
1822
1823 static void
sctp_hs_cwnd_update_after_sack(struct sctp_tcb * stcb,struct sctp_association * asoc,int accum_moved,int reneged_all SCTP_UNUSED,int will_exit)1824 sctp_hs_cwnd_update_after_sack(struct sctp_tcb *stcb,
1825 struct sctp_association *asoc,
1826 int accum_moved, int reneged_all SCTP_UNUSED, int will_exit)
1827 {
1828 struct sctp_nets *net;
1829 /******************************/
1830 /* update cwnd and Early FR */
1831 /******************************/
1832 TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
1833
1834 #ifdef JANA_CMT_FAST_RECOVERY
1835 /*
1836 * CMT fast recovery code. Need to debug.
1837 */
1838 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) {
1839 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) ||
1840 SCTP_TSN_GE(net->pseudo_cumack,net->fast_recovery_tsn)) {
1841 net->will_exit_fast_recovery = 1;
1842 }
1843 }
1844 #endif
1845 /* if nothing was acked on this destination skip it */
1846 if (net->net_ack == 0) {
1847 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
1848 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK);
1849 }
1850 continue;
1851 }
1852 #ifdef JANA_CMT_FAST_RECOVERY
1853 /* CMT fast recovery code
1854 */
1855 /*
1856 if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery && net->will_exit_fast_recovery == 0) {
1857 @@@ Do something
1858 }
1859 else if (sctp_cmt_on_off == 0 && asoc->fast_retran_loss_recovery && will_exit == 0) {
1860 */
1861 #endif
1862
1863 if (asoc->fast_retran_loss_recovery &&
1864 (will_exit == 0) &&
1865 (asoc->sctp_cmt_on_off == 0)) {
1866 /*
1867 * If we are in loss recovery we skip any cwnd
1868 * update
1869 */
1870 return;
1871 }
1872 /*
1873 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has
1874 * moved.
1875 */
1876 if (accum_moved ||
1877 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) {
1878 /* If the cumulative ack moved we can proceed */
1879 if (net->cwnd <= net->ssthresh) {
1880 /* We are in slow start */
1881 if (net->flight_size + net->net_ack >= net->cwnd) {
1882
1883 sctp_hs_cwnd_increase(stcb, net);
1884
1885 } else {
1886 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
1887 sctp_log_cwnd(stcb, net, net->net_ack,
1888 SCTP_CWND_LOG_NOADV_SS);
1889 }
1890 }
1891 } else {
1892 /* We are in congestion avoidance */
1893 net->partial_bytes_acked += net->net_ack;
1894 if ((net->flight_size + net->net_ack >= net->cwnd) &&
1895 (net->partial_bytes_acked >= net->cwnd)) {
1896 net->partial_bytes_acked -= net->cwnd;
1897 net->cwnd += net->mtu;
1898 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
1899 sctp_log_cwnd(stcb, net, net->mtu,
1900 SCTP_CWND_LOG_FROM_CA);
1901 }
1902 } else {
1903 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
1904 sctp_log_cwnd(stcb, net, net->net_ack,
1905 SCTP_CWND_LOG_NOADV_CA);
1906 }
1907 }
1908 }
1909 } else {
1910 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
1911 sctp_log_cwnd(stcb, net, net->mtu,
1912 SCTP_CWND_LOG_NO_CUMACK);
1913 }
1914 }
1915 }
1916 }
1917
1918
1919 /*
1920 * H-TCP congestion control. The algorithm is detailed in:
1921 * R.N.Shorten, D.J.Leith:
1922 * "H-TCP: TCP for high-speed and long-distance networks"
1923 * Proc. PFLDnet, Argonne, 2004.
1924 * http://www.hamilton.ie/net/htcp3.pdf
1925 */
1926
1927
1928 static int use_rtt_scaling = 1;
1929 static int use_bandwidth_switch = 1;
1930
1931 static inline int
between(uint32_t seq1,uint32_t seq2,uint32_t seq3)1932 between(uint32_t seq1, uint32_t seq2, uint32_t seq3)
1933 {
1934 return (seq3 - seq2 >= seq1 - seq2);
1935 }
1936
1937 static inline uint32_t
htcp_cong_time(struct htcp * ca)1938 htcp_cong_time(struct htcp *ca)
1939 {
1940 return (sctp_get_tick_count() - ca->last_cong);
1941 }
1942
1943 static inline uint32_t
htcp_ccount(struct htcp * ca)1944 htcp_ccount(struct htcp *ca)
1945 {
1946 return (htcp_cong_time(ca)/ca->minRTT);
1947 }
1948
1949 static inline void
htcp_reset(struct htcp * ca)1950 htcp_reset(struct htcp *ca)
1951 {
1952 ca->undo_last_cong = ca->last_cong;
1953 ca->undo_maxRTT = ca->maxRTT;
1954 ca->undo_old_maxB = ca->old_maxB;
1955 ca->last_cong = sctp_get_tick_count();
1956 }
1957
1958 #ifdef SCTP_NOT_USED
1959
1960 static uint32_t
htcp_cwnd_undo(struct sctp_tcb * stcb,struct sctp_nets * net)1961 htcp_cwnd_undo(struct sctp_tcb *stcb, struct sctp_nets *net)
1962 {
1963 net->cc_mod.htcp_ca.last_cong = net->cc_mod.htcp_ca.undo_last_cong;
1964 net->cc_mod.htcp_ca.maxRTT = net->cc_mod.htcp_ca.undo_maxRTT;
1965 net->cc_mod.htcp_ca.old_maxB = net->cc_mod.htcp_ca.undo_old_maxB;
1966 return (max(net->cwnd, ((net->ssthresh/net->mtu<<7)/net->cc_mod.htcp_ca.beta)*net->mtu));
1967 }
1968
1969 #endif
1970
1971 static inline void
measure_rtt(struct sctp_nets * net)1972 measure_rtt(struct sctp_nets *net)
1973 {
1974 uint32_t srtt = net->lastsa>>SCTP_RTT_SHIFT;
1975
1976 /* keep track of minimum RTT seen so far, minRTT is zero at first */
1977 if (net->cc_mod.htcp_ca.minRTT > srtt || !net->cc_mod.htcp_ca.minRTT)
1978 net->cc_mod.htcp_ca.minRTT = srtt;
1979
1980 /* max RTT */
1981 if (net->fast_retran_ip == 0 && net->ssthresh < 0xFFFF && htcp_ccount(&net->cc_mod.htcp_ca) > 3) {
1982 if (net->cc_mod.htcp_ca.maxRTT < net->cc_mod.htcp_ca.minRTT)
1983 net->cc_mod.htcp_ca.maxRTT = net->cc_mod.htcp_ca.minRTT;
1984 if (net->cc_mod.htcp_ca.maxRTT < srtt && srtt <= net->cc_mod.htcp_ca.maxRTT+MSEC_TO_TICKS(20))
1985 net->cc_mod.htcp_ca.maxRTT = srtt;
1986 }
1987 }
1988
1989 static void
measure_achieved_throughput(struct sctp_nets * net)1990 measure_achieved_throughput(struct sctp_nets *net)
1991 {
1992 uint32_t now = sctp_get_tick_count();
1993
1994 if (net->fast_retran_ip == 0)
1995 net->cc_mod.htcp_ca.bytes_acked = net->net_ack;
1996
1997 if (!use_bandwidth_switch)
1998 return;
1999
2000 /* achieved throughput calculations */
2001 /* JRS - not 100% sure of this statement */
2002 if (net->fast_retran_ip == 1) {
2003 net->cc_mod.htcp_ca.bytecount = 0;
2004 net->cc_mod.htcp_ca.lasttime = now;
2005 return;
2006 }
2007
2008 net->cc_mod.htcp_ca.bytecount += net->net_ack;
2009 if ((net->cc_mod.htcp_ca.bytecount >= net->cwnd - (((net->cc_mod.htcp_ca.alpha >> 7) ? (net->cc_mod.htcp_ca.alpha >> 7) : 1) * net->mtu)) &&
2010 (now - net->cc_mod.htcp_ca.lasttime >= net->cc_mod.htcp_ca.minRTT) &&
2011 (net->cc_mod.htcp_ca.minRTT > 0)) {
2012 uint32_t cur_Bi = net->cc_mod.htcp_ca.bytecount/net->mtu*hz/(now - net->cc_mod.htcp_ca.lasttime);
2013
2014 if (htcp_ccount(&net->cc_mod.htcp_ca) <= 3) {
2015 /* just after backoff */
2016 net->cc_mod.htcp_ca.minB = net->cc_mod.htcp_ca.maxB = net->cc_mod.htcp_ca.Bi = cur_Bi;
2017 } else {
2018 net->cc_mod.htcp_ca.Bi = (3*net->cc_mod.htcp_ca.Bi + cur_Bi)/4;
2019 if (net->cc_mod.htcp_ca.Bi > net->cc_mod.htcp_ca.maxB)
2020 net->cc_mod.htcp_ca.maxB = net->cc_mod.htcp_ca.Bi;
2021 if (net->cc_mod.htcp_ca.minB > net->cc_mod.htcp_ca.maxB)
2022 net->cc_mod.htcp_ca.minB = net->cc_mod.htcp_ca.maxB;
2023 }
2024 net->cc_mod.htcp_ca.bytecount = 0;
2025 net->cc_mod.htcp_ca.lasttime = now;
2026 }
2027 }
2028
2029 static inline void
htcp_beta_update(struct htcp * ca,uint32_t minRTT,uint32_t maxRTT)2030 htcp_beta_update(struct htcp *ca, uint32_t minRTT, uint32_t maxRTT)
2031 {
2032 if (use_bandwidth_switch) {
2033 uint32_t maxB = ca->maxB;
2034 uint32_t old_maxB = ca->old_maxB;
2035 ca->old_maxB = ca->maxB;
2036
2037 if (!between(5*maxB, 4*old_maxB, 6*old_maxB)) {
2038 ca->beta = BETA_MIN;
2039 ca->modeswitch = 0;
2040 return;
2041 }
2042 }
2043
2044 if (ca->modeswitch && minRTT > (uint32_t)MSEC_TO_TICKS(10) && maxRTT) {
2045 ca->beta = (minRTT<<7)/maxRTT;
2046 if (ca->beta < BETA_MIN)
2047 ca->beta = BETA_MIN;
2048 else if (ca->beta > BETA_MAX)
2049 ca->beta = BETA_MAX;
2050 } else {
2051 ca->beta = BETA_MIN;
2052 ca->modeswitch = 1;
2053 }
2054 }
2055
2056 static inline void
htcp_alpha_update(struct htcp * ca)2057 htcp_alpha_update(struct htcp *ca)
2058 {
2059 uint32_t minRTT = ca->minRTT;
2060 uint32_t factor = 1;
2061 uint32_t diff = htcp_cong_time(ca);
2062
2063 if (diff > (uint32_t)hz) {
2064 diff -= hz;
2065 factor = 1+ ( 10*diff + ((diff/2)*(diff/2)/hz))/hz;
2066 }
2067
2068 if (use_rtt_scaling && minRTT) {
2069 uint32_t scale = (hz<<3)/(10*minRTT);
2070 scale = min(max(scale, 1U<<2), 10U<<3); /* clamping ratio to interval [0.5,10]<<3 */
2071 factor = (factor<<3)/scale;
2072 if (!factor)
2073 factor = 1;
2074 }
2075
2076 ca->alpha = 2*factor*((1<<7)-ca->beta);
2077 if (!ca->alpha)
2078 ca->alpha = ALPHA_BASE;
2079 }
2080
2081 /* After we have the rtt data to calculate beta, we'd still prefer to wait one
2082 * rtt before we adjust our beta to ensure we are working from a consistent
2083 * data.
2084 *
2085 * This function should be called when we hit a congestion event since only at
2086 * that point do we really have a real sense of maxRTT (the queues en route
2087 * were getting just too full now).
2088 */
2089 static void
htcp_param_update(struct sctp_nets * net)2090 htcp_param_update(struct sctp_nets *net)
2091 {
2092 uint32_t minRTT = net->cc_mod.htcp_ca.minRTT;
2093 uint32_t maxRTT = net->cc_mod.htcp_ca.maxRTT;
2094
2095 htcp_beta_update(&net->cc_mod.htcp_ca, minRTT, maxRTT);
2096 htcp_alpha_update(&net->cc_mod.htcp_ca);
2097
2098 /* add slowly fading memory for maxRTT to accommodate routing changes etc */
2099 if (minRTT > 0 && maxRTT > minRTT)
2100 net->cc_mod.htcp_ca.maxRTT = minRTT + ((maxRTT-minRTT)*95)/100;
2101 }
2102
2103 static uint32_t
htcp_recalc_ssthresh(struct sctp_nets * net)2104 htcp_recalc_ssthresh(struct sctp_nets *net)
2105 {
2106 htcp_param_update(net);
2107 return (max(((net->cwnd/net->mtu * net->cc_mod.htcp_ca.beta) >> 7)*net->mtu, 2U*net->mtu));
2108 }
2109
2110 static void
htcp_cong_avoid(struct sctp_tcb * stcb,struct sctp_nets * net)2111 htcp_cong_avoid(struct sctp_tcb *stcb, struct sctp_nets *net)
2112 {
2113 /*-
2114 * How to handle these functions?
2115 * if (!tcp_is_cwnd_limited(sk, in_flight)) RRS - good question.
2116 * return;
2117 */
2118 if (net->cwnd <= net->ssthresh) {
2119 /* We are in slow start */
2120 if (net->flight_size + net->net_ack >= net->cwnd) {
2121 if (net->net_ack > (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable))) {
2122 net->cwnd += (net->mtu * SCTP_BASE_SYSCTL(sctp_L2_abc_variable));
2123 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
2124 sctp_log_cwnd(stcb, net, net->mtu,
2125 SCTP_CWND_LOG_FROM_SS);
2126 }
2127
2128 } else {
2129 net->cwnd += net->net_ack;
2130 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
2131 sctp_log_cwnd(stcb, net, net->net_ack,
2132 SCTP_CWND_LOG_FROM_SS);
2133 }
2134
2135 }
2136 } else {
2137 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
2138 sctp_log_cwnd(stcb, net, net->net_ack,
2139 SCTP_CWND_LOG_NOADV_SS);
2140 }
2141 }
2142 } else {
2143 measure_rtt(net);
2144
2145 /* In dangerous area, increase slowly.
2146 * In theory this is net->cwnd += alpha / net->cwnd
2147 */
2148 /* What is snd_cwnd_cnt?? */
2149 if (((net->partial_bytes_acked/net->mtu * net->cc_mod.htcp_ca.alpha) >> 7)*net->mtu >= net->cwnd) {
2150 /*-
2151 * Does SCTP have a cwnd clamp?
2152 * if (net->snd_cwnd < net->snd_cwnd_clamp) - Nope (RRS).
2153 */
2154 net->cwnd += net->mtu;
2155 net->partial_bytes_acked = 0;
2156 htcp_alpha_update(&net->cc_mod.htcp_ca);
2157 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
2158 sctp_log_cwnd(stcb, net, net->mtu,
2159 SCTP_CWND_LOG_FROM_CA);
2160 }
2161 } else {
2162 net->partial_bytes_acked += net->net_ack;
2163 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
2164 sctp_log_cwnd(stcb, net, net->net_ack,
2165 SCTP_CWND_LOG_NOADV_CA);
2166 }
2167 }
2168
2169 net->cc_mod.htcp_ca.bytes_acked = net->mtu;
2170 }
2171 }
2172
2173 #ifdef SCTP_NOT_USED
2174 /* Lower bound on congestion window. */
2175 static uint32_t
htcp_min_cwnd(struct sctp_tcb * stcb,struct sctp_nets * net)2176 htcp_min_cwnd(struct sctp_tcb *stcb, struct sctp_nets *net)
2177 {
2178 return (net->ssthresh);
2179 }
2180 #endif
2181
2182 static void
htcp_init(struct sctp_nets * net)2183 htcp_init(struct sctp_nets *net)
2184 {
2185 memset(&net->cc_mod.htcp_ca, 0, sizeof(struct htcp));
2186 net->cc_mod.htcp_ca.alpha = ALPHA_BASE;
2187 net->cc_mod.htcp_ca.beta = BETA_MIN;
2188 net->cc_mod.htcp_ca.bytes_acked = net->mtu;
2189 net->cc_mod.htcp_ca.last_cong = sctp_get_tick_count();
2190 }
2191
2192 static void
sctp_htcp_set_initial_cc_param(struct sctp_tcb * stcb,struct sctp_nets * net)2193 sctp_htcp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net)
2194 {
2195 /*
2196 * We take the max of the burst limit times a MTU or the
2197 * INITIAL_CWND. We then limit this to 4 MTU's of sending.
2198 */
2199 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND));
2200 net->ssthresh = stcb->asoc.peers_rwnd;
2201 htcp_init(net);
2202
2203 if (SCTP_BASE_SYSCTL(sctp_logging_level) & (SCTP_CWND_MONITOR_ENABLE|SCTP_CWND_LOGGING_ENABLE)) {
2204 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION);
2205 }
2206 }
2207
2208 static void
sctp_htcp_cwnd_update_after_sack(struct sctp_tcb * stcb,struct sctp_association * asoc,int accum_moved,int reneged_all SCTP_UNUSED,int will_exit)2209 sctp_htcp_cwnd_update_after_sack(struct sctp_tcb *stcb,
2210 struct sctp_association *asoc,
2211 int accum_moved, int reneged_all SCTP_UNUSED, int will_exit)
2212 {
2213 struct sctp_nets *net;
2214
2215 /******************************/
2216 /* update cwnd and Early FR */
2217 /******************************/
2218 TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
2219
2220 #ifdef JANA_CMT_FAST_RECOVERY
2221 /*
2222 * CMT fast recovery code. Need to debug.
2223 */
2224 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) {
2225 if (SCTP_TSN_GE(asoc->last_acked_seq, net->fast_recovery_tsn) ||
2226 SCTP_TSN_GE(net->pseudo_cumack,net->fast_recovery_tsn)) {
2227 net->will_exit_fast_recovery = 1;
2228 }
2229 }
2230 #endif
2231 /* if nothing was acked on this destination skip it */
2232 if (net->net_ack == 0) {
2233 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
2234 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK);
2235 }
2236 continue;
2237 }
2238 #ifdef JANA_CMT_FAST_RECOVERY
2239 /* CMT fast recovery code
2240 */
2241 /*
2242 if (sctp_cmt_on_off > 0 && net->fast_retran_loss_recovery && net->will_exit_fast_recovery == 0) {
2243 @@@ Do something
2244 }
2245 else if (sctp_cmt_on_off == 0 && asoc->fast_retran_loss_recovery && will_exit == 0) {
2246 */
2247 #endif
2248
2249 if (asoc->fast_retran_loss_recovery &&
2250 will_exit == 0 &&
2251 (asoc->sctp_cmt_on_off == 0)) {
2252 /*
2253 * If we are in loss recovery we skip any cwnd
2254 * update
2255 */
2256 return;
2257 }
2258 /*
2259 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has
2260 * moved.
2261 */
2262 if (accum_moved ||
2263 ((asoc->sctp_cmt_on_off > 0) && net->new_pseudo_cumack)) {
2264 htcp_cong_avoid(stcb, net);
2265 measure_achieved_throughput(net);
2266 } else {
2267 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_LOGGING_ENABLE) {
2268 sctp_log_cwnd(stcb, net, net->mtu,
2269 SCTP_CWND_LOG_NO_CUMACK);
2270 }
2271 }
2272 }
2273 }
2274
2275 static void
sctp_htcp_cwnd_update_after_fr(struct sctp_tcb * stcb,struct sctp_association * asoc)2276 sctp_htcp_cwnd_update_after_fr(struct sctp_tcb *stcb,
2277 struct sctp_association *asoc)
2278 {
2279 struct sctp_nets *net;
2280 /*
2281 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off > 0) &&
2282 * (net->fast_retran_loss_recovery == 0)))
2283 */
2284 TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
2285 if ((asoc->fast_retran_loss_recovery == 0) ||
2286 (asoc->sctp_cmt_on_off > 0)) {
2287 /* out of a RFC2582 Fast recovery window? */
2288 if (net->net_ack > 0) {
2289 /*
2290 * per section 7.2.3, are there any
2291 * destinations that had a fast retransmit
2292 * to them. If so what we need to do is
2293 * adjust ssthresh and cwnd.
2294 */
2295 struct sctp_tmit_chunk *lchk;
2296 int old_cwnd = net->cwnd;
2297
2298 /* JRS - reset as if state were changed */
2299 htcp_reset(&net->cc_mod.htcp_ca);
2300 net->ssthresh = htcp_recalc_ssthresh(net);
2301 net->cwnd = net->ssthresh;
2302 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
2303 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd),
2304 SCTP_CWND_LOG_FROM_FR);
2305 }
2306 lchk = TAILQ_FIRST(&asoc->send_queue);
2307
2308 net->partial_bytes_acked = 0;
2309 /* Turn on fast recovery window */
2310 asoc->fast_retran_loss_recovery = 1;
2311 if (lchk == NULL) {
2312 /* Mark end of the window */
2313 asoc->fast_recovery_tsn = asoc->sending_seq - 1;
2314 } else {
2315 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
2316 }
2317
2318 /*
2319 * CMT fast recovery -- per destination
2320 * recovery variable.
2321 */
2322 net->fast_retran_loss_recovery = 1;
2323
2324 if (lchk == NULL) {
2325 /* Mark end of the window */
2326 net->fast_recovery_tsn = asoc->sending_seq - 1;
2327 } else {
2328 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
2329 }
2330
2331 sctp_timer_stop(SCTP_TIMER_TYPE_SEND,
2332 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA+SCTP_LOC_32);
2333 sctp_timer_start(SCTP_TIMER_TYPE_SEND,
2334 stcb->sctp_ep, stcb, net);
2335 }
2336 } else if (net->net_ack > 0) {
2337 /*
2338 * Mark a peg that we WOULD have done a cwnd
2339 * reduction but RFC2582 prevented this action.
2340 */
2341 SCTP_STAT_INCR(sctps_fastretransinrtt);
2342 }
2343 }
2344 }
2345
2346 static void
sctp_htcp_cwnd_update_after_timeout(struct sctp_tcb * stcb,struct sctp_nets * net)2347 sctp_htcp_cwnd_update_after_timeout(struct sctp_tcb *stcb,
2348 struct sctp_nets *net)
2349 {
2350 int old_cwnd = net->cwnd;
2351
2352 /* JRS - reset as if the state were being changed to timeout */
2353 htcp_reset(&net->cc_mod.htcp_ca);
2354 net->ssthresh = htcp_recalc_ssthresh(net);
2355 net->cwnd = net->mtu;
2356 net->partial_bytes_acked = 0;
2357 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
2358 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX);
2359 }
2360 }
2361
2362 static void
sctp_htcp_cwnd_update_after_ecn_echo(struct sctp_tcb * stcb,struct sctp_nets * net,int in_window,int num_pkt_lost SCTP_UNUSED)2363 sctp_htcp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb,
2364 struct sctp_nets *net, int in_window, int num_pkt_lost SCTP_UNUSED)
2365 {
2366 int old_cwnd;
2367 old_cwnd = net->cwnd;
2368
2369 /* JRS - reset hctp as if state changed */
2370 if (in_window == 0) {
2371 htcp_reset(&net->cc_mod.htcp_ca);
2372 SCTP_STAT_INCR(sctps_ecnereducedcwnd);
2373 net->ssthresh = htcp_recalc_ssthresh(net);
2374 if (net->ssthresh < net->mtu) {
2375 net->ssthresh = net->mtu;
2376 /* here back off the timer as well, to slow us down */
2377 net->RTO <<= 1;
2378 }
2379 net->cwnd = net->ssthresh;
2380 if (SCTP_BASE_SYSCTL(sctp_logging_level) & SCTP_CWND_MONITOR_ENABLE) {
2381 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT);
2382 }
2383 }
2384 }
2385
2386 struct sctp_cc_functions sctp_cc_functions[] = {
2387 {
2388 #if defined(__Windows__) || defined(__Userspace_os_Windows)
2389 sctp_set_initial_cc_param,
2390 sctp_cwnd_update_after_sack,
2391 sctp_cwnd_update_exit_pf_common,
2392 sctp_cwnd_update_after_fr,
2393 sctp_cwnd_update_after_timeout,
2394 sctp_cwnd_update_after_ecn_echo,
2395 sctp_cwnd_update_after_packet_dropped,
2396 sctp_cwnd_update_after_output,
2397 #else
2398 .sctp_set_initial_cc_param = sctp_set_initial_cc_param,
2399 .sctp_cwnd_update_after_sack = sctp_cwnd_update_after_sack,
2400 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common,
2401 .sctp_cwnd_update_after_fr = sctp_cwnd_update_after_fr,
2402 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout,
2403 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_after_ecn_echo,
2404 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped,
2405 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output,
2406 #endif
2407 },
2408 {
2409 #if defined(__Windows__) || defined(__Userspace_os_Windows)
2410 sctp_set_initial_cc_param,
2411 sctp_hs_cwnd_update_after_sack,
2412 sctp_cwnd_update_exit_pf_common,
2413 sctp_hs_cwnd_update_after_fr,
2414 sctp_cwnd_update_after_timeout,
2415 sctp_cwnd_update_after_ecn_echo,
2416 sctp_cwnd_update_after_packet_dropped,
2417 sctp_cwnd_update_after_output,
2418 #else
2419 .sctp_set_initial_cc_param = sctp_set_initial_cc_param,
2420 .sctp_cwnd_update_after_sack = sctp_hs_cwnd_update_after_sack,
2421 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common,
2422 .sctp_cwnd_update_after_fr = sctp_hs_cwnd_update_after_fr,
2423 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout,
2424 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_after_ecn_echo,
2425 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped,
2426 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output,
2427 #endif
2428 },
2429 {
2430 #if defined(__Windows__) || defined(__Userspace_os_Windows)
2431 sctp_htcp_set_initial_cc_param,
2432 sctp_htcp_cwnd_update_after_sack,
2433 sctp_cwnd_update_exit_pf_common,
2434 sctp_htcp_cwnd_update_after_fr,
2435 sctp_htcp_cwnd_update_after_timeout,
2436 sctp_htcp_cwnd_update_after_ecn_echo,
2437 sctp_cwnd_update_after_packet_dropped,
2438 sctp_cwnd_update_after_output,
2439 #else
2440 .sctp_set_initial_cc_param = sctp_htcp_set_initial_cc_param,
2441 .sctp_cwnd_update_after_sack = sctp_htcp_cwnd_update_after_sack,
2442 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common,
2443 .sctp_cwnd_update_after_fr = sctp_htcp_cwnd_update_after_fr,
2444 .sctp_cwnd_update_after_timeout = sctp_htcp_cwnd_update_after_timeout,
2445 .sctp_cwnd_update_after_ecn_echo = sctp_htcp_cwnd_update_after_ecn_echo,
2446 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped,
2447 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output,
2448 #endif
2449 },
2450 {
2451 #if defined(__Windows__) || defined(__Userspace_os_Windows)
2452 sctp_set_rtcc_initial_cc_param,
2453 sctp_cwnd_update_rtcc_after_sack,
2454 sctp_cwnd_update_exit_pf_common,
2455 sctp_cwnd_update_after_fr,
2456 sctp_cwnd_update_after_timeout,
2457 sctp_cwnd_update_rtcc_after_ecn_echo,
2458 sctp_cwnd_update_after_packet_dropped,
2459 sctp_cwnd_update_after_output,
2460 sctp_cwnd_update_rtcc_packet_transmitted,
2461 sctp_cwnd_update_rtcc_tsn_acknowledged,
2462 sctp_cwnd_new_rtcc_transmission_begins,
2463 sctp_cwnd_prepare_rtcc_net_for_sack,
2464 sctp_cwnd_rtcc_socket_option,
2465 sctp_rtt_rtcc_calculated
2466 #else
2467 .sctp_set_initial_cc_param = sctp_set_rtcc_initial_cc_param,
2468 .sctp_cwnd_update_after_sack = sctp_cwnd_update_rtcc_after_sack,
2469 .sctp_cwnd_update_exit_pf = sctp_cwnd_update_exit_pf_common,
2470 .sctp_cwnd_update_after_fr = sctp_cwnd_update_after_fr,
2471 .sctp_cwnd_update_after_timeout = sctp_cwnd_update_after_timeout,
2472 .sctp_cwnd_update_after_ecn_echo = sctp_cwnd_update_rtcc_after_ecn_echo,
2473 .sctp_cwnd_update_after_packet_dropped = sctp_cwnd_update_after_packet_dropped,
2474 .sctp_cwnd_update_after_output = sctp_cwnd_update_after_output,
2475 .sctp_cwnd_update_packet_transmitted = sctp_cwnd_update_rtcc_packet_transmitted,
2476 .sctp_cwnd_update_tsn_acknowledged = sctp_cwnd_update_rtcc_tsn_acknowledged,
2477 .sctp_cwnd_new_transmission_begins = sctp_cwnd_new_rtcc_transmission_begins,
2478 .sctp_cwnd_prepare_net_for_sack = sctp_cwnd_prepare_rtcc_net_for_sack,
2479 .sctp_cwnd_socket_option = sctp_cwnd_rtcc_socket_option,
2480 .sctp_rtt_calculated = sctp_rtt_rtcc_calculated
2481 #endif
2482 }
2483 };
2484