1 /*-
2 * Copyright (c) 2001-2007, by Cisco Systems, Inc. All rights reserved.
3 *
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions are met:
6 *
7 * a) Redistributions of source code must retain the above copyright notice,
8 * this list of conditions and the following disclaimer.
9 *
10 * b) Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in
12 * the documentation and/or other materials provided with the distribution.
13 *
14 * c) Neither the name of Cisco Systems, Inc. nor the names of its
15 * contributors may be used to endorse or promote products derived
16 * from this software without specific prior written permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
20 * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
22 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
28 * THE POSSIBILITY OF SUCH DAMAGE.
29 */
30
31 #include <netinet/sctp_os.h>
32 #include <netinet/sctp_var.h>
33 #include <netinet/sctp_sysctl.h>
34 #include <netinet/sctp_pcb.h>
35 #include <netinet/sctp_header.h>
36 #include <netinet/sctputil.h>
37 #include <netinet/sctp_output.h>
38 #include <netinet/sctp_input.h>
39 #include <netinet/sctp_indata.h>
40 #include <netinet/sctp_uio.h>
41 #include <netinet/sctp_timer.h>
42 #include <netinet/sctp_auth.h>
43 #include <netinet/sctp_asconf.h>
44 #include <netinet/sctp_cc_functions.h>
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 void
48 sctp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net)
49 {
50 /*
51 * We take the max of the burst limit times a MTU or the
52 * INITIAL_CWND. We then limit this to 4 MTU's of sending.
53 */
54 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND));
55 /* we always get at LEAST 2 MTU's */
56 if (net->cwnd < (2 * net->mtu)) {
57 net->cwnd = 2 * net->mtu;
58 }
59 net->ssthresh = stcb->asoc.peers_rwnd;
60
61 if (sctp_logging_level & (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) {
62 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION);
63 }
64 }
65
66 void
67 sctp_cwnd_update_after_fr(struct sctp_tcb *stcb,
68 struct sctp_association *asoc)
69 {
70 struct sctp_nets *net;
71
72 /*-
73 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off == 1) &&
74 * (net->fast_retran_loss_recovery == 0)))
75 */
76 TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
77 if ((asoc->fast_retran_loss_recovery == 0) || (sctp_cmt_on_off == 1)) {
78 /* out of a RFC2582 Fast recovery window? */
79 if (net->net_ack > 0) {
80 /*
81 * per section 7.2.3, are there any
82 * destinations that had a fast retransmit
83 * to them. If so what we need to do is
84 * adjust ssthresh and cwnd.
85 */
86 struct sctp_tmit_chunk *lchk;
87 int old_cwnd = net->cwnd;
88
89 net->ssthresh = net->cwnd / 2;
90 if (net->ssthresh < (net->mtu * 2)) {
91 net->ssthresh = 2 * net->mtu;
92 }
93 net->cwnd = net->ssthresh;
94 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
95 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd),
96 SCTP_CWND_LOG_FROM_FR);
97 }
98 lchk = TAILQ_FIRST(&asoc->send_queue);
99
100 net->partial_bytes_acked = 0;
101 /* Turn on fast recovery window */
102 asoc->fast_retran_loss_recovery = 1;
103 if (lchk == NULL) {
104 /* Mark end of the window */
105 asoc->fast_recovery_tsn = asoc->sending_seq - 1;
106 } else {
107 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
108 }
109
110 /*
111 * CMT fast recovery -- per destination
112 * recovery variable.
113 */
114 net->fast_retran_loss_recovery = 1;
115
116 if (lchk == NULL) {
117 /* Mark end of the window */
118 net->fast_recovery_tsn = asoc->sending_seq - 1;
119 } else {
120 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
121 }
122
123 /*
124 * Disable Nonce Sum Checking and store the
125 * resync tsn
126 */
127 asoc->nonce_sum_check = 0;
128 asoc->nonce_resync_tsn = asoc->fast_recovery_tsn + 1;
129
130 sctp_timer_stop(SCTP_TIMER_TYPE_SEND,
131 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32);
132 sctp_timer_start(SCTP_TIMER_TYPE_SEND,
133 stcb->sctp_ep, stcb, net);
134 }
135 } else if (net->net_ack > 0) {
136 /*
137 * Mark a peg that we WOULD have done a cwnd
138 * reduction but RFC2582 prevented this action.
139 */
140 SCTP_STAT_INCR(sctps_fastretransinrtt);
141 }
142 }
143 }
144
145 void
146 sctp_cwnd_update_after_sack(struct sctp_tcb *stcb,
147 struct sctp_association *asoc,
148 int accum_moved, int reneged_all, int will_exit)
149 {
150 struct sctp_nets *net;
151
152 /******************************/
153 /* update cwnd and Early FR */
154 /******************************/
155 TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
156
157 #ifdef JANA_CMT_FAST_RECOVERY
158 /*
159 * CMT fast recovery code. Need to debug.
160 */
161 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) {
162 if (compare_with_wrap(asoc->last_acked_seq,
163 net->fast_recovery_tsn, MAX_TSN) ||
164 (asoc->last_acked_seq == net->fast_recovery_tsn) ||
165 compare_with_wrap(net->pseudo_cumack, net->fast_recovery_tsn, MAX_TSN) ||
166 (net->pseudo_cumack == net->fast_recovery_tsn)) {
167 net->will_exit_fast_recovery = 1;
168 }
169 }
170 #endif
171 if (sctp_early_fr) {
172 /*
173 * So, first of all do we need to have a Early FR
174 * timer running?
175 */
176 if (((TAILQ_FIRST(&asoc->sent_queue)) &&
177 (net->ref_count > 1) &&
178 (net->flight_size < net->cwnd)) ||
179 (reneged_all)) {
180 /*
181 * yes, so in this case stop it if its
182 * running, and then restart it. Reneging
183 * all is a special case where we want to
184 * run the Early FR timer and then force the
185 * last few unacked to be sent, causing us
186 * to illicit a sack with gaps to force out
187 * the others.
188 */
189 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
190 SCTP_STAT_INCR(sctps_earlyfrstpidsck2);
191 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
192 SCTP_FROM_SCTP_INDATA + SCTP_LOC_20);
193 }
194 SCTP_STAT_INCR(sctps_earlyfrstrid);
195 sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net);
196 } else {
197 /* No, stop it if its running */
198 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
199 SCTP_STAT_INCR(sctps_earlyfrstpidsck3);
200 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
201 SCTP_FROM_SCTP_INDATA + SCTP_LOC_21);
202 }
203 }
204 }
205 /* if nothing was acked on this destination skip it */
206 if (net->net_ack == 0) {
207 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
208 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK);
209 }
210 continue;
211 }
212 if (net->net_ack2 > 0) {
213 /*
214 * Karn's rule applies to clearing error count, this
215 * is optional.
216 */
217 net->error_count = 0;
218 if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) ==
219 SCTP_ADDR_NOT_REACHABLE) {
220 /* addr came good */
221 net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE;
222 net->dest_state |= SCTP_ADDR_REACHABLE;
223 sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb,
224 SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED);
225 /* now was it the primary? if so restore */
226 if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) {
227 (void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net);
228 }
229 }
230 /*
231 * JRS 5/14/07 - If CMT PF is on and the destination
232 * is in PF state, set the destination to active
233 * state and set the cwnd to one or two MTU's based
234 * on whether PF1 or PF2 is being used.
235 *
236 * Should we stop any running T3 timer here?
237 */
238 if (sctp_cmt_on_off && sctp_cmt_pf && ((net->dest_state & SCTP_ADDR_PF) ==
239 SCTP_ADDR_PF)) {
240 net->dest_state &= ~SCTP_ADDR_PF;
241 net->cwnd = net->mtu * sctp_cmt_pf;
242 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n",
243 net, net->cwnd);
244 /*
245 * Since the cwnd value is explicitly set,
246 * skip the code that updates the cwnd
247 * value.
248 */
249 goto skip_cwnd_update;
250 }
251 }
252 #ifdef JANA_CMT_FAST_RECOVERY
253 /*
254 * CMT fast recovery code
255 */
256 /*
257 * if (sctp_cmt_on_off == 1 &&
258 * net->fast_retran_loss_recovery &&
259 * net->will_exit_fast_recovery == 0) { @@@ Do something }
260 * else if (sctp_cmt_on_off == 0 &&
261 * asoc->fast_retran_loss_recovery && will_exit == 0) {
262 */
263 #endif
264
265 if (asoc->fast_retran_loss_recovery && will_exit == 0 && sctp_cmt_on_off == 0) {
266 /*
267 * If we are in loss recovery we skip any cwnd
268 * update
269 */
270 goto skip_cwnd_update;
271 }
272 /*
273 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has
274 * moved.
275 */
276 if (accum_moved || (sctp_cmt_on_off && net->new_pseudo_cumack)) {
277 /* If the cumulative ack moved we can proceed */
278 if (net->cwnd <= net->ssthresh) {
279 /* We are in slow start */
280 if (net->flight_size + net->net_ack >=
281 net->cwnd) {
282 if (net->net_ack > (net->mtu * sctp_L2_abc_variable)) {
283 net->cwnd += (net->mtu * sctp_L2_abc_variable);
284 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
285 sctp_log_cwnd(stcb, net, net->mtu,
286 SCTP_CWND_LOG_FROM_SS);
287 }
288 } else {
289 net->cwnd += net->net_ack;
290 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
291 sctp_log_cwnd(stcb, net, net->net_ack,
292 SCTP_CWND_LOG_FROM_SS);
293 }
294 }
295 } else {
296 unsigned int dif;
297
298 dif = net->cwnd - (net->flight_size +
299 net->net_ack);
300 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
301 sctp_log_cwnd(stcb, net, net->net_ack,
302 SCTP_CWND_LOG_NOADV_SS);
303 }
304 }
305 } else {
306 /* We are in congestion avoidance */
307 if (net->flight_size + net->net_ack >=
308 net->cwnd) {
309 /*
310 * add to pba only if we had a
311 * cwnd's worth (or so) in flight OR
312 * the burst limit was applied.
313 */
314 net->partial_bytes_acked +=
315 net->net_ack;
316
317 /*
318 * Do we need to increase (if pba is
319 * > cwnd)?
320 */
321 if (net->partial_bytes_acked >=
322 net->cwnd) {
323 if (net->cwnd <
324 net->partial_bytes_acked) {
325 net->partial_bytes_acked -=
326 net->cwnd;
327 } else {
328 net->partial_bytes_acked =
329 0;
330 }
331 net->cwnd += net->mtu;
332 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
333 sctp_log_cwnd(stcb, net, net->mtu,
334 SCTP_CWND_LOG_FROM_CA);
335 }
336 } else {
337 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
338 sctp_log_cwnd(stcb, net, net->net_ack,
339 SCTP_CWND_LOG_NOADV_CA);
340 }
341 }
342 } else {
343 unsigned int dif;
344
345 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
346 sctp_log_cwnd(stcb, net, net->net_ack,
347 SCTP_CWND_LOG_NOADV_CA);
348 }
349 dif = net->cwnd - (net->flight_size +
350 net->net_ack);
351 }
352 }
353 } else {
354 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
355 sctp_log_cwnd(stcb, net, net->mtu,
356 SCTP_CWND_LOG_NO_CUMACK);
357 }
358 }
359 skip_cwnd_update:
360 /*
361 * NOW, according to Karn's rule do we need to restore the
362 * RTO timer back? Check our net_ack2. If not set then we
363 * have a ambiguity.. i.e. all data ack'd was sent to more
364 * than one place.
365 */
366 if (net->net_ack2) {
367 /* restore any doubled timers */
368 net->RTO = ((net->lastsa >> 2) + net->lastsv) >> 1;
369 if (net->RTO < stcb->asoc.minrto) {
370 net->RTO = stcb->asoc.minrto;
371 }
372 if (net->RTO > stcb->asoc.maxrto) {
373 net->RTO = stcb->asoc.maxrto;
374 }
375 }
376 }
377 }
378
379 void
380 sctp_cwnd_update_after_timeout(struct sctp_tcb *stcb,
381 struct sctp_nets *net)
382 {
383 int old_cwnd = net->cwnd;
384
385 net->ssthresh = net->cwnd >> 1;
386 if (net->ssthresh < (net->mtu << 1)) {
387 net->ssthresh = (net->mtu << 1);
388 }
389 net->cwnd = net->mtu;
390 /* floor of 1 mtu */
391 if (net->cwnd < net->mtu)
392 net->cwnd = net->mtu;
393 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
394 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX);
395 }
396 net->partial_bytes_acked = 0;
397 }
398
399 struct sctp_hs_raise_drop {
400 int32_t cwnd;
401 int32_t increase;
402 int32_t drop_percent;
403 };
404
405 #define SCTP_HS_TABLE_SIZE 73
406
407 struct sctp_hs_raise_drop sctp_cwnd_adjust[SCTP_HS_TABLE_SIZE] = {
408 {38, 1, 50}, /* 0 */
409 {118, 2, 44}, /* 1 */
410 {221, 3, 41}, /* 2 */
411 {347, 4, 38}, /* 3 */
412 {495, 5, 37}, /* 4 */
413 {663, 6, 35}, /* 5 */
414 {851, 7, 34}, /* 6 */
415 {1058, 8, 33}, /* 7 */
416 {1284, 9, 32}, /* 8 */
417 {1529, 10, 31}, /* 9 */
418 {1793, 11, 30}, /* 10 */
419 {2076, 12, 29}, /* 11 */
420 {2378, 13, 28}, /* 12 */
421 {2699, 14, 28}, /* 13 */
422 {3039, 15, 27}, /* 14 */
423 {3399, 16, 27}, /* 15 */
424 {3778, 17, 26}, /* 16 */
425 {4177, 18, 26}, /* 17 */
426 {4596, 19, 25}, /* 18 */
427 {5036, 20, 25}, /* 19 */
428 {5497, 21, 24}, /* 20 */
429 {5979, 22, 24}, /* 21 */
430 {6483, 23, 23}, /* 22 */
431 {7009, 24, 23}, /* 23 */
432 {7558, 25, 22}, /* 24 */
433 {8130, 26, 22}, /* 25 */
434 {8726, 27, 22}, /* 26 */
435 {9346, 28, 21}, /* 27 */
436 {9991, 29, 21}, /* 28 */
437 {10661, 30, 21}, /* 29 */
438 {11358, 31, 20}, /* 30 */
439 {12082, 32, 20}, /* 31 */
440 {12834, 33, 20}, /* 32 */
441 {13614, 34, 19}, /* 33 */
442 {14424, 35, 19}, /* 34 */
443 {15265, 36, 19}, /* 35 */
444 {16137, 37, 19}, /* 36 */
445 {17042, 38, 18}, /* 37 */
446 {17981, 39, 18}, /* 38 */
447 {18955, 40, 18}, /* 39 */
448 {19965, 41, 17}, /* 40 */
449 {21013, 42, 17}, /* 41 */
450 {22101, 43, 17}, /* 42 */
451 {23230, 44, 17}, /* 43 */
452 {24402, 45, 16}, /* 44 */
453 {25618, 46, 16}, /* 45 */
454 {26881, 47, 16}, /* 46 */
455 {28193, 48, 16}, /* 47 */
456 {29557, 49, 15}, /* 48 */
457 {30975, 50, 15}, /* 49 */
458 {32450, 51, 15}, /* 50 */
459 {33986, 52, 15}, /* 51 */
460 {35586, 53, 14}, /* 52 */
461 {37253, 54, 14}, /* 53 */
462 {38992, 55, 14}, /* 54 */
463 {40808, 56, 14}, /* 55 */
464 {42707, 57, 13}, /* 56 */
465 {44694, 58, 13}, /* 57 */
466 {46776, 59, 13}, /* 58 */
467 {48961, 60, 13}, /* 59 */
468 {51258, 61, 13}, /* 60 */
469 {53677, 62, 12}, /* 61 */
470 {56230, 63, 12}, /* 62 */
471 {58932, 64, 12}, /* 63 */
472 {61799, 65, 12}, /* 64 */
473 {64851, 66, 11}, /* 65 */
474 {68113, 67, 11}, /* 66 */
475 {71617, 68, 11}, /* 67 */
476 {75401, 69, 10}, /* 68 */
477 {79517, 70, 10}, /* 69 */
478 {84035, 71, 10}, /* 70 */
479 {89053, 72, 10}, /* 71 */
480 {94717, 73, 9} /* 72 */
481 };
482
483 static void
484 sctp_hs_cwnd_increase(struct sctp_tcb *stcb, struct sctp_nets *net)
485 {
486 int cur_val, i, indx, incr;
487
488 cur_val = net->cwnd >> 10;
489 indx = SCTP_HS_TABLE_SIZE - 1;
490 #ifdef SCTP_DEBUG
491 printf("HS CC CAlled.\n");
492 #endif
493 if (cur_val < sctp_cwnd_adjust[0].cwnd) {
494 /* normal mode */
495 if (net->net_ack > net->mtu) {
496 net->cwnd += net->mtu;
497 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
498 sctp_log_cwnd(stcb, net, net->mtu, SCTP_CWND_LOG_FROM_SS);
499 }
500 } else {
501 net->cwnd += net->net_ack;
502 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
503 sctp_log_cwnd(stcb, net, net->net_ack, SCTP_CWND_LOG_FROM_SS);
504 }
505 }
506 } else {
507 for (i = net->last_hs_used; i < SCTP_HS_TABLE_SIZE; i++) {
508 if (cur_val < sctp_cwnd_adjust[i].cwnd) {
509 indx = i;
510 break;
511 }
512 }
513 net->last_hs_used = indx;
514 incr = ((sctp_cwnd_adjust[indx].increase) << 10);
515 net->cwnd += incr;
516 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
517 sctp_log_cwnd(stcb, net, incr, SCTP_CWND_LOG_FROM_SS);
518 }
519 }
520 }
521
522 static void
523 sctp_hs_cwnd_decrease(struct sctp_tcb *stcb, struct sctp_nets *net)
524 {
525 int cur_val, i, indx;
526 int old_cwnd = net->cwnd;
527
528 cur_val = net->cwnd >> 10;
529 indx = net->last_hs_used;
530 if (cur_val < sctp_cwnd_adjust[0].cwnd) {
531 /* normal mode */
532 net->ssthresh = net->cwnd / 2;
533 if (net->ssthresh < (net->mtu * 2)) {
534 net->ssthresh = 2 * net->mtu;
535 }
536 net->cwnd = net->ssthresh;
537 } else {
538 /* drop by the proper amount */
539 net->ssthresh = net->cwnd - (int)((net->cwnd / 100) *
540 sctp_cwnd_adjust[net->last_hs_used].drop_percent);
541 net->cwnd = net->ssthresh;
542 /* now where are we */
543 indx = net->last_hs_used;
544 cur_val = net->cwnd >> 10;
545 /* reset where we are in the table */
546 if (cur_val < sctp_cwnd_adjust[0].cwnd) {
547 /* feel out of hs */
548 net->last_hs_used = 0;
549 } else {
550 for (i = indx; i >= 1; i--) {
551 if (cur_val > sctp_cwnd_adjust[i - 1].cwnd) {
552 break;
553 }
554 }
555 net->last_hs_used = indx;
556 }
557 }
558 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
559 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_FR);
560 }
561 }
562
563 void
564 sctp_hs_cwnd_update_after_fr(struct sctp_tcb *stcb,
565 struct sctp_association *asoc)
566 {
567 struct sctp_nets *net;
568
569 /*
570 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off == 1) &&
571 * (net->fast_retran_loss_recovery == 0)))
572 */
573 TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
574 if ((asoc->fast_retran_loss_recovery == 0) || (sctp_cmt_on_off == 1)) {
575 /* out of a RFC2582 Fast recovery window? */
576 if (net->net_ack > 0) {
577 /*
578 * per section 7.2.3, are there any
579 * destinations that had a fast retransmit
580 * to them. If so what we need to do is
581 * adjust ssthresh and cwnd.
582 */
583 struct sctp_tmit_chunk *lchk;
584
585 sctp_hs_cwnd_decrease(stcb, net);
586
587 lchk = TAILQ_FIRST(&asoc->send_queue);
588
589 net->partial_bytes_acked = 0;
590 /* Turn on fast recovery window */
591 asoc->fast_retran_loss_recovery = 1;
592 if (lchk == NULL) {
593 /* Mark end of the window */
594 asoc->fast_recovery_tsn = asoc->sending_seq - 1;
595 } else {
596 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
597 }
598
599 /*
600 * CMT fast recovery -- per destination
601 * recovery variable.
602 */
603 net->fast_retran_loss_recovery = 1;
604
605 if (lchk == NULL) {
606 /* Mark end of the window */
607 net->fast_recovery_tsn = asoc->sending_seq - 1;
608 } else {
609 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
610 }
611
612 /*
613 * Disable Nonce Sum Checking and store the
614 * resync tsn
615 */
616 asoc->nonce_sum_check = 0;
617 asoc->nonce_resync_tsn = asoc->fast_recovery_tsn + 1;
618
619 sctp_timer_stop(SCTP_TIMER_TYPE_SEND,
620 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32);
621 sctp_timer_start(SCTP_TIMER_TYPE_SEND,
622 stcb->sctp_ep, stcb, net);
623 }
624 } else if (net->net_ack > 0) {
625 /*
626 * Mark a peg that we WOULD have done a cwnd
627 * reduction but RFC2582 prevented this action.
628 */
629 SCTP_STAT_INCR(sctps_fastretransinrtt);
630 }
631 }
632 }
633
634 void
635 sctp_hs_cwnd_update_after_sack(struct sctp_tcb *stcb,
636 struct sctp_association *asoc,
637 int accum_moved, int reneged_all, int will_exit)
638 {
639 struct sctp_nets *net;
640
641 /******************************/
642 /* update cwnd and Early FR */
643 /******************************/
644 TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
645
646 #ifdef JANA_CMT_FAST_RECOVERY
647 /*
648 * CMT fast recovery code. Need to debug.
649 */
650 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) {
651 if (compare_with_wrap(asoc->last_acked_seq,
652 net->fast_recovery_tsn, MAX_TSN) ||
653 (asoc->last_acked_seq == net->fast_recovery_tsn) ||
654 compare_with_wrap(net->pseudo_cumack, net->fast_recovery_tsn, MAX_TSN) ||
655 (net->pseudo_cumack == net->fast_recovery_tsn)) {
656 net->will_exit_fast_recovery = 1;
657 }
658 }
659 #endif
660 if (sctp_early_fr) {
661 /*
662 * So, first of all do we need to have a Early FR
663 * timer running?
664 */
665 if (((TAILQ_FIRST(&asoc->sent_queue)) &&
666 (net->ref_count > 1) &&
667 (net->flight_size < net->cwnd)) ||
668 (reneged_all)) {
669 /*
670 * yes, so in this case stop it if its
671 * running, and then restart it. Reneging
672 * all is a special case where we want to
673 * run the Early FR timer and then force the
674 * last few unacked to be sent, causing us
675 * to illicit a sack with gaps to force out
676 * the others.
677 */
678 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
679 SCTP_STAT_INCR(sctps_earlyfrstpidsck2);
680 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
681 SCTP_FROM_SCTP_INDATA + SCTP_LOC_20);
682 }
683 SCTP_STAT_INCR(sctps_earlyfrstrid);
684 sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net);
685 } else {
686 /* No, stop it if its running */
687 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
688 SCTP_STAT_INCR(sctps_earlyfrstpidsck3);
689 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
690 SCTP_FROM_SCTP_INDATA + SCTP_LOC_21);
691 }
692 }
693 }
694 /* if nothing was acked on this destination skip it */
695 if (net->net_ack == 0) {
696 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
697 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK);
698 }
699 continue;
700 }
701 if (net->net_ack2 > 0) {
702 /*
703 * Karn's rule applies to clearing error count, this
704 * is optional.
705 */
706 net->error_count = 0;
707 if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) ==
708 SCTP_ADDR_NOT_REACHABLE) {
709 /* addr came good */
710 net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE;
711 net->dest_state |= SCTP_ADDR_REACHABLE;
712 sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb,
713 SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED);
714 /* now was it the primary? if so restore */
715 if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) {
716 (void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net);
717 }
718 }
719 /*
720 * JRS 5/14/07 - If CMT PF is on and the destination
721 * is in PF state, set the destination to active
722 * state and set the cwnd to one or two MTU's based
723 * on whether PF1 or PF2 is being used.
724 *
725 * Should we stop any running T3 timer here?
726 */
727 if (sctp_cmt_on_off && sctp_cmt_pf && ((net->dest_state & SCTP_ADDR_PF) ==
728 SCTP_ADDR_PF)) {
729 net->dest_state &= ~SCTP_ADDR_PF;
730 net->cwnd = net->mtu * sctp_cmt_pf;
731 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n",
732 net, net->cwnd);
733 /*
734 * Since the cwnd value is explicitly set,
735 * skip the code that updates the cwnd
736 * value.
737 */
738 goto skip_cwnd_update;
739 }
740 }
741 #ifdef JANA_CMT_FAST_RECOVERY
742 /*
743 * CMT fast recovery code
744 */
745 /*
746 * if (sctp_cmt_on_off == 1 &&
747 * net->fast_retran_loss_recovery &&
748 * net->will_exit_fast_recovery == 0) { @@@ Do something }
749 * else if (sctp_cmt_on_off == 0 &&
750 * asoc->fast_retran_loss_recovery && will_exit == 0) {
751 */
752 #endif
753
754 if (asoc->fast_retran_loss_recovery && will_exit == 0 && sctp_cmt_on_off == 0) {
755 /*
756 * If we are in loss recovery we skip any cwnd
757 * update
758 */
759 goto skip_cwnd_update;
760 }
761 /*
762 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has
763 * moved.
764 */
765 if (accum_moved || (sctp_cmt_on_off && net->new_pseudo_cumack)) {
766 /* If the cumulative ack moved we can proceed */
767 if (net->cwnd <= net->ssthresh) {
768 /* We are in slow start */
769 if (net->flight_size + net->net_ack >=
770 net->cwnd) {
771
772 sctp_hs_cwnd_increase(stcb, net);
773
774 } else {
775 unsigned int dif;
776
777 dif = net->cwnd - (net->flight_size +
778 net->net_ack);
779 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
780 sctp_log_cwnd(stcb, net, net->net_ack,
781 SCTP_CWND_LOG_NOADV_SS);
782 }
783 }
784 } else {
785 /* We are in congestion avoidance */
786 if (net->flight_size + net->net_ack >=
787 net->cwnd) {
788 /*
789 * add to pba only if we had a
790 * cwnd's worth (or so) in flight OR
791 * the burst limit was applied.
792 */
793 net->partial_bytes_acked +=
794 net->net_ack;
795
796 /*
797 * Do we need to increase (if pba is
798 * > cwnd)?
799 */
800 if (net->partial_bytes_acked >=
801 net->cwnd) {
802 if (net->cwnd <
803 net->partial_bytes_acked) {
804 net->partial_bytes_acked -=
805 net->cwnd;
806 } else {
807 net->partial_bytes_acked =
808 0;
809 }
810 net->cwnd += net->mtu;
811 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
812 sctp_log_cwnd(stcb, net, net->mtu,
813 SCTP_CWND_LOG_FROM_CA);
814 }
815 } else {
816 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
817 sctp_log_cwnd(stcb, net, net->net_ack,
818 SCTP_CWND_LOG_NOADV_CA);
819 }
820 }
821 } else {
822 unsigned int dif;
823
824 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
825 sctp_log_cwnd(stcb, net, net->net_ack,
826 SCTP_CWND_LOG_NOADV_CA);
827 }
828 dif = net->cwnd - (net->flight_size +
829 net->net_ack);
830 }
831 }
832 } else {
833 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
834 sctp_log_cwnd(stcb, net, net->mtu,
835 SCTP_CWND_LOG_NO_CUMACK);
836 }
837 }
838 skip_cwnd_update:
839 /*
840 * NOW, according to Karn's rule do we need to restore the
841 * RTO timer back? Check our net_ack2. If not set then we
842 * have a ambiguity.. i.e. all data ack'd was sent to more
843 * than one place.
844 */
845 if (net->net_ack2) {
846 /* restore any doubled timers */
847 net->RTO = ((net->lastsa >> 2) + net->lastsv) >> 1;
848 if (net->RTO < stcb->asoc.minrto) {
849 net->RTO = stcb->asoc.minrto;
850 }
851 if (net->RTO > stcb->asoc.maxrto) {
852 net->RTO = stcb->asoc.maxrto;
853 }
854 }
855 }
856 }
857
858 void
859 sctp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb,
860 struct sctp_nets *net)
861 {
862 int old_cwnd;
863
864 old_cwnd = net->cwnd;
865
866 SCTP_STAT_INCR(sctps_ecnereducedcwnd);
867 net->ssthresh = net->cwnd / 2;
868 if (net->ssthresh < net->mtu) {
869 net->ssthresh = net->mtu;
870 /* here back off the timer as well, to slow us down */
871 net->RTO <<= 1;
872 }
873 net->cwnd = net->ssthresh;
874 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
875 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT);
876 }
877 }
878
879 void
880 sctp_cwnd_update_after_packet_dropped(struct sctp_tcb *stcb,
881 struct sctp_nets *net, struct sctp_pktdrop_chunk *cp,
882 uint32_t * bottle_bw, uint32_t * on_queue)
883 {
884 uint32_t bw_avail;
885 int rtt, incr;
886 int old_cwnd = net->cwnd;
887
888 /* need real RTT for this calc */
889 rtt = ((net->lastsa >> 2) + net->lastsv) >> 1;
890 /* get bottle neck bw */
891 *bottle_bw = ntohl(cp->bottle_bw);
892 /* and whats on queue */
893 *on_queue = ntohl(cp->current_onq);
894 /*
895 * adjust the on-queue if our flight is more it could be that the
896 * router has not yet gotten data "in-flight" to it
897 */
898 if (*on_queue < net->flight_size)
899 *on_queue = net->flight_size;
900 /* calculate the available space */
901 bw_avail = (*bottle_bw * rtt) / 1000;
902 if (bw_avail > *bottle_bw) {
903 /*
904 * Cap the growth to no more than the bottle neck. This can
905 * happen as RTT slides up due to queues. It also means if
906 * you have more than a 1 second RTT with a empty queue you
907 * will be limited to the bottle_bw per second no matter if
908 * other points have 1/2 the RTT and you could get more
909 * out...
910 */
911 bw_avail = *bottle_bw;
912 }
913 if (*on_queue > bw_avail) {
914 /*
915 * No room for anything else don't allow anything else to be
916 * "added to the fire".
917 */
918 int seg_inflight, seg_onqueue, my_portion;
919
920 net->partial_bytes_acked = 0;
921
922 /* how much are we over queue size? */
923 incr = *on_queue - bw_avail;
924 if (stcb->asoc.seen_a_sack_this_pkt) {
925 /*
926 * undo any cwnd adjustment that the sack might have
927 * made
928 */
929 net->cwnd = net->prev_cwnd;
930 }
931 /* Now how much of that is mine? */
932 seg_inflight = net->flight_size / net->mtu;
933 seg_onqueue = *on_queue / net->mtu;
934 my_portion = (incr * seg_inflight) / seg_onqueue;
935
936 /* Have I made an adjustment already */
937 if (net->cwnd > net->flight_size) {
938 /*
939 * for this flight I made an adjustment we need to
940 * decrease the portion by a share our previous
941 * adjustment.
942 */
943 int diff_adj;
944
945 diff_adj = net->cwnd - net->flight_size;
946 if (diff_adj > my_portion)
947 my_portion = 0;
948 else
949 my_portion -= diff_adj;
950 }
951 /*
952 * back down to the previous cwnd (assume we have had a sack
953 * before this packet). minus what ever portion of the
954 * overage is my fault.
955 */
956 net->cwnd -= my_portion;
957
958 /* we will NOT back down more than 1 MTU */
959 if (net->cwnd <= net->mtu) {
960 net->cwnd = net->mtu;
961 }
962 /* force into CA */
963 net->ssthresh = net->cwnd - 1;
964 } else {
965 /*
966 * Take 1/4 of the space left or max burst up .. whichever
967 * is less.
968 */
969 incr = min((bw_avail - *on_queue) >> 2,
970 stcb->asoc.max_burst * net->mtu);
971 net->cwnd += incr;
972 }
973 if (net->cwnd > bw_avail) {
974 /* We can't exceed the pipe size */
975 net->cwnd = bw_avail;
976 }
977 if (net->cwnd < net->mtu) {
978 /* We always have 1 MTU */
979 net->cwnd = net->mtu;
980 }
981 if (net->cwnd - old_cwnd != 0) {
982 /* log only changes */
983 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
984 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd),
985 SCTP_CWND_LOG_FROM_SAT);
986 }
987 }
988 }
989
990 void
991 sctp_cwnd_update_after_output(struct sctp_tcb *stcb,
992 struct sctp_nets *net, int burst_limit)
993 {
994 int old_cwnd;
995
996 if (net->ssthresh < net->cwnd)
997 net->ssthresh = net->cwnd;
998 old_cwnd = net->cwnd;
999 net->cwnd = (net->flight_size + (burst_limit * net->mtu));
1000
1001 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1002 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_BRST);
1003 }
1004 }
1005
1006 void
1007 sctp_cwnd_update_after_fr_timer(struct sctp_inpcb *inp,
1008 struct sctp_tcb *stcb, struct sctp_nets *net)
1009 {
1010 int old_cwnd;
1011
1012 old_cwnd = net->cwnd;
1013
1014 sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_EARLY_FR_TMR, SCTP_SO_NOT_LOCKED);
1015 /*
1016 * make a small adjustment to cwnd and force to CA.
1017 */
1018 if (net->cwnd > net->mtu)
1019 /* drop down one MTU after sending */
1020 net->cwnd -= net->mtu;
1021 if (net->cwnd < net->ssthresh)
1022 /* still in SS move to CA */
1023 net->ssthresh = net->cwnd - 1;
1024 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1025 sctp_log_cwnd(stcb, net, (old_cwnd - net->cwnd), SCTP_CWND_LOG_FROM_FR);
1026 }
1027 }
1028
1029 /*
1030 * H-TCP congestion control. The algorithm is detailed in:
1031 * R.N.Shorten, D.J.Leith:
1032 * "H-TCP: TCP for high-speed and long-distance networks"
1033 * Proc. PFLDnet, Argonne, 2004.
1034 * http://www.hamilton.ie/net/htcp3.pdf
1035 */
1036
1037
1038 static int use_rtt_scaling = 1;
1039 static int use_bandwidth_switch = 1;
1040
1041 static inline int
1042 between(uint32_t seq1, uint32_t seq2, uint32_t seq3)
1043 {
1044 return seq3 - seq2 >= seq1 - seq2;
1045 }
1046
1047 static inline uint32_t
1048 htcp_cong_time(struct htcp *ca)
1049 {
1050 return sctp_get_tick_count() - ca->last_cong;
1051 }
1052
1053 static inline uint32_t
1054 htcp_ccount(struct htcp *ca)
1055 {
1056 return htcp_cong_time(ca) / ca->minRTT;
1057 }
1058
1059 static inline void
1060 htcp_reset(struct htcp *ca)
1061 {
1062 ca->undo_last_cong = ca->last_cong;
1063 ca->undo_maxRTT = ca->maxRTT;
1064 ca->undo_old_maxB = ca->old_maxB;
1065 ca->last_cong = sctp_get_tick_count();
1066 }
1067
1068 #ifdef SCTP_NOT_USED
1069
1070 static uint32_t
1071 htcp_cwnd_undo(struct sctp_tcb *stcb, struct sctp_nets *net)
1072 {
1073 net->htcp_ca.last_cong = net->htcp_ca.undo_last_cong;
1074 net->htcp_ca.maxRTT = net->htcp_ca.undo_maxRTT;
1075 net->htcp_ca.old_maxB = net->htcp_ca.undo_old_maxB;
1076 return max(net->cwnd, ((net->ssthresh / net->mtu << 7) / net->htcp_ca.beta) * net->mtu);
1077 }
1078
1079 #endif
1080
1081 static inline void
1082 measure_rtt(struct sctp_tcb *stcb, struct sctp_nets *net)
1083 {
1084 uint32_t srtt = net->lastsa >> 3;
1085
1086 /* keep track of minimum RTT seen so far, minRTT is zero at first */
1087 if (net->htcp_ca.minRTT > srtt || !net->htcp_ca.minRTT)
1088 net->htcp_ca.minRTT = srtt;
1089
1090 /* max RTT */
1091 if (net->fast_retran_ip == 0 && net->ssthresh < 0xFFFF && htcp_ccount(&net->htcp_ca) > 3) {
1092 if (net->htcp_ca.maxRTT < net->htcp_ca.minRTT)
1093 net->htcp_ca.maxRTT = net->htcp_ca.minRTT;
1094 if (net->htcp_ca.maxRTT < srtt && srtt <= net->htcp_ca.maxRTT + MSEC_TO_TICKS(20))
1095 net->htcp_ca.maxRTT = srtt;
1096 }
1097 }
1098
1099 static void
1100 measure_achieved_throughput(struct sctp_tcb *stcb, struct sctp_nets *net)
1101 {
1102 uint32_t now = sctp_get_tick_count();
1103
1104 if (net->fast_retran_ip == 0)
1105 net->htcp_ca.bytes_acked = net->net_ack;
1106
1107 if (!use_bandwidth_switch)
1108 return;
1109
1110 /* achieved throughput calculations */
1111 /* JRS - not 100% sure of this statement */
1112 if (net->fast_retran_ip == 1) {
1113 net->htcp_ca.bytecount = 0;
1114 net->htcp_ca.lasttime = now;
1115 return;
1116 }
1117 net->htcp_ca.bytecount += net->net_ack;
1118
1119 if (net->htcp_ca.bytecount >= net->cwnd - ((net->htcp_ca.alpha >> 7 ? : 1) * net->mtu)
1120 && now - net->htcp_ca.lasttime >= net->htcp_ca.minRTT
1121 && net->htcp_ca.minRTT > 0) {
1122 uint32_t cur_Bi = net->htcp_ca.bytecount / net->mtu * hz / (now - net->htcp_ca.lasttime);
1123
1124 if (htcp_ccount(&net->htcp_ca) <= 3) {
1125 /* just after backoff */
1126 net->htcp_ca.minB = net->htcp_ca.maxB = net->htcp_ca.Bi = cur_Bi;
1127 } else {
1128 net->htcp_ca.Bi = (3 * net->htcp_ca.Bi + cur_Bi) / 4;
1129 if (net->htcp_ca.Bi > net->htcp_ca.maxB)
1130 net->htcp_ca.maxB = net->htcp_ca.Bi;
1131 if (net->htcp_ca.minB > net->htcp_ca.maxB)
1132 net->htcp_ca.minB = net->htcp_ca.maxB;
1133 }
1134 net->htcp_ca.bytecount = 0;
1135 net->htcp_ca.lasttime = now;
1136 }
1137 }
1138
1139 static inline void
1140 htcp_beta_update(struct htcp *ca, uint32_t minRTT, uint32_t maxRTT)
1141 {
1142 if (use_bandwidth_switch) {
1143 uint32_t maxB = ca->maxB;
1144 uint32_t old_maxB = ca->old_maxB;
1145
1146 ca->old_maxB = ca->maxB;
1147
1148 if (!between(5 * maxB, 4 * old_maxB, 6 * old_maxB)) {
1149 ca->beta = BETA_MIN;
1150 ca->modeswitch = 0;
1151 return;
1152 }
1153 }
1154 if (ca->modeswitch && minRTT > (uint32_t) MSEC_TO_TICKS(10) && maxRTT) {
1155 ca->beta = (minRTT << 7) / maxRTT;
1156 if (ca->beta < BETA_MIN)
1157 ca->beta = BETA_MIN;
1158 else if (ca->beta > BETA_MAX)
1159 ca->beta = BETA_MAX;
1160 } else {
1161 ca->beta = BETA_MIN;
1162 ca->modeswitch = 1;
1163 }
1164 }
1165
1166 static inline void
1167 htcp_alpha_update(struct htcp *ca)
1168 {
1169 uint32_t minRTT = ca->minRTT;
1170 uint32_t factor = 1;
1171 uint32_t diff = htcp_cong_time(ca);
1172
1173 if (diff > (uint32_t) hz) {
1174 diff -= hz;
1175 factor = 1 + (10 * diff + ((diff / 2) * (diff / 2) / hz)) / hz;
1176 }
1177 if (use_rtt_scaling && minRTT) {
1178 uint32_t scale = (hz << 3) / (10 * minRTT);
1179
1180 scale = min(max(scale, 1U << 2), 10U << 3); /* clamping ratio to
1181 * interval [0.5,10]<<3 */
1182 factor = (factor << 3) / scale;
1183 if (!factor)
1184 factor = 1;
1185 }
1186 ca->alpha = 2 * factor * ((1 << 7) - ca->beta);
1187 if (!ca->alpha)
1188 ca->alpha = ALPHA_BASE;
1189 }
1190
1191 /* After we have the rtt data to calculate beta, we'd still prefer to wait one
1192 * rtt before we adjust our beta to ensure we are working from a consistent
1193 * data.
1194 *
1195 * This function should be called when we hit a congestion event since only at
1196 * that point do we really have a real sense of maxRTT (the queues en route
1197 * were getting just too full now).
1198 */
1199 static void
1200 htcp_param_update(struct sctp_tcb *stcb, struct sctp_nets *net)
1201 {
1202 uint32_t minRTT = net->htcp_ca.minRTT;
1203 uint32_t maxRTT = net->htcp_ca.maxRTT;
1204
1205 htcp_beta_update(&net->htcp_ca, minRTT, maxRTT);
1206 htcp_alpha_update(&net->htcp_ca);
1207
1208 /*
1209 * add slowly fading memory for maxRTT to accommodate routing
1210 * changes etc
1211 */
1212 if (minRTT > 0 && maxRTT > minRTT)
1213 net->htcp_ca.maxRTT = minRTT + ((maxRTT - minRTT) * 95) / 100;
1214 }
1215
1216 static uint32_t
1217 htcp_recalc_ssthresh(struct sctp_tcb *stcb, struct sctp_nets *net)
1218 {
1219 htcp_param_update(stcb, net);
1220 return max(((net->cwnd / net->mtu * net->htcp_ca.beta) >> 7) * net->mtu, 2U * net->mtu);
1221 }
1222
1223 static void
1224 htcp_cong_avoid(struct sctp_tcb *stcb, struct sctp_nets *net)
1225 {
1226 /*-
1227 * How to handle these functions?
1228 * if (!tcp_is_cwnd_limited(sk, in_flight)) RRS - good question.
1229 * return;
1230 */
1231 if (net->cwnd <= net->ssthresh) {
1232 /* We are in slow start */
1233 if (net->flight_size + net->net_ack >= net->cwnd) {
1234 if (net->net_ack > (net->mtu * sctp_L2_abc_variable)) {
1235 net->cwnd += (net->mtu * sctp_L2_abc_variable);
1236 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1237 sctp_log_cwnd(stcb, net, net->mtu,
1238 SCTP_CWND_LOG_FROM_SS);
1239 }
1240 } else {
1241 net->cwnd += net->net_ack;
1242 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1243 sctp_log_cwnd(stcb, net, net->net_ack,
1244 SCTP_CWND_LOG_FROM_SS);
1245 }
1246 }
1247 } else {
1248 unsigned int dif;
1249
1250 dif = net->cwnd - (net->flight_size +
1251 net->net_ack);
1252 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
1253 sctp_log_cwnd(stcb, net, net->net_ack,
1254 SCTP_CWND_LOG_NOADV_SS);
1255 }
1256 }
1257 } else {
1258 measure_rtt(stcb, net);
1259
1260 /*
1261 * In dangerous area, increase slowly. In theory this is
1262 * net->cwnd += alpha / net->cwnd
1263 */
1264 /* What is snd_cwnd_cnt?? */
1265 if (((net->partial_bytes_acked / net->mtu * net->htcp_ca.alpha) >> 7) * net->mtu >= net->cwnd) {
1266 /*-
1267 * Does SCTP have a cwnd clamp?
1268 * if (net->snd_cwnd < net->snd_cwnd_clamp) - Nope (RRS).
1269 */
1270 net->cwnd += net->mtu;
1271 net->partial_bytes_acked = 0;
1272 htcp_alpha_update(&net->htcp_ca);
1273 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1274 sctp_log_cwnd(stcb, net, net->mtu,
1275 SCTP_CWND_LOG_FROM_CA);
1276 }
1277 } else {
1278 net->partial_bytes_acked += net->net_ack;
1279 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
1280 sctp_log_cwnd(stcb, net, net->net_ack,
1281 SCTP_CWND_LOG_NOADV_CA);
1282 }
1283 }
1284
1285 net->htcp_ca.bytes_acked = net->mtu;
1286 }
1287 }
1288
1289 #ifdef SCTP_NOT_USED
1290 /* Lower bound on congestion window. */
1291 static uint32_t
1292 htcp_min_cwnd(struct sctp_tcb *stcb, struct sctp_nets *net)
1293 {
1294 return net->ssthresh;
1295 }
1296
1297 #endif
1298
1299 static void
1300 htcp_init(struct sctp_tcb *stcb, struct sctp_nets *net)
1301 {
1302 memset(&net->htcp_ca, 0, sizeof(struct htcp));
1303 net->htcp_ca.alpha = ALPHA_BASE;
1304 net->htcp_ca.beta = BETA_MIN;
1305 net->htcp_ca.bytes_acked = net->mtu;
1306 net->htcp_ca.last_cong = sctp_get_tick_count();
1307 }
1308
1309 void
1310 sctp_htcp_set_initial_cc_param(struct sctp_tcb *stcb, struct sctp_nets *net)
1311 {
1312 /*
1313 * We take the max of the burst limit times a MTU or the
1314 * INITIAL_CWND. We then limit this to 4 MTU's of sending.
1315 */
1316 net->cwnd = min((net->mtu * 4), max((2 * net->mtu), SCTP_INITIAL_CWND));
1317 /* we always get at LEAST 2 MTU's */
1318 if (net->cwnd < (2 * net->mtu)) {
1319 net->cwnd = 2 * net->mtu;
1320 }
1321 net->ssthresh = stcb->asoc.peers_rwnd;
1322 htcp_init(stcb, net);
1323
1324 if (sctp_logging_level & (SCTP_CWND_MONITOR_ENABLE | SCTP_CWND_LOGGING_ENABLE)) {
1325 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_INITIALIZATION);
1326 }
1327 }
1328
1329 void
1330 sctp_htcp_cwnd_update_after_sack(struct sctp_tcb *stcb,
1331 struct sctp_association *asoc,
1332 int accum_moved, int reneged_all, int will_exit)
1333 {
1334 struct sctp_nets *net;
1335
1336 /******************************/
1337 /* update cwnd and Early FR */
1338 /******************************/
1339 TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
1340
1341 #ifdef JANA_CMT_FAST_RECOVERY
1342 /*
1343 * CMT fast recovery code. Need to debug.
1344 */
1345 if (net->fast_retran_loss_recovery && net->new_pseudo_cumack) {
1346 if (compare_with_wrap(asoc->last_acked_seq,
1347 net->fast_recovery_tsn, MAX_TSN) ||
1348 (asoc->last_acked_seq == net->fast_recovery_tsn) ||
1349 compare_with_wrap(net->pseudo_cumack, net->fast_recovery_tsn, MAX_TSN) ||
1350 (net->pseudo_cumack == net->fast_recovery_tsn)) {
1351 net->will_exit_fast_recovery = 1;
1352 }
1353 }
1354 #endif
1355 if (sctp_early_fr) {
1356 /*
1357 * So, first of all do we need to have a Early FR
1358 * timer running?
1359 */
1360 if (((TAILQ_FIRST(&asoc->sent_queue)) &&
1361 (net->ref_count > 1) &&
1362 (net->flight_size < net->cwnd)) ||
1363 (reneged_all)) {
1364 /*
1365 * yes, so in this case stop it if its
1366 * running, and then restart it. Reneging
1367 * all is a special case where we want to
1368 * run the Early FR timer and then force the
1369 * last few unacked to be sent, causing us
1370 * to illicit a sack with gaps to force out
1371 * the others.
1372 */
1373 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
1374 SCTP_STAT_INCR(sctps_earlyfrstpidsck2);
1375 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
1376 SCTP_FROM_SCTP_INDATA + SCTP_LOC_20);
1377 }
1378 SCTP_STAT_INCR(sctps_earlyfrstrid);
1379 sctp_timer_start(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net);
1380 } else {
1381 /* No, stop it if its running */
1382 if (SCTP_OS_TIMER_PENDING(&net->fr_timer.timer)) {
1383 SCTP_STAT_INCR(sctps_earlyfrstpidsck3);
1384 sctp_timer_stop(SCTP_TIMER_TYPE_EARLYFR, stcb->sctp_ep, stcb, net,
1385 SCTP_FROM_SCTP_INDATA + SCTP_LOC_21);
1386 }
1387 }
1388 }
1389 /* if nothing was acked on this destination skip it */
1390 if (net->net_ack == 0) {
1391 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
1392 sctp_log_cwnd(stcb, net, 0, SCTP_CWND_LOG_FROM_SACK);
1393 }
1394 continue;
1395 }
1396 if (net->net_ack2 > 0) {
1397 /*
1398 * Karn's rule applies to clearing error count, this
1399 * is optional.
1400 */
1401 net->error_count = 0;
1402 if ((net->dest_state & SCTP_ADDR_NOT_REACHABLE) ==
1403 SCTP_ADDR_NOT_REACHABLE) {
1404 /* addr came good */
1405 net->dest_state &= ~SCTP_ADDR_NOT_REACHABLE;
1406 net->dest_state |= SCTP_ADDR_REACHABLE;
1407 sctp_ulp_notify(SCTP_NOTIFY_INTERFACE_UP, stcb,
1408 SCTP_RECEIVED_SACK, (void *)net, SCTP_SO_NOT_LOCKED);
1409 /* now was it the primary? if so restore */
1410 if (net->dest_state & SCTP_ADDR_WAS_PRIMARY) {
1411 (void)sctp_set_primary_addr(stcb, (struct sockaddr *)NULL, net);
1412 }
1413 }
1414 /*
1415 * JRS 5/14/07 - If CMT PF is on and the destination
1416 * is in PF state, set the destination to active
1417 * state and set the cwnd to one or two MTU's based
1418 * on whether PF1 or PF2 is being used.
1419 *
1420 * Should we stop any running T3 timer here?
1421 */
1422 if (sctp_cmt_on_off && sctp_cmt_pf && ((net->dest_state & SCTP_ADDR_PF) ==
1423 SCTP_ADDR_PF)) {
1424 net->dest_state &= ~SCTP_ADDR_PF;
1425 net->cwnd = net->mtu * sctp_cmt_pf;
1426 SCTPDBG(SCTP_DEBUG_INDATA1, "Destination %p moved from PF to reachable with cwnd %d.\n",
1427 net, net->cwnd);
1428 /*
1429 * Since the cwnd value is explicitly set,
1430 * skip the code that updates the cwnd
1431 * value.
1432 */
1433 goto skip_cwnd_update;
1434 }
1435 }
1436 #ifdef JANA_CMT_FAST_RECOVERY
1437 /*
1438 * CMT fast recovery code
1439 */
1440 /*
1441 * if (sctp_cmt_on_off == 1 &&
1442 * net->fast_retran_loss_recovery &&
1443 * net->will_exit_fast_recovery == 0) { @@@ Do something }
1444 * else if (sctp_cmt_on_off == 0 &&
1445 * asoc->fast_retran_loss_recovery && will_exit == 0) {
1446 */
1447 #endif
1448
1449 if (asoc->fast_retran_loss_recovery && will_exit == 0 && sctp_cmt_on_off == 0) {
1450 /*
1451 * If we are in loss recovery we skip any cwnd
1452 * update
1453 */
1454 goto skip_cwnd_update;
1455 }
1456 /*
1457 * CMT: CUC algorithm. Update cwnd if pseudo-cumack has
1458 * moved.
1459 */
1460 if (accum_moved || (sctp_cmt_on_off && net->new_pseudo_cumack)) {
1461 htcp_cong_avoid(stcb, net);
1462 measure_achieved_throughput(stcb, net);
1463 } else {
1464 if (sctp_logging_level & SCTP_CWND_LOGGING_ENABLE) {
1465 sctp_log_cwnd(stcb, net, net->mtu,
1466 SCTP_CWND_LOG_NO_CUMACK);
1467 }
1468 }
1469 skip_cwnd_update:
1470 /*
1471 * NOW, according to Karn's rule do we need to restore the
1472 * RTO timer back? Check our net_ack2. If not set then we
1473 * have a ambiguity.. i.e. all data ack'd was sent to more
1474 * than one place.
1475 */
1476 if (net->net_ack2) {
1477 /* restore any doubled timers */
1478 net->RTO = ((net->lastsa >> 2) + net->lastsv) >> 1;
1479 if (net->RTO < stcb->asoc.minrto) {
1480 net->RTO = stcb->asoc.minrto;
1481 }
1482 if (net->RTO > stcb->asoc.maxrto) {
1483 net->RTO = stcb->asoc.maxrto;
1484 }
1485 }
1486 }
1487 }
1488
1489 void
1490 sctp_htcp_cwnd_update_after_fr(struct sctp_tcb *stcb,
1491 struct sctp_association *asoc)
1492 {
1493 struct sctp_nets *net;
1494
1495 /*
1496 * CMT fast recovery code. Need to debug. ((sctp_cmt_on_off == 1) &&
1497 * (net->fast_retran_loss_recovery == 0)))
1498 */
1499 TAILQ_FOREACH(net, &asoc->nets, sctp_next) {
1500 if ((asoc->fast_retran_loss_recovery == 0) || (sctp_cmt_on_off == 1)) {
1501 /* out of a RFC2582 Fast recovery window? */
1502 if (net->net_ack > 0) {
1503 /*
1504 * per section 7.2.3, are there any
1505 * destinations that had a fast retransmit
1506 * to them. If so what we need to do is
1507 * adjust ssthresh and cwnd.
1508 */
1509 struct sctp_tmit_chunk *lchk;
1510 int old_cwnd = net->cwnd;
1511
1512 /* JRS - reset as if state were changed */
1513 htcp_reset(&net->htcp_ca);
1514 net->ssthresh = htcp_recalc_ssthresh(stcb, net);
1515 net->cwnd = net->ssthresh;
1516 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1517 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd),
1518 SCTP_CWND_LOG_FROM_FR);
1519 }
1520 lchk = TAILQ_FIRST(&asoc->send_queue);
1521
1522 net->partial_bytes_acked = 0;
1523 /* Turn on fast recovery window */
1524 asoc->fast_retran_loss_recovery = 1;
1525 if (lchk == NULL) {
1526 /* Mark end of the window */
1527 asoc->fast_recovery_tsn = asoc->sending_seq - 1;
1528 } else {
1529 asoc->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
1530 }
1531
1532 /*
1533 * CMT fast recovery -- per destination
1534 * recovery variable.
1535 */
1536 net->fast_retran_loss_recovery = 1;
1537
1538 if (lchk == NULL) {
1539 /* Mark end of the window */
1540 net->fast_recovery_tsn = asoc->sending_seq - 1;
1541 } else {
1542 net->fast_recovery_tsn = lchk->rec.data.TSN_seq - 1;
1543 }
1544
1545 /*
1546 * Disable Nonce Sum Checking and store the
1547 * resync tsn
1548 */
1549 asoc->nonce_sum_check = 0;
1550 asoc->nonce_resync_tsn = asoc->fast_recovery_tsn + 1;
1551
1552 sctp_timer_stop(SCTP_TIMER_TYPE_SEND,
1553 stcb->sctp_ep, stcb, net, SCTP_FROM_SCTP_INDATA + SCTP_LOC_32);
1554 sctp_timer_start(SCTP_TIMER_TYPE_SEND,
1555 stcb->sctp_ep, stcb, net);
1556 }
1557 } else if (net->net_ack > 0) {
1558 /*
1559 * Mark a peg that we WOULD have done a cwnd
1560 * reduction but RFC2582 prevented this action.
1561 */
1562 SCTP_STAT_INCR(sctps_fastretransinrtt);
1563 }
1564 }
1565 }
1566
1567 void
1568 sctp_htcp_cwnd_update_after_timeout(struct sctp_tcb *stcb,
1569 struct sctp_nets *net)
1570 {
1571 int old_cwnd = net->cwnd;
1572
1573 /* JRS - reset as if the state were being changed to timeout */
1574 htcp_reset(&net->htcp_ca);
1575 net->ssthresh = htcp_recalc_ssthresh(stcb, net);
1576 net->cwnd = net->mtu;
1577 /* floor of 1 mtu */
1578 if (net->cwnd < net->mtu)
1579 net->cwnd = net->mtu;
1580 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1581 sctp_log_cwnd(stcb, net, net->cwnd - old_cwnd, SCTP_CWND_LOG_FROM_RTX);
1582 }
1583 net->partial_bytes_acked = 0;
1584 }
1585
1586 void
1587 sctp_htcp_cwnd_update_after_fr_timer(struct sctp_inpcb *inp,
1588 struct sctp_tcb *stcb, struct sctp_nets *net)
1589 {
1590 int old_cwnd;
1591
1592 old_cwnd = net->cwnd;
1593
1594 sctp_chunk_output(inp, stcb, SCTP_OUTPUT_FROM_EARLY_FR_TMR, SCTP_SO_NOT_LOCKED);
1595 net->htcp_ca.last_cong = sctp_get_tick_count();
1596 /*
1597 * make a small adjustment to cwnd and force to CA.
1598 */
1599 if (net->cwnd > net->mtu)
1600 /* drop down one MTU after sending */
1601 net->cwnd -= net->mtu;
1602 if (net->cwnd < net->ssthresh)
1603 /* still in SS move to CA */
1604 net->ssthresh = net->cwnd - 1;
1605 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1606 sctp_log_cwnd(stcb, net, (old_cwnd - net->cwnd), SCTP_CWND_LOG_FROM_FR);
1607 }
1608 }
1609
1610 void
1611 sctp_htcp_cwnd_update_after_ecn_echo(struct sctp_tcb *stcb,
1612 struct sctp_nets *net)
1613 {
1614 int old_cwnd;
1615
1616 old_cwnd = net->cwnd;
1617
1618 /* JRS - reset hctp as if state changed */
1619 htcp_reset(&net->htcp_ca);
1620 SCTP_STAT_INCR(sctps_ecnereducedcwnd);
1621 net->ssthresh = htcp_recalc_ssthresh(stcb, net);
1622 if (net->ssthresh < net->mtu) {
1623 net->ssthresh = net->mtu;
1624 /* here back off the timer as well, to slow us down */
1625 net->RTO <<= 1;
1626 }
1627 net->cwnd = net->ssthresh;
1628 if (sctp_logging_level & SCTP_CWND_MONITOR_ENABLE) {
1629 sctp_log_cwnd(stcb, net, (net->cwnd - old_cwnd), SCTP_CWND_LOG_FROM_SAT);
1630 }
1631 }
Cache object: 98e136e998455870ba22eeedad2c0772
|